blob: 9c33f542dc7785da628c2943c3626c878e0df07b [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/slab.h>
26#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040027#include <linux/fdtable.h>
Hugh Dickinsba92a432008-07-25 01:45:43 -070028#include <linux/mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/stat.h>
30#include <linux/fcntl.h>
31#include <linux/smp_lock.h>
Hugh Dickinsba92a432008-07-25 01:45:43 -070032#include <linux/swap.h>
Neil Horman74aadce2007-10-16 23:26:35 -070033#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/init.h>
Hugh Dickinsca5b1722008-07-28 15:46:18 -070035#include <linux/pagemap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/highmem.h>
37#include <linux/spinlock.h>
38#include <linux/key.h>
39#include <linux/personality.h>
40#include <linux/binfmts.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/utsname.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080042#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/module.h>
44#include <linux/namei.h>
45#include <linux/proc_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/mount.h>
47#include <linux/security.h>
48#include <linux/syscalls.h>
Jay Lan8f0ab512006-09-30 23:28:59 -070049#include <linux/tsacct_kern.h>
Matt Helsley9f460802005-11-07 00:59:16 -080050#include <linux/cn_proc.h>
Al Viro473ae302006-04-26 14:04:08 -040051#include <linux/audit.h>
Roland McGrath6341c392008-07-25 19:45:44 -070052#include <linux/tracehook.h>
Johannes Berg5f4123b2008-07-09 10:28:40 +020053#include <linux/kmod.h>
Eric Paris6110e3a2008-12-17 13:53:20 -050054#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56#include <asm/uaccess.h>
57#include <asm/mmu_context.h>
Ollie Wildb6a2fea2007-07-19 01:48:16 -070058#include <asm/tlb.h>
David Howellsa6f76f22008-11-14 10:39:24 +110059#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Linus Torvalds1da177e2005-04-16 15:20:36 -070061int core_uses_pid;
Dan Aloni71ce92f2007-05-16 22:11:16 -070062char core_pattern[CORENAME_MAX_SIZE] = "core";
Alan Coxd6e71142005-06-23 00:09:43 -070063int suid_dumpable = 0;
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/* The maximal length of core_pattern is also specified in sysctl.c */
66
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070067static LIST_HEAD(formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070068static DEFINE_RWLOCK(binfmt_lock);
69
70int register_binfmt(struct linux_binfmt * fmt)
71{
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 if (!fmt)
73 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 write_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070075 list_add(&fmt->lh, &formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 write_unlock(&binfmt_lock);
77 return 0;
78}
79
80EXPORT_SYMBOL(register_binfmt);
81
Alexey Dobriyanf6b450d2007-10-16 23:26:04 -070082void unregister_binfmt(struct linux_binfmt * fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083{
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 write_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070085 list_del(&fmt->lh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 write_unlock(&binfmt_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087}
88
89EXPORT_SYMBOL(unregister_binfmt);
90
91static inline void put_binfmt(struct linux_binfmt * fmt)
92{
93 module_put(fmt->module);
94}
95
96/*
97 * Note that a shared library must be both readable and executable due to
98 * security reasons.
99 *
100 * Also note that we take the address to load from from the file itself.
101 */
102asmlinkage long sys_uselib(const char __user * library)
103{
Al Viro964bd182008-07-26 03:33:14 -0400104 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 struct nameidata nd;
Al Viro964bd182008-07-26 03:33:14 -0400106 char *tmp = getname(library);
107 int error = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
Al Viro964bd182008-07-26 03:33:14 -0400109 if (!IS_ERR(tmp)) {
110 error = path_lookup_open(AT_FDCWD, tmp,
111 LOOKUP_FOLLOW, &nd,
112 FMODE_READ|FMODE_EXEC);
113 putname(tmp);
114 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 if (error)
116 goto out;
117
118 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -0800119 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 goto exit;
121
Al Viro30524472008-07-22 00:02:33 -0400122 error = -EACCES;
123 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
124 goto exit;
125
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200126 error = inode_permission(nd.path.dentry->d_inode,
127 MAY_READ | MAY_EXEC | MAY_OPEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 if (error)
129 goto exit;
130
Andi Kleenabe8be32008-02-08 04:20:23 -0800131 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 error = PTR_ERR(file);
133 if (IS_ERR(file))
134 goto out;
135
Eric Paris6110e3a2008-12-17 13:53:20 -0500136 fsnotify_open(file->f_path.dentry);
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 error = -ENOEXEC;
139 if(file->f_op) {
140 struct linux_binfmt * fmt;
141
142 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -0700143 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 if (!fmt->load_shlib)
145 continue;
146 if (!try_module_get(fmt->module))
147 continue;
148 read_unlock(&binfmt_lock);
149 error = fmt->load_shlib(file);
150 read_lock(&binfmt_lock);
151 put_binfmt(fmt);
152 if (error != -ENOEXEC)
153 break;
154 }
155 read_unlock(&binfmt_lock);
156 }
157 fput(file);
158out:
159 return error;
160exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -0700161 release_open_intent(&nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800162 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 goto out;
164}
165
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700166#ifdef CONFIG_MMU
167
168static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
169 int write)
170{
171 struct page *page;
172 int ret;
173
174#ifdef CONFIG_STACK_GROWSUP
175 if (write) {
176 ret = expand_stack_downwards(bprm->vma, pos);
177 if (ret < 0)
178 return NULL;
179 }
180#endif
181 ret = get_user_pages(current, bprm->mm, pos,
182 1, write, 1, &page, NULL);
183 if (ret <= 0)
184 return NULL;
185
186 if (write) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700187 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800188 struct rlimit *rlim;
189
190 /*
191 * We've historically supported up to 32 pages (ARG_MAX)
192 * of argument strings even with small stacks
193 */
194 if (size <= ARG_MAX)
195 return page;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700196
197 /*
198 * Limit to 1/4-th the stack size for the argv+env strings.
