blob: c466fec5de2005cee77acfc7a2f120afefcd33f8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Matt Helsley9f460802005-11-07 00:59:16 -080051#include <linux/cn_proc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53#include <asm/uaccess.h>
54#include <asm/mmu_context.h>
55
56#ifdef CONFIG_KMOD
57#include <linux/kmod.h>
58#endif
59
60int core_uses_pid;
61char core_pattern[65] = "core";
Alan Coxd6e71142005-06-23 00:09:43 -070062int suid_dumpable = 0;
63
64EXPORT_SYMBOL(suid_dumpable);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/* The maximal length of core_pattern is also specified in sysctl.c */
66
67static struct linux_binfmt *formats;
68static DEFINE_RWLOCK(binfmt_lock);
69
70int register_binfmt(struct linux_binfmt * fmt)
71{
72 struct linux_binfmt ** tmp = &formats;
73
74 if (!fmt)
75 return -EINVAL;
76 if (fmt->next)
77 return -EBUSY;
78 write_lock(&binfmt_lock);
79 while (*tmp) {
80 if (fmt == *tmp) {
81 write_unlock(&binfmt_lock);
82 return -EBUSY;
83 }
84 tmp = &(*tmp)->next;
85 }
86 fmt->next = formats;
87 formats = fmt;
88 write_unlock(&binfmt_lock);
89 return 0;
90}
91
92EXPORT_SYMBOL(register_binfmt);
93
94int unregister_binfmt(struct linux_binfmt * fmt)
95{
96 struct linux_binfmt ** tmp = &formats;
97
98 write_lock(&binfmt_lock);
99 while (*tmp) {
100 if (fmt == *tmp) {
101 *tmp = fmt->next;
102 write_unlock(&binfmt_lock);
103 return 0;
104 }
105 tmp = &(*tmp)->next;
106 }
107 write_unlock(&binfmt_lock);
108 return -EINVAL;
109}
110
111EXPORT_SYMBOL(unregister_binfmt);
112
113static inline void put_binfmt(struct linux_binfmt * fmt)
114{
115 module_put(fmt->module);
116}
117
118/*
119 * Note that a shared library must be both readable and executable due to
120 * security reasons.
121 *
122 * Also note that we take the address to load from from the file itself.
123 */
124asmlinkage long sys_uselib(const char __user * library)
125{
126 struct file * file;
127 struct nameidata nd;
128 int error;
129
Trond Myklebust834f2a42005-10-18 14:20:16 -0700130 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 if (error)
132 goto out;
133
134 error = -EINVAL;
135 if (!S_ISREG(nd.dentry->d_inode->i_mode))
136 goto exit;
137
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800138 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 if (error)
140 goto exit;
141
Trond Myklebust834f2a42005-10-18 14:20:16 -0700142 file = nameidata_to_filp(&nd, O_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 error = PTR_ERR(file);
144 if (IS_ERR(file))
145 goto out;
146
147 error = -ENOEXEC;
148 if(file->f_op) {
149 struct linux_binfmt * fmt;
150
151 read_lock(&binfmt_lock);
152 for (fmt = formats ; fmt ; fmt = fmt->next) {
153 if (!fmt->load_shlib)
154 continue;
155 if (!try_module_get(fmt->module))
156 continue;
157 read_unlock(&binfmt_lock);
158 error = fmt->load_shlib(file);
159 read_lock(&binfmt_lock);
160 put_binfmt(fmt);
161 if (error != -ENOEXEC)
162 break;
163 }
164 read_unlock(&binfmt_lock);
165 }
166 fput(file);
167out:
168 return error;
169exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -0700170 release_open_intent(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 path_release(&nd);
172 goto out;
173}
174
175/*
176 * count() counts the number of strings in array ARGV.
177 */
178static int count(char __user * __user * argv, int max)
179{
180 int i = 0;
181
182 if (argv != NULL) {
183 for (;;) {
184 char __user * p;
185
186 if (get_user(p, argv))
187 return -EFAULT;
188 if (!p)
189 break;
190 argv++;
191 if(++i > max)
192 return -E2BIG;
193 cond_resched();
194 }
195 }
196 return i;
197}
198
199/*
200 * 'copy_strings()' copies argument/environment strings from user
201 * memory to free pages in kernel mem. These are in a format ready
202 * to be put directly into the top of new user memory.
203 */
Adrian Bunk75c96f82005-05-05 16:16:09 -0700204static int copy_strings(int argc, char __user * __user * argv,
205 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206{
207 struct page *kmapped_page = NULL;
208 char *kaddr = NULL;
209 int ret;
210
211 while (argc-- > 0) {
212 char __user *str;
213 int len;
214 unsigned long pos;
215
216 if (get_user(str, argv+argc) ||
217 !(len = strnlen_user(str, bprm->p))) {
218 ret = -EFAULT;
219 goto out;
220 }
221
222 if (bprm->p < len) {
223 ret = -E2BIG;
224 goto out;
225 }
226
227 bprm->p -= len;
228 /* XXX: add architecture specific overflow check here. */
229 pos = bprm->p;
230
231 while (len > 0) {
232 int i, new, err;
233 int offset, bytes_to_copy;
234 struct page *page;
235
236 offset = pos % PAGE_SIZE;
237 i = pos/PAGE_SIZE;
238 page = bprm->page[i];
239 new = 0;
240 if (!page) {
241 page = alloc_page(GFP_HIGHUSER);
242 bprm->page[i] = page;
243 if (!page) {
244 ret = -ENOMEM;
245 goto out;
246 }
247 new = 1;
248 }
249
250 if (page != kmapped_page) {
251 if (kmapped_page)
252 kunmap(kmapped_page);
253 kmapped_page = page;
254 kaddr = kmap(kmapped_page);
255 }
256 if (new && offset)
257 memset(kaddr, 0, offset);
258 bytes_to_copy = PAGE_SIZE - offset;
259 if (bytes_to_copy > len) {
260 bytes_to_copy = len;
261 if (new)
262 memset(kaddr+offset+len, 0,
263 PAGE_SIZE-offset-len);
264 }
265 err = copy_from_user(kaddr+offset, str, bytes_to_copy);
266 if (err) {
267 ret = -EFAULT;
268 goto out;
269 }
270
271 pos += bytes_to_copy;
272 str += bytes_to_copy;
273 len -= bytes_to_copy;
274 }
275 }
276 ret = 0;
277out:
278 if (kmapped_page)
279 kunmap(kmapped_page);
280 return ret;
281}
282
283/*
284 * Like copy_strings, but get argv and its values from kernel memory.
