Linux-2.6.12-rc2
[linux-3.10.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/config.h>
26 #include <linux/slab.h>
27 #include <linux/file.h>
28 #include <linux/mman.h>
29 #include <linux/a.out.h>
30 #include <linux/stat.h>
31 #include <linux/fcntl.h>
32 #include <linux/smp_lock.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/highmem.h>
36 #include <linux/spinlock.h>
37 #include <linux/key.h>
38 #include <linux/personality.h>
39 #include <linux/binfmts.h>
40 #include <linux/swap.h>
41 #include <linux/utsname.h>
42 #include <linux/module.h>
43 #include <linux/namei.h>
44 #include <linux/proc_fs.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/rmap.h>
50 #include <linux/acct.h>
51
52 #include <asm/uaccess.h>
53 #include <asm/mmu_context.h>
54
55 #ifdef CONFIG_KMOD
56 #include <linux/kmod.h>
57 #endif
58
59 int core_uses_pid;
60 char core_pattern[65] = "core";
61 /* The maximal length of core_pattern is also specified in sysctl.c */
62
63 static struct linux_binfmt *formats;
64 static DEFINE_RWLOCK(binfmt_lock);
65
66 int register_binfmt(struct linux_binfmt * fmt)
67 {
68         struct linux_binfmt ** tmp = &formats;
69
70         if (!fmt)
71                 return -EINVAL;
72         if (fmt->next)
73                 return -EBUSY;
74         write_lock(&binfmt_lock);
75         while (*tmp) {
76                 if (fmt == *tmp) {
77                         write_unlock(&binfmt_lock);
78                         return -EBUSY;
79                 }
80                 tmp = &(*tmp)->next;
81         }
82         fmt->next = formats;
83         formats = fmt;
84         write_unlock(&binfmt_lock);
85         return 0;       
86 }
87
88 EXPORT_SYMBOL(register_binfmt);
89
90 int unregister_binfmt(struct linux_binfmt * fmt)
91 {
92         struct linux_binfmt ** tmp = &formats;
93
94         write_lock(&binfmt_lock);
95         while (*tmp) {
96                 if (fmt == *tmp) {
97                         *tmp = fmt->next;
98                         write_unlock(&binfmt_lock);
99                         return 0;
100                 }
101                 tmp = &(*tmp)->next;
102         }
103         write_unlock(&binfmt_lock);
104         return -EINVAL;
105 }
106
107 EXPORT_SYMBOL(unregister_binfmt);
108
109 static inline void put_binfmt(struct linux_binfmt * fmt)
110 {
111         module_put(fmt->module);
112 }
113
114 /*
115  * Note that a shared library must be both readable and executable due to
116  * security reasons.
117  *
118  * Also note that we take the address to load from from the file itself.
119  */
120 asmlinkage long sys_uselib(const char __user * library)
121 {
122         struct file * file;
123         struct nameidata nd;
124         int error;
125
126         nd.intent.open.flags = FMODE_READ;
127         error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
128         if (error)
129                 goto out;
130
131         error = -EINVAL;
132         if (!S_ISREG(nd.dentry->d_inode->i_mode))
133                 goto exit;
134
135         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
136         if (error)
137                 goto exit;
138
139         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
140         error = PTR_ERR(file);
141         if (IS_ERR(file))
142                 goto out;
143
144         error = -ENOEXEC;
145         if(file->f_op) {
146                 struct linux_binfmt * fmt;
147
148                 read_lock(&binfmt_lock);
149                 for (fmt = formats ; fmt ; fmt = fmt->next) {
150                         if (!fmt->load_shlib)
151                                 continue;
152                         if (!try_module_get(fmt->module))
153                                 continue;
154                         read_unlock(&binfmt_lock);
155                         error = fmt->load_shlib(file);
156                         read_lock(&binfmt_lock);
157                         put_binfmt(fmt);
158                         if (error != -ENOEXEC)
159                                 break;
160                 }
161                 read_unlock(&binfmt_lock);
162         }
163         fput(file);
164 out:
165         return error;
166 exit:
167         path_release(&nd);
168         goto out;
169 }
170
171 /*
172  * count() counts the number of strings in array ARGV.
173  */
174 static int count(char __user * __user * argv, int max)
175 {
176         int i = 0;
177
178         if (argv != NULL) {
179                 for (;;) {
180                         char __user * p;
181
182                         if (get_user(p, argv))
183                                 return -EFAULT;
184                         if (!p)
185                                 break;
186                         argv++;
187                         if(++i > max)
188                                 return -E2BIG;
189                         cond_resched();
190                 }
191         }
192         return i;
193 }
194
195 /*
196  * 'copy_strings()' copies argument/environment strings from user
197  * memory to free pages in kernel mem. These are in a format ready
198  * to be put directly into the top of new user memory.
