integrity: path_check update
[linux-2.6.git] / fs / exec.c
1 /*
2  *  linux/fs/exec.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  * #!-checking implemented by tytso.
9  */
10 /*
11  * Demand-loading implemented 01.12.91 - no need to read anything but
12  * the header into memory. The inode of the executable is put into
13  * "current->executable", and page faults do the actual loading. Clean.
14  *
15  * Once more I can proudly say that linux stood up to being changed: it
16  * was less than 2 hours work to get demand-loading completely implemented.
17  *
18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
19  * current->executable is only used by the procfs.  This allows a dispatch
20  * table to check for several different types  of binary formats.  We keep
21  * trying until we recognize the file or we run out of supported binary
22  * formats. 
23  */
24
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/mm.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/ima.h>
49 #include <linux/syscalls.h>
50 #include <linux/tsacct_kern.h>
51 #include <linux/cn_proc.h>
52 #include <linux/audit.h>
53 #include <linux/tracehook.h>
54 #include <linux/kmod.h>
55 #include <linux/fsnotify.h>
56 #include <linux/fs_struct.h>
57
58 #include <asm/uaccess.h>
59 #include <asm/mmu_context.h>
60 #include <asm/tlb.h>
61 #include "internal.h"
62
63 int core_uses_pid;
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 int suid_dumpable = 0;
66
67 /* The maximal length of core_pattern is also specified in sysctl.c */
68
69 static LIST_HEAD(formats);
70 static DEFINE_RWLOCK(binfmt_lock);
71
72 int __register_binfmt(struct linux_binfmt * fmt, int insert)
73 {
74         if (!fmt)
75                 return -EINVAL;
76         write_lock(&binfmt_lock);
77         insert ? list_add(&fmt->lh, &formats) :
78                  list_add_tail(&fmt->lh, &formats);
79         write_unlock(&binfmt_lock);
80         return 0;       
81 }
82
83 EXPORT_SYMBOL(__register_binfmt);
84
85 void unregister_binfmt(struct linux_binfmt * fmt)
86 {
87         write_lock(&binfmt_lock);
88         list_del(&fmt->lh);
89         write_unlock(&binfmt_lock);
90 }
91
92 EXPORT_SYMBOL(unregister_binfmt);
93
94 static inline void put_binfmt(struct linux_binfmt * fmt)
95 {
96         module_put(fmt->module);
97 }
98
99 /*
100  * Note that a shared library must be both readable and executable due to
101  * security reasons.
102  *
103  * Also note that we take the address to load from from the file itself.
104  */
105 SYSCALL_DEFINE1(uselib, const char __user *, library)
106 {
107         struct file *file;
108         struct nameidata nd;
109         char *tmp = getname(library);
110         int error = PTR_ERR(tmp);
111
112         if (!IS_ERR(tmp)) {
113                 error = path_lookup_open(AT_FDCWD, tmp,
114                                          LOOKUP_FOLLOW, &nd,
115                                          FMODE_READ|FMODE_EXEC);
116                 putname(tmp);
117         }
118         if (error)
119                 goto out;
120
121         error = -EINVAL;
122         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
123                 goto exit;
124
125         error = -EACCES;
126         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
127                 goto exit;
128
129         error = inode_permission(nd.path.dentry->d_inode,
130                                  MAY_READ | MAY_EXEC | MAY_OPEN);
131         if (error)
132                 goto exit;
133         error = ima_path_check(&nd.path, MAY_READ | MAY_EXEC | MAY_OPEN,
134                                IMA_COUNT_UPDATE);
135         if (error)
136                 goto exit;
137
138         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
139         error = PTR_ERR(file);
140         if (IS_ERR(file))
141                 goto out;
142
143         fsnotify_open(file->f_path.dentry);
144
145         error = -ENOEXEC;
146         if(file->f_op) {
147                 struct linux_binfmt * fmt;
148
149                 read_lock(&binfmt_lock);
150                 list_for_each_entry(fmt, &formats, lh) {
151                         if (!fmt->load_shlib)
152                                 continue;
153                         if (!try_module_get(fmt->module))
154                                 continue;
155                         read_unlock(&binfmt_lock);
156                         error = fmt->load_shlib(file);
157                         read_lock(&binfmt_lock);
158                         put_binfmt(fmt);
159                         if (error != -ENOEXEC)
160                                 break;
161                 }
162                 read_unlock(&binfmt_lock);
163         }
164         fput(file);
165 out:
166         return error;
167 exit:
168         release_open_intent(&nd);
169         path_put(&nd.path);
170         goto out;
171 }
172
173 #ifdef CONFIG_MMU
174
175 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
176                 int write)
177 {
178         struct page *page;
179         int ret;
180
181 #ifdef CONFIG_STACK_GROWSUP
182         if (write) {
183                 ret = expand_stack_downwards(bprm->vma, pos);
184                 if (ret < 0)
185                         return NULL;
186         }
187 #endif
188         ret = get_user_pages(current, bprm->mm, pos,
189                         1, write, 1, &page, NULL);
190         if (ret <= 0)
191                 return NULL;
192
193         if (write) {
194                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
195                 struct rlimit *rlim;
196
197                 /*
198                  * We've historically supported up to 32 pages (ARG_MAX)
199                  * of argument strings even with small stacks
200                  */
201                 if (size <= ARG_MAX)
202                         return page;
203
204                 /*
205                  * Limit to 1/4-th the stack size for the argv+env strings.
206                  * This ensures that:
207                  *  - the remaining binfmt code will not run out of stack space,
208                  *  - the program will have a reasonable amount of stack left
209                  *    to work from.
210                  */
211                 rlim = current->signal->rlim;
212                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
213                         put_page(page);
214                         return NULL;
215                 }
216         }
217
218         return page;
219 }
220
221 static void put_arg_page(struct page *page)
222 {
223         put_page(page);
224 }
225
226 static void free_arg_page(struct linux_binprm *bprm, int i)
227 {
228 }
229
230 static void free_arg_pages(struct linux_binprm *bprm)
231 {
232 }
233
234 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
235                 struct page *page)
236 {
237         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
238 }
239
240 static int __bprm_mm_init(struct linux_binprm *bprm)
241 {
242         int err;
243         struct vm_area_struct *vma = NULL;
244         struct mm_struct *mm = bprm->mm;
245
246         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
247         if (!vma)
248                 return -ENOMEM;
249
250         down_write(&mm->mmap_sem);
251         vma->vm_mm = mm;
252
253         /*
254          * Place the stack at the largest stack address the architecture
255          * supports. Later, we'll move this to an appropriate place. We don't
256          * use STACK_TOP because that can depend on attributes which aren't
257          * configured yet.