285 */
286int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
287{
288 int r;
289 mm_segment_t oldfs = get_fs();
290 set_fs(KERNEL_DS);
291 r = copy_strings(argc, (char __user * __user *)argv, bprm);
292 set_fs(oldfs);
293 return r;
294}
295
296EXPORT_SYMBOL(copy_strings_kernel);
297
298#ifdef CONFIG_MMU
299/*
300 * This routine is used to map in a page into an address space: needed by
301 * execve() for the initial stack and environment pages.
302 *
303 * vma->vm_mm->mmap_sem is held for writing.
304 */
305void install_arg_page(struct vm_area_struct *vma,
306 struct page *page, unsigned long address)
307{
308 struct mm_struct *mm = vma->vm_mm;
309 pgd_t * pgd;
310 pud_t * pud;
311 pmd_t * pmd;
312 pte_t * pte;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700313 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
315 if (unlikely(anon_vma_prepare(vma)))
Hugh Dickinsc74df322005-10-29 18:16:23 -0700316 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
318 flush_dcache_page(page);
319 pgd = pgd_offset(mm, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 pud = pud_alloc(mm, pgd, address);
321 if (!pud)
322 goto out;
323 pmd = pmd_alloc(mm, pud, address);
324 if (!pmd)
325 goto out;
Hugh Dickinsc74df322005-10-29 18:16:23 -0700326 pte = pte_alloc_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 if (!pte)
328 goto out;
329 if (!pte_none(*pte)) {
Hugh Dickinsc74df322005-10-29 18:16:23 -0700330 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 goto out;
332 }
Hugh Dickins42946212005-10-29 18:16:05 -0700333 inc_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 lru_cache_add_active(page);
335 set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
336 page, vma->vm_page_prot))));
337 page_add_anon_rmap(page, vma, address);
Hugh Dickinsc74df322005-10-29 18:16:23 -0700338 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
340 /* no need for flush_tlb */
341 return;
342out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 __free_page(page);
344 force_sig(SIGKILL, current);
345}
346
347#define EXTRA_STACK_VM_PAGES 20 /* random */
348
349int setup_arg_pages(struct linux_binprm *bprm,
350 unsigned long stack_top,
351 int executable_stack)
352{
353 unsigned long stack_base;
354 struct vm_area_struct *mpnt;
355 struct mm_struct *mm = current->mm;
356 int i, ret;
357 long arg_size;
358
359#ifdef CONFIG_STACK_GROWSUP
360 /* Move the argument and environment strings to the bottom of the
361 * stack space.
362 */
363 int offset, j;
364 char *to, *from;
365
366 /* Start by shifting all the pages down */
367 i = 0;
368 for (j = 0; j < MAX_ARG_PAGES; j++) {
369 struct page *page = bprm->page[j];
370 if (!page)
371 continue;
372 bprm->page[i++] = page;
373 }
374
375 /* Now move them within their pages */
376 offset = bprm->p % PAGE_SIZE;
377 to = kmap(bprm->page[0]);
378 for (j = 1; j < i; j++) {
379 memmove(to, to + offset, PAGE_SIZE - offset);
380 from = kmap(bprm->page[j]);
381 memcpy(to + PAGE_SIZE - offset, from, offset);
382 kunmap(bprm->page[j - 1]);
383 to = from;
384 }
385 memmove(to, to + offset, PAGE_SIZE - offset);
386 kunmap(bprm->page[j - 1]);
387
388 /* Limit stack size to 1GB */
389 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
390 if (stack_base > (1 << 30))
391 stack_base = 1 << 30;
392 stack_base = PAGE_ALIGN(stack_top - stack_base);
393
394 /* Adjust bprm->p to point to the end of the strings. */
395 bprm->p = stack_base + PAGE_SIZE * i - offset;
396
397 mm->arg_start = stack_base;
398 arg_size = i << PAGE_SHIFT;
399
400 /* zero pages that were copied above */
401 while (i < MAX_ARG_PAGES)
402 bprm->page[i++] = NULL;
403#else
404 stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
405 stack_base = PAGE_ALIGN(stack_base);
406 bprm->p += stack_base;
407 mm->arg_start = bprm->p;
408 arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
409#endif
410
411 arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
412
413 if (bprm->loader)
414 bprm->loader += stack_base;
415 bprm->exec += stack_base;
416
417 mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
418 if (!mpnt)
419 return -ENOMEM;
420
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 memset(mpnt, 0, sizeof(*mpnt));
422
423 down_write(&mm->mmap_sem);
424 {
425 mpnt->vm_mm = mm;
426#ifdef CONFIG_STACK_GROWSUP
427 mpnt->vm_start = stack_base;
428 mpnt->vm_end = stack_base + arg_size;
429#else
430 mpnt->vm_end = stack_top;
431 mpnt->vm_start = mpnt->vm_end - arg_size;
432#endif
433 /* Adjust stack execute permissions; explicitly enable
434 * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
435 * and leave alone (arch default) otherwise. */
436 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
437 mpnt->vm_flags = VM_STACK_FLAGS | VM_EXEC;
438 else if (executable_stack == EXSTACK_DISABLE_X)
439 mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
440 else
441 mpnt->vm_flags = VM_STACK_FLAGS;
442 mpnt->vm_flags |= mm->def_flags;
443 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