199 * This ensures that:
200 * - the remaining binfmt code will not run out of stack space,
201 * - the program will have a reasonable amount of stack left
202 * to work from.
203 */
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800204 rlim = current->signal->rlim;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700205 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
206 put_page(page);
207 return NULL;
208 }
209 }
210
211 return page;
212}
213
214static void put_arg_page(struct page *page)
215{
216 put_page(page);
217}
218
219static void free_arg_page(struct linux_binprm *bprm, int i)
220{
221}
222
223static void free_arg_pages(struct linux_binprm *bprm)
224{
225}
226
227static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
228 struct page *page)
229{
230 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
231}
232
233static int __bprm_mm_init(struct linux_binprm *bprm)
234{
235 int err = -ENOMEM;
236 struct vm_area_struct *vma = NULL;
237 struct mm_struct *mm = bprm->mm;
238
239 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
240 if (!vma)
241 goto err;
242
243 down_write(&mm->mmap_sem);
244 vma->vm_mm = mm;
245
246 /*
247 * Place the stack at the largest stack address the architecture
248 * supports. Later, we'll move this to an appropriate place. We don't
249 * use STACK_TOP because that can depend on attributes which aren't
250 * configured yet.
251 */
252 vma->vm_end = STACK_TOP_MAX;
253 vma->vm_start = vma->vm_end - PAGE_SIZE;
254
255 vma->vm_flags = VM_STACK_FLAGS;
Coly Li3ed75eb2007-10-18 23:39:15 -0700256 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700257 err = insert_vm_struct(mm, vma);
258 if (err) {
259 up_write(&mm->mmap_sem);
260 goto err;
261 }
262
263 mm->stack_vm = mm->total_vm = 1;
264 up_write(&mm->mmap_sem);
265
266 bprm->p = vma->vm_end - sizeof(void *);
267
268 return 0;
269
270err:
271 if (vma) {
272 bprm->vma = NULL;
273 kmem_cache_free(vm_area_cachep, vma);
274 }
275
276 return err;
277}
278
279static bool valid_arg_len(struct linux_binprm *bprm, long len)
280{
281 return len <= MAX_ARG_STRLEN;
282}
283
284#else
285
286static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
287 int write)
288{
289 struct page *page;
290
291 page = bprm->page[pos / PAGE_SIZE];
292 if (!page && write) {
293 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
294 if (!page)
295 return NULL;
296 bprm->page[pos / PAGE_SIZE] = page;
297 }
298
299 return page;
300}
301
302static void put_arg_page(struct page *page)
303{
304}
305
306static void free_arg_page(struct linux_binprm *bprm, int i)
307{
308 if (bprm->page[i]) {
309 __free_page(bprm->page[i]);
310 bprm->page[i] = NULL;
311 }
312}
313
314static void free_arg_pages(struct linux_binprm *bprm)
315{
316 int i;
317
318 for (i = 0; i < MAX_ARG_PAGES; i++)
319 free_arg_page(bprm, i);
320}
321
322static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
323 struct page *page)
324{
325}
326
327static int __bprm_mm_init(struct linux_binprm *bprm)
328{
329 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
330 return 0;
331}
332
333static bool valid_arg_len(struct linux_binprm *bprm, long len)
334{
335 return len <= bprm->p;
336}
337
338#endif /* CONFIG_MMU */
339
340/*
341 * Create a new mm_struct and populate it with a temporary stack
342 * vm_area_struct. We don't have enough context at this point to set the stack
343 * flags, permissions, and offset, so we use temporary values. We'll update
344 * them later in setup_arg_pages().
345 */
346int bprm_mm_init(struct linux_binprm *bprm)
347{
348 int err;
349 struct mm_struct *mm = NULL;
350
351 bprm->mm = mm = mm_alloc();
352 err = -ENOMEM;
353 if (!mm)
354 goto err;
355
356 err = init_new_context(current, mm);
357 if (err)
358 goto err;
359
360 err = __bprm_mm_init(bprm);
361 if (err)
362 goto err;
363
364 return 0;
365
366err:
367 if (mm) {
368 bprm->mm = NULL;
369 mmdrop(mm);
370 }
371
372 return err;
373}
374
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375/*
376 * count() counts the number of strings in array ARGV.