199  */
200 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
201 {
202         struct page *kmapped_page = NULL;
203         char *kaddr = NULL;
204         int ret;
205
206         while (argc-- > 0) {
207                 char __user *str;
208                 int len;
209                 unsigned long pos;
210
211                 if (get_user(str, argv+argc) ||
212                                 !(len = strnlen_user(str, bprm->p))) {
213                         ret = -EFAULT;
214                         goto out;
215                 }
216
217                 if (bprm->p < len)  {
218                         ret = -E2BIG;
219                         goto out;
220                 }
221
222                 bprm->p -= len;
223                 /* XXX: add architecture specific overflow check here. */
224                 pos = bprm->p;
225
226                 while (len > 0) {
227                         int i, new, err;
228                         int offset, bytes_to_copy;
229                         struct page *page;
230
231                         offset = pos % PAGE_SIZE;
232                         i = pos/PAGE_SIZE;
233                         page = bprm->page[i];
234                         new = 0;
235                         if (!page) {
236                                 page = alloc_page(GFP_HIGHUSER);
237                                 bprm->page[i] = page;
238                                 if (!page) {
239                                         ret = -ENOMEM;
240                                         goto out;
241                                 }
242                                 new = 1;
243                         }
244
245                         if (page != kmapped_page) {
246                                 if (kmapped_page)
247                                         kunmap(kmapped_page);
248                                 kmapped_page = page;
249                                 kaddr = kmap(kmapped_page);
250                         }
251                         if (new && offset)
252                                 memset(kaddr, 0, offset);
253                         bytes_to_copy = PAGE_SIZE - offset;
254                         if (bytes_to_copy > len) {
255                                 bytes_to_copy = len;
256                                 if (new)
257                                         memset(kaddr+offset+len, 0,
258                                                 PAGE_SIZE-offset-len);
259                         }
260                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
261                         if (err) {
262                                 ret = -EFAULT;
263                                 goto out;
264                         }
265
266                         pos += bytes_to_copy;
267                         str += bytes_to_copy;
268                         len -= bytes_to_copy;
269                 }
270         }
271         ret = 0;
272 out:
273         if (kmapped_page)
274                 kunmap(kmapped_page);
275         return ret;
276 }
277
278 /*
279  * Like copy_strings, but get argv and its values from kernel memory.
280  */
281 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
282 {
283         int r;
284         mm_segment_t oldfs = get_fs();
285         set_fs(KERNEL_DS);
286         r = copy_strings(argc, (char __user * __user *)argv, bprm);
287         set_fs(oldfs);
288         return r;
289 }
290
291 EXPORT_SYMBOL(copy_strings_kernel);
292
293 #ifdef CONFIG_MMU
294 /*
295  * This routine is used to map in a page into an address space: needed by
296  * execve() for the initial stack and environment pages.
297  *
298  * vma->vm_mm->mmap_sem is held for writing.
299  */
300 void install_arg_page(struct vm_area_struct *vma,
301                         struct page *page, unsigned long address)
302 {
303         struct mm_struct *mm = vma->vm_mm;
304         pgd_t * pgd;
305         pud_t * pud;
306         pmd_t * pmd;
307         pte_t * pte;
308
309         if (unlikely(anon_vma_prepare(vma)))
310                 goto out_sig;
311
312         flush_dcache_page(page);
313         pgd = pgd_offset(mm, address);
314
315         spin_lock(&mm->page_table_lock);
316         pud = pud_alloc(mm, pgd, address);
317         if (!pud)
318                 goto out;
319         pmd = pmd_alloc(mm, pud, address);
320         if (!pmd)
321                 goto out;
322         pte = pte_alloc_map(mm, pmd, address);
323         if (!pte)
324                 goto out;
325         if (!pte_none(*pte)) {
326                 pte_unmap(pte);
327                 goto out;
328         }
329         inc_mm_counter(mm, rss);
330         lru_cache_add_active(page);
331         set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
332                                         page, vma->vm_page_prot))));
333         page_add_anon_rmap(page, vma, address);
334         pte_unmap(pte);
335         spin_unlock(&mm->page_table_lock);
336
337         /* no need for flush_tlb */
338         return;
339 out:
340         spin_unlock(&mm->page_table_lock);
341 out_sig:
342         __free_page(page);
343         force_sig(SIGKILL, current);
344 }
345
346 #define EXTRA_STACK_VM_PAGES    20      /* random */
347
348 int setup_arg_pages(struct linux_binprm *bprm,
349                     unsigned long stack_top,
350                     int executable_stack)
351 {
352         unsigned long stack_base;
353         struct vm_area_struct *mpnt;
354         struct mm_struct *mm = current->mm;
355         int i, ret;
356         long arg_size;
357
358 #ifdef CONFIG_STACK_GROWSUP
359         /* Move the argument and environment strings to the bottom of the
360          * stack space.
361          */
362         int offset, j;
363         char *to, *from;
364
365         /* Start by shifting all the pages down */
366         i = 0;
367         for (j = 0; j < MAX_ARG_PAGES; j++) {
368                 struct page *page = bprm->page[j];
369                 if (!page)
370                         continue;
371                 bprm->page[i++] = page;
372         }
373
374         /* Now move them within their pages */
375         offset = bprm->p % PAGE_SIZE;
376         to = kmap(bprm->page[0]);
377         for (j = 1; j < i; j++) {
378                 memmove(to, to + offset, PAGE_SIZE - offset);
379                 from = kmap(bprm->page[j]);
380                 memcpy(to + PAGE_SIZE - offset, from, offset);
381                 kunmap(bprm->page[j - 1]);
382                 to = from;
383         }
384         memmove(to, to + offset, PAGE_SIZE - offset);
385         kunmap(bprm->page[j - 1]);
386
387         /* Limit stack size to 1GB */
388         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
389         if (stack_base > (1 << 30))
390                 stack_base = 1 << 30;
391         stack_base = PAGE_ALIGN(stack_top - stack_base);
392
393         /* Adjust bprm->p to point to the end of the strings. */
394         bprm->p = stack_base + PAGE_SIZE * i - offset;
395
396         mm->arg_start = stack_base;
397         arg_size = i << PAGE_SHIFT;
398
399         /* zero pages that were copied above */
400         while (i < MAX_ARG_PAGES)
401                 bprm->page[i++] = NULL;
402 #else
403         stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