444 if ((ret = insert_vm_struct(mm, mpnt))) {
445 up_write(&mm->mmap_sem);
446 kmem_cache_free(vm_area_cachep, mpnt);
447 return ret;
448 }
449 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
450 }
451
452 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
453 struct page *page = bprm->page[i];
454 if (page) {
455 bprm->page[i] = NULL;
456 install_arg_page(mpnt, page, stack_base);
457 }
458 stack_base += PAGE_SIZE;
459 }
460 up_write(&mm->mmap_sem);
461
462 return 0;
463}
464
465EXPORT_SYMBOL(setup_arg_pages);
466
467#define free_arg_pages(bprm) do { } while (0)
468
469#else
470
471static inline void free_arg_pages(struct linux_binprm *bprm)
472{
473 int i;
474
475 for (i = 0; i < MAX_ARG_PAGES; i++) {
476 if (bprm->page[i])
477 __free_page(bprm->page[i]);
478 bprm->page[i] = NULL;
479 }
480}
481
482#endif /* CONFIG_MMU */
483
484struct file *open_exec(const char *name)
485{
486 struct nameidata nd;
487 int err;
488 struct file *file;
489
Trond Myklebust834f2a42005-10-18 14:20:16 -0700490 err = path_lookup_open(name, LOOKUP_FOLLOW, &nd, FMODE_READ);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 file = ERR_PTR(err);
492
493 if (!err) {
494 struct inode *inode = nd.dentry->d_inode;
495 file = ERR_PTR(-EACCES);
496 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
497 S_ISREG(inode->i_mode)) {
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800498 int err = vfs_permission(&nd, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 if (!err && !(inode->i_mode & 0111))
500 err = -EACCES;
501 file = ERR_PTR(err);
502 if (!err) {
Trond Myklebust834f2a42005-10-18 14:20:16 -0700503 file = nameidata_to_filp(&nd, O_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 if (!IS_ERR(file)) {
505 err = deny_write_access(file);
506 if (err) {
507 fput(file);
508 file = ERR_PTR(err);
509 }
510 }
511out:
512 return file;
513 }
514 }
Trond Myklebust834f2a42005-10-18 14:20:16 -0700515 release_open_intent(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 path_release(&nd);
517 }
518 goto out;
519}
520
521EXPORT_SYMBOL(open_exec);
522
523int kernel_read(struct file *file, unsigned long offset,
524 char *addr, unsigned long count)
525{
526 mm_segment_t old_fs;
527 loff_t pos = offset;
528 int result;
529
530 old_fs = get_fs();
531 set_fs(get_ds());
532 /* The cast to a user pointer is valid due to the set_fs() */
533 result = vfs_read(file, (void __user *)addr, count, &pos);
534 set_fs(old_fs);
535 return result;
536}
537
538EXPORT_SYMBOL(kernel_read);
539
540static int exec_mmap(struct mm_struct *mm)
541{
542 struct task_struct *tsk;
543 struct mm_struct * old_mm, *active_mm;
544
545 /* Notify parent that we're no longer interested in the old VM */
546 tsk = current;
547 old_mm = current->mm;
548 mm_release(tsk, old_mm);
549
550 if (old_mm) {
551 /*
552 * Make sure that if there is a core dump in progress
553 * for the old mm, we get out and die instead of going
554 * through with the exec. We must hold mmap_sem around
555 * checking core_waiters and changing tsk->mm. The
556 * core-inducing thread will increment core_waiters for
557 * each thread whose ->mm == old_mm.
558 */
559 down_read(&old_mm->mmap_sem);
560 if (unlikely(old_mm->core_waiters)) {
561 up_read(&old_mm->mmap_sem);
562 return -EINTR;
563 }
564 }
565 task_lock(tsk);
566 active_mm = tsk->active_mm;
567 tsk->mm = mm;
568 tsk->active_mm = mm;
569 activate_mm(active_mm, mm);
570 task_unlock(tsk);
571 arch_pick_mmap_layout(mm);
572 if (old_mm) {
573 up_read(&old_mm->mmap_sem);
574 if (active_mm != old_mm) BUG();
575 mmput(old_mm);
576 return 0;
577 }
578 mmdrop(active_mm);
579 return 0;
580}
581
582/*
583 * This function makes sure the current process has its own signal table,
584 * so that flush_signal_handlers can later reset the handlers without
585 * disturbing other processes. (Other processes might share the signal
586 * table via the CLONE_SIGHAND option to clone().)
587 */
588static inline int de_thread(struct task_struct *tsk)
589{
590 struct signal_struct *sig = tsk->signal;
591 struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
592 spinlock_t *lock = &oldsighand->siglock;
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300593 struct task_struct *leader = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 int count;
595
596 /*
597 * If we don't share sighandlers, then we aren't sharing anything
598 * and we can just re-use it all.
599 */
600 if (atomic_read(&oldsighand->count) <= 1) {
601 BUG_ON(atomic_read(&sig->count) != 1);
602 exit_itimers(sig);
603 return 0;
604 }
605
606 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
607 if (!newsighand)
608 return -ENOMEM;
609
610 if (thread_group_empty(current))
611 goto no_thread_group;
612
613 /*
614 * Kill all other threads in the thread group.
615 * We must hold tasklist_lock to call zap_other_threads.
616 */
617 read_lock(&tasklist_lock);
618 spin_lock_irq(lock);
619 if (sig->flags & SIGNAL_GROUP_EXIT) {
620 /*
621 * Another group action in progress, just
622 * return so that the signal is processed.