377 */
378static int count(char __user * __user * argv, int max)
379{
380 int i = 0;
381
382 if (argv != NULL) {
383 for (;;) {
384 char __user * p;
385
386 if (get_user(p, argv))
387 return -EFAULT;
388 if (!p)
389 break;
390 argv++;
Jason Baron362e6662008-10-15 22:01:52 -0700391 if (i++ >= max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 return -E2BIG;
393 cond_resched();
394 }
395 }
396 return i;
397}
398
399/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700400 * 'copy_strings()' copies argument/environment strings from the old
401 * processes's memory to the new process's stack. The call to get_user_pages()
402 * ensures the destination page is created and not swapped out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 */
Adrian Bunk75c96f82005-05-05 16:16:09 -0700404static int copy_strings(int argc, char __user * __user * argv,
405 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
407 struct page *kmapped_page = NULL;
408 char *kaddr = NULL;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700409 unsigned long kpos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 int ret;
411
412 while (argc-- > 0) {
413 char __user *str;
414 int len;
415 unsigned long pos;
416
417 if (get_user(str, argv+argc) ||
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700418 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 ret = -EFAULT;
420 goto out;
421 }
422
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700423 if (!valid_arg_len(bprm, len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 ret = -E2BIG;
425 goto out;
426 }
427
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700428 /* We're going to work our way backwords. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 pos = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700430 str += len;
431 bprm->p -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
433 while (len > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 int offset, bytes_to_copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
436 offset = pos % PAGE_SIZE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700437 if (offset == 0)
438 offset = PAGE_SIZE;
439
440 bytes_to_copy = offset;
441 if (bytes_to_copy > len)
442 bytes_to_copy = len;
443
444 offset -= bytes_to_copy;
445 pos -= bytes_to_copy;
446 str -= bytes_to_copy;
447 len -= bytes_to_copy;
448
449 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
450 struct page *page;
451
452 page = get_arg_page(bprm, pos, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 if (!page) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700454 ret = -E2BIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 goto out;
456 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700458 if (kmapped_page) {
459 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700461 put_arg_page(kmapped_page);
462 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 kmapped_page = page;
464 kaddr = kmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700465 kpos = pos & PAGE_MASK;
466 flush_arg_page(bprm, kpos, kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700468 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 ret = -EFAULT;
470 goto out;
471 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 }
473 }
474 ret = 0;
475out:
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700476 if (kmapped_page) {
477 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700479 put_arg_page(kmapped_page);
480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 return ret;
482}
483
484/*
485 * Like copy_strings, but get argv and its values from kernel memory.
486 */
487int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
488{
489 int r;
490 mm_segment_t oldfs = get_fs();
491 set_fs(KERNEL_DS);
492 r = copy_strings(argc, (char __user * __user *)argv, bprm);
493 set_fs(oldfs);
494 return r;
495}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496EXPORT_SYMBOL(copy_strings_kernel);
497
498#ifdef CONFIG_MMU
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700499
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700501 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
502 * the binfmt code determines where the new stack should reside, we shift it to
503 * its final location. The process proceeds as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 *
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700505 * 1) Use shift to calculate the new vma endpoints.
506 * 2) Extend vma to cover both the old and new ranges. This ensures the
507 * arguments passed to subsequent functions are consistent.
508 * 3) Move vma's page tables to the new range.
509 * 4) Free up any cleared pgd range.
510 * 5) Shrink the vma to cover only the new range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700512static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513{
514 struct mm_struct *mm = vma->vm_mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700515 unsigned long old_start = vma->vm_start;
516 unsigned long old_end = vma->vm_end;
517 unsigned long length = old_end - old_start;
518 unsigned long new_start = old_start - shift;
519 unsigned long new_end = old_end - shift;
520 struct mmu_gather *tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700522 BUG_ON(new_start > new_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700524 /*
525 * ensure there are no vmas between where we want to go
526 * and where we are
527 */
528 if (vma != find_vma(mm, new_start))
529 return -EFAULT;
530
531 /*
532 * cover the whole range: [new_start, old_end)
533 */
534 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
535
536 /*
537 * move the page tables downwards, on failure we rely on
538 * process cleanup to remove whatever mess we made.
539 */
540 if (length != move_page_tables(vma, old_start,
541 vma, new_start, length))
542 return -ENOMEM;
543
544 lru_add_drain();
545 tlb = tlb_gather_mmu(mm, 0);
546 if (new_end > old_start) {
547 /*
548 * when the old and new regions overlap clear from new_end.
549 */
Jan Beulich42b77722008-07-23 21:27:10 -0700550 free_pgd_range(tlb, new_end, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700551 vma->vm_next ? vma->vm_next->vm_start : 0);
552 } else {
553 /*
554 * otherwise, clean from old_start; this is done to not touch
555 * the address space in [new_end, old_start) some architectures
556 * have constraints on va-space that make this illegal (IA64) -
557 * for the others its just a little faster.
558 */
Jan Beulich42b77722008-07-23 21:27:10 -0700559 free_pgd_range(tlb, old_start, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700560 vma->vm_next ? vma->vm_next->vm_start : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700562 tlb_finish_mmu(tlb, new_end, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700564 /*
565 * shrink the vma to just the new range.
566 */
567 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
568
569 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570}
571
572#define EXTRA_STACK_VM_PAGES 20 /* random */
573
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700574/*
575 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
576 * the stack is optionally relocated, and some extra space is added.
