This commit was manufactured by cvs2svn to create tag
[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/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/personality.h>
38 #include <linux/binfmts.h>
39 #include <linux/swap.h>
40 #include <linux/utsname.h>
41 #include <linux/module.h>
42 #include <linux/namei.h>
43 #include <linux/proc_fs.h>
44 #include <linux/ptrace.h>
45 #include <linux/mount.h>
46 #include <linux/security.h>
47 #include <linux/syscalls.h>
48 #include <linux/rmap.h>
49 #include <linux/ckrm.h>
50
51 #include <asm/uaccess.h>
52 #include <asm/mmu_context.h>
53
54 #ifdef CONFIG_KMOD
55 #include <linux/kmod.h>
56 #endif
57
58 int core_uses_pid;
59 char core_pattern[65] = "core";
60 /* The maximal length of core_pattern is also specified in sysctl.c */
61
62 static struct linux_binfmt *formats;
63 static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
64
65 int register_binfmt(struct linux_binfmt * fmt)
66 {
67         struct linux_binfmt ** tmp = &formats;
68
69         if (!fmt)
70                 return -EINVAL;
71         if (fmt->next)
72                 return -EBUSY;
73         write_lock(&binfmt_lock);
74         while (*tmp) {
75                 if (fmt == *tmp) {
76                         write_unlock(&binfmt_lock);
77                         return -EBUSY;
78                 }
79                 tmp = &(*tmp)->next;
80         }
81         fmt->next = formats;
82         formats = fmt;
83         write_unlock(&binfmt_lock);
84         return 0;       
85 }
86
87 EXPORT_SYMBOL(register_binfmt);
88
89 int unregister_binfmt(struct linux_binfmt * fmt)
90 {
91         struct linux_binfmt ** tmp = &formats;
92
93         write_lock(&binfmt_lock);
94         while (*tmp) {
95                 if (fmt == *tmp) {
96                         *tmp = fmt->next;
97                         write_unlock(&binfmt_lock);
98                         return 0;
99                 }
100                 tmp = &(*tmp)->next;
101         }
102         write_unlock(&binfmt_lock);
103         return -EINVAL;
104 }
105
106 EXPORT_SYMBOL(unregister_binfmt);
107
108 static inline void put_binfmt(struct linux_binfmt * fmt)
109 {
110         module_put(fmt->module);
111 }
112
113 /*
114  * Note that a shared library must be both readable and executable due to
115  * security reasons.
116  *
117  * Also note that we take the address to load from from the file itself.
118  */
119 asmlinkage long sys_uselib(const char __user * library)
120 {
121         struct file * file;
122         struct nameidata nd;
123         int error;
124
125         nd.intent.open.flags = FMODE_READ;
126         error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
127         if (error)
128                 goto out;
129
130         error = -EINVAL;
131         if (!S_ISREG(nd.dentry->d_inode->i_mode))
132                 goto exit;
133
134         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
135         if (error)
136                 goto exit;
137
138         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
139         error = PTR_ERR(file);
140         if (IS_ERR(file))
141                 goto out;
142
143         error = -ENOEXEC;
144         if(file->f_op) {
145                 struct linux_binfmt * fmt;
146
147                 read_lock(&binfmt_lock);
148                 for (fmt = formats ; fmt ; fmt = fmt->next) {
149                         if (!fmt->load_shlib)
150                                 continue;
151                         if (!try_module_get(fmt->module))
152                                 continue;
153                         read_unlock(&binfmt_lock);
154                         error = fmt->load_shlib(file);
155                         read_lock(&binfmt_lock);
156                         put_binfmt(fmt);
157                         if (error != -ENOEXEC)
158                                 break;
159                 }
160                 read_unlock(&binfmt_lock);
161         }
162         fput(file);
163 out:
164         return error;
165 exit:
166         path_release(&nd);
167         goto out;
168 }
169
170 /*
171  * count() counts the number of strings in array ARGV.
172  */
173 static int count(char __user * __user * argv, int max)
174 {
175         int i = 0;
176
177         if (argv != NULL) {
178                 for (;;) {
179                         char __user * p;
180
181                         if (get_user(p, argv))
182                                 return -EFAULT;
183                         if (!p)
184                                 break;
185                         argv++;
186                         if(++i > max)
187                                 return -E2BIG;
188                 }
189         }
190         return i;
191 }
192
193 /*
194  * 'copy_strings()' copies argument/environment strings from user
195  * memory to free pages in kernel mem. These are in a format ready
196  * to be put directly into the top of new user memory.