404         stack_base = PAGE_ALIGN(stack_base);
405         bprm->p += stack_base;
406         mm->arg_start = bprm->p;
407         arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
408 #endif
409
410         arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
411
412         if (bprm->loader)
413                 bprm->loader += stack_base;
414         bprm->exec += stack_base;
415
416         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
417         if (!mpnt)
418                 return -ENOMEM;
419
420         if (security_vm_enough_memory(arg_size >> PAGE_SHIFT)) {
421                 kmem_cache_free(vm_area_cachep, mpnt);
422                 return -ENOMEM;
423         }
424
425         memset(mpnt, 0, sizeof(*mpnt));
426
427         down_write(&mm->mmap_sem);
428         {
429                 mpnt->vm_mm = mm;
430 #ifdef CONFIG_STACK_GROWSUP
431                 mpnt->vm_start = stack_base;
432                 mpnt->vm_end = stack_base + arg_size;
433 #else
434                 mpnt->vm_end = stack_top;
435                 mpnt->vm_start = mpnt->vm_end - arg_size;
436 #endif
437                 /* Adjust stack execute permissions; explicitly enable
438                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
439                  * and leave alone (arch default) otherwise. */
440                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
441                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
442                 else if (executable_stack == EXSTACK_DISABLE_X)
443                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
444                 else
445                         mpnt->vm_flags = VM_STACK_FLAGS;
446                 mpnt->vm_flags |= mm->def_flags;
447                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
448                 if ((ret = insert_vm_struct(mm, mpnt))) {
449                         up_write(&mm->mmap_sem);
450                         kmem_cache_free(vm_area_cachep, mpnt);
451                         return ret;
452                 }
453                 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
454         }
455
456         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
457                 struct page *page = bprm->page[i];
458                 if (page) {
459                         bprm->page[i] = NULL;
460                         install_arg_page(mpnt, page, stack_base);
461                 }
462                 stack_base += PAGE_SIZE;
463         }
464         up_write(&mm->mmap_sem);
465         
466         return 0;
467 }
468
469 EXPORT_SYMBOL(setup_arg_pages);
470
471 #define free_arg_pages(bprm) do { } while (0)
472
473 #else
474
475 static inline void free_arg_pages(struct linux_binprm *bprm)
476 {
477         int i;
478
479         for (i = 0; i < MAX_ARG_PAGES; i++) {
480                 if (bprm->page[i])
481                         __free_page(bprm->page[i]);
482                 bprm->page[i] = NULL;
483         }
484 }
485
486 #endif /* CONFIG_MMU */
487
488 struct file *open_exec(const char *name)
489 {
490         struct nameidata nd;
491         int err;
492         struct file *file;
493
494         nd.intent.open.flags = FMODE_READ;
495         err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
496         file = ERR_PTR(err);
497
498         if (!err) {
499                 struct inode *inode = nd.dentry->d_inode;
500                 file = ERR_PTR(-EACCES);
501                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
502                     S_ISREG(inode->i_mode)) {
503                         int err = permission(inode, MAY_EXEC, &nd);
504                         if (!err && !(inode->i_mode & 0111))
505                                 err = -EACCES;
506                         file = ERR_PTR(err);
507                         if (!err) {
508                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
509                                 if (!IS_ERR(file)) {
510                                         err = deny_write_access(file);
511                                         if (err) {
512                                                 fput(file);
513                                                 file = ERR_PTR(err);
514                                         }
515                                 }
516 out:
517                                 return file;
518                         }
519                 }
520                 path_release(&nd);
521         }
522         goto out;
523 }
524
525 EXPORT_SYMBOL(open_exec);
526
527 int kernel_read(struct file *file, unsigned long offset,
528         char *addr, unsigned long count)
529 {
530         mm_segment_t old_fs;
531         loff_t pos = offset;
532         int result;
533
534         old_fs = get_fs();
535         set_fs(get_ds());
536         /* The cast to a user pointer is valid due to the set_fs() */
537         result = vfs_read(file, (void __user *)addr, count, &pos);
538         set_fs(old_fs);
539         return result;
540 }
541
542 EXPORT_SYMBOL(kernel_read);
543
544 static int exec_mmap(struct mm_struct *mm)
545 {
546         struct task_struct *tsk;
547         struct mm_struct * old_mm, *active_mm;
548
549         /* Notify parent that we're no longer interested in the old VM */
550         tsk = current;
551         old_mm = current->mm;
552         mm_release(tsk, old_mm);
553
554         if (old_mm) {
555                 /*
556                  * Make sure that if there is a core dump in progress
557                  * for the old mm, we get out and die instead of going
558                  * through with the exec.  We must hold mmap_sem around
559                  * checking core_waiters and changing tsk->mm.  The
560                  * core-inducing thread will increment core_waiters for
561                  * each thread whose ->mm == old_mm.
562                  */
563                 down_read(&old_mm->mmap_sem);
564                 if (unlikely(old_mm->core_waiters)) {
565                         up_read(&old_mm->mmap_sem);
566                         return -EINTR;
567                 }
568         }
569         task_lock(tsk);
570         active_mm = tsk->active_mm;
571         tsk->mm = mm;
572         tsk->active_mm = mm;
573         activate_mm(active_mm, mm);
574         task_unlock(tsk);
575         arch_pick_mmap_layout(mm);
576         if (old_mm) {
577                 up_read(&old_mm->mmap_sem);
578                 if (active_mm != old_mm) BUG();
579                 mmput(old_mm);
580                 return 0;
581         }
582         mmdrop(active_mm);
583         return 0;
584 }
585
586 /*
587  * This function makes sure the current process has its own signal table,
588  * so that flush_signal_handlers can later reset the handlers without
589  * disturbing other processes.  (Other processes might share the signal
590  * table via the CLONE_SIGHAND option to clone().)
591  */
592 static inline int de_thread(struct task_struct *tsk)
593 {
594         struct signal_struct *sig = tsk->signal;
595         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
596         spinlock_t *lock = &oldsighand->siglock;
597         int count;
598
599         /*
600          * If we don't share sighandlers, then we aren't sharing anything
601          * and we can just re-use it all.