258          */
259         vma->vm_end = STACK_TOP_MAX;
260         vma->vm_start = vma->vm_end - PAGE_SIZE;
261         vma->vm_flags = VM_STACK_FLAGS;
262         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
263         err = insert_vm_struct(mm, vma);
264         if (err)
265                 goto err;
266
267         mm->stack_vm = mm->total_vm = 1;
268         up_write(&mm->mmap_sem);
269         bprm->p = vma->vm_end - sizeof(void *);
270         return 0;
271 err:
272         up_write(&mm->mmap_sem);
273         bprm->vma = NULL;
274         kmem_cache_free(vm_area_cachep, vma);
275         return err;
276 }
277
278 static bool valid_arg_len(struct linux_binprm *bprm, long len)
279 {
280         return len <= MAX_ARG_STRLEN;
281 }
282
283 #else
284
285 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
286                 int write)
287 {
288         struct page *page;
289
290         page = bprm->page[pos / PAGE_SIZE];
291         if (!page && write) {
292                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
293                 if (!page)
294                         return NULL;
295                 bprm->page[pos / PAGE_SIZE] = page;
296         }
297
298         return page;
299 }
300
301 static void put_arg_page(struct page *page)
302 {
303 }
304
305 static void free_arg_page(struct linux_binprm *bprm, int i)
306 {
307         if (bprm->page[i]) {
308                 __free_page(bprm->page[i]);
309                 bprm->page[i] = NULL;
310         }
311 }
312
313 static void free_arg_pages(struct linux_binprm *bprm)
314 {
315         int i;
316
317         for (i = 0; i < MAX_ARG_PAGES; i++)
318                 free_arg_page(bprm, i);
319 }
320
321 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
322                 struct page *page)
323 {
324 }
325
326 static int __bprm_mm_init(struct linux_binprm *bprm)
327 {
328         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
329         return 0;
330 }
331
332 static bool valid_arg_len(struct linux_binprm *bprm, long len)
333 {
334         return len <= bprm->p;
335 }
336
337 #endif /* CONFIG_MMU */
338
339 /*
340  * Create a new mm_struct and populate it with a temporary stack
341  * vm_area_struct.  We don't have enough context at this point to set the stack
342  * flags, permissions, and offset, so we use temporary values.  We'll update
343  * them later in setup_arg_pages().
344  */
345 int bprm_mm_init(struct linux_binprm *bprm)
346 {
347         int err;
348         struct mm_struct *mm = NULL;
349
350         bprm->mm = mm = mm_alloc();
351         err = -ENOMEM;
352         if (!mm)
353                 goto err;
354
355         err = init_new_context(current, mm);
356         if (err)
357                 goto err;
358
359         err = __bprm_mm_init(bprm);
360         if (err)
361                 goto err;
362
363         return 0;
364
365 err:
366         if (mm) {
367                 bprm->mm = NULL;
368                 mmdrop(mm);
369         }
370
371         return err;
372 }
373
374 /*
375  * count() counts the number of strings in array ARGV.
376  */
377 static int count(char __user * __user * argv, int max)
378 {
379         int i = 0;
380
381         if (argv != NULL) {
382                 for (;;) {
383                         char __user * p;
384
385                         if (get_user(p, argv))
386                                 return -EFAULT;
387                         if (!p)
388                                 break;
389                         argv++;
390                         if (i++ >= max)
391                                 return -E2BIG;
392                         cond_resched();
393                 }
394         }
395         return i;
396 }
397
398 /*
399  * 'copy_strings()' copies argument/environment strings from the old
400  * processes's memory to the new process's stack.  The call to get_user_pages()
401  * ensures the destination page is created and not swapped out.
402  */
403 static int copy_strings(int argc, char __user * __user * argv,
404                         struct linux_binprm *bprm)
405 {
406         struct page *kmapped_page = NULL;
407         char *kaddr = NULL;
408         unsigned long kpos = 0;
409         int ret;
410
411         while (argc-- > 0) {
412                 char __user *str;
413                 int len;
414                 unsigned long pos;
415
416                 if (get_user(str, argv+argc) ||
417                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
418                         ret = -EFAULT;
419                         goto out;
420                 }
421
422                 if (!valid_arg_len(bprm, len)) {
423                         ret = -E2BIG;
424                         goto out;
425                 }
426
427                 /* We're going to work our way backwords. */
428                 pos = bprm->p;
429                 str += len;
430                 bprm->p -= len;
431
432                 while (len > 0) {
433                         int offset, bytes_to_copy;
434
435                         offset = pos % PAGE_SIZE;
436                         if (offset == 0)
437                                 offset = PAGE_SIZE;
438
439                         bytes_to_copy = offset;
440                         if (bytes_to_copy > len)
441                                 bytes_to_copy = len;
442
443                         offset -= bytes_to_copy;
444                         pos -= bytes_to_copy;
445                         str -= bytes_to_copy;
446                         len -= bytes_to_copy;
447
448                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
449                                 struct page *page;
450
451                                 page = get_arg_page(bprm, pos, 1);
452                                 if (!page) {
453                                         ret = -E2BIG;
454                                         goto out;
455                                 }
456
457                                 if (kmapped_page) {
458                                         flush_kernel_dcache_page(kmapped_page);
459                                         kunmap(kmapped_page);
460                                         put_arg_page(kmapped_page);
461                                 }
462                                 kmapped_page = page;
463                                 kaddr = kmap(kmapped_page);
464                                 kpos = pos & PAGE_MASK;
465                                 flush_arg_page(bprm, kpos, kmapped_page);
466                         }
467                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
468                                 ret = -EFAULT;
469                                 goto out;
470                         }
471                 }
472         }
473         ret = 0;
474 out:
475         if (kmapped_page) {
476                 flush_kernel_dcache_page(kmapped_page);
477                 kunmap(kmapped_page);
478                 put_arg_page(kmapped_page);
479         }
480         return ret;
481 }
482
483 /*
484  * Like copy_strings, but get argv and its values from kernel memory.
485  */
486 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
487 {
488         int r;
489         mm_segment_t oldfs = get_fs();
490         set_fs(KERNEL_DS);
491         r = copy_strings(argc, (char __user * __user *)argv, bprm);
492         set_fs(oldfs);
493         return r;
494 }
495 EXPORT_SYMBOL(copy_strings_kernel);
496
497 #ifdef CONFIG_MMU
498
499 /*
500  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
501  * the binfmt code determines where the new stack should reside, we shift it to
502  * its final location.  The process proceeds as follows:
503  *
504  * 1) Use shift to calculate the new vma endpoints.
505  * 2) Extend vma to cover both the old and new ranges.  This ensures the
506  *    arguments passed to subsequent functions are consistent.
507  * 3) Move vma's page tables to the new range.
508  * 4) Free up any cleared pgd range.
509  * 5) Shrink the vma to cover only the new range.
510  */
511 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
512 {
513         struct mm_struct *mm = vma->vm_mm;
514         unsigned long old_start = vma->vm_start;
515         unsigned long old_end = vma->vm_end;
516         unsigned long length = old_end - old_start;
517         unsigned long new_start = old_start - shift;
518         unsigned long new_end = old_end - shift;
519         struct mmu_gather *tlb;
520
521         BUG_ON(new_start > new_end);
522
523         /*
524          * ensure there are no vmas between where we want to go
525          * and where we are
526          */
527         if (vma != find_vma(mm, new_start))
528                 return -EFAULT;
529
530         /*
531          * cover the whole range: [new_start, old_end)
532          */
533         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
534
535         /*
536          * move the page tables downwards, on failure we rely on
537          * process cleanup to remove whatever mess we made.