623 */
624 spin_unlock_irq(lock);
625 read_unlock(&tasklist_lock);
626 kmem_cache_free(sighand_cachep, newsighand);
627 return -EAGAIN;
628 }
629 zap_other_threads(current);
630 read_unlock(&tasklist_lock);
631
632 /*
633 * Account for the thread group leader hanging around:
634 */
Oleg Nesterov9e4e23b2005-10-30 15:01:37 -0800635 count = 1;
636 if (!thread_group_leader(current)) {
637 count = 2;
Roland McGrath53231252005-07-12 13:58:27 -0700638 /*
639 * The SIGALRM timer survives the exec, but needs to point
640 * at us as the new group leader now. We have a race with
641 * a timer firing now getting the old leader, so we need to
642 * synchronize with any firing (by calling del_timer_sync)
643 * before we can safely let the old group leader die.
644 */
645 sig->real_timer.data = (unsigned long)current;
Oleg Nesterov932aeaf2005-10-30 15:02:17 -0800646 spin_unlock_irq(lock);
Roland McGrath53231252005-07-12 13:58:27 -0700647 if (del_timer_sync(&sig->real_timer))
648 add_timer(&sig->real_timer);
Oleg Nesterov932aeaf2005-10-30 15:02:17 -0800649 spin_lock_irq(lock);
Roland McGrath53231252005-07-12 13:58:27 -0700650 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 while (atomic_read(&sig->count) > count) {
652 sig->group_exit_task = current;
653 sig->notify_count = count;
654 __set_current_state(TASK_UNINTERRUPTIBLE);
655 spin_unlock_irq(lock);
656 schedule();
657 spin_lock_irq(lock);
658 }
659 sig->group_exit_task = NULL;
660 sig->notify_count = 0;
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 (!thread_group_leader(current)) {
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300669 struct task_struct *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 struct dentry *proc_dentry1, *proc_dentry2;
671 unsigned long exit_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 */
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300678 leader = current->group_leader;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 while (leader->exit_state != EXIT_ZOMBIE)
680 yield();
681
682 spin_lock(&leader->proc_lock);
683 spin_lock(&current->proc_lock);
684 proc_dentry1 = proc_pid_unhash(current);
685 proc_dentry2 = proc_pid_unhash(leader);
686 write_lock_irq(&tasklist_lock);
687
Linus Torvaldsc2a0f592005-06-18 13:06:22 -0700688 BUG_ON(leader->tgid != current->tgid);
689 BUG_ON(current->pid == current->tgid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 /*
691 * An exec() starts a new thread group with the
692 * TGID of the previous thread group. Rehash the
693 * two threads with a switched PID, and release
694 * the former thread group leader:
695 */
696 ptrace = leader->ptrace;
697 parent = leader->parent;
698 if (unlikely(ptrace) && unlikely(parent == current)) {
699 /*
700 * Joker was ptracing his own group leader,
701 * and now he wants to be his own parent!
702 * We can't have that.
703 */
704 ptrace = 0;
705 }
706
707 ptrace_unlink(current);
708 ptrace_unlink(leader);
709 remove_parent(current);
710 remove_parent(leader);
711
712 switch_exec_pids(leader, current);
713
714 current->parent = current->real_parent = leader->real_parent;
715 leader->parent = leader->real_parent = child_reaper;
716 current->group_leader = current;
717 leader->group_leader = leader;
718
719 add_parent(current, current->parent);
720 add_parent(leader, leader->parent);
721 if (ptrace) {
722 current->ptrace = ptrace;
723 __ptrace_link(current, parent);
724 }
725
726 list_del(&current->tasks);
727 list_add_tail(&current->tasks, &init_task.tasks);
728 current->exit_signal = SIGCHLD;
729 exit_state = leader->exit_state;
730
731 write_unlock_irq(&tasklist_lock);
732 spin_unlock(&leader->proc_lock);
733 spin_unlock(&current->proc_lock);
734 proc_pid_flush(proc_dentry1);
735 proc_pid_flush(proc_dentry2);
736
Linus Torvaldsc2a0f592005-06-18 13:06:22 -0700737 BUG_ON(exit_state != EXIT_ZOMBIE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 }
739
740 /*
Alexander Nybergfb085cf2005-09-14 18:54:06 +0200741 * There may be one thread left which is just exiting,
742 * but it's safe to stop telling the group to kill themselves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 */
744 sig->flags = 0;
745
746no_thread_group:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 exit_itimers(sig);
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300748 if (leader)
749 release_task(leader);
750
751 BUG_ON(atomic_read(&sig->count) != 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
753 if (atomic_read(&oldsighand->count) == 1) {
754 /*
755 * Now that we nuked the rest of the thread group,
756 * it turns out we are not sharing sighand any more either.
757 * So we can just keep it.
758 */
759 kmem_cache_free(sighand_cachep, newsighand);
760 } else {
761 /*
762 * Move our state over to newsighand and switch it in.