197  */
198 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
199 {
200         struct page *kmapped_page = NULL;
201         char *kaddr = NULL;
202         int ret;
203
204         while (argc-- > 0) {
205                 char __user *str;
206                 int len;
207                 unsigned long pos;
208
209                 if (get_user(str, argv+argc) ||
210                                 !(len = strnlen_user(str, bprm->p))) {
211                         ret = -EFAULT;
212                         goto out;
213                 }
214
215                 if (bprm->p < len)  {
216                         ret = -E2BIG;
217                         goto out;
218                 }
219
220                 bprm->p -= len;
221                 /* XXX: add architecture specific overflow check here. */
222                 pos = bprm->p;
223
224                 while (len > 0) {
225                         int i, new, err;
226                         int offset, bytes_to_copy;
227                         struct page *page;
228
229                         offset = pos % PAGE_SIZE;
230                         i = pos/PAGE_SIZE;
231                         page = bprm->page[i];
232                         new = 0;
233                         if (!page) {
234                                 page = alloc_page(GFP_HIGHUSER);
235                                 bprm->page[i] = page;
236                                 if (!page) {
237                                         ret = -ENOMEM;
238                                         goto out;
239                                 }
240                                 new = 1;
241                         }
242
243                         if (page != kmapped_page) {
244                                 if (kmapped_page)
245                                         kunmap(kmapped_page);
246                                 kmapped_page = page;
247                                 kaddr = kmap(kmapped_page);
248                         }
249                         if (new && offset)
250                                 memset(kaddr, 0, offset);
251                         bytes_to_copy = PAGE_SIZE - offset;
252                         if (bytes_to_copy > len) {
253                                 bytes_to_copy = len;
254                                 if (new)
255                                         memset(kaddr+offset+len, 0,
256                                                 PAGE_SIZE-offset-len);
257                         }
258                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
259                         if (err) {
260                                 ret = -EFAULT;
261                                 goto out;
262                         }
263
264                         pos += bytes_to_copy;
265                         str += bytes_to_copy;
266                         len -= bytes_to_copy;
267                 }
268         }
269         ret = 0;
270 out:
271         if (kmapped_page)
272                 kunmap(kmapped_page);
273         return ret;
274 }
275
276 /*
277  * Like copy_strings, but get argv and its values from kernel memory.
278  */
279 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
280 {
281         int r;
282         mm_segment_t oldfs = get_fs();
283         set_fs(KERNEL_DS);
284         r = copy_strings(argc, (char __user * __user *)argv, bprm);
285         set_fs(oldfs);
286         return r;
287 }
288
289 EXPORT_SYMBOL(copy_strings_kernel);
290
291 #ifdef CONFIG_MMU
292 /*
293  * This routine is used to map in a page into an address space: needed by
294  * execve() for the initial stack and environment pages.
295  *
296  * vma->vm_mm->mmap_sem is held for writing.
297  */
298 void install_arg_page(struct vm_area_struct *vma,
299                         struct page *page, unsigned long address)
300 {
301         struct mm_struct *mm = vma->vm_mm;
302         pgd_t * pgd;
303         pmd_t * pmd;
304         pte_t * pte;
305
306         if (unlikely(anon_vma_prepare(vma)))
307                 goto out_sig;
308
309         flush_dcache_page(page);
310         pgd = pgd_offset(mm, address);
311
312         spin_lock(&mm->page_table_lock);
313         pmd = pmd_alloc(mm, pgd, address);
314         if (!pmd)
315                 goto out;
316         pte = pte_alloc_map(mm, pmd, address);
317         if (!pte)
318                 goto out;
319         if (!pte_none(*pte)) {
320                 pte_unmap(pte);
321                 goto out;
322         }
323         // mm->rss++;
324         vx_rsspages_inc(mm);
325         lru_cache_add_active(page);
326         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(
327                                         page, vma->vm_page_prot))));
328         page_add_anon_rmap(page, vma, address);
329         pte_unmap(pte);
330         spin_unlock(&mm->page_table_lock);
331
332         /* no need for flush_tlb */
333         return;
334 out:
335         spin_unlock(&mm->page_table_lock);
336 out_sig:
337         __free_page(page);
338         force_sig(SIGKILL, current);
339 }
340
341 int setup_arg_pages(struct linux_binprm *bprm, int executable_stack)
342 {
343         unsigned long stack_base;
344         struct vm_area_struct *mpnt;
345         struct mm_struct *mm = current->mm;
346         int i;
347         long arg_size;
348
349 #ifdef CONFIG_STACK_GROWSUP
350         /* Move the argument and environment strings to the bottom of the
351          * stack space.
352          */
353         int offset, j;
354         char *to, *from;
355
356         /* Start by shifting all the pages down */
357         i = 0;
358         for (j = 0; j < MAX_ARG_PAGES; j++) {
359                 struct page *page = bprm->page[j];
360                 if (!page)
361                         continue;
362                 bprm->page[i++] = page;
363         }
364
365         /* Now move them within their pages */
366         offset = bprm->p % PAGE_SIZE;
367         to = kmap(bprm->page[0]);
368         for (j = 1; j < i; j++) {
369                 memmove(to, to + offset, PAGE_SIZE - offset);
370                 from = kmap(bprm->page[j]);
371                 memcpy(to + PAGE_SIZE - offset, from, offset);
372                 kunmap(bprm->page[j - 1]);
373                 to = from;
374         }
375         memmove(to, to + offset, PAGE_SIZE - offset);
376         kunmap(bprm->page[j - 1]);
377
378         /* Adjust bprm->p to point to the end of the strings. */
379         bprm->p = PAGE_SIZE * i - offset;
380
381         /* Limit stack size to 1GB */
382         stack_base = current->rlim[RLIMIT_STACK].rlim_max;
383         if (stack_base > (1 << 30))
384                 stack_base = 1 << 30;
385         stack_base = PAGE_ALIGN(STACK_TOP - stack_base);
386