538          */
539         if (length != move_page_tables(vma, old_start,
540                                        vma, new_start, length))
541                 return -ENOMEM;
542
543         lru_add_drain();
544         tlb = tlb_gather_mmu(mm, 0);
545         if (new_end > old_start) {
546                 /*
547                  * when the old and new regions overlap clear from new_end.
548                  */
549                 free_pgd_range(tlb, new_end, old_end, new_end,
550                         vma->vm_next ? vma->vm_next->vm_start : 0);
551         } else {
552                 /*
553                  * otherwise, clean from old_start; this is done to not touch
554                  * the address space in [new_end, old_start) some architectures
555                  * have constraints on va-space that make this illegal (IA64) -
556                  * for the others its just a little faster.
557                  */
558                 free_pgd_range(tlb, old_start, old_end, new_end,
559                         vma->vm_next ? vma->vm_next->vm_start : 0);
560         }
561         tlb_finish_mmu(tlb, new_end, old_end);
562
563         /*
564          * shrink the vma to just the new range.
565          */
566         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
567
568         return 0;
569 }
570
571 #define EXTRA_STACK_VM_PAGES    20      /* random */
572
573 /*
574  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
575  * the stack is optionally relocated, and some extra space is added.
576  */
577 int setup_arg_pages(struct linux_binprm *bprm,
578                     unsigned long stack_top,
579                     int executable_stack)
580 {
581         unsigned long ret;
582         unsigned long stack_shift;
583         struct mm_struct *mm = current->mm;
584         struct vm_area_struct *vma = bprm->vma;
585         struct vm_area_struct *prev = NULL;
586         unsigned long vm_flags;
587         unsigned long stack_base;
588
589 #ifdef CONFIG_STACK_GROWSUP
590         /* Limit stack size to 1GB */
591         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
592         if (stack_base > (1 << 30))
593                 stack_base = 1 << 30;
594
595         /* Make sure we didn't let the argument array grow too large. */
596         if (vma->vm_end - vma->vm_start > stack_base)
597                 return -ENOMEM;
598
599         stack_base = PAGE_ALIGN(stack_top - stack_base);
600
601         stack_shift = vma->vm_start - stack_base;
602         mm->arg_start = bprm->p - stack_shift;
603         bprm->p = vma->vm_end - stack_shift;
604 #else
605         stack_top = arch_align_stack(stack_top);
606         stack_top = PAGE_ALIGN(stack_top);
607         stack_shift = vma->vm_end - stack_top;
608
609         bprm->p -= stack_shift;
610         mm->arg_start = bprm->p;
611 #endif
612
613         if (bprm->loader)
614                 bprm->loader -= stack_shift;
615         bprm->exec -= stack_shift;
616
617         down_write(&mm->mmap_sem);
618         vm_flags = VM_STACK_FLAGS;
619
620         /*
621          * Adjust stack execute permissions; explicitly enable for
622          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
623          * (arch default) otherwise.
624          */
625         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
626                 vm_flags |= VM_EXEC;
627         else if (executable_stack == EXSTACK_DISABLE_X)
628                 vm_flags &= ~VM_EXEC;
629         vm_flags |= mm->def_flags;
630
631         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
632                         vm_flags);
633         if (ret)
634                 goto out_unlock;
635         BUG_ON(prev != vma);
636
637         /* Move stack pages down in memory. */
638         if (stack_shift) {
639                 ret = shift_arg_pages(vma, stack_shift);
640                 if (ret) {
641                         up_write(&mm->mmap_sem);
642                         return ret;
643                 }
644         }
645
646 #ifdef CONFIG_STACK_GROWSUP
647         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
648 #else
649         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
650 #endif
651         ret = expand_stack(vma, stack_base);
652         if (ret)
653                 ret = -EFAULT;
654
655 out_unlock:
656         up_write(&mm->mmap_sem);
657         return 0;
658 }
659 EXPORT_SYMBOL(setup_arg_pages);
660
661 #endif /* CONFIG_MMU */
662
663 struct file *open_exec(const char *name)
664 {
665         struct nameidata nd;
666         struct file *file;
667         int err;
668
669         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
670                                 FMODE_READ|FMODE_EXEC);
671         if (err)
672                 goto out;
673
674         err = -EACCES;
675         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
676                 goto out_path_put;
677
678         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
679                 goto out_path_put;
680
681         err = inode_permission(nd.path.dentry->d_inode, MAY_EXEC | MAY_OPEN);
682         if (err)
683                 goto out_path_put;
684         err = ima_path_check(&nd.path, MAY_EXEC | MAY_OPEN, IMA_COUNT_UPDATE);
685         if (err)
686                 goto out_path_put;
687
688         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
689         if (IS_ERR(file))
690                 return file;
691
692         fsnotify_open(file->f_path.dentry);
693
694         err = deny_write_access(file);
695         if (err) {
696                 fput(file);
697                 goto out;
698         }
699
700         return file;
701
702  out_path_put:
703         release_open_intent(&nd);
704         path_put(&nd.path);
705  out:
706         return ERR_PTR(err);
707 }
708 EXPORT_SYMBOL(open_exec);
709
710 int kernel_read(struct file *file, unsigned long offset,
711         char *addr, unsigned long count)
712 {
713         mm_segment_t old_fs;
714         loff_t pos = offset;
715         int result;
716
717         old_fs = get_fs();
718         set_fs(get_ds());
719         /* The cast to a user pointer is valid due to the set_fs() */
720         result = vfs_read(file, (void __user *)addr, count, &pos);
721         set_fs(old_fs);
722         return result;
723 }
724
725 EXPORT_SYMBOL(kernel_read);
726
727 static int exec_mmap(struct mm_struct *mm)
728 {
729         struct task_struct *tsk;
730         struct mm_struct * old_mm, *active_mm;
731
732         /* Notify parent that we're no longer interested in the old VM */
733         tsk = current;
734         old_mm = current->mm;
735         mm_release(tsk, old_mm);
736
737         if (old_mm) {
738                 /*
739                  * Make sure that if there is a core dump in progress
740                  * for the old mm, we get out and die instead of going
741                  * through with the exec.  We must hold mmap_sem around
742                  * checking core_state and changing tsk->mm.
743                  */
744                 down_read(&old_mm->mmap_sem);
745                 if (unlikely(old_mm->core_state)) {
746                         up_read(&old_mm->mmap_sem);
747                         return -EINTR;
748                 }
749         }
750         task_lock(tsk);
751         active_mm = tsk->active_mm;
752         tsk->mm = mm;
753         tsk->active_mm = mm;
754         activate_mm(active_mm, mm);
755         task_unlock(tsk);
756         arch_pick_mmap_layout(mm);
757         if (old_mm) {
758                 up_read(&old_mm->mmap_sem);
759                 BUG_ON(active_mm != old_mm);
760                 mm_update_next_owner(old_mm);
761                 mmput(old_mm);
762                 return 0;
763         }
764         mmdrop(active_mm);
765         return 0;
766 }
767
768 /*
769  * This function makes sure the current process has its own signal table,
770  * so that flush_signal_handlers can later reset the handlers without
771  * disturbing other processes.  (Other processes might share the signal
772  * table via the CLONE_SIGHAND option to clone().)