577 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578int setup_arg_pages(struct linux_binprm *bprm,
579 unsigned long stack_top,
580 int executable_stack)
581{
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700582 unsigned long ret;
583 unsigned long stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 struct mm_struct *mm = current->mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700585 struct vm_area_struct *vma = bprm->vma;
586 struct vm_area_struct *prev = NULL;
587 unsigned long vm_flags;
588 unsigned long stack_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590#ifdef CONFIG_STACK_GROWSUP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 /* Limit stack size to 1GB */
592 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
593 if (stack_base > (1 << 30))
594 stack_base = 1 << 30;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700596 /* Make sure we didn't let the argument array grow too large. */
597 if (vma->vm_end - vma->vm_start > stack_base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 return -ENOMEM;
599
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700600 stack_base = PAGE_ALIGN(stack_top - stack_base);
601
602 stack_shift = vma->vm_start - stack_base;
603 mm->arg_start = bprm->p - stack_shift;
604 bprm->p = vma->vm_end - stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605#else
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700606 stack_top = arch_align_stack(stack_top);
607 stack_top = PAGE_ALIGN(stack_top);
608 stack_shift = vma->vm_end - stack_top;
609
610 bprm->p -= stack_shift;
611 mm->arg_start = bprm->p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612#endif
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700613
614 if (bprm->loader)
615 bprm->loader -= stack_shift;
616 bprm->exec -= stack_shift;
617
618 down_write(&mm->mmap_sem);
Hugh Dickins96a8e132008-07-10 21:19:20 +0100619 vm_flags = VM_STACK_FLAGS;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700620
621 /*
622 * Adjust stack execute permissions; explicitly enable for
623 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
624 * (arch default) otherwise.
625 */
626 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
627 vm_flags |= VM_EXEC;
628 else if (executable_stack == EXSTACK_DISABLE_X)
629 vm_flags &= ~VM_EXEC;
630 vm_flags |= mm->def_flags;
631
632 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
633 vm_flags);
634 if (ret)
635 goto out_unlock;
636 BUG_ON(prev != vma);
637
638 /* Move stack pages down in memory. */
639 if (stack_shift) {
640 ret = shift_arg_pages(vma, stack_shift);
641 if (ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 up_write(&mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 return ret;
644 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 }
646
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700647#ifdef CONFIG_STACK_GROWSUP
648 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
649#else
650 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
651#endif
652 ret = expand_stack(vma, stack_base);
653 if (ret)
654 ret = -EFAULT;
655
656out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 up_write(&mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 return 0;
659}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660EXPORT_SYMBOL(setup_arg_pages);
661
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662#endif /* CONFIG_MMU */
663
664struct file *open_exec(const char *name)
665{
666 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 struct file *file;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200668 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200670 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
671 FMODE_READ|FMODE_EXEC);
672 if (err)
673 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200675 err = -EACCES;
676 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
677 goto out_path_put;
678
Al Viro30524472008-07-22 00:02:33 -0400679 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
680 goto out_path_put;
681
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200682 err = inode_permission(nd.path.dentry->d_inode, MAY_EXEC | MAY_OPEN);
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200683 if (err)
684 goto out_path_put;
685
686 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
687 if (IS_ERR(file))
688 return file;
689
Eric Paris6110e3a2008-12-17 13:53:20 -0500690 fsnotify_open(file->f_path.dentry);
691
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200692 err = deny_write_access(file);
693 if (err) {
694 fput(file);
695 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200698 return file;
699
700 out_path_put:
701 release_open_intent(&nd);
702 path_put(&nd.path);
703 out:
704 return ERR_PTR(err);
705}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706EXPORT_SYMBOL(open_exec);
707
708int kernel_read(struct file *file, unsigned long offset,
709 char *addr, unsigned long count)
710{
711 mm_segment_t old_fs;
712 loff_t pos = offset;
713 int result;
714
715 old_fs = get_fs();
716 set_fs(get_ds());
717 /* The cast to a user pointer is valid due to the set_fs() */
718 result = vfs_read(file, (void __user *)addr, count, &pos);
719 set_fs(old_fs);
720 return result;
721}
722
723EXPORT_SYMBOL(kernel_read);
724
725static int exec_mmap(struct mm_struct *mm)
726{
727 struct task_struct *tsk;
728 struct mm_struct * old_mm, *active_mm;
729
730 /* Notify parent that we're no longer interested in the old VM */
731 tsk = current;
732 old_mm = current->mm;
733 mm_release(tsk, old_mm);
734
735 if (old_mm) {
736 /*
737 * Make sure that if there is a core dump in progress
738 * for the old mm, we get out and die instead of going
739 * through with the exec. We must hold mmap_sem around
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700740 * checking core_state and changing tsk->mm.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 */
742 down_read(&old_mm->mmap_sem);
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700743 if (unlikely(old_mm->core_state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 up_read(&old_mm->mmap_sem);
745 return -EINTR;
746 }
747 }
748 task_lock(tsk);
749 active_mm = tsk->active_mm;
750 tsk->mm = mm;
751 tsk->active_mm = mm;
752 activate_mm(active_mm, mm);
753 task_unlock(tsk);
754 arch_pick_mmap_layout(mm);
755 if (old_mm) {
756 up_read(&old_mm->mmap_sem);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +0200757 BUG_ON(active_mm != old_mm);
Balbir Singh31a78f22008-09-28 23:09:31 +0100758 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 mmput(old_mm);
760 return 0;
761 }
762 mmdrop(active_mm);
763 return 0;
764}
765
766/*
767 * This function makes sure the current process has its own signal table,
768 * so that flush_signal_handlers can later reset the handlers without
769 * disturbing other processes. (Other processes might share the signal
770 * table via the CLONE_SIGHAND option to clone().)
771 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800772static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773{
774 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700775 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 spinlock_t *lock = &oldsighand->siglock;
777 int count;
778
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700779 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 goto no_thread_group;
781
782 /*
783 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800786 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 /*
788 * Another group action in progress, just
789 * return so that the signal is processed.