387         mm->arg_start = stack_base;
388         arg_size = i << PAGE_SHIFT;
389
390         /* zero pages that were copied above */
391         while (i < MAX_ARG_PAGES)
392                 bprm->page[i++] = NULL;
393 #else
394 #ifdef __HAVE_ARCH_ALIGN_STACK
395         stack_base = arch_align_stack(STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE);
396         stack_base = PAGE_ALIGN(stack_base);
397 #else
398         stack_base = STACK_TOP - MAX_ARG_PAGES * PAGE_SIZE;
399 #endif
400         mm->arg_start = bprm->p + stack_base;
401         arg_size = STACK_TOP - (PAGE_MASK & (unsigned long) mm->arg_start);
402 #endif
403
404         bprm->p += stack_base;
405         if (bprm->loader)
406                 bprm->loader += stack_base;
407         bprm->exec += stack_base;
408
409         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
410         if (!mpnt)
411                 return -ENOMEM;
412
413         if (security_vm_enough_memory(arg_size >> PAGE_SHIFT) ||
414                 !vx_vmpages_avail(mm, arg_size >> PAGE_SHIFT)) {
415                 kmem_cache_free(vm_area_cachep, mpnt);
416                 return -ENOMEM;
417         }
418
419         memset(mpnt, 0, sizeof(*mpnt));
420
421         down_write(&mm->mmap_sem);
422         {
423                 mpnt->vm_mm = mm;
424 #ifdef CONFIG_STACK_GROWSUP
425                 mpnt->vm_start = stack_base;
426                 mpnt->vm_end = PAGE_MASK &
427                         (PAGE_SIZE - 1 + (unsigned long) bprm->p);
428 #else
429                 mpnt->vm_start = PAGE_MASK & (unsigned long) bprm->p;
430                 mpnt->vm_end = STACK_TOP;
431 #endif
432                 /* Adjust stack execute permissions; explicitly enable
433                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
434                  * and leave alone (arch default) otherwise. */
435                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
436                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
437                 else if (executable_stack == EXSTACK_DISABLE_X)
438                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
439                 else
440                         mpnt->vm_flags = VM_STACK_FLAGS;
441                 mpnt->vm_flags |= mm->def_flags;
442                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
443                 insert_vm_struct(mm, mpnt);
444                 // mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
445                 vx_vmpages_sub(mm, mm->total_vm -
446                         ((mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT));
447         }
448
449         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
450                 struct page *page = bprm->page[i];
451                 if (page) {
452                         bprm->page[i] = NULL;
453                         install_arg_page(mpnt, page, stack_base);
454                 }
455                 stack_base += PAGE_SIZE;
456         }
457         up_write(&mm->mmap_sem);
458         
459         return 0;
460 }
461
462 EXPORT_SYMBOL(setup_arg_pages);
463
464 #define free_arg_pages(bprm) do { } while (0)
465
466 #else
467
468 static inline void free_arg_pages(struct linux_binprm *bprm)
469 {
470         int i;
471
472         for (i = 0; i < MAX_ARG_PAGES; i++) {
473                 if (bprm->page[i])
474                         __free_page(bprm->page[i]);
475                 bprm->page[i] = NULL;
476         }
477 }
478
479 #endif /* CONFIG_MMU */
480
481 struct file *open_exec(const char *name)
482 {
483         struct nameidata nd;
484         int err;
485         struct file *file;
486
487         nd.intent.open.flags = FMODE_READ;
488         err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
489         file = ERR_PTR(err);
490
491         if (!err) {
492                 struct inode *inode = nd.dentry->d_inode;
493                 file = ERR_PTR(-EACCES);
494                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
495                     S_ISREG(inode->i_mode)) {
496                         int err = permission(inode, MAY_EXEC, &nd);
497                         if (!err && !(inode->i_mode & 0111))
498                                 err = -EACCES;
499                         file = ERR_PTR(err);
500                         if (!err) {
501                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
502                                 if (!IS_ERR(file)) {
503                                         err = deny_write_access(file);
504                                         if (err) {
505                                                 fput(file);
506                                                 file = ERR_PTR(err);
507                                         }
508                                 }
509 out:
510                                 return file;
511                         }
512                 }
513                 path_release(&nd);
514         }
515         goto out;
516 }
517
518 EXPORT_SYMBOL(open_exec);
519
520 int kernel_read(struct file *file, unsigned long offset,
521         char *addr, unsigned long count)
522 {
523         mm_segment_t old_fs;
524         loff_t pos = offset;
525         int result;
526
527         old_fs = get_fs();
528         set_fs(get_ds());
529         /* The cast to a user pointer is valid due to the set_fs() */
530         result = vfs_read(file, (void __user *)addr, count, &pos);
531         set_fs(old_fs);
532         return result;
533 }
534
535 EXPORT_SYMBOL(kernel_read);
536
537 static int exec_mmap(struct mm_struct *mm)
538 {
539         struct task_struct *tsk;
540         struct mm_struct * old_mm, *active_mm;
541
542         /* Add it to the list of mm's */
543         spin_lock(&mmlist_lock);
544         list_add(&mm->mmlist, &init_mm.mmlist);
545         mmlist_nr++;
546         spin_unlock(&mmlist_lock);
547
548         /* Notify parent that we're no longer interested in the old VM */
549         tsk = current;
550         old_mm = current->mm;
551         mm_release(tsk, old_mm);
552
553         task_lock(tsk);
554         active_mm = tsk->active_mm;
555         tsk->mm = mm;
556         tsk->active_mm = mm;
557         activate_mm(active_mm, mm);
558         task_unlock(tsk);
559         if (old_mm) {
560                 if (active_mm != old_mm) BUG();
561                 mmput(old_mm);
562                 return 0;
563         }
564         mmdrop(active_mm);
565         return 0;
566 }
567
568 /*
569  * This function makes sure the current process has its own signal table,
570  * so that flush_signal_handlers can later reset the handlers without
571  * disturbing other processes.  (Other processes might share the signal
572  * table via the CLONE_SIGHAND option to clone().)