773  */
774 static int de_thread(struct task_struct *tsk)
775 {
776         struct signal_struct *sig = tsk->signal;
777         struct sighand_struct *oldsighand = tsk->sighand;
778         spinlock_t *lock = &oldsighand->siglock;
779         int count;
780
781         if (thread_group_empty(tsk))
782                 goto no_thread_group;
783
784         /*
785          * Kill all other threads in the thread group.
786          */
787         spin_lock_irq(lock);
788         if (signal_group_exit(sig)) {
789                 /*
790                  * Another group action in progress, just
791                  * return so that the signal is processed.
792                  */
793                 spin_unlock_irq(lock);
794                 return -EAGAIN;
795         }
796         sig->group_exit_task = tsk;
797         zap_other_threads(tsk);
798
799         /* Account for the thread group leader hanging around: */
800         count = thread_group_leader(tsk) ? 1 : 2;
801         sig->notify_count = count;
802         while (atomic_read(&sig->count) > count) {
803                 __set_current_state(TASK_UNINTERRUPTIBLE);
804                 spin_unlock_irq(lock);
805                 schedule();
806                 spin_lock_irq(lock);
807         }
808         spin_unlock_irq(lock);
809
810         /*
811          * At this point all other threads have exited, all we have to
812          * do is to wait for the thread group leader to become inactive,
813          * and to assume its PID:
814          */
815         if (!thread_group_leader(tsk)) {
816                 struct task_struct *leader = tsk->group_leader;
817
818                 sig->notify_count = -1; /* for exit_notify() */
819                 for (;;) {
820                         write_lock_irq(&tasklist_lock);
821                         if (likely(leader->exit_state))
822                                 break;
823                         __set_current_state(TASK_UNINTERRUPTIBLE);
824                         write_unlock_irq(&tasklist_lock);
825                         schedule();
826                 }
827
828                 /*
829                  * The only record we have of the real-time age of a
830                  * process, regardless of execs it's done, is start_time.
831                  * All the past CPU time is accumulated in signal_struct
832                  * from sister threads now dead.  But in this non-leader
833                  * exec, nothing survives from the original leader thread,
834                  * whose birth marks the true age of this process now.
835                  * When we take on its identity by switching to its PID, we
836                  * also take its birthdate (always earlier than our own).
837                  */
838                 tsk->start_time = leader->start_time;
839
840                 BUG_ON(!same_thread_group(leader, tsk));
841                 BUG_ON(has_group_leader_pid(tsk));
842                 /*
843                  * An exec() starts a new thread group with the
844                  * TGID of the previous thread group. Rehash the
845                  * two threads with a switched PID, and release
846                  * the former thread group leader:
847                  */
848
849                 /* Become a process group leader with the old leader's pid.
850                  * The old leader becomes a thread of the this thread group.
851                  * Note: The old leader also uses this pid until release_task
852                  *       is called.  Odd but simple and correct.
853                  */
854                 detach_pid(tsk, PIDTYPE_PID);
855                 tsk->pid = leader->pid;
856                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
857                 transfer_pid(leader, tsk, PIDTYPE_PGID);
858                 transfer_pid(leader, tsk, PIDTYPE_SID);
859                 list_replace_rcu(&leader->tasks, &tsk->tasks);
860
861                 tsk->group_leader = tsk;
862                 leader->group_leader = tsk;
863
864                 tsk->exit_signal = SIGCHLD;
865
866                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
867                 leader->exit_state = EXIT_DEAD;
868                 write_unlock_irq(&tasklist_lock);
869
870                 release_task(leader);
871         }
872
873         sig->group_exit_task = NULL;
874         sig->notify_count = 0;
875
876 no_thread_group:
877         exit_itimers(sig);
878         flush_itimer_signals();
879
880         if (atomic_read(&oldsighand->count) != 1) {
881                 struct sighand_struct *newsighand;
882                 /*
883                  * This ->sighand is shared with the CLONE_SIGHAND
884                  * but not CLONE_THREAD task, switch to the new one.
885                  */
886                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
887                 if (!newsighand)
888                         return -ENOMEM;
889
890                 atomic_set(&newsighand->count, 1);
891                 memcpy(newsighand->action, oldsighand->action,
892                        sizeof(newsighand->action));
893
894                 write_lock_irq(&tasklist_lock);
895                 spin_lock(&oldsighand->siglock);
896                 rcu_assign_pointer(tsk->sighand, newsighand);
897                 spin_unlock(&oldsighand->siglock);
898                 write_unlock_irq(&tasklist_lock);
899
900                 __cleanup_sighand(oldsighand);
901         }
902
903         BUG_ON(!thread_group_leader(tsk));
904         return 0;
905 }
906
907 /*
908  * These functions flushes out all traces of the currently running executable
909  * so that a new one can be started
910  */
911 static void flush_old_files(struct files_struct * files)
912 {
913         long j = -1;
914         struct fdtable *fdt;
915
916         spin_lock(&files->file_lock);
917         for (;;) {
918                 unsigned long set, i;
919
920                 j++;
921                 i = j * __NFDBITS;
922                 fdt = files_fdtable(files);
923                 if (i >= fdt->max_fds)
924                         break;
925                 set = fdt->close_on_exec->fds_bits[j];
926                 if (!set)
927                         continue;
928                 fdt->close_on_exec->fds_bits[j] = 0;
929                 spin_unlock(&files->file_lock);
930                 for ( ; set ; i++,set >>= 1) {
931                         if (set & 1) {
932                                 sys_close(i);
933                         }
934                 }
935                 spin_lock(&files->file_lock);
936
937         }
938         spin_unlock(&files->file_lock);
939 }
940
941 char *get_task_comm(char *buf, struct task_struct *tsk)
942 {
943         /* buf must be at least sizeof(tsk->comm) in size */
944         task_lock(tsk);
945         strncpy(buf, tsk->comm, sizeof(tsk->comm));
946         task_unlock(tsk);
947         return buf;
948 }
949
950 void set_task_comm(struct task_struct *tsk, char *buf)
951 {
952         task_lock(tsk);
953         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
954         task_unlock(tsk);
955 }
956
957 int flush_old_exec(struct linux_binprm * bprm)
958 {
959         char * name;
960         int i, ch, retval;
961         char tcomm[sizeof(current->comm)];
962
963         /*
964          * Make sure we have a private signal table and that
965          * we are unassociated from the previous thread group.