790 */
791 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 return -EAGAIN;
793 }
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800794 sig->group_exit_task = tsk;
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700795 zap_other_threads(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
Oleg Nesterovfea9d172008-02-08 04:19:19 -0800797 /* Account for the thread group leader hanging around: */
798 count = thread_group_leader(tsk) ? 1 : 2;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700799 sig->notify_count = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 while (atomic_read(&sig->count) > count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 __set_current_state(TASK_UNINTERRUPTIBLE);
802 spin_unlock_irq(lock);
803 schedule();
804 spin_lock_irq(lock);
805 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 spin_unlock_irq(lock);
807
808 /*
809 * At this point all other threads have exited, all we have to
810 * do is to wait for the thread group leader to become inactive,
811 * and to assume its PID:
812 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700813 if (!thread_group_leader(tsk)) {
Oleg Nesterov81879262008-12-01 14:18:16 -0800814 struct task_struct *leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700815
Oleg Nesterov2800d8d2008-04-30 00:53:12 -0700816 sig->notify_count = -1; /* for exit_notify() */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700817 for (;;) {
818 write_lock_irq(&tasklist_lock);
819 if (likely(leader->exit_state))
820 break;
821 __set_current_state(TASK_UNINTERRUPTIBLE);
822 write_unlock_irq(&tasklist_lock);
823 schedule();
824 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
Roland McGrathf5e90282006-04-10 22:54:16 -0700826 /*
827 * The only record we have of the real-time age of a
828 * process, regardless of execs it's done, is start_time.
829 * All the past CPU time is accumulated in signal_struct
830 * from sister threads now dead. But in this non-leader
831 * exec, nothing survives from the original leader thread,
832 * whose birth marks the true age of this process now.
833 * When we take on its identity by switching to its PID, we
834 * also take its birthdate (always earlier than our own).
835 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700836 tsk->start_time = leader->start_time;
Roland McGrathf5e90282006-04-10 22:54:16 -0700837
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700838 BUG_ON(!same_thread_group(leader, tsk));
839 BUG_ON(has_group_leader_pid(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 /*
841 * An exec() starts a new thread group with the
842 * TGID of the previous thread group. Rehash the
843 * two threads with a switched PID, and release
844 * the former thread group leader:
845 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800846
847 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700848 * The old leader becomes a thread of the this thread group.
849 * Note: The old leader also uses this pid until release_task
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800850 * is called. Odd but simple and correct.
851 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700852 detach_pid(tsk, PIDTYPE_PID);
853 tsk->pid = leader->pid;
Sukadev Bhattiprolu3743ca02007-10-18 23:39:51 -0700854 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700855 transfer_pid(leader, tsk, PIDTYPE_PGID);
856 transfer_pid(leader, tsk, PIDTYPE_SID);
857 list_replace_rcu(&leader->tasks, &tsk->tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700859 tsk->group_leader = tsk;
860 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -0600861
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700862 tsk->exit_signal = SIGCHLD;
Oleg Nesterov962b5642005-11-23 13:37:43 -0800863
864 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
865 leader->exit_state = EXIT_DEAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 write_unlock_irq(&tasklist_lock);
Oleg Nesterov81879262008-12-01 14:18:16 -0800867
868 release_task(leader);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800869 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700871 sig->group_exit_task = NULL;
872 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
874no_thread_group:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 exit_itimers(sig);
Oleg Nesterovcbaffba2008-05-26 20:55:42 +0400876 flush_itimer_signals();
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300877
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700878 if (atomic_read(&oldsighand->count) != 1) {
879 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700881 * This ->sighand is shared with the CLONE_SIGHAND
882 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700884 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
885 if (!newsighand)
886 return -ENOMEM;
887
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 atomic_set(&newsighand->count, 1);
889 memcpy(newsighand->action, oldsighand->action,
890 sizeof(newsighand->action));
891
892 write_lock_irq(&tasklist_lock);
893 spin_lock(&oldsighand->siglock);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700894 rcu_assign_pointer(tsk->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 spin_unlock(&oldsighand->siglock);
896 write_unlock_irq(&tasklist_lock);
897
Davide Libenzifba2afa2007-05-10 22:23:13 -0700898 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 }
900
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700901 BUG_ON(!thread_group_leader(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 return 0;
903}
Oleg Nesterov0840a902007-10-16 23:27:22 -0700904
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905/*
906 * These functions flushes out all traces of the currently running executable
907 * so that a new one can be started
908 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800909static void flush_old_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910{
911 long j = -1;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700912 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
914 spin_lock(&files->file_lock);
915 for (;;) {
916 unsigned long set, i;
917
918 j++;
919 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700920 fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800921 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700923 set = fdt->close_on_exec->fds_bits[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 if (!set)
925 continue;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700926 fdt->close_on_exec->fds_bits[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 spin_unlock(&files->file_lock);
928 for ( ; set ; i++,set >>= 1) {
929 if (set & 1) {
930 sys_close(i);
931 }
932 }
933 spin_lock(&files->file_lock);
934
935 }
936 spin_unlock(&files->file_lock);
937}
938
Andrew Morton59714d62008-02-04 22:27:21 -0800939char *get_task_comm(char *buf, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
941 /* buf must be at least sizeof(tsk->comm) in size */
942 task_lock(tsk);
943 strncpy(buf, tsk->comm, sizeof(tsk->comm));
944 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -0800945 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946}
947
948void set_task_comm(struct task_struct *tsk, char *buf)
949{
950 task_lock(tsk);
951 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
952 task_unlock(tsk);
953}
954
955int flush_old_exec(struct linux_binprm * bprm)
956{
957 char * name;
958 int i, ch, retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 char tcomm[sizeof(current->comm)];
960
961 /*
962 * Make sure we have a private signal table and that
963 * we are unassociated from the previous thread group.