602          */
603         if (atomic_read(&oldsighand->count) <= 1) {
604                 BUG_ON(atomic_read(&sig->count) != 1);
605                 exit_itimers(sig);
606                 return 0;
607         }
608
609         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
610         if (!newsighand)
611                 return -ENOMEM;
612
613         if (thread_group_empty(current))
614                 goto no_thread_group;
615
616         /*
617          * Kill all other threads in the thread group.
618          * We must hold tasklist_lock to call zap_other_threads.
619          */
620         read_lock(&tasklist_lock);
621         spin_lock_irq(lock);
622         if (sig->flags & SIGNAL_GROUP_EXIT) {
623                 /*
624                  * Another group action in progress, just
625                  * return so that the signal is processed.
626                  */
627                 spin_unlock_irq(lock);
628                 read_unlock(&tasklist_lock);
629                 kmem_cache_free(sighand_cachep, newsighand);
630                 return -EAGAIN;
631         }
632         zap_other_threads(current);
633         read_unlock(&tasklist_lock);
634
635         /*
636          * Account for the thread group leader hanging around:
637          */
638         count = 2;
639         if (thread_group_leader(current))
640                 count = 1;
641         while (atomic_read(&sig->count) > count) {
642                 sig->group_exit_task = current;
643                 sig->notify_count = count;
644                 __set_current_state(TASK_UNINTERRUPTIBLE);
645                 spin_unlock_irq(lock);
646                 schedule();
647                 spin_lock_irq(lock);
648         }
649         sig->group_exit_task = NULL;
650         sig->notify_count = 0;
651         spin_unlock_irq(lock);
652
653         /*
654          * At this point all other threads have exited, all we have to
655          * do is to wait for the thread group leader to become inactive,
656          * and to assume its PID:
657          */
658         if (!thread_group_leader(current)) {
659                 struct task_struct *leader = current->group_leader, *parent;
660                 struct dentry *proc_dentry1, *proc_dentry2;
661                 unsigned long exit_state, ptrace;
662
663                 /*
664                  * Wait for the thread group leader to be a zombie.
665                  * It should already be zombie at this point, most
666                  * of the time.
667                  */
668                 while (leader->exit_state != EXIT_ZOMBIE)
669                         yield();
670
671                 spin_lock(&leader->proc_lock);
672                 spin_lock(&current->proc_lock);
673                 proc_dentry1 = proc_pid_unhash(current);
674                 proc_dentry2 = proc_pid_unhash(leader);
675                 write_lock_irq(&tasklist_lock);
676
677                 if (leader->tgid != current->tgid)
678                         BUG();
679                 if (current->pid == current->tgid)
680                         BUG();
681                 /*
682                  * An exec() starts a new thread group with the
683                  * TGID of the previous thread group. Rehash the
684                  * two threads with a switched PID, and release
685                  * the former thread group leader:
686                  */
687                 ptrace = leader->ptrace;
688                 parent = leader->parent;
689                 if (unlikely(ptrace) && unlikely(parent == current)) {
690                         /*
691                          * Joker was ptracing his own group leader,
692                          * and now he wants to be his own parent!
693                          * We can't have that.
694                          */
695                         ptrace = 0;
696                 }
697
698                 ptrace_unlink(current);
699                 ptrace_unlink(leader);
700                 remove_parent(current);
701                 remove_parent(leader);
702
703                 switch_exec_pids(leader, current);
704
705                 current->parent = current->real_parent = leader->real_parent;
706                 leader->parent = leader->real_parent = child_reaper;
707                 current->group_leader = current;
708                 leader->group_leader = leader;
709
710                 add_parent(current, current->parent);
711                 add_parent(leader, leader->parent);
712                 if (ptrace) {
713                         current->ptrace = ptrace;
714                         __ptrace_link(current, parent);
715                 }
716
717                 list_del(&current->tasks);
718                 list_add_tail(&current->tasks, &init_task.tasks);
719                 current->exit_signal = SIGCHLD;
720                 exit_state = leader->exit_state;
721
722                 write_unlock_irq(&tasklist_lock);
723                 spin_unlock(&leader->proc_lock);
724                 spin_unlock(&current->proc_lock);
725                 proc_pid_flush(proc_dentry1);
726                 proc_pid_flush(proc_dentry2);
727
728                 if (exit_state != EXIT_ZOMBIE)
729                         BUG();
730                 release_task(leader);
731         }
732
733         /*
734          * Now there are really no other threads at all,
735          * so it's safe to stop telling them to kill themselves.
736          */
737         sig->flags = 0;
738
739 no_thread_group:
740         BUG_ON(atomic_read(&sig->count) != 1);
741         exit_itimers(sig);
742
743         if (atomic_read(&oldsighand->count) == 1) {
744                 /*
745                  * Now that we nuked the rest of the thread group,
746                  * it turns out we are not sharing sighand any more either.
747                  * So we can just keep it.
748                  */
749                 kmem_cache_free(sighand_cachep, newsighand);
750         } else {
751                 /*
752                  * Move our state over to newsighand and switch it in.