966          */
967         retval = de_thread(current);
968         if (retval)
969                 goto out;
970
971         set_mm_exe_file(bprm->mm, bprm->file);
972
973         /*
974          * Release all of the old mmap stuff
975          */
976         retval = exec_mmap(bprm->mm);
977         if (retval)
978                 goto out;
979
980         bprm->mm = NULL;                /* We're using it now */
981
982         /* This is the point of no return */
983         current->sas_ss_sp = current->sas_ss_size = 0;
984
985         if (current_euid() == current_uid() && current_egid() == current_gid())
986                 set_dumpable(current->mm, 1);
987         else
988                 set_dumpable(current->mm, suid_dumpable);
989
990         name = bprm->filename;
991
992         /* Copies the binary name from after last slash */
993         for (i=0; (ch = *(name++)) != '\0';) {
994                 if (ch == '/')
995                         i = 0; /* overwrite what we wrote */
996                 else
997                         if (i < (sizeof(tcomm) - 1))
998                                 tcomm[i++] = ch;
999         }
1000         tcomm[i] = '\0';
1001         set_task_comm(current, tcomm);
1002
1003         current->flags &= ~PF_RANDOMIZE;
1004         flush_thread();
1005
1006         /* Set the new mm task size. We have to do that late because it may
1007          * depend on TIF_32BIT which is only updated in flush_thread() on
1008          * some architectures like powerpc
1009          */
1010         current->mm->task_size = TASK_SIZE;
1011
1012         /* install the new credentials */
1013         if (bprm->cred->uid != current_euid() ||
1014             bprm->cred->gid != current_egid()) {
1015                 current->pdeath_signal = 0;
1016         } else if (file_permission(bprm->file, MAY_READ) ||
1017                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1018                 set_dumpable(current->mm, suid_dumpable);
1019         }
1020
1021         current->personality &= ~bprm->per_clear;
1022
1023         /* An exec changes our domain. We are no longer part of the thread
1024            group */
1025
1026         current->self_exec_id++;
1027                         
1028         flush_signal_handlers(current, 0);
1029         flush_old_files(current->files);
1030
1031         return 0;
1032
1033 out:
1034         return retval;
1035 }
1036
1037 EXPORT_SYMBOL(flush_old_exec);
1038
1039 /*
1040  * install the new credentials for this executable
1041  */
1042 void install_exec_creds(struct linux_binprm *bprm)
1043 {
1044         security_bprm_committing_creds(bprm);
1045
1046         commit_creds(bprm->cred);
1047         bprm->cred = NULL;
1048
1049         /* cred_guard_mutex must be held at least to this point to prevent
1050          * ptrace_attach() from altering our determination of the task's
1051          * credentials; any time after this it may be unlocked */
1052
1053         security_bprm_committed_creds(bprm);
1054 }
1055 EXPORT_SYMBOL(install_exec_creds);
1056
1057 /*
1058  * determine how safe it is to execute the proposed program
1059  * - the caller must hold current->cred_guard_mutex to protect against
1060  *   PTRACE_ATTACH
1061  */
1062 int check_unsafe_exec(struct linux_binprm *bprm)
1063 {
1064         struct task_struct *p = current, *t;
1065         unsigned n_fs;
1066         int res = 0;
1067
1068         bprm->unsafe = tracehook_unsafe_exec(p);
1069
1070         n_fs = 1;
1071         write_lock(&p->fs->lock);
1072         rcu_read_lock();
1073         for (t = next_thread(p); t != p; t = next_thread(t)) {
1074                 if (t->fs == p->fs)
1075                         n_fs++;
1076         }
1077         rcu_read_unlock();
1078
1079         if (p->fs->users > n_fs) {
1080                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1081         } else {
1082                 res = -EAGAIN;
1083                 if (!p->fs->in_exec) {
1084                         p->fs->in_exec = 1;
1085                         res = 1;
1086                 }
1087         }
1088         write_unlock(&p->fs->lock);
1089
1090         return res;
1091 }
1092
1093 /* 
1094  * Fill the binprm structure from the inode. 
1095  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1096  *
1097  * This may be called multiple times for binary chains (scripts for example).
1098  */
1099 int prepare_binprm(struct linux_binprm *bprm)
1100 {
1101         umode_t mode;
1102         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1103         int retval;
1104
1105         mode = inode->i_mode;
1106         if (bprm->file->f_op == NULL)
1107                 return -EACCES;
1108
1109         /* clear any previous set[ug]id data from a previous binary */
1110         bprm->cred->euid = current_euid();
1111         bprm->cred->egid = current_egid();
1112
1113         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1114                 /* Set-uid? */
1115                 if (mode & S_ISUID) {
1116                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1117                         bprm->cred->euid = inode->i_uid;
1118                 }
1119
1120                 /* Set-gid? */
1121                 /*
1122                  * If setgid is set but no group execute bit then this
1123                  * is a candidate for mandatory locking, not a setgid
1124                  * executable.
1125                  */
1126                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1127                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1128                         bprm->cred->egid = inode->i_gid;
1129                 }
1130         }
1131
1132         /* fill in binprm security blob */
1133         retval = security_bprm_set_creds(bprm);
1134         if (retval)
1135                 return retval;
1136         bprm->cred_prepared = 1;
1137
1138         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1139         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1140 }
1141
1142 EXPORT_SYMBOL(prepare_binprm);
1143
1144 /*
1145  * Arguments are '\0' separated strings found at the location bprm->p
1146  * points to; chop off the first by relocating brpm->p to right after
1147  * the first '\0' encountered.
1148  */
1149 int remove_arg_zero(struct linux_binprm *bprm)
1150 {
1151         int ret = 0;
1152         unsigned long offset;
1153         char *kaddr;
1154         struct page *page;
1155
1156         if (!bprm->argc)
1157                 return 0;
1158
1159         do {
1160                 offset = bprm->p & ~PAGE_MASK;
1161                 page = get_arg_page(bprm, bprm->p, 0);
1162                 if (!page) {
1163                         ret = -EFAULT;
1164                         goto out;
1165                 }
1166                 kaddr = kmap_atomic(page, KM_USER0);
1167
1168                 for (; offset < PAGE_SIZE && kaddr[offset];
1169                                 offset++, bprm->p++)
1170                         ;
1171
1172                 kunmap_atomic(kaddr, KM_USER0);
1173                 put_arg_page(page);
1174
1175                 if (offset == PAGE_SIZE)
1176                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1177         } while (offset == PAGE_SIZE);
1178
1179         bprm->p++;
1180         bprm->argc--;
1181         ret = 0;
1182
1183 out:
1184         return ret;
1185 }
1186 EXPORT_SYMBOL(remove_arg_zero);
1187
1188 /*
1189  * cycle the list of binary formats handler, until one recognizes the image
1190  */
1191 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1192 {
1193         unsigned int depth = bprm->recursion_depth;
1194         int try,retval;
1195         struct linux_binfmt *fmt;
1196
1197         retval = security_bprm_check(bprm);
1198         if (retval)
1199                 return retval;
1200         retval = ima_bprm_check(bprm);
1201         if (retval)
1202                 return retval;
1203
1204         /* kernel module loader fixup */
1205         /* so we don't try to load run modprobe in kernel space. */
1206         set_fs(USER_DS);
1207
1208         retval = audit_bprm(bprm);
1209         if (retval)
1210                 return retval;
1211
1212         retval = -ENOENT;
1213         for (try=0; try<2; try++) {
1214                 read_lock(&binfmt_lock);
1215                 list_for_each_entry(fmt, &formats, lh) {
1216                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1217                         if (!fn)
1218                                 continue;
1219                         if (!try_module_get(fmt->module))
1220                                 continue;
1221                         read_unlock(&binfmt_lock);
1222                         retval = fn(bprm, regs);
1223                         /*
1224                          * Restore the depth counter to its starting value
1225                          * in this call, so we don't have to rely on every
1226                          * load_binary function to restore it on return.