964 */
965 retval = de_thread(current);
966 if (retval)
967 goto out;
968
Matt Helsley925d1c42008-04-29 01:01:36 -0700969 set_mm_exe_file(bprm->mm, bprm->file);
970
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 * Release all of the old mmap stuff
973 */
974 retval = exec_mmap(bprm->mm);
975 if (retval)
Al Virofd8328b2008-04-22 05:11:59 -0400976 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
978 bprm->mm = NULL; /* We're using it now */
979
980 /* This is the point of no return */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 current->sas_ss_sp = current->sas_ss_size = 0;
982
David Howellsda9592e2008-11-14 10:39:05 +1100983 if (current_euid() == current_uid() && current_egid() == current_gid())
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -0700984 set_dumpable(current->mm, 1);
Alan Coxd6e71142005-06-23 00:09:43 -0700985 else
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -0700986 set_dumpable(current->mm, suid_dumpable);
Alan Coxd6e71142005-06-23 00:09:43 -0700987
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 name = bprm->filename;
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -0700989
990 /* Copies the binary name from after last slash */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 for (i=0; (ch = *(name++)) != '\0';) {
992 if (ch == '/')
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -0700993 i = 0; /* overwrite what we wrote */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 else
995 if (i < (sizeof(tcomm) - 1))
996 tcomm[i++] = ch;
997 }
998 tcomm[i] = '\0';
999 set_task_comm(current, tcomm);
1000
1001 current->flags &= ~PF_RANDOMIZE;
1002 flush_thread();
1003
Benjamin Herrenschmidt0551fbd2006-02-28 16:59:19 -08001004 /* Set the new mm task size. We have to do that late because it may
1005 * depend on TIF_32BIT which is only updated in flush_thread() on
1006 * some architectures like powerpc
1007 */
1008 current->mm->task_size = TASK_SIZE;
1009
David Howellsa6f76f22008-11-14 10:39:24 +11001010 /* install the new credentials */
1011 if (bprm->cred->uid != current_euid() ||
1012 bprm->cred->gid != current_egid()) {
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001013 current->pdeath_signal = 0;
1014 } else if (file_permission(bprm->file, MAY_READ) ||
David Howellsa6f76f22008-11-14 10:39:24 +11001015 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001016 set_dumpable(current->mm, suid_dumpable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 }
1018
David Howellsa6f76f22008-11-14 10:39:24 +11001019 current->personality &= ~bprm->per_clear;
1020
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 /* An exec changes our domain. We are no longer part of the thread
1022 group */
1023
1024 current->self_exec_id++;
1025
1026 flush_signal_handlers(current, 0);
1027 flush_old_files(current->files);
1028
1029 return 0;
1030
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031out:
1032 return retval;
1033}
1034
1035EXPORT_SYMBOL(flush_old_exec);
1036
David Howellsa6f76f22008-11-14 10:39:24 +11001037/*
1038 * install the new credentials for this executable
1039 */
1040void install_exec_creds(struct linux_binprm *bprm)
1041{
1042 security_bprm_committing_creds(bprm);
1043
1044 commit_creds(bprm->cred);
1045 bprm->cred = NULL;
1046
1047 /* cred_exec_mutex must be held at least to this point to prevent
1048 * ptrace_attach() from altering our determination of the task's
1049 * credentials; any time after this it may be unlocked */
1050
1051 security_bprm_committed_creds(bprm);
1052}
1053EXPORT_SYMBOL(install_exec_creds);
1054
1055/*
1056 * determine how safe it is to execute the proposed program
1057 * - the caller must hold current->cred_exec_mutex to protect against
1058 * PTRACE_ATTACH
1059 */
1060void check_unsafe_exec(struct linux_binprm *bprm)
1061{
1062 struct task_struct *p = current;
1063
1064 bprm->unsafe = tracehook_unsafe_exec(p);
1065
1066 if (atomic_read(&p->fs->count) > 1 ||
1067 atomic_read(&p->files->count) > 1 ||
1068 atomic_read(&p->sighand->count) > 1)
1069 bprm->unsafe |= LSM_UNSAFE_SHARE;
1070}
1071
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072/*
1073 * Fill the binprm structure from the inode.
1074 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
David Howellsa6f76f22008-11-14 10:39:24 +11001075 *
1076 * This may be called multiple times for binary chains (scripts for example).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 */
1078int prepare_binprm(struct linux_binprm *bprm)
1079{
David Howellsa6f76f22008-11-14 10:39:24 +11001080 umode_t mode;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001081 struct inode * inode = bprm->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 int retval;
1083
1084 mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 if (bprm->file->f_op == NULL)
1086 return -EACCES;
1087
David Howellsa6f76f22008-11-14 10:39:24 +11001088 /* clear any previous set[ug]id data from a previous binary */
1089 bprm->cred->euid = current_euid();
1090 bprm->cred->egid = current_egid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
David Howellsa6f76f22008-11-14 10:39:24 +11001092 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 /* Set-uid? */
1094 if (mode & S_ISUID) {
David Howellsa6f76f22008-11-14 10:39:24 +11001095 bprm->per_clear |= PER_CLEAR_ON_SETID;
1096 bprm->cred->euid = inode->i_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 }
1098
1099 /* Set-gid? */
1100 /*
1101 * If setgid is set but no group execute bit then this
1102 * is a candidate for mandatory locking, not a setgid
1103 * executable.