573  */
574 static inline int de_thread(struct task_struct *tsk)
575 {
576         struct signal_struct *newsig, *oldsig = tsk->signal;
577         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
578         spinlock_t *lock = &oldsighand->siglock;
579         int count;
580
581         /*
582          * If we don't share sighandlers, then we aren't sharing anything
583          * and we can just re-use it all.
584          */
585         if (atomic_read(&oldsighand->count) <= 1)
586                 return 0;
587
588         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
589         if (!newsighand)
590                 return -ENOMEM;
591
592         spin_lock_init(&newsighand->siglock);
593         atomic_set(&newsighand->count, 1);
594         memcpy(newsighand->action, oldsighand->action, sizeof(newsighand->action));
595
596         /*
597          * See if we need to allocate a new signal structure
598          */
599         newsig = NULL;
600         if (atomic_read(&oldsig->count) > 1) {
601                 newsig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
602                 if (!newsig) {
603                         kmem_cache_free(sighand_cachep, newsighand);
604                         return -ENOMEM;
605                 }
606                 atomic_set(&newsig->count, 1);
607                 newsig->group_exit = 0;
608                 newsig->group_exit_code = 0;
609                 newsig->group_exit_task = NULL;
610                 newsig->group_stop_count = 0;
611                 newsig->curr_target = NULL;
612                 init_sigpending(&newsig->shared_pending);
613                 INIT_LIST_HEAD(&newsig->posix_timers);
614
615                 newsig->tty = oldsig->tty;
616                 newsig->pgrp = oldsig->pgrp;
617                 newsig->session = oldsig->session;
618                 newsig->leader = oldsig->leader;
619                 newsig->tty_old_pgrp = oldsig->tty_old_pgrp;
620         }
621
622         if (thread_group_empty(current))
623                 goto no_thread_group;
624
625         /*
626          * Kill all other threads in the thread group.
627          * We must hold tasklist_lock to call zap_other_threads.
628          */
629         read_lock(&tasklist_lock);
630         spin_lock_irq(lock);
631         if (oldsig->group_exit) {
632                 /*
633                  * Another group action in progress, just
634                  * return so that the signal is processed.
635                  */
636                 spin_unlock_irq(lock);
637                 read_unlock(&tasklist_lock);
638                 kmem_cache_free(sighand_cachep, newsighand);
639                 if (newsig)
640                         kmem_cache_free(signal_cachep, newsig);
641                 return -EAGAIN;
642         }
643         oldsig->group_exit = 1;
644         zap_other_threads(current);
645         read_unlock(&tasklist_lock);
646
647         /*
648          * Account for the thread group leader hanging around:
649          */
650         count = 2;
651         if (current->pid == current->tgid)
652                 count = 1;
653         while (atomic_read(&oldsig->count) > count) {
654                 oldsig->group_exit_task = current;
655                 oldsig->notify_count = count;
656                 __set_current_state(TASK_UNINTERRUPTIBLE);
657                 spin_unlock_irq(lock);
658                 schedule();
659                 spin_lock_irq(lock);
660         }
661         spin_unlock_irq(lock);
662
663         /*
664          * At this point all other threads have exited, all we have to
665          * do is to wait for the thread group leader to become inactive,
666          * and to assume its PID:
667          */
668         if (current->pid != current->tgid) {
669                 struct task_struct *leader = current->group_leader, *parent;
670                 struct dentry *proc_dentry1, *proc_dentry2;
671                 unsigned long state, ptrace;
672
673                 /*
674                  * Wait for the thread group leader to be a zombie.
675                  * It should already be zombie at this point, most
676                  * of the time.
677                  */
678                 while (leader->state != TASK_ZOMBIE)
679                         yield();
680
681                 spin_lock(&leader->proc_lock);
682                 spin_lock(&current->proc_lock);
683                 proc_dentry1 = proc_pid_unhash(current);
684                 proc_dentry2 = proc_pid_unhash(leader);
685                 write_lock_irq(&tasklist_lock);
686
687                 if (leader->tgid != current->tgid)
688                         BUG();
689                 if (current->pid == current->tgid)
690                         BUG();
691                 /*
692                  * An exec() starts a new thread group with the
693                  * TGID of the previous thread group. Rehash the
694                  * two threads with a switched PID, and release
695                  * the former thread group leader:
696                  */
697                 ptrace = leader->ptrace;
698                 parent = leader->parent;
699
700                 ptrace_unlink(current);
701                 ptrace_unlink(leader);
702                 remove_parent(current);
703                 remove_parent(leader);
704
705                 switch_exec_pids(leader, current);
706
707                 current->parent = current->real_parent = leader->real_parent;
708                 leader->parent = leader->real_parent = child_reaper;
709                 current->group_leader = current;
710                 leader->group_leader = leader;
711
712                 add_parent(current, current->parent);
713                 add_parent(leader, leader->parent);
714                 if (ptrace) {
715                         current->ptrace = ptrace;
716                         __ptrace_link(current, parent);
717                 }
718
719                 list_del(&current->tasks);
720                 list_add_tail(&current->tasks, &init_task.tasks);
721                 current->exit_signal = SIGCHLD;
722                 state = leader->state;
723
724                 write_unlock_irq(&tasklist_lock);
725                 spin_unlock(&leader->proc_lock);
726                 spin_unlock(&current->proc_lock);