1227                          */
1228                         bprm->recursion_depth = depth;
1229                         if (retval >= 0) {
1230                                 if (depth == 0)
1231                                         tracehook_report_exec(fmt, bprm, regs);
1232                                 put_binfmt(fmt);
1233                                 allow_write_access(bprm->file);
1234                                 if (bprm->file)
1235                                         fput(bprm->file);
1236                                 bprm->file = NULL;
1237                                 current->did_exec = 1;
1238                                 proc_exec_connector(current);
1239                                 return retval;
1240                         }
1241                         read_lock(&binfmt_lock);
1242                         put_binfmt(fmt);
1243                         if (retval != -ENOEXEC || bprm->mm == NULL)
1244                                 break;
1245                         if (!bprm->file) {
1246                                 read_unlock(&binfmt_lock);
1247                                 return retval;
1248                         }
1249                 }
1250                 read_unlock(&binfmt_lock);
1251                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1252                         break;
1253 #ifdef CONFIG_MODULES
1254                 } else {
1255 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1256                         if (printable(bprm->buf[0]) &&
1257                             printable(bprm->buf[1]) &&
1258                             printable(bprm->buf[2]) &&
1259                             printable(bprm->buf[3]))
1260                                 break; /* -ENOEXEC */
1261                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1262 #endif
1263                 }
1264         }
1265         return retval;
1266 }
1267
1268 EXPORT_SYMBOL(search_binary_handler);
1269
1270 void free_bprm(struct linux_binprm *bprm)
1271 {
1272         free_arg_pages(bprm);
1273         if (bprm->cred)
1274                 abort_creds(bprm->cred);
1275         kfree(bprm);
1276 }
1277
1278 /*
1279  * sys_execve() executes a new program.
1280  */
1281 int do_execve(char * filename,
1282         char __user *__user *argv,
1283         char __user *__user *envp,
1284         struct pt_regs * regs)
1285 {
1286         struct linux_binprm *bprm;
1287         struct file *file;
1288         struct files_struct *displaced;
1289         bool clear_in_exec;
1290         int retval;
1291
1292         retval = unshare_files(&displaced);
1293         if (retval)
1294                 goto out_ret;
1295
1296         retval = -ENOMEM;
1297         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1298         if (!bprm)
1299                 goto out_files;
1300
1301         retval = mutex_lock_interruptible(&current->cred_guard_mutex);
1302         if (retval < 0)
1303                 goto out_free;
1304         current->in_execve = 1;
1305
1306         retval = -ENOMEM;
1307         bprm->cred = prepare_exec_creds();
1308         if (!bprm->cred)
1309                 goto out_unlock;
1310
1311         retval = check_unsafe_exec(bprm);
1312         if (retval < 0)
1313                 goto out_unlock;
1314         clear_in_exec = retval;
1315
1316         file = open_exec(filename);
1317         retval = PTR_ERR(file);
1318         if (IS_ERR(file))
1319                 goto out_unmark;
1320
1321         sched_exec();
1322
1323         bprm->file = file;
1324         bprm->filename = filename;
1325         bprm->interp = filename;
1326
1327         retval = bprm_mm_init(bprm);
1328         if (retval)
1329                 goto out_file;
1330
1331         bprm->argc = count(argv, MAX_ARG_STRINGS);
1332         if ((retval = bprm->argc) < 0)
1333                 goto out;
1334
1335         bprm->envc = count(envp, MAX_ARG_STRINGS);
1336         if ((retval = bprm->envc) < 0)
1337                 goto out;
1338
1339         retval = prepare_binprm(bprm);
1340         if (retval < 0)
1341                 goto out;
1342
1343         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1344         if (retval < 0)
1345                 goto out;
1346
1347         bprm->exec = bprm->p;
1348         retval = copy_strings(bprm->envc, envp, bprm);
1349         if (retval < 0)
1350                 goto out;
1351
1352         retval = copy_strings(bprm->argc, argv, bprm);
1353         if (retval < 0)
1354                 goto out;
1355
1356         current->flags &= ~PF_KTHREAD;
1357         retval = search_binary_handler(bprm,regs);
1358         if (retval < 0)
1359                 goto out;
1360
1361         /* execve succeeded */
1362         current->fs->in_exec = 0;
1363         current->in_execve = 0;
1364         mutex_unlock(&current->cred_guard_mutex);
1365         acct_update_integrals(current);
1366         free_bprm(bprm);
1367         if (displaced)
1368                 put_files_struct(displaced);
1369         return retval;
1370
1371 out:
1372         if (bprm->mm)
1373                 mmput (bprm->mm);
1374
1375 out_file:
1376         if (bprm->file) {
1377                 allow_write_access(bprm->file);
1378                 fput(bprm->file);
1379         }
1380
1381 out_unmark:
1382         if (clear_in_exec)
1383                 current->fs->in_exec = 0;
1384
1385 out_unlock:
1386         current->in_execve = 0;
1387         mutex_unlock(&current->cred_guard_mutex);
1388
1389 out_free:
1390         free_bprm(bprm);
1391
1392 out_files:
1393         if (displaced)
1394                 reset_files_struct(displaced);
1395 out_ret:
1396         return retval;
1397 }
1398
1399 int set_binfmt(struct linux_binfmt *new)
1400 {
1401         struct linux_binfmt *old = current->binfmt;
1402
1403         if (new) {
1404                 if (!try_module_get(new->module))
1405                         return -1;
1406         }
1407         current->binfmt = new;
1408         if (old)
1409                 module_put(old->module);
1410         return 0;
1411 }
1412
1413 EXPORT_SYMBOL(set_binfmt);
1414
1415 /* format_corename will inspect the pattern parameter, and output a
1416  * name into corename, which must have space for at least
1417  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1418  */
1419 static int format_corename(char *corename, long signr)
1420 {
1421         const struct cred *cred = current_cred();
1422         const char *pat_ptr = core_pattern;
1423         int ispipe = (*pat_ptr == '|');
1424         char *out_ptr = corename;
1425         char *const out_end = corename + CORENAME_MAX_SIZE;
1426         int rc;
1427         int pid_in_pattern = 0;
1428
1429         /* Repeat as long as we have more pattern to process and more output
1430            space */
1431         while (*pat_ptr) {
1432                 if (*pat_ptr != '%') {
1433                         if (out_ptr == out_end)
1434                                 goto out;
1435                         *out_ptr++ = *pat_ptr++;
1436                 } else {
1437                         switch (*++pat_ptr) {
1438                         case 0:
1439                                 goto out;
1440                         /* Double percent, output one percent */
1441                         case '%':
1442                                 if (out_ptr == out_end)
1443                                         goto out;
1444                                 *out_ptr++ = '%';
1445                                 break;
1446                         /* pid */
1447                         case 'p':
1448                                 pid_in_pattern = 1;
1449                                 rc = snprintf(out_ptr, out_end - out_ptr,
1450                                               "%d", task_tgid_vnr(current));
1451                                 if (rc > out_end - out_ptr)
1452                                         goto out;
1453                                 out_ptr += rc;
1454                                 break;
1455                         /* uid */
1456                         case 'u':
1457                                 rc = snprintf(out_ptr, out_end - out_ptr,
1458                                               "%d", cred->uid);
1459                                 if (rc > out_end - out_ptr)
1460                                         goto out;
1461                                 out_ptr += rc;
1462                                 break;
1463                         /* gid */
1464                         case 'g':
1465                                 rc = snprintf(out_ptr, out_end - out_ptr,
1466                                               "%d", cred->gid);
1467                                 if (rc > out_end - out_ptr)
1468                                         goto out;
1469                                 out_ptr += rc;