1104 */
1105 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
David Howellsa6f76f22008-11-14 10:39:24 +11001106 bprm->per_clear |= PER_CLEAR_ON_SETID;
1107 bprm->cred->egid = inode->i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 }
1109 }
1110
1111 /* fill in binprm security blob */
David Howellsa6f76f22008-11-14 10:39:24 +11001112 retval = security_bprm_set_creds(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 if (retval)
1114 return retval;
David Howellsa6f76f22008-11-14 10:39:24 +11001115 bprm->cred_prepared = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
David Howellsa6f76f22008-11-14 10:39:24 +11001117 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1118 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119}
1120
1121EXPORT_SYMBOL(prepare_binprm);
1122
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001123/*
1124 * Arguments are '\0' separated strings found at the location bprm->p
1125 * points to; chop off the first by relocating brpm->p to right after
1126 * the first '\0' encountered.
1127 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001128int remove_arg_zero(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129{
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001130 int ret = 0;
1131 unsigned long offset;
1132 char *kaddr;
1133 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001135 if (!bprm->argc)
1136 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001138 do {
1139 offset = bprm->p & ~PAGE_MASK;
1140 page = get_arg_page(bprm, bprm->p, 0);
1141 if (!page) {
1142 ret = -EFAULT;
1143 goto out;
1144 }
1145 kaddr = kmap_atomic(page, KM_USER0);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001146
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001147 for (; offset < PAGE_SIZE && kaddr[offset];
1148 offset++, bprm->p++)
1149 ;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001150
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001151 kunmap_atomic(kaddr, KM_USER0);
1152 put_arg_page(page);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001153
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001154 if (offset == PAGE_SIZE)
1155 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1156 } while (offset == PAGE_SIZE);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001157
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001158 bprm->p++;
1159 bprm->argc--;
1160 ret = 0;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001161
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001162out:
1163 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165EXPORT_SYMBOL(remove_arg_zero);
1166
1167/*
1168 * cycle the list of binary formats handler, until one recognizes the image
1169 */
1170int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1171{
Roland McGrath85f33462008-12-09 19:36:38 -08001172 unsigned int depth = bprm->recursion_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 int try,retval;
1174 struct linux_binfmt *fmt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 retval = security_bprm_check(bprm);
1177 if (retval)
1178 return retval;
1179
1180 /* kernel module loader fixup */
1181 /* so we don't try to load run modprobe in kernel space. */
1182 set_fs(USER_DS);
Al Viro473ae302006-04-26 14:04:08 -04001183
1184 retval = audit_bprm(bprm);
1185 if (retval)
1186 return retval;
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 retval = -ENOENT;
1189 for (try=0; try<2; try++) {
1190 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -07001191 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1193 if (!fn)
1194 continue;
1195 if (!try_module_get(fmt->module))
1196 continue;
1197 read_unlock(&binfmt_lock);
1198 retval = fn(bprm, regs);
Roland McGrath85f33462008-12-09 19:36:38 -08001199 /*
1200 * Restore the depth counter to its starting value
1201 * in this call, so we don't have to rely on every
1202 * load_binary function to restore it on return.
1203 */
1204 bprm->recursion_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 if (retval >= 0) {
Roland McGrath85f33462008-12-09 19:36:38 -08001206 if (depth == 0)
1207 tracehook_report_exec(fmt, bprm, regs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 put_binfmt(fmt);
1209 allow_write_access(bprm->file);
1210 if (bprm->file)
1211 fput(bprm->file);
1212 bprm->file = NULL;
1213 current->did_exec = 1;
Matt Helsley9f460802005-11-07 00:59:16 -08001214 proc_exec_connector(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 return retval;
1216 }
1217 read_lock(&binfmt_lock);
1218 put_binfmt(fmt);
1219 if (retval != -ENOEXEC || bprm->mm == NULL)
1220 break;
1221 if (!bprm->file) {
1222 read_unlock(&binfmt_lock);
1223 return retval;
1224 }
1225 }
1226 read_unlock(&binfmt_lock);
1227 if (retval != -ENOEXEC || bprm->mm == NULL) {
1228 break;
Johannes Berg5f4123b2008-07-09 10:28:40 +02001229#ifdef CONFIG_MODULES
1230 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1232 if (printable(bprm->buf[0]) &&
1233 printable(bprm->buf[1]) &&
1234 printable(bprm->buf[2]) &&
1235 printable(bprm->buf[3]))
1236 break; /* -ENOEXEC */
1237 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1238#endif
1239 }
1240 }
1241 return retval;
1242}
1243
1244EXPORT_SYMBOL(search_binary_handler);
1245
Al Viro08a6fac2008-05-10 16:38:25 -04001246void free_bprm(struct linux_binprm *bprm)
1247{
1248 free_arg_pages(bprm);
David Howellsa6f76f22008-11-14 10:39:24 +11001249 if (bprm->cred)
1250 abort_creds(bprm->cred);
Al Viro08a6fac2008-05-10 16:38:25 -04001251 kfree(bprm);
1252}
1253
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254/*
1255 * sys_execve() executes a new program.