727                 proc_pid_flush(proc_dentry1);
728                 proc_pid_flush(proc_dentry2);
729
730                 if (state != TASK_ZOMBIE)
731                         BUG();
732                 release_task(leader);
733         }
734
735 no_thread_group:
736
737         write_lock_irq(&tasklist_lock);
738         spin_lock(&oldsighand->siglock);
739         spin_lock(&newsighand->siglock);
740
741         if (current == oldsig->curr_target)
742                 oldsig->curr_target = next_thread(current);
743         if (newsig)
744                 current->signal = newsig;
745         current->sighand = newsighand;
746         init_sigpending(&current->pending);
747         recalc_sigpending();
748
749         spin_unlock(&newsighand->siglock);
750         spin_unlock(&oldsighand->siglock);
751         write_unlock_irq(&tasklist_lock);
752
753         if (newsig && atomic_dec_and_test(&oldsig->count))
754                 kmem_cache_free(signal_cachep, oldsig);
755
756         if (atomic_dec_and_test(&oldsighand->count))
757                 kmem_cache_free(sighand_cachep, oldsighand);
758
759         if (!thread_group_empty(current))
760                 BUG();
761         if (current->tgid != current->pid)
762                 BUG();
763         return 0;
764 }
765         
766 /*
767  * These functions flushes out all traces of the currently running executable
768  * so that a new one can be started
769  */
770
771 static inline void flush_old_files(struct files_struct * files)
772 {
773         long j = -1;
774
775         spin_lock(&files->file_lock);
776         for (;;) {
777                 unsigned long set, i;
778
779                 j++;
780                 i = j * __NFDBITS;
781                 if (i >= files->max_fds || i >= files->max_fdset)
782                         break;
783                 set = files->close_on_exec->fds_bits[j];
784                 if (!set)
785                         continue;
786                 files->close_on_exec->fds_bits[j] = 0;
787                 spin_unlock(&files->file_lock);
788                 for ( ; set ; i++,set >>= 1) {
789                         if (set & 1) {
790                                 sys_close(i);
791                         }
792                 }
793                 spin_lock(&files->file_lock);
794
795         }
796         spin_unlock(&files->file_lock);
797 }
798
799 int flush_old_exec(struct linux_binprm * bprm)
800 {
801         char * name;
802         int i, ch, retval;
803         struct files_struct *files;
804
805         /*
806          * Make sure we have a private signal table and that
807          * we are unassociated from the previous thread group.
808          */
809         retval = de_thread(current);
810         if (retval)
811                 goto out;
812
813         /*
814          * Make sure we have private file handles. Ask the
815          * fork helper to do the work for us and the exit
816          * helper to do the cleanup of the old one.
817          */
818         files = current->files;         /* refcounted so safe to hold */
819         retval = unshare_files();
820         if (retval)
821                 goto out;
822         /*
823          * Release all of the old mmap stuff
824          */
825         retval = exec_mmap(bprm->mm);
826         if (retval)
827                 goto mmap_failed;
828
829         bprm->mm = NULL;                /* We're using it now */
830
831         /* This is the point of no return */
832         steal_locks(files);
833         put_files_struct(files);
834
835         current->sas_ss_sp = current->sas_ss_size = 0;
836
837         if (current->euid == current->uid && current->egid == current->gid)
838                 current->mm->dumpable = 1;
839         name = bprm->filename;
840         for (i=0; (ch = *(name++)) != '\0';) {
841                 if (ch == '/')
842                         i = 0;
843                 else
844                         if (i < 15)
845                                 current->comm[i++] = ch;
846         }
847         current->comm[i] = '\0';
848
849         current->flags &= ~PF_RELOCEXEC;
850         flush_thread();
851
852         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid || 
853             permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
854             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP))
855                 current->mm->dumpable = 0;
856
857         /* An exec changes our domain. We are no longer part of the thread
858            group */
859
860         current->self_exec_id++;
861                         
862         flush_signal_handlers(current, 0);
863         flush_old_files(current->files);
864
865         return 0;
866
867 mmap_failed:
868         put_files_struct(current->files);
869         current->files = files;
870 out:
871         return retval;
872 }
873
874 EXPORT_SYMBOL(flush_old_exec);
875
876 /* 
877  * Fill the binprm structure from the inode. 
878  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
879  */
880 int prepare_binprm(struct linux_binprm *bprm)
881 {
882         int mode;
883         struct inode * inode = bprm->file->f_dentry->d_inode;
884         int retval;
885
886         mode = inode->i_mode;
887         /*
888          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
889          * vfs_permission lets a non-executable through
890          */
891         if (!(mode & 0111))     /* with at least _one_ execute bit set */
892                 return -EACCES;
893         if (bprm->file->f_op == NULL)
894                 return -EACCES;
895
896         bprm->e_uid = current->euid;
897         bprm->e_gid = current->egid;
898
899         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
900                 /* Set-uid? */
901                 if (mode & S_ISUID) {
902                         bprm->e_uid = inode->i_uid;
903 #ifdef __i386__
904                         /* reset personality */
905                         current->personality = PER_LINUX;
906 #endif
907                 }
908
909                 /* Set-gid? */
910                 /*
911                  * If setgid is set but no group execute bit then this
912                  * is a candidate for mandatory locking, not a setgid
913                  * executable.