1470                                 break;
1471                         /* signal that caused the coredump */
1472                         case 's':
1473                                 rc = snprintf(out_ptr, out_end - out_ptr,
1474                                               "%ld", signr);
1475                                 if (rc > out_end - out_ptr)
1476                                         goto out;
1477                                 out_ptr += rc;
1478                                 break;
1479                         /* UNIX time of coredump */
1480                         case 't': {
1481                                 struct timeval tv;
1482                                 do_gettimeofday(&tv);
1483                                 rc = snprintf(out_ptr, out_end - out_ptr,
1484                                               "%lu", tv.tv_sec);
1485                                 if (rc > out_end - out_ptr)
1486                                         goto out;
1487                                 out_ptr += rc;
1488                                 break;
1489                         }
1490                         /* hostname */
1491                         case 'h':
1492                                 down_read(&uts_sem);
1493                                 rc = snprintf(out_ptr, out_end - out_ptr,
1494                                               "%s", utsname()->nodename);
1495                                 up_read(&uts_sem);
1496                                 if (rc > out_end - out_ptr)
1497                                         goto out;
1498                                 out_ptr += rc;
1499                                 break;
1500                         /* executable */
1501                         case 'e':
1502                                 rc = snprintf(out_ptr, out_end - out_ptr,
1503                                               "%s", current->comm);
1504                                 if (rc > out_end - out_ptr)
1505                                         goto out;
1506                                 out_ptr += rc;
1507                                 break;
1508                         /* core limit size */
1509                         case 'c':
1510                                 rc = snprintf(out_ptr, out_end - out_ptr,
1511                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1512                                 if (rc > out_end - out_ptr)
1513                                         goto out;
1514                                 out_ptr += rc;
1515                                 break;
1516                         default:
1517                                 break;
1518                         }
1519                         ++pat_ptr;
1520                 }
1521         }
1522         /* Backward compatibility with core_uses_pid:
1523          *
1524          * If core_pattern does not include a %p (as is the default)
1525          * and core_uses_pid is set, then .%pid will be appended to
1526          * the filename. Do not do this for piped commands. */
1527         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1528                 rc = snprintf(out_ptr, out_end - out_ptr,
1529                               ".%d", task_tgid_vnr(current));
1530                 if (rc > out_end - out_ptr)
1531                         goto out;
1532                 out_ptr += rc;
1533         }
1534 out:
1535         *out_ptr = 0;
1536         return ispipe;
1537 }
1538
1539 static int zap_process(struct task_struct *start)
1540 {
1541         struct task_struct *t;
1542         int nr = 0;
1543
1544         start->signal->flags = SIGNAL_GROUP_EXIT;
1545         start->signal->group_stop_count = 0;
1546
1547         t = start;
1548         do {
1549                 if (t != current && t->mm) {
1550                         sigaddset(&t->pending.signal, SIGKILL);
1551                         signal_wake_up(t, 1);
1552                         nr++;
1553                 }
1554         } while_each_thread(start, t);
1555
1556         return nr;
1557 }
1558
1559 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1560                                 struct core_state *core_state, int exit_code)
1561 {
1562         struct task_struct *g, *p;
1563         unsigned long flags;
1564         int nr = -EAGAIN;
1565
1566         spin_lock_irq(&tsk->sighand->siglock);
1567         if (!signal_group_exit(tsk->signal)) {
1568                 mm->core_state = core_state;
1569                 tsk->signal->group_exit_code = exit_code;
1570                 nr = zap_process(tsk);
1571         }
1572         spin_unlock_irq(&tsk->sighand->siglock);
1573         if (unlikely(nr < 0))
1574                 return nr;
1575
1576         if (atomic_read(&mm->mm_users) == nr + 1)
1577                 goto done;
1578         /*
1579          * We should find and kill all tasks which use this mm, and we should
1580          * count them correctly into ->nr_threads. We don't take tasklist
1581          * lock, but this is safe wrt:
1582          *
1583          * fork:
1584          *      None of sub-threads can fork after zap_process(leader). All
1585          *      processes which were created before this point should be
1586          *      visible to zap_threads() because copy_process() adds the new
1587          *      process to the tail of init_task.tasks list, and lock/unlock
1588          *      of ->siglock provides a memory barrier.
1589          *
1590          * do_exit:
1591          *      The caller holds mm->mmap_sem. This means that the task which
1592          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1593          *      its ->mm.
1594          *
1595          * de_thread:
1596          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1597          *      we must see either old or new leader, this does not matter.
1598          *      However, it can change p->sighand, so lock_task_sighand(p)
1599          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1600          *      it can't fail.
1601          *
1602          *      Note also that "g" can be the old leader with ->mm == NULL
1603          *      and already unhashed and thus removed from ->thread_group.
1604          *      This is OK, __unhash_process()->list_del_rcu() does not
1605          *      clear the ->next pointer, we will find the new leader via
1606          *      next_thread().
1607          */
1608         rcu_read_lock();
1609         for_each_process(g) {
1610                 if (g == tsk->group_leader)
1611                         continue;
1612                 if (g->flags & PF_KTHREAD)
1613                         continue;
1614                 p = g;
1615                 do {
1616                         if (p->mm) {
1617                                 if (unlikely(p->mm == mm)) {
1618                                         lock_task_sighand(p, &flags);
1619                                         nr += zap_process(p);
1620                                         unlock_task_sighand(p, &flags);
1621                                 }
1622                                 break;
1623                         }
1624                 } while_each_thread(g, p);
1625         }
1626         rcu_read_unlock();
1627 done:
1628         atomic_set(&core_state->nr_threads, nr);
1629         return nr;
1630 }
1631
1632 static int coredump_wait(int exit_code, struct core_state *core_state)
1633 {
1634         struct task_struct *tsk = current;
1635         struct mm_struct *mm = tsk->mm;
1636         struct completion *vfork_done;
1637         int core_waiters;
1638
1639         init_completion(&core_state->startup);
1640         core_state->dumper.task = tsk;
1641         core_state->dumper.next = NULL;
1642         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1643         up_write(&mm->mmap_sem);
1644
1645         if (unlikely(core_waiters < 0))
1646                 goto fail;
1647
1648         /*
1649          * Make sure nobody is waiting for us to release the VM,
1650          * otherwise we can deadlock when we wait on each other
1651          */
1652         vfork_done = tsk->vfork_done;
1653         if (vfork_done) {
1654                 tsk->vfork_done = NULL;
1655                 complete(vfork_done);
1656         }
1657
1658         if (core_waiters)
1659                 wait_for_completion(&core_state->startup);
1660 fail:
1661         return core_waiters;
1662 }
1663
1664 static void coredump_finish(struct mm_struct *mm)
1665 {
1666         struct core_thread *curr, *next;
1667         struct task_struct *task;
1668
1669         next = mm->core_state->dumper.next;
1670         while ((curr = next) != NULL) {
1671                 next = curr->next;
1672                 task = curr->task;
1673                 /*
1674                  * see exit_mm(), curr->task must not see
1675                  * ->task == NULL before we read ->next.