753                  */
754                 spin_lock_init(&newsighand->siglock);
755                 atomic_set(&newsighand->count, 1);
756                 memcpy(newsighand->action, oldsighand->action,
757                        sizeof(newsighand->action));
758
759                 write_lock_irq(&tasklist_lock);
760                 spin_lock(&oldsighand->siglock);
761                 spin_lock(&newsighand->siglock);
762
763                 current->sighand = newsighand;
764                 recalc_sigpending();
765
766                 spin_unlock(&newsighand->siglock);
767                 spin_unlock(&oldsighand->siglock);
768                 write_unlock_irq(&tasklist_lock);
769
770                 if (atomic_dec_and_test(&oldsighand->count))
771                         kmem_cache_free(sighand_cachep, oldsighand);
772         }
773
774         if (!thread_group_empty(current))
775                 BUG();
776         if (!thread_group_leader(current))
777                 BUG();
778         return 0;
779 }
780         
781 /*
782  * These functions flushes out all traces of the currently running executable
783  * so that a new one can be started
784  */
785
786 static inline void flush_old_files(struct files_struct * files)
787 {
788         long j = -1;
789
790         spin_lock(&files->file_lock);
791         for (;;) {
792                 unsigned long set, i;
793
794                 j++;
795                 i = j * __NFDBITS;
796                 if (i >= files->max_fds || i >= files->max_fdset)
797                         break;
798                 set = files->close_on_exec->fds_bits[j];
799                 if (!set)
800                         continue;
801                 files->close_on_exec->fds_bits[j] = 0;
802                 spin_unlock(&files->file_lock);
803                 for ( ; set ; i++,set >>= 1) {
804                         if (set & 1) {
805                                 sys_close(i);
806                         }
807                 }
808                 spin_lock(&files->file_lock);
809
810         }
811         spin_unlock(&files->file_lock);
812 }
813
814 void get_task_comm(char *buf, struct task_struct *tsk)
815 {
816         /* buf must be at least sizeof(tsk->comm) in size */
817         task_lock(tsk);
818         strncpy(buf, tsk->comm, sizeof(tsk->comm));
819         task_unlock(tsk);
820 }
821
822 void set_task_comm(struct task_struct *tsk, char *buf)
823 {
824         task_lock(tsk);
825         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
826         task_unlock(tsk);
827 }
828
829 int flush_old_exec(struct linux_binprm * bprm)
830 {
831         char * name;
832         int i, ch, retval;
833         struct files_struct *files;
834         char tcomm[sizeof(current->comm)];
835
836         /*
837          * Make sure we have a private signal table and that
838          * we are unassociated from the previous thread group.
839          */
840         retval = de_thread(current);
841         if (retval)
842                 goto out;
843
844         /*
845          * Make sure we have private file handles. Ask the
846          * fork helper to do the work for us and the exit
847          * helper to do the cleanup of the old one.
848          */
849         files = current->files;         /* refcounted so safe to hold */
850         retval = unshare_files();
851         if (retval)
852                 goto out;
853         /*
854          * Release all of the old mmap stuff
855          */
856         retval = exec_mmap(bprm->mm);
857         if (retval)
858                 goto mmap_failed;
859
860         bprm->mm = NULL;                /* We're using it now */
861
862         /* This is the point of no return */
863         steal_locks(files);
864         put_files_struct(files);
865
866         current->sas_ss_sp = current->sas_ss_size = 0;
867
868         if (current->euid == current->uid && current->egid == current->gid)
869                 current->mm->dumpable = 1;
870         name = bprm->filename;
871         for (i=0; (ch = *(name++)) != '\0';) {
872                 if (ch == '/')
873                         i = 0;
874                 else
875                         if (i < (sizeof(tcomm) - 1))
876                                 tcomm[i++] = ch;
877         }
878         tcomm[i] = '\0';
879         set_task_comm(current, tcomm);
880
881         current->flags &= ~PF_RANDOMIZE;
882         flush_thread();
883
884         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid || 
885             permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
886             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
887                 suid_keys(current);
888                 current->mm->dumpable = 0;
889         }
890
891         /* An exec changes our domain. We are no longer part of the thread
892            group */
893
894         current->self_exec_id++;
895                         
896         flush_signal_handlers(current, 0);
897         flush_old_files(current->files);
898
899         return 0;
900
901 mmap_failed:
902         put_files_struct(current->files);
903         current->files = files;
904 out:
905         return retval;
906 }
907
908 EXPORT_SYMBOL(flush_old_exec);
909
910 /* 
911  * Fill the binprm structure from the inode. 
912  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
913  */
914 int prepare_binprm(struct linux_binprm *bprm)
915 {
916         int mode;
917         struct inode * inode = bprm->file->f_dentry->d_inode;
918         int retval;
919
920         mode = inode->i_mode;
921         /*
922          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
923          * generic_permission lets a non-executable through
924          */
925         if (!(mode & 0111))     /* with at least _one_ execute bit set */
926                 return -EACCES;
927         if (bprm->file->f_op == NULL)
928                 return -EACCES;
929
930         bprm->e_uid = current->euid;
931         bprm->e_gid = current->egid;
932
933         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
934                 /* Set-uid? */
935                 if (mode & S_ISUID) {
936                         current->personality &= ~PER_CLEAR_ON_SETID;
937                         bprm->e_uid = inode->i_uid;
938                 }
939
940                 /* Set-gid? */
941                 /*
942                  * If setgid is set but no group execute bit then this
943                  * is a candidate for mandatory locking, not a setgid
944                  * executable.