914                  */
915                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
916                         bprm->e_gid = inode->i_gid;
917 #ifdef __i386__
918                         /* reset personality */
919                         current->personality = PER_LINUX;
920 #endif
921                 }
922         }
923
924         /* fill in binprm security blob */
925         retval = security_bprm_set(bprm);
926         if (retval)
927                 return retval;
928
929         memset(bprm->buf,0,BINPRM_BUF_SIZE);
930         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
931 }
932
933 EXPORT_SYMBOL(prepare_binprm);
934
935 static inline int unsafe_exec(struct task_struct *p)
936 {
937         int unsafe = 0;
938         if (p->ptrace & PT_PTRACED) {
939                 if (p->ptrace & PT_PTRACE_CAP)
940                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
941                 else
942                         unsafe |= LSM_UNSAFE_PTRACE;
943         }
944         if (atomic_read(&p->fs->count) > 1 ||
945             atomic_read(&p->files->count) > 1 ||
946             atomic_read(&p->sighand->count) > 1)
947                 unsafe |= LSM_UNSAFE_SHARE;
948
949         return unsafe;
950 }
951
952 void compute_creds(struct linux_binprm *bprm)
953 {
954         int unsafe;
955         task_lock(current);
956         unsafe = unsafe_exec(current);
957         security_bprm_apply_creds(bprm, unsafe);
958         task_unlock(current);
959 }
960
961 EXPORT_SYMBOL(compute_creds);
962
963 void remove_arg_zero(struct linux_binprm *bprm)
964 {
965         if (bprm->argc) {
966                 unsigned long offset;
967                 char * kaddr;
968                 struct page *page;
969
970                 offset = bprm->p % PAGE_SIZE;
971                 goto inside;
972
973                 while (bprm->p++, *(kaddr+offset++)) {
974                         if (offset != PAGE_SIZE)
975                                 continue;
976                         offset = 0;
977                         kunmap_atomic(kaddr, KM_USER0);
978 inside:
979                         page = bprm->page[bprm->p/PAGE_SIZE];
980                         kaddr = kmap_atomic(page, KM_USER0);
981                 }
982                 kunmap_atomic(kaddr, KM_USER0);
983                 bprm->argc--;
984         }
985 }
986
987 EXPORT_SYMBOL(remove_arg_zero);
988
989 /*
990  * cycle the list of binary formats handler, until one recognizes the image
991  */
992 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
993 {
994         int try,retval=0;
995         struct linux_binfmt *fmt;
996 #ifdef __alpha__
997         /* handle /sbin/loader.. */
998         {
999             struct exec * eh = (struct exec *) bprm->buf;
1000
1001             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1002                 (eh->fh.f_flags & 0x3000) == 0x3000)
1003             {
1004                 struct file * file;
1005                 unsigned long loader;
1006
1007                 allow_write_access(bprm->file);
1008                 fput(bprm->file);
1009                 bprm->file = NULL;
1010
1011                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1012
1013                 file = open_exec("/sbin/loader");
1014                 retval = PTR_ERR(file);
1015                 if (IS_ERR(file))
1016                         return retval;
1017
1018                 /* Remember if the application is TASO.  */
1019                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1020
1021                 bprm->file = file;
1022                 bprm->loader = loader;
1023                 retval = prepare_binprm(bprm);
1024                 if (retval<0)
1025                         return retval;
1026                 /* should call search_binary_handler recursively here,
1027                    but it does not matter */
1028             }
1029         }
1030 #endif
1031         retval = security_bprm_check(bprm);
1032         if (retval)
1033                 return retval;
1034
1035         /* kernel module loader fixup */
1036         /* so we don't try to load run modprobe in kernel space. */
1037         set_fs(USER_DS);
1038         for (try=0; try<2; try++) {
1039                 read_lock(&binfmt_lock);
1040                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1041                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1042                         if (!fn)
1043                                 continue;
1044                         if (!try_module_get(fmt->module))
1045                                 continue;
1046                         read_unlock(&binfmt_lock);
1047                         retval = fn(bprm, regs);
1048                         if (retval >= 0) {
1049                                 put_binfmt(fmt);
1050                                 allow_write_access(bprm->file);
1051                                 if (bprm->file)
1052                                         fput(bprm->file);
1053                                 bprm->file = NULL;
1054                                 current->did_exec = 1;
1055                                 ckrm_cb_exec(bprm->filename);
1056                                 return retval;
1057                         }
1058                         read_lock(&binfmt_lock);
1059                         put_binfmt(fmt);
1060                         if (retval != -ENOEXEC || bprm->mm == NULL)
1061                                 break;
1062                         if (!bprm->file) {
1063                                 read_unlock(&binfmt_lock);
1064                                 return retval;
1065                         }
1066                 }
1067                 read_unlock(&binfmt_lock);
1068                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1069                         break;
1070 #ifdef CONFIG_KMOD
1071                 }else{
1072 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1073                         if (printable(bprm->buf[0]) &&
1074                             printable(bprm->buf[1]) &&
1075                             printable(bprm->buf[2]) &&
1076                             printable(bprm->buf[3]))
1077                                 break; /* -ENOEXEC */
1078                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1079 #endif
1080                 }
1081         }
1082         return retval;
1083 }
1084
1085 EXPORT_SYMBOL(search_binary_handler);
1086
1087 /*
1088  * sys_execve() executes a new program.