1676                  */
1677                 smp_mb();
1678                 curr->task = NULL;
1679                 wake_up_process(task);
1680         }
1681
1682         mm->core_state = NULL;
1683 }
1684
1685 /*
1686  * set_dumpable converts traditional three-value dumpable to two flags and
1687  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1688  * these bits are not changed atomically.  So get_dumpable can observe the
1689  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1690  * return either old dumpable or new one by paying attention to the order of
1691  * modifying the bits.
1692  *
1693  * dumpable |   mm->flags (binary)
1694  * old  new | initial interim  final
1695  * ---------+-----------------------
1696  *  0    1  |   00      01      01
1697  *  0    2  |   00      10(*)   11
1698  *  1    0  |   01      00      00
1699  *  1    2  |   01      11      11
1700  *  2    0  |   11      10(*)   00
1701  *  2    1  |   11      11      01
1702  *
1703  * (*) get_dumpable regards interim value of 10 as 11.
1704  */
1705 void set_dumpable(struct mm_struct *mm, int value)
1706 {
1707         switch (value) {
1708         case 0:
1709                 clear_bit(MMF_DUMPABLE, &mm->flags);
1710                 smp_wmb();
1711                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1712                 break;
1713         case 1:
1714                 set_bit(MMF_DUMPABLE, &mm->flags);
1715                 smp_wmb();
1716                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1717                 break;
1718         case 2:
1719                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1720                 smp_wmb();
1721                 set_bit(MMF_DUMPABLE, &mm->flags);
1722                 break;
1723         }
1724 }
1725
1726 int get_dumpable(struct mm_struct *mm)
1727 {
1728         int ret;
1729
1730         ret = mm->flags & 0x3;
1731         return (ret >= 2) ? 2 : ret;
1732 }
1733
1734 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1735 {
1736         struct core_state core_state;
1737         char corename[CORENAME_MAX_SIZE + 1];
1738         struct mm_struct *mm = current->mm;
1739         struct linux_binfmt * binfmt;
1740         struct inode * inode;
1741         struct file * file;
1742         const struct cred *old_cred;
1743         struct cred *cred;
1744         int retval = 0;
1745         int flag = 0;
1746         int ispipe = 0;
1747         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1748         char **helper_argv = NULL;
1749         int helper_argc = 0;
1750         char *delimit;
1751
1752         audit_core_dumps(signr);
1753
1754         binfmt = current->binfmt;
1755         if (!binfmt || !binfmt->core_dump)
1756                 goto fail;
1757
1758         cred = prepare_creds();
1759         if (!cred) {
1760                 retval = -ENOMEM;
1761                 goto fail;
1762         }
1763
1764         down_write(&mm->mmap_sem);
1765         /*
1766          * If another thread got here first, or we are not dumpable, bail out.
1767          */
1768         if (mm->core_state || !get_dumpable(mm)) {
1769                 up_write(&mm->mmap_sem);
1770                 put_cred(cred);
1771                 goto fail;
1772         }
1773
1774         /*
1775          *      We cannot trust fsuid as being the "true" uid of the
1776          *      process nor do we know its entire history. We only know it
1777          *      was tainted so we dump it as root in mode 2.
1778          */
1779         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1780                 flag = O_EXCL;          /* Stop rewrite attacks */
1781                 cred->fsuid = 0;        /* Dump root private */
1782         }
1783
1784         retval = coredump_wait(exit_code, &core_state);
1785         if (retval < 0) {
1786                 put_cred(cred);
1787                 goto fail;
1788         }
1789
1790         old_cred = override_creds(cred);
1791
1792         /*
1793          * Clear any false indication of pending signals that might
1794          * be seen by the filesystem code called to write the core file.
1795          */
1796         clear_thread_flag(TIF_SIGPENDING);
1797
1798         /*
1799          * lock_kernel() because format_corename() is controlled by sysctl, which
1800          * uses lock_kernel()
1801          */
1802         lock_kernel();
1803         ispipe = format_corename(corename, signr);
1804         unlock_kernel();
1805         /*
1806          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1807          * to a pipe.  Since we're not writing directly to the filesystem
1808          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1809          * created unless the pipe reader choses to write out the core file
1810          * at which point file size limits and permissions will be imposed
1811          * as it does with any other process
1812          */
1813         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1814                 goto fail_unlock;
1815
1816         if (ispipe) {
1817                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1818                 if (!helper_argv) {
1819                         printk(KERN_WARNING "%s failed to allocate memory\n",
1820                                __func__);
1821                         goto fail_unlock;
1822                 }
1823                 /* Terminate the string before the first option */
1824                 delimit = strchr(corename, ' ');
1825                 if (delimit)
1826                         *delimit = '\0';
1827                 delimit = strrchr(helper_argv[0], '/');
1828                 if (delimit)
1829                         delimit++;
1830                 else
1831                         delimit = helper_argv[0];
1832                 if (!strcmp(delimit, current->comm)) {
1833                         printk(KERN_NOTICE "Recursive core dump detected, "
1834                                         "aborting\n");
1835                         goto fail_unlock;
1836                 }
1837
1838                 core_limit = RLIM_INFINITY;
1839
1840                 /* SIGPIPE can happen, but it's just never processed */
1841                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1842                                 &file)) {
1843                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1844                                corename);
1845                         goto fail_unlock;
1846                 }
1847         } else
1848                 file = filp_open(corename,
1849                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1850                                  0600);
1851         if (IS_ERR(file))
1852                 goto fail_unlock;
1853         inode = file->f_path.dentry->d_inode;
1854         if (inode->i_nlink > 1)
1855                 goto close_fail;        /* multiple links - don't dump */
1856         if (!ispipe && d_unhashed(file->f_path.dentry))
1857                 goto close_fail;
1858
1859         /* AK: actually i see no reason to not allow this for named pipes etc.,
1860            but keep the previous behaviour for now. */
1861         if (!ispipe && !S_ISREG(inode->i_mode))
1862                 goto close_fail;
1863         /*
1864          * Dont allow local users get cute and trick others to coredump
1865          * into their pre-created files:
1866          */
1867         if (inode->i_uid != current_fsuid())
1868                 goto close_fail;
1869         if (!file->f_op)
1870                 goto close_fail;
1871         if (!file->f_op->write)
1872                 goto close_fail;
1873         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1874                 goto close_fail;
1875
1876         retval = binfmt->core_dump(signr, regs, file, core_limit);
1877
1878         if (retval)
1879                 current->signal->group_exit_code |= 0x80;
1880 close_fail:
1881         filp_close(file, NULL);
1882 fail_unlock:
1883         if (helper_argv)
1884                 argv_free(helper_argv);
1885
1886         revert_creds(old_cred);
1887         put_cred(cred);
1888         coredump_finish(mm);
1889 fail:
1890         return;
1891 }