763 */
764 spin_lock_init(&newsighand->siglock);
765 atomic_set(&newsighand->count, 1);
766 memcpy(newsighand->action, oldsighand->action,
767 sizeof(newsighand->action));
768
769 write_lock_irq(&tasklist_lock);
770 spin_lock(&oldsighand->siglock);
771 spin_lock(&newsighand->siglock);
772
773 current->sighand = newsighand;
774 recalc_sigpending();
775
776 spin_unlock(&newsighand->siglock);
777 spin_unlock(&oldsighand->siglock);
778 write_unlock_irq(&tasklist_lock);
779
780 if (atomic_dec_and_test(&oldsighand->count))
781 kmem_cache_free(sighand_cachep, oldsighand);
782 }
783
Linus Torvaldsc2a0f592005-06-18 13:06:22 -0700784 BUG_ON(!thread_group_leader(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 return 0;
786}
787
788/*
789 * These functions flushes out all traces of the currently running executable
790 * so that a new one can be started
791 */
792
793static inline void flush_old_files(struct files_struct * files)
794{
795 long j = -1;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700796 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
798 spin_lock(&files->file_lock);
799 for (;;) {
800 unsigned long set, i;
801
802 j++;
803 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700804 fdt = files_fdtable(files);
805 if (i >= fdt->max_fds || i >= fdt->max_fdset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700807 set = fdt->close_on_exec->fds_bits[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 if (!set)
809 continue;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700810 fdt->close_on_exec->fds_bits[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 spin_unlock(&files->file_lock);
812 for ( ; set ; i++,set >>= 1) {
813 if (set & 1) {
814 sys_close(i);
815 }
816 }
817 spin_lock(&files->file_lock);
818
819 }
820 spin_unlock(&files->file_lock);
821}
822
823void get_task_comm(char *buf, struct task_struct *tsk)
824{
825 /* buf must be at least sizeof(tsk->comm) in size */
826 task_lock(tsk);
827 strncpy(buf, tsk->comm, sizeof(tsk->comm));
828 task_unlock(tsk);
829}
830
831void set_task_comm(struct task_struct *tsk, char *buf)
832{
833 task_lock(tsk);
834 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
835 task_unlock(tsk);
836}
837
838int flush_old_exec(struct linux_binprm * bprm)
839{
840 char * name;
841 int i, ch, retval;
842 struct files_struct *files;
843 char tcomm[sizeof(current->comm)];
844
845 /*
846 * Make sure we have a private signal table and that
847 * we are unassociated from the previous thread group.
848 */
849 retval = de_thread(current);
850 if (retval)
851 goto out;
852
853 /*
854 * Make sure we have private file handles. Ask the
855 * fork helper to do the work for us and the exit
856 * helper to do the cleanup of the old one.
857 */
858 files = current->files; /* refcounted so safe to hold */
859 retval = unshare_files();
860 if (retval)
861 goto out;
862 /*
863 * Release all of the old mmap stuff
864 */
865 retval = exec_mmap(bprm->mm);
866 if (retval)
867 goto mmap_failed;
868
869 bprm->mm = NULL; /* We're using it now */
870
871 /* This is the point of no return */
872 steal_locks(files);
873 put_files_struct(files);
874
875 current->sas_ss_sp = current->sas_ss_size = 0;
876
877 if (current->euid == current->uid && current->egid == current->gid)
878 current->mm->dumpable = 1;
Alan Coxd6e71142005-06-23 00:09:43 -0700879 else
880 current->mm->dumpable = suid_dumpable;
881
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 name = bprm->filename;
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -0700883
884 /* Copies the binary name from after last slash */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 for (i=0; (ch = *(name++)) != '\0';) {
886 if (ch == '/')
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -0700887 i = 0; /* overwrite what we wrote */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 else
889 if (i < (sizeof(tcomm) - 1))
890 tcomm[i++] = ch;
891 }
892 tcomm[i] = '\0';
893 set_task_comm(current, tcomm);
894
895 current->flags &= ~PF_RANDOMIZE;
896 flush_thread();
897
898 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800899 file_permission(bprm->file, MAY_READ) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
901 suid_keys(current);
Alan Coxd6e71142005-06-23 00:09:43 -0700902 current->mm->dumpable = suid_dumpable;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 }
904
905 /* An exec changes our domain. We are no longer part of the thread
906 group */
907
908 current->self_exec_id++;
909
910 flush_signal_handlers(current, 0);
911 flush_old_files(current->files);
912
913 return 0;
914
915mmap_failed:
916 put_files_struct(current->files);
917 current->files = files;
918out:
919 return retval;
920}
921
922EXPORT_SYMBOL(flush_old_exec);
923
924/*
925 * Fill the binprm structure from the inode.
926 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
927 */
928int prepare_binprm(struct linux_binprm *bprm)
929{
930 int mode;
931 struct inode * inode = bprm->file->f_dentry->d_inode;
932 int retval;
933
934 mode = inode->i_mode;
935 /*
936 * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
937 * generic_permission lets a non-executable through
938 */
939 if (!(mode & 0111)) /* with at least _one_ execute bit set */
940 return -EACCES;
941 if (bprm->file->f_op == NULL)
942 return -EACCES;
943
944 bprm->e_uid = current->euid;
945 bprm->e_gid = current->egid;
946
947 if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
948 /* Set-uid? */
949 if (mode & S_ISUID) {
950 current->personality &= ~PER_CLEAR_ON_SETID;
951 bprm->e_uid = inode->i_uid;
952 }
953
954 /* Set-gid? */
955 /*
956 * If setgid is set but no group execute bit then this
957 * is a candidate for mandatory locking, not a setgid
958 * executable.