945                  */
946                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
947                         current->personality &= ~PER_CLEAR_ON_SETID;
948                         bprm->e_gid = inode->i_gid;
949                 }
950         }
951
952         /* fill in binprm security blob */
953         retval = security_bprm_set(bprm);
954         if (retval)
955                 return retval;
956
957         memset(bprm->buf,0,BINPRM_BUF_SIZE);
958         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
959 }
960
961 EXPORT_SYMBOL(prepare_binprm);
962
963 static inline int unsafe_exec(struct task_struct *p)
964 {
965         int unsafe = 0;
966         if (p->ptrace & PT_PTRACED) {
967                 if (p->ptrace & PT_PTRACE_CAP)
968                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
969                 else
970                         unsafe |= LSM_UNSAFE_PTRACE;
971         }
972         if (atomic_read(&p->fs->count) > 1 ||
973             atomic_read(&p->files->count) > 1 ||
974             atomic_read(&p->sighand->count) > 1)
975                 unsafe |= LSM_UNSAFE_SHARE;
976
977         return unsafe;
978 }
979
980 void compute_creds(struct linux_binprm *bprm)
981 {
982         int unsafe;
983
984         if (bprm->e_uid != current->uid)
985                 suid_keys(current);
986         exec_keys(current);
987
988         task_lock(current);
989         unsafe = unsafe_exec(current);
990         security_bprm_apply_creds(bprm, unsafe);
991         task_unlock(current);
992         security_bprm_post_apply_creds(bprm);
993 }
994
995 EXPORT_SYMBOL(compute_creds);
996
997 void remove_arg_zero(struct linux_binprm *bprm)
998 {
999         if (bprm->argc) {
1000                 unsigned long offset;
1001                 char * kaddr;
1002                 struct page *page;
1003
1004                 offset = bprm->p % PAGE_SIZE;
1005                 goto inside;
1006
1007                 while (bprm->p++, *(kaddr+offset++)) {
1008                         if (offset != PAGE_SIZE)
1009                                 continue;
1010                         offset = 0;
1011                         kunmap_atomic(kaddr, KM_USER0);
1012 inside:
1013                         page = bprm->page[bprm->p/PAGE_SIZE];
1014                         kaddr = kmap_atomic(page, KM_USER0);
1015                 }
1016                 kunmap_atomic(kaddr, KM_USER0);
1017                 bprm->argc--;
1018         }
1019 }
1020
1021 EXPORT_SYMBOL(remove_arg_zero);
1022
1023 /*
1024  * cycle the list of binary formats handler, until one recognizes the image
1025  */
1026 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1027 {
1028         int try,retval;
1029         struct linux_binfmt *fmt;
1030 #ifdef __alpha__
1031         /* handle /sbin/loader.. */
1032         {
1033             struct exec * eh = (struct exec *) bprm->buf;
1034
1035             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1036                 (eh->fh.f_flags & 0x3000) == 0x3000)
1037             {
1038                 struct file * file;
1039                 unsigned long loader;
1040
1041                 allow_write_access(bprm->file);
1042                 fput(bprm->file);
1043                 bprm->file = NULL;
1044
1045                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1046
1047                 file = open_exec("/sbin/loader");
1048                 retval = PTR_ERR(file);
1049                 if (IS_ERR(file))
1050                         return retval;
1051
1052                 /* Remember if the application is TASO.  */
1053                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1054
1055                 bprm->file = file;
1056                 bprm->loader = loader;
1057                 retval = prepare_binprm(bprm);
1058                 if (retval<0)
1059                         return retval;
1060                 /* should call search_binary_handler recursively here,
1061                    but it does not matter */
1062             }
1063         }
1064 #endif
1065         retval = security_bprm_check(bprm);
1066         if (retval)
1067                 return retval;
1068
1069         /* kernel module loader fixup */
1070         /* so we don't try to load run modprobe in kernel space. */
1071         set_fs(USER_DS);
1072         retval = -ENOENT;
1073         for (try=0; try<2; try++) {
1074                 read_lock(&binfmt_lock);
1075                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1076                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1077                         if (!fn)
1078                                 continue;
1079                         if (!try_module_get(fmt->module))
1080                                 continue;
1081                         read_unlock(&binfmt_lock);
1082                         retval = fn(bprm, regs);
1083                         if (retval >= 0) {
1084                                 put_binfmt(fmt);
1085                                 allow_write_access(bprm->file);
1086                                 if (bprm->file)
1087                                         fput(bprm->file);
1088                                 bprm->file = NULL;
1089                                 current->did_exec = 1;
1090                                 return retval;
1091                         }
1092                         read_lock(&binfmt_lock);
1093                         put_binfmt(fmt);
1094                         if (retval != -ENOEXEC || bprm->mm == NULL)
1095                                 break;
1096                         if (!bprm->file) {
1097                                 read_unlock(&binfmt_lock);
1098                                 return retval;
1099                         }
1100                 }
1101                 read_unlock(&binfmt_lock);
1102                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1103                         break;
1104 #ifdef CONFIG_KMOD
1105                 }else{
1106 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1107                         if (printable(bprm->buf[0]) &&
1108                             printable(bprm->buf[1]) &&
1109                             printable(bprm->buf[2]) &&
1110                             printable(bprm->buf[3]))
1111                                 break; /* -ENOEXEC */
1112                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1113 #endif
1114                 }
1115         }
1116         return retval;
1117 }
1118
1119 EXPORT_SYMBOL(search_binary_handler);
1120
1121 /*
1122  * sys_execve() executes a new program.