1089  */
1090 int do_execve(char * filename,
1091         char __user *__user *argv,
1092         char __user *__user *envp,
1093         struct pt_regs * regs)
1094 {
1095         struct linux_binprm bprm;
1096         struct file *file;
1097         int retval;
1098         int i;
1099
1100         file = open_exec(filename);
1101
1102         retval = PTR_ERR(file);
1103         if (IS_ERR(file))
1104                 return retval;
1105
1106         sched_balance_exec();
1107
1108         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1109         memset(bprm.page, 0, MAX_ARG_PAGES*sizeof(bprm.page[0]));
1110
1111         bprm.file = file;
1112         bprm.filename = filename;
1113         bprm.interp = filename;
1114         bprm.interp_flags = 0;
1115         bprm.interp_data = 0;
1116         bprm.sh_bang = 0;
1117         bprm.loader = 0;
1118         bprm.exec = 0;
1119         bprm.security = NULL;
1120         bprm.mm = mm_alloc();
1121         retval = -ENOMEM;
1122         if (!bprm.mm)
1123                 goto out_file;
1124
1125         retval = init_new_context(current, bprm.mm);
1126         if (retval < 0)
1127                 goto out_mm;
1128
1129         bprm.argc = count(argv, bprm.p / sizeof(void *));
1130         if ((retval = bprm.argc) < 0)
1131                 goto out_mm;
1132
1133         bprm.envc = count(envp, bprm.p / sizeof(void *));
1134         if ((retval = bprm.envc) < 0)
1135                 goto out_mm;
1136
1137         retval = security_bprm_alloc(&bprm);
1138         if (retval)
1139                 goto out;
1140
1141         retval = prepare_binprm(&bprm);
1142         if (retval < 0)
1143                 goto out;
1144
1145         retval = copy_strings_kernel(1, &bprm.filename, &bprm);
1146         if (retval < 0)
1147                 goto out;
1148
1149         bprm.exec = bprm.p;
1150         retval = copy_strings(bprm.envc, envp, &bprm);
1151         if (retval < 0)
1152                 goto out;
1153
1154         retval = copy_strings(bprm.argc, argv, &bprm);
1155         if (retval < 0)
1156                 goto out;
1157
1158         retval = search_binary_handler(&bprm,regs);
1159         if (retval >= 0) {
1160                 free_arg_pages(&bprm);
1161
1162                 /* execve success */
1163                 security_bprm_free(&bprm);
1164                 return retval;
1165         }
1166
1167 out:
1168         /* Something went wrong, return the inode and free the argument pages*/
1169         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1170                 struct page * page = bprm.page[i];
1171                 if (page)
1172                         __free_page(page);
1173         }
1174
1175         if (bprm.security)
1176                 security_bprm_free(&bprm);
1177
1178 out_mm:
1179         if (bprm.mm)
1180                 mmdrop(bprm.mm);
1181
1182 out_file:
1183         if (bprm.file) {
1184                 allow_write_access(bprm.file);
1185                 fput(bprm.file);
1186         }
1187         return retval;
1188 }
1189
1190 EXPORT_SYMBOL(do_execve);
1191
1192 int set_binfmt(struct linux_binfmt *new)
1193 {
1194         struct linux_binfmt *old = current->binfmt;
1195
1196         if (new) {
1197                 if (!try_module_get(new->module))
1198                         return -1;
1199         }
1200         current->binfmt = new;
1201         if (old)
1202                 module_put(old->module);
1203         return 0;
1204 }
1205
1206 EXPORT_SYMBOL(set_binfmt);
1207
1208 #define CORENAME_MAX_SIZE 64
1209
1210 /* format_corename will inspect the pattern parameter, and output a
1211  * name into corename, which must have space for at least
1212  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1213  */
1214 void format_corename(char *corename, const char *pattern, long signr)
1215 {
1216         const char *pat_ptr = pattern;
1217         char *out_ptr = corename;
1218         char *const out_end = corename + CORENAME_MAX_SIZE;
1219         int rc;
1220         int pid_in_pattern = 0;
1221
1222         /* Repeat as long as we have more pattern to process and more output
1223            space */
1224         while (*pat_ptr) {
1225                 if (*pat_ptr != '%') {
1226                         if (out_ptr == out_end)
1227                                 goto out;
1228                         *out_ptr++ = *pat_ptr++;
1229                 } else {
1230                         switch (*++pat_ptr) {
1231                         case 0:
1232                                 goto out;
1233                         /* Double percent, output one percent */
1234                         case '%':
1235                                 if (out_ptr == out_end)
1236                                         goto out;
1237                                 *out_ptr++ = '%';
1238                                 break;
1239                         /* pid */
1240                         case 'p':
1241                                 pid_in_pattern = 1;
1242                                 rc = snprintf(out_ptr, out_end - out_ptr,
1243                                               "%d", current->tgid);
1244                                 if (rc > out_end - out_ptr)
1245                                         goto out;
1246                                 out_ptr += rc;
1247                                 break;
1248                         /* uid */
1249                         case 'u':
1250                                 rc = snprintf(out_ptr, out_end - out_ptr,
1251                                               "%d", current->uid);
1252                                 if (rc > out_end - out_ptr)
1253                                         goto out;
1254                                 out_ptr += rc;
1255                                 break;
1256                         /* gid */
1257                         case 'g':
1258                                 rc = snprintf(out_ptr, out_end - out_ptr,
1259                                               "%d", current->gid);
1260                                 if (rc > out_end - out_ptr)
1261                                         goto out;
1262                                 out_ptr += rc;
1263                                 break;
1264                         /* signal that caused the coredump */
1265                         case 's':