959 */
960 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
961 current->personality &= ~PER_CLEAR_ON_SETID;
962 bprm->e_gid = inode->i_gid;
963 }
964 }
965
966 /* fill in binprm security blob */
967 retval = security_bprm_set(bprm);
968 if (retval)
969 return retval;
970
971 memset(bprm->buf,0,BINPRM_BUF_SIZE);
972 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
973}
974
975EXPORT_SYMBOL(prepare_binprm);
976
977static inline int unsafe_exec(struct task_struct *p)
978{
979 int unsafe = 0;
980 if (p->ptrace & PT_PTRACED) {
981 if (p->ptrace & PT_PTRACE_CAP)
982 unsafe |= LSM_UNSAFE_PTRACE_CAP;
983 else
984 unsafe |= LSM_UNSAFE_PTRACE;
985 }
986 if (atomic_read(&p->fs->count) > 1 ||
987 atomic_read(&p->files->count) > 1 ||
988 atomic_read(&p->sighand->count) > 1)
989 unsafe |= LSM_UNSAFE_SHARE;
990
991 return unsafe;
992}
993
994void compute_creds(struct linux_binprm *bprm)
995{
996 int unsafe;
997
998 if (bprm->e_uid != current->uid)
999 suid_keys(current);
1000 exec_keys(current);
1001
1002 task_lock(current);
1003 unsafe = unsafe_exec(current);
1004 security_bprm_apply_creds(bprm, unsafe);
1005 task_unlock(current);
1006 security_bprm_post_apply_creds(bprm);
1007}
1008
1009EXPORT_SYMBOL(compute_creds);
1010
1011void remove_arg_zero(struct linux_binprm *bprm)
1012{
1013 if (bprm->argc) {
1014 unsigned long offset;
1015 char * kaddr;
1016 struct page *page;
1017
1018 offset = bprm->p % PAGE_SIZE;
1019 goto inside;
1020
1021 while (bprm->p++, *(kaddr+offset++)) {
1022 if (offset != PAGE_SIZE)
1023 continue;
1024 offset = 0;
1025 kunmap_atomic(kaddr, KM_USER0);
1026inside:
1027 page = bprm->page[bprm->p/PAGE_SIZE];
1028 kaddr = kmap_atomic(page, KM_USER0);
1029 }
1030 kunmap_atomic(kaddr, KM_USER0);
1031 bprm->argc--;
1032 }
1033}
1034
1035EXPORT_SYMBOL(remove_arg_zero);
1036
1037/*
1038 * cycle the list of binary formats handler, until one recognizes the image
1039 */
1040int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1041{
1042 int try,retval;
1043 struct linux_binfmt *fmt;
1044#ifdef __alpha__
1045 /* handle /sbin/loader.. */
1046 {
1047 struct exec * eh = (struct exec *) bprm->buf;
1048
1049 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1050 (eh->fh.f_flags & 0x3000) == 0x3000)
1051 {
1052 struct file * file;
1053 unsigned long loader;
1054
1055 allow_write_access(bprm->file);
1056 fput(bprm->file);
1057 bprm->file = NULL;
1058
1059 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1060
1061 file = open_exec("/sbin/loader");
1062 retval = PTR_ERR(file);
1063 if (IS_ERR(file))
1064 return retval;
1065
1066 /* Remember if the application is TASO. */
1067 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1068
1069 bprm->file = file;
1070 bprm->loader = loader;
1071 retval = prepare_binprm(bprm);
1072 if (retval<0)
1073 return retval;
1074 /* should call search_binary_handler recursively here,
1075 but it does not matter */
1076 }
1077 }
1078#endif
1079 retval = security_bprm_check(bprm);
1080 if (retval)
1081 return retval;
1082
1083 /* kernel module loader fixup */
1084 /* so we don't try to load run modprobe in kernel space. */
1085 set_fs(USER_DS);
1086 retval = -ENOENT;
1087 for (try=0; try<2; try++) {
1088 read_lock(&binfmt_lock);
1089 for (fmt = formats ; fmt ; fmt = fmt->next) {
1090 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1091 if (!fn)
1092 continue;
1093 if (!try_module_get(fmt->module))
1094 continue;
1095 read_unlock(&binfmt_lock);
1096 retval = fn(bprm, regs);
1097 if (retval >= 0) {
1098 put_binfmt(fmt);
1099 allow_write_access(bprm->file);
1100 if (bprm->file)
1101 fput(bprm->file);
1102 bprm->file = NULL;
1103 current->did_exec = 1;
Matt Helsley9f460802005-11-07 00:59:16 -08001104 proc_exec_connector(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 return retval;
1106 }
1107 read_lock(&binfmt_lock);
1108 put_binfmt(fmt);
1109 if (retval != -ENOEXEC || bprm->mm == NULL)
1110 break;
1111 if (!bprm->file) {
1112 read_unlock(&binfmt_lock);
1113 return retval;
1114 }
1115 }
1116 read_unlock(&binfmt_lock);
1117 if (retval != -ENOEXEC || bprm->mm == NULL) {
1118 break;
1119#ifdef CONFIG_KMOD
1120 }else{
1121#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1122 if (printable(bprm->buf[0]) &&
1123 printable(bprm->buf[1]) &&
1124 printable(bprm->buf[2]) &&
1125 printable(bprm->buf[3]))
1126 break; /* -ENOEXEC */
1127 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1128#endif
1129 }
1130 }
1131 return retval;
1132}
1133
1134EXPORT_SYMBOL(search_binary_handler);
1135
1136/*
1137 * sys_execve() executes a new program.