1123  */
1124 int do_execve(char * filename,
1125         char __user *__user *argv,
1126         char __user *__user *envp,
1127         struct pt_regs * regs)
1128 {
1129         struct linux_binprm *bprm;
1130         struct file *file;
1131         int retval;
1132         int i;
1133
1134         retval = -ENOMEM;
1135         bprm = kmalloc(sizeof(*bprm), GFP_KERNEL);
1136         if (!bprm)
1137                 goto out_ret;
1138         memset(bprm, 0, sizeof(*bprm));
1139
1140         file = open_exec(filename);
1141         retval = PTR_ERR(file);
1142         if (IS_ERR(file))
1143                 goto out_kfree;
1144
1145         sched_exec();
1146
1147         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1148
1149         bprm->file = file;
1150         bprm->filename = filename;
1151         bprm->interp = filename;
1152         bprm->mm = mm_alloc();
1153         retval = -ENOMEM;
1154         if (!bprm->mm)
1155                 goto out_file;
1156
1157         retval = init_new_context(current, bprm->mm);
1158         if (retval < 0)
1159                 goto out_mm;
1160
1161         bprm->argc = count(argv, bprm->p / sizeof(void *));
1162         if ((retval = bprm->argc) < 0)
1163                 goto out_mm;
1164
1165         bprm->envc = count(envp, bprm->p / sizeof(void *));
1166         if ((retval = bprm->envc) < 0)
1167                 goto out_mm;
1168
1169         retval = security_bprm_alloc(bprm);
1170         if (retval)
1171                 goto out;
1172
1173         retval = prepare_binprm(bprm);
1174         if (retval < 0)
1175                 goto out;
1176
1177         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1178         if (retval < 0)
1179                 goto out;
1180
1181         bprm->exec = bprm->p;
1182         retval = copy_strings(bprm->envc, envp, bprm);
1183         if (retval < 0)
1184                 goto out;
1185
1186         retval = copy_strings(bprm->argc, argv, bprm);
1187         if (retval < 0)
1188                 goto out;
1189
1190         retval = search_binary_handler(bprm,regs);
1191         if (retval >= 0) {
1192                 free_arg_pages(bprm);
1193
1194                 /* execve success */
1195                 security_bprm_free(bprm);
1196                 acct_update_integrals(current);
1197                 update_mem_hiwater(current);
1198                 kfree(bprm);
1199                 return retval;
1200         }
1201
1202 out:
1203         /* Something went wrong, return the inode and free the argument pages*/
1204         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1205                 struct page * page = bprm->page[i];
1206                 if (page)
1207                         __free_page(page);
1208         }
1209
1210         if (bprm->security)
1211                 security_bprm_free(bprm);
1212
1213 out_mm:
1214         if (bprm->mm)
1215                 mmdrop(bprm->mm);
1216
1217 out_file:
1218         if (bprm->file) {
1219                 allow_write_access(bprm->file);
1220                 fput(bprm->file);
1221         }
1222
1223 out_kfree:
1224         kfree(bprm);
1225
1226 out_ret:
1227         return retval;
1228 }
1229
1230 int set_binfmt(struct linux_binfmt *new)
1231 {
1232         struct linux_binfmt *old = current->binfmt;
1233
1234         if (new) {
1235                 if (!try_module_get(new->module))
1236                         return -1;
1237         }
1238         current->binfmt = new;
1239         if (old)
1240                 module_put(old->module);
1241         return 0;
1242 }
1243
1244 EXPORT_SYMBOL(set_binfmt);
1245
1246 #define CORENAME_MAX_SIZE 64
1247
1248 /* format_corename will inspect the pattern parameter, and output a
1249  * name into corename, which must have space for at least
1250  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1251  */
1252 static void format_corename(char *corename, const char *pattern, long signr)
1253 {
1254         const char *pat_ptr = pattern;
1255         char *out_ptr = corename;
1256         char *const out_end = corename + CORENAME_MAX_SIZE;
1257         int rc;
1258         int pid_in_pattern = 0;
1259
1260         /* Repeat as long as we have more pattern to process and more output
1261            space */
1262         while (*pat_ptr) {
1263                 if (*pat_ptr != '%') {
1264                         if (out_ptr == out_end)
1265                                 goto out;
1266                         *out_ptr++ = *pat_ptr++;
1267                 } else {
1268                         switch (*++pat_ptr) {
1269                         case 0:
1270                                 goto out;
1271                         /* Double percent, output one percent */
1272                         case '%':
1273                                 if (out_ptr == out_end)
1274                                         goto out;
1275                                 *out_ptr++ = '%';
1276                                 break;
1277                         /* pid */
1278                         case 'p':
1279                                 pid_in_pattern = 1;
1280                                 rc = snprintf(out_ptr, out_end - out_ptr,
1281                                               "%d", current->tgid);
1282                                 if (rc > out_end - out_ptr)
1283                                         goto out;
1284                                 out_ptr += rc;
1285                                 break;
1286                         /* uid */
1287                         case 'u':
1288                                 rc = snprintf(out_ptr, out_end - out_ptr,
1289                                               "%d", current->uid);
1290                                 if (rc > out_end - out_ptr)
1291                                         goto out;
1292                                 out_ptr += rc;
1293                                 break;
1294                         /* gid */
1295                         case 'g':
1296                                 rc = snprintf(out_ptr, out_end - out_ptr,
1297                                               "%d", current->gid);
1298                                 if (rc > out_end - out_ptr)
1299                                         goto out;
1300                                 out_ptr += rc;
1301                                 break;
1302                         /* signal that caused the coredump */
1303                         case 's':
1304                                 rc = snprintf(out_ptr, out_end - out_ptr,
1305                                               "%ld", signr);
1306                                 if (rc > out_end - out_ptr)
1307                                         goto out;
1308                                 out_ptr += rc;
1309                                 break;
1310                         /* UNIX time of coredump */
1311                         case 't': {
1312                                 struct timeval tv;
1313                                 do_gettimeofday(&tv);
1314                                 rc = snprintf(out_ptr, out_end - out_ptr,
1315                                               "%lu", tv.tv_sec);