1266                                 rc = snprintf(out_ptr, out_end - out_ptr,
1267                                               "%ld", signr);
1268                                 if (rc > out_end - out_ptr)
1269                                         goto out;
1270                                 out_ptr += rc;
1271                                 break;
1272                         /* UNIX time of coredump */
1273                         case 't': {
1274                                 struct timeval tv;
1275                                 do_gettimeofday(&tv);
1276                                 rc = snprintf(out_ptr, out_end - out_ptr,
1277                                               "%lu", tv.tv_sec);
1278                                 if (rc > out_end - out_ptr)
1279                                         goto out;
1280                                 out_ptr += rc;
1281                                 break;
1282                         }
1283                         /* hostname */
1284                         case 'h':
1285                                 down_read(&uts_sem);
1286                                 rc = snprintf(out_ptr, out_end - out_ptr,
1287                                               "%s", system_utsname.nodename);
1288                                 up_read(&uts_sem);
1289                                 if (rc > out_end - out_ptr)
1290                                         goto out;
1291                                 out_ptr += rc;
1292                                 break;
1293                         /* executable */
1294                         case 'e':
1295                                 rc = snprintf(out_ptr, out_end - out_ptr,
1296                                               "%s", current->comm);
1297                                 if (rc > out_end - out_ptr)
1298                                         goto out;
1299                                 out_ptr += rc;
1300                                 break;
1301                         default:
1302                                 break;
1303                         }
1304                         ++pat_ptr;
1305                 }
1306         }
1307         /* Backward compatibility with core_uses_pid:
1308          *
1309          * If core_pattern does not include a %p (as is the default)
1310          * and core_uses_pid is set, then .%pid will be appended to
1311          * the filename */
1312         if (!pid_in_pattern
1313             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1314                 rc = snprintf(out_ptr, out_end - out_ptr,
1315                               ".%d", current->tgid);
1316                 if (rc > out_end - out_ptr)
1317                         goto out;
1318                 out_ptr += rc;
1319         }
1320       out:
1321         *out_ptr = 0;
1322 }
1323
1324 static void zap_threads (struct mm_struct *mm)
1325 {
1326         struct task_struct *g, *p;
1327         struct task_struct *tsk = current;
1328         struct completion *vfork_done = tsk->vfork_done;
1329
1330         /*
1331          * Make sure nobody is waiting for us to release the VM,
1332          * otherwise we can deadlock when we wait on each other
1333          */
1334         if (vfork_done) {
1335                 tsk->vfork_done = NULL;
1336                 complete(vfork_done);
1337         }
1338
1339         read_lock(&tasklist_lock);
1340         do_each_thread(g,p)
1341                 if (mm == p->mm && p != tsk) {
1342                         force_sig_specific(SIGKILL, p);
1343                         mm->core_waiters++;
1344                 }
1345         while_each_thread(g,p);
1346
1347         read_unlock(&tasklist_lock);
1348 }
1349
1350 static void coredump_wait(struct mm_struct *mm)
1351 {
1352         DECLARE_COMPLETION(startup_done);
1353
1354         mm->core_waiters++; /* let other threads block */
1355         mm->core_startup_done = &startup_done;
1356
1357         /* give other threads a chance to run: */
1358         yield();
1359
1360         zap_threads(mm);
1361         if (--mm->core_waiters) {
1362                 up_write(&mm->mmap_sem);
1363                 wait_for_completion(&startup_done);
1364         } else
1365                 up_write(&mm->mmap_sem);
1366         BUG_ON(mm->core_waiters);
1367 }
1368
1369 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1370 {
1371         char corename[CORENAME_MAX_SIZE + 1];
1372         struct mm_struct *mm = current->mm;
1373         struct linux_binfmt * binfmt;
1374         struct inode * inode;
1375         struct file * file;
1376         int retval = 0;
1377
1378         lock_kernel();
1379         binfmt = current->binfmt;
1380         if (!binfmt || !binfmt->core_dump)
1381                 goto fail;
1382         down_write(&mm->mmap_sem);
1383         if (!mm->dumpable) {
1384                 up_write(&mm->mmap_sem);
1385                 goto fail;
1386         }
1387         mm->dumpable = 0;
1388         init_completion(&mm->core_done);
1389         current->signal->group_exit = 1;
1390         current->signal->group_exit_code = exit_code;
1391         coredump_wait(mm);
1392
1393         if (current->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1394                 goto fail_unlock;
1395
1396         format_corename(corename, core_pattern, signr);
1397         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
1398         if (IS_ERR(file))
1399                 goto fail_unlock;
1400         inode = file->f_dentry->d_inode;
1401         if (inode->i_nlink > 1)
1402                 goto close_fail;        /* multiple links - don't dump */
1403         if (d_unhashed(file->f_dentry))
1404                 goto close_fail;
1405
1406         if (!S_ISREG(inode->i_mode))
1407                 goto close_fail;
1408         if (!file->f_op)
1409                 goto close_fail;
1410         if (!file->f_op->write)
1411                 goto close_fail;
1412         if (do_truncate(file->f_dentry, 0) != 0)
1413                 goto close_fail;
1414
1415         retval = binfmt->core_dump(signr, regs, file);
1416
1417         current->signal->group_exit_code |= 0x80;
1418 close_fail:
1419         filp_close(file, NULL);
1420 fail_unlock:
1421         complete_all(&mm->core_done);
1422 fail:
1423         unlock_kernel();
1424         return retval;
1425 }