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