1138 */
1139int do_execve(char * filename,
1140 char __user *__user *argv,
1141 char __user *__user *envp,
1142 struct pt_regs * regs)
1143{
1144 struct linux_binprm *bprm;
1145 struct file *file;
1146 int retval;
1147 int i;
1148
1149 retval = -ENOMEM;
1150 bprm = kmalloc(sizeof(*bprm), GFP_KERNEL);
1151 if (!bprm)
1152 goto out_ret;
1153 memset(bprm, 0, sizeof(*bprm));
1154
1155 file = open_exec(filename);
1156 retval = PTR_ERR(file);
1157 if (IS_ERR(file))
1158 goto out_kfree;
1159
1160 sched_exec();
1161
1162 bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1163
1164 bprm->file = file;
1165 bprm->filename = filename;
1166 bprm->interp = filename;
1167 bprm->mm = mm_alloc();
1168 retval = -ENOMEM;
1169 if (!bprm->mm)
1170 goto out_file;
1171
1172 retval = init_new_context(current, bprm->mm);
1173 if (retval < 0)
1174 goto out_mm;
1175
1176 bprm->argc = count(argv, bprm->p / sizeof(void *));
1177 if ((retval = bprm->argc) < 0)
1178 goto out_mm;
1179
1180 bprm->envc = count(envp, bprm->p / sizeof(void *));
1181 if ((retval = bprm->envc) < 0)
1182 goto out_mm;
1183
1184 retval = security_bprm_alloc(bprm);
1185 if (retval)
1186 goto out;
1187
1188 retval = prepare_binprm(bprm);
1189 if (retval < 0)
1190 goto out;
1191
1192 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1193 if (retval < 0)
1194 goto out;
1195
1196 bprm->exec = bprm->p;
1197 retval = copy_strings(bprm->envc, envp, bprm);
1198 if (retval < 0)
1199 goto out;
1200
1201 retval = copy_strings(bprm->argc, argv, bprm);
1202 if (retval < 0)
1203 goto out;
1204
1205 retval = search_binary_handler(bprm,regs);
1206 if (retval >= 0) {
1207 free_arg_pages(bprm);
1208
1209 /* execve success */
1210 security_bprm_free(bprm);
1211 acct_update_integrals(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 kfree(bprm);
1213 return retval;
1214 }
1215
1216out:
1217 /* Something went wrong, return the inode and free the argument pages*/
1218 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1219 struct page * page = bprm->page[i];
1220 if (page)
1221 __free_page(page);
1222 }
1223
1224 if (bprm->security)
1225 security_bprm_free(bprm);
1226
1227out_mm:
1228 if (bprm->mm)
1229 mmdrop(bprm->mm);
1230
1231out_file:
1232 if (bprm->file) {
1233 allow_write_access(bprm->file);
1234 fput(bprm->file);
1235 }
1236
1237out_kfree:
1238 kfree(bprm);
1239
1240out_ret:
1241 return retval;
1242}
1243
1244int set_binfmt(struct linux_binfmt *new)
1245{
1246 struct linux_binfmt *old = current->binfmt;
1247
1248 if (new) {
1249 if (!try_module_get(new->module))
1250 return -1;
1251 }
1252 current->binfmt = new;
1253 if (old)
1254 module_put(old->module);
1255 return 0;
1256}
1257
1258EXPORT_SYMBOL(set_binfmt);
1259
1260#define CORENAME_MAX_SIZE 64
1261
1262/* format_corename will inspect the pattern parameter, and output a
1263 * name into corename, which must have space for at least
1264 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1265 */
1266static void format_corename(char *corename, const char *pattern, long signr)
1267{
1268 const char *pat_ptr = pattern;
1269 char *out_ptr = corename;
1270 char *const out_end = corename + CORENAME_MAX_SIZE;
1271 int rc;
1272 int pid_in_pattern = 0;
1273
1274 /* Repeat as long as we have more pattern to process and more output
1275 space */
1276 while (*pat_ptr) {
1277 if (*pat_ptr != '%') {
1278 if (out_ptr == out_end)
1279 goto out;
1280 *out_ptr++ = *pat_ptr++;
1281 } else {
1282 switch (*++pat_ptr) {
1283 case 0:
1284 goto out;
1285 /* Double percent, output one percent */
1286 case '%':
1287 if (out_ptr == out_end)
1288 goto out;
1289 *out_ptr++ = '%';
1290 break;
1291 /* pid */
1292 case 'p':
1293 pid_in_pattern = 1;
1294 rc = snprintf(out_ptr, out_end - out_ptr,
1295 "%d", current->tgid);
1296 if (rc > out_end - out_ptr)
1297 goto out;
1298 out_ptr += rc;
1299 break;
1300 /* uid */
1301 case 'u':
1302 rc = snprintf(out_ptr, out_end - out_ptr,
1303 "%d", current->uid);
1304 if (rc > out_end - out_ptr)
1305 goto out;
1306 out_ptr += rc;
1307 break;
1308 /* gid */
1309 case 'g':
1310 rc = snprintf(out_ptr, out_end - out_ptr,
1311 "%d", current->gid);
1312 if (rc > out_end - out_ptr)
1313 goto out;
1314 out_ptr += rc;
1315 break;
1316 /* signal that caused the coredump */
1317 case 's':
1318 rc = snprintf(out_ptr, out_end - out_ptr,
1319 "%ld", signr);
1320 if (rc > out_end - out_ptr)
1321 goto out;
1322 out_ptr += rc;
1323 break;
1324 /* UNIX time of coredump */
1325 case 't': {
1326 struct timeval tv;
1327 do_gettimeofday(&tv);
1328 rc = snprintf(out_ptr, out_end - out_ptr,
1329 "%lu", tv.tv_sec);
1330 if (rc > out_end - out_ptr)
1331 goto out;
1332 out_ptr += rc;
1333 break;
1334 }
1335 /* hostname */
1336 case 'h':
1337 down_read(&uts_sem);
1338 rc = snprintf(out_ptr, out_end - out_ptr,
1339 "%s", system_utsname.nodename);
1340 up_read(&uts_sem);
1341 if (rc > out_end - out_ptr)
1342 goto out;
1343 out_ptr += rc;
1344 break;
1345 /* executable */
1346 case 'e':
1347 rc = snprintf(out_ptr, out_end - out_ptr,
1348 "%s", current->comm);
1349 if (rc > out_end - out_ptr)
1350 goto out;
1351 out_ptr += rc;
1352 break;
1353 default:
1354 break;
1355 }
1356 ++pat_ptr;
1357 }
1358 }
1359 /* Backward compatibility with core_uses_pid:
1360 *
1361 * If core_pattern does not include a %p (as is the default)
1362 * and core_uses_pid is set, then .%pid will be appended to
1363 * the filename */
1364 if (!pid_in_pattern
1365 && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1366 rc = snprintf(out_ptr, out_end - out_ptr,
1367 ".%d", current->tgid);
1368 if (rc > out_end - out_ptr)
1369 goto out;
1370 out_ptr += rc;
1371 }
1372 out:
1373 *out_ptr = 0;
1374}
1375
1376static void zap_threads (struct mm_struct *mm)
1377{
1378 struct task_struct *g, *p;
1379 struct task_struct *tsk = current;
1380 struct completion *vfork_done = tsk->vfork_done;
1381 int traced = 0;
1382
1383 /*
1384 * Make sure nobody is waiting for us to release the VM,
1385 * otherwise we can deadlock when we wait on each other
1386 */
1387 if (vfork_done) {
1388 tsk->vfork_done = NULL;
1389 complete(vfork_done);
1390 }
1391
1392 read_lock(&tasklist_lock);
1393 do_each_thread(g,p)
1394 if (mm == p->mm && p != tsk) {
1395 force_sig_specific(SIGKILL, p);
1396 mm->core_waiters++;
1397 if (unlikely(p->ptrace) &&
1398 unlikely(p->parent->mm == mm))
1399 traced = 1;
1400 }
1401 while_each_thread(g,p);
1402
1403 read_unlock(&tasklist_lock);
1404
1405 if (unlikely(traced)) {
1406 /*
1407 * We are zapping a thread and the thread it ptraces.