1256 */
1257int do_execve(char * filename,
1258 char __user *__user *argv,
1259 char __user *__user *envp,
1260 struct pt_regs * regs)
1261{
1262 struct linux_binprm *bprm;
1263 struct file *file;
Al Viro3b125382008-04-22 05:31:30 -04001264 struct files_struct *displaced;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
Al Viro3b125382008-04-22 05:31:30 -04001267 retval = unshare_files(&displaced);
Al Virofd8328b2008-04-22 05:11:59 -04001268 if (retval)
1269 goto out_ret;
1270
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 retval = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001272 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 if (!bprm)
Al Virofd8328b2008-04-22 05:11:59 -04001274 goto out_files;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
David Howellsa6f76f22008-11-14 10:39:24 +11001276 retval = mutex_lock_interruptible(&current->cred_exec_mutex);
1277 if (retval < 0)
1278 goto out_free;
1279
1280 retval = -ENOMEM;
1281 bprm->cred = prepare_exec_creds();
1282 if (!bprm->cred)
1283 goto out_unlock;
1284 check_unsafe_exec(bprm);
1285
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 file = open_exec(filename);
1287 retval = PTR_ERR(file);
1288 if (IS_ERR(file))
David Howellsa6f76f22008-11-14 10:39:24 +11001289 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291 sched_exec();
1292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 bprm->file = file;
1294 bprm->filename = filename;
1295 bprm->interp = filename;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001296
1297 retval = bprm_mm_init(bprm);
1298 if (retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 goto out_file;
1300
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001301 bprm->argc = count(argv, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 if ((retval = bprm->argc) < 0)
David Howellsa6f76f22008-11-14 10:39:24 +11001303 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001305 bprm->envc = count(envp, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 if ((retval = bprm->envc) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 goto out;
1308
1309 retval = prepare_binprm(bprm);
1310 if (retval < 0)
1311 goto out;
1312
1313 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1314 if (retval < 0)
1315 goto out;
1316
1317 bprm->exec = bprm->p;
1318 retval = copy_strings(bprm->envc, envp, bprm);
1319 if (retval < 0)
1320 goto out;
1321
1322 retval = copy_strings(bprm->argc, argv, bprm);
1323 if (retval < 0)
1324 goto out;
1325
Oleg Nesterov7b34e422008-07-25 01:47:37 -07001326 current->flags &= ~PF_KTHREAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 retval = search_binary_handler(bprm,regs);
David Howellsa6f76f22008-11-14 10:39:24 +11001328 if (retval < 0)
1329 goto out;
1330
1331 /* execve succeeded */
1332 mutex_unlock(&current->cred_exec_mutex);
1333 acct_update_integrals(current);
1334 free_bprm(bprm);
1335 if (displaced)
1336 put_files_struct(displaced);
1337 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
1339out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 if (bprm->mm)
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001341 mmput (bprm->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
1343out_file:
1344 if (bprm->file) {
1345 allow_write_access(bprm->file);
1346 fput(bprm->file);
1347 }
David Howellsa6f76f22008-11-14 10:39:24 +11001348
1349out_unlock:
1350 mutex_unlock(&current->cred_exec_mutex);
1351
1352out_free:
Al Viro08a6fac2008-05-10 16:38:25 -04001353 free_bprm(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
Al Virofd8328b2008-04-22 05:11:59 -04001355out_files:
Al Viro3b125382008-04-22 05:31:30 -04001356 if (displaced)
1357 reset_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358out_ret:
1359 return retval;
1360}
1361
1362int set_binfmt(struct linux_binfmt *new)
1363{
1364 struct linux_binfmt *old = current->binfmt;
1365
1366 if (new) {
1367 if (!try_module_get(new->module))
1368 return -1;
1369 }
1370 current->binfmt = new;
1371 if (old)
1372 module_put(old->module);
1373 return 0;
1374}
1375
1376EXPORT_SYMBOL(set_binfmt);
1377
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378/* format_corename will inspect the pattern parameter, and output a
1379 * name into corename, which must have space for at least
1380 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1381 */
Oleg Nesterov64093242008-10-18 20:28:22 -07001382static int format_corename(char *corename, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383{
David Howells86a264a2008-11-14 10:39:18 +11001384 const struct cred *cred = current_cred();
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001385 const char *pat_ptr = core_pattern;
1386 int ispipe = (*pat_ptr == '|');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 char *out_ptr = corename;
1388 char *const out_end = corename + CORENAME_MAX_SIZE;
1389 int rc;
1390 int pid_in_pattern = 0;
1391
1392 /* Repeat as long as we have more pattern to process and more output
1393 space */
1394 while (*pat_ptr) {
1395 if (*pat_ptr != '%') {
1396 if (out_ptr == out_end)
1397 goto out;
1398 *out_ptr++ = *pat_ptr++;
1399 } else {
1400 switch (*++pat_ptr) {
1401 case 0:
1402 goto out;
1403 /* Double percent, output one percent */
1404 case '%':
1405 if (out_ptr == out_end)
1406 goto out;
1407 *out_ptr++ = '%';
1408 break;
1409 /* pid */
1410 case 'p':
1411 pid_in_pattern = 1;
1412 rc = snprintf(out_ptr, out_end - out_ptr,
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001413 "%d", task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 if (rc > out_end - out_ptr)
1415 goto out;
1416 out_ptr += rc;
1417 break;
1418 /* uid */
1419 case 'u':
1420 rc = snprintf(out_ptr, out_end - out_ptr,
David Howells86a264a2008-11-14 10:39:18 +11001421 "%d", cred->uid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 if (rc > out_end - out_ptr)
1423 goto out;
1424 out_ptr += rc;
1425 break;
1426 /* gid */
1427 case 'g':
1428 rc = snprintf(out_ptr, out_end - out_ptr,
David