1316                                 if (rc > out_end - out_ptr)
1317                                         goto out;
1318                                 out_ptr += rc;
1319                                 break;
1320                         }
1321                         /* hostname */
1322                         case 'h':
1323                                 down_read(&uts_sem);
1324                                 rc = snprintf(out_ptr, out_end - out_ptr,
1325                                               "%s", system_utsname.nodename);
1326                                 up_read(&uts_sem);
1327                                 if (rc > out_end - out_ptr)
1328                                         goto out;
1329                                 out_ptr += rc;
1330                                 break;
1331                         /* executable */
1332                         case 'e':
1333                                 rc = snprintf(out_ptr, out_end - out_ptr,
1334                                               "%s", current->comm);
1335                                 if (rc > out_end - out_ptr)
1336                                         goto out;
1337                                 out_ptr += rc;
1338                                 break;
1339                         default:
1340                                 break;
1341                         }
1342                         ++pat_ptr;
1343                 }
1344         }
1345         /* Backward compatibility with core_uses_pid:
1346          *
1347          * If core_pattern does not include a %p (as is the default)
1348          * and core_uses_pid is set, then .%pid will be appended to
1349          * the filename */
1350         if (!pid_in_pattern
1351             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1352                 rc = snprintf(out_ptr, out_end - out_ptr,
1353                               ".%d", current->tgid);
1354                 if (rc > out_end - out_ptr)
1355                         goto out;
1356                 out_ptr += rc;
1357         }
1358       out:
1359         *out_ptr = 0;
1360 }
1361
1362 static void zap_threads (struct mm_struct *mm)
1363 {
1364         struct task_struct *g, *p;
1365         struct task_struct *tsk = current;
1366         struct completion *vfork_done = tsk->vfork_done;
1367         int traced = 0;
1368
1369         /*
1370          * Make sure nobody is waiting for us to release the VM,
1371          * otherwise we can deadlock when we wait on each other
1372          */
1373         if (vfork_done) {
1374                 tsk->vfork_done = NULL;
1375                 complete(vfork_done);
1376         }
1377
1378         read_lock(&tasklist_lock);
1379         do_each_thread(g,p)
1380                 if (mm == p->mm && p != tsk) {
1381                         force_sig_specific(SIGKILL, p);
1382                         mm->core_waiters++;
1383                         if (unlikely(p->ptrace) &&
1384                             unlikely(p->parent->mm == mm))
1385                                 traced = 1;
1386                 }
1387         while_each_thread(g,p);
1388
1389         read_unlock(&tasklist_lock);
1390
1391         if (unlikely(traced)) {
1392                 /*
1393                  * We are zapping a thread and the thread it ptraces.
1394                  * If the tracee went into a ptrace stop for exit tracing,
1395                  * we could deadlock since the tracer is waiting for this
1396                  * coredump to finish.  Detach them so they can both die.
1397                  */
1398                 write_lock_irq(&tasklist_lock);
1399                 do_each_thread(g,p) {
1400                         if (mm == p->mm && p != tsk &&
1401                             p->ptrace && p->parent->mm == mm) {
1402                                 __ptrace_unlink(p);
1403                         }
1404                 } while_each_thread(g,p);
1405                 write_unlock_irq(&tasklist_lock);
1406         }
1407 }
1408
1409 static void coredump_wait(struct mm_struct *mm)
1410 {
1411         DECLARE_COMPLETION(startup_done);
1412
1413         mm->core_waiters++; /* let other threads block */
1414         mm->core_startup_done = &startup_done;
1415
1416         /* give other threads a chance to run: */
1417         yield();
1418
1419         zap_threads(mm);
1420         if (--mm->core_waiters) {
1421                 up_write(&mm->mmap_sem);
1422                 wait_for_completion(&startup_done);
1423         } else
1424                 up_write(&mm->mmap_sem);
1425         BUG_ON(mm->core_waiters);
1426 }
1427
1428 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1429 {
1430         char corename[CORENAME_MAX_SIZE + 1];
1431         struct mm_struct *mm = current->mm;
1432         struct linux_binfmt * binfmt;
1433         struct inode * inode;
1434         struct file * file;
1435         int retval = 0;
1436
1437         binfmt = current->binfmt;
1438         if (!binfmt || !binfmt->core_dump)
1439                 goto fail;
1440         down_write(&mm->mmap_sem);
1441         if (!mm->dumpable) {
1442                 up_write(&mm->mmap_sem);
1443                 goto fail;
1444         }
1445         mm->dumpable = 0;
1446         init_completion(&mm->core_done);
1447         spin_lock_irq(&current->sighand->siglock);
1448         current->signal->flags = SIGNAL_GROUP_EXIT;
1449         current->signal->group_exit_code = exit_code;
1450         spin_unlock_irq(&current->sighand->siglock);
1451         coredump_wait(mm);
1452
1453         /*
1454          * Clear any false indication of pending signals that might
1455          * be seen by the filesystem code called to write the core file.
1456          */
1457         current->signal->group_stop_count = 0;
1458         clear_thread_flag(TIF_SIGPENDING);
1459
1460         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1461                 goto fail_unlock;
1462
1463         /*
1464          * lock_kernel() because format_corename() is controlled by sysctl, which
1465          * uses lock_kernel()
1466          */
1467         lock_kernel();
1468         format_corename(corename, core_pattern, signr);
1469         unlock_kernel();
1470         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
1471         if (IS_ERR(file))
1472                 goto fail_unlock;
1473         inode = file->f_dentry->d_inode;
1474         if (inode->i_nlink > 1)
1475                 goto close_fail;        /* multiple links - don't dump */
1476         if (d_unhashed(file->f_dentry))
1477                 goto close_fail;
1478
1479         if (!S_ISREG(inode->i_mode))
1480                 goto close_fail;
1481         if (!file->f_op)
1482                 goto close_fail;
1483         if (!file->f_op->write)
1484                 goto close_fail;
1485         if (do_truncate(file->f_dentry, 0) != 0)
1486                 goto close_fail;
1487
1488         retval = binfmt->core_dump(signr, regs, file);
1489
1490         if (retval)
1491                 current->signal->group_exit_code |= 0x80;
1492 close_fail:
1493         filp_close(file, NULL);
1494 fail_unlock:
1495         complete_all(&mm->core_done);
1496 fail:
1497         return retval;
1498 }