1408 * If the tracee went into a ptrace stop for exit tracing,
1409 * we could deadlock since the tracer is waiting for this
1410 * coredump to finish. Detach them so they can both die.
1411 */
1412 write_lock_irq(&tasklist_lock);
1413 do_each_thread(g,p) {
1414 if (mm == p->mm && p != tsk &&
1415 p->ptrace && p->parent->mm == mm) {
1416 __ptrace_unlink(p);
1417 }
1418 } while_each_thread(g,p);
1419 write_unlock_irq(&tasklist_lock);
1420 }
1421}
1422
1423static void coredump_wait(struct mm_struct *mm)
1424{
1425 DECLARE_COMPLETION(startup_done);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001426 int core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 mm->core_startup_done = &startup_done;
1429
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 zap_threads(mm);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001431 core_waiters = mm->core_waiters;
1432 up_write(&mm->mmap_sem);
1433
1434 if (core_waiters)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 wait_for_completion(&startup_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 BUG_ON(mm->core_waiters);
1437}
1438
1439int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1440{
1441 char corename[CORENAME_MAX_SIZE + 1];
1442 struct mm_struct *mm = current->mm;
1443 struct linux_binfmt * binfmt;
1444 struct inode * inode;
1445 struct file * file;
1446 int retval = 0;
Alan Coxd6e71142005-06-23 00:09:43 -07001447 int fsuid = current->fsuid;
1448 int flag = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
1450 binfmt = current->binfmt;
1451 if (!binfmt || !binfmt->core_dump)
1452 goto fail;
1453 down_write(&mm->mmap_sem);
1454 if (!mm->dumpable) {
1455 up_write(&mm->mmap_sem);
1456 goto fail;
1457 }
Alan Coxd6e71142005-06-23 00:09:43 -07001458
1459 /*
1460 * We cannot trust fsuid as being the "true" uid of the
1461 * process nor do we know its entire history. We only know it
1462 * was tainted so we dump it as root in mode 2.
1463 */
1464 if (mm->dumpable == 2) { /* Setuid core dump mode */
1465 flag = O_EXCL; /* Stop rewrite attacks */
1466 current->fsuid = 0; /* Dump root private */
1467 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 mm->dumpable = 0;
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001469
1470 retval = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 spin_lock_irq(&current->sighand->siglock);
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001472 if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
1473 current->signal->flags = SIGNAL_GROUP_EXIT;
1474 current->signal->group_exit_code = exit_code;
1475 retval = 0;
1476 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 spin_unlock_irq(&current->sighand->siglock);
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001478 if (retval) {
1479 up_write(&mm->mmap_sem);
1480 goto fail;
1481 }
1482
1483 init_completion(&mm->core_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 coredump_wait(mm);
1485
1486 /*
1487 * Clear any false indication of pending signals that might
1488 * be seen by the filesystem code called to write the core file.
1489 */
1490 current->signal->group_stop_count = 0;
1491 clear_thread_flag(TIF_SIGPENDING);
1492
1493 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1494 goto fail_unlock;
1495
1496 /*
1497 * lock_kernel() because format_corename() is controlled by sysctl, which
1498 * uses lock_kernel()
1499 */
1500 lock_kernel();
1501 format_corename(corename, core_pattern, signr);
1502 unlock_kernel();
Alan Coxd6e71142005-06-23 00:09:43 -07001503 file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 if (IS_ERR(file))
1505 goto fail_unlock;
1506 inode = file->f_dentry->d_inode;
1507 if (inode->i_nlink > 1)
1508 goto close_fail; /* multiple links - don't dump */
1509 if (d_unhashed(file->f_dentry))
1510 goto close_fail;
1511
1512 if (!S_ISREG(inode->i_mode))
1513 goto close_fail;
1514 if (!file->f_op)
1515 goto close_fail;
1516 if (!file->f_op->write)
1517 goto close_fail;
Miklos Szeredicc4e69d2005-11-07 00:59:49 -08001518 if (do_truncate(file->f_dentry, 0, file) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 goto close_fail;
1520
1521 retval = binfmt->core_dump(signr, regs, file);
1522
1523 if (retval)
1524 current->signal->group_exit_code |= 0x80;
1525close_fail:
1526 filp_close(file, NULL);
1527fail_unlock:
Alan Coxd6e71142005-06-23 00:09:43 -07001528 current->fsuid = fsuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 complete_all(&mm->core_done);
1530fail:
1531 return retval;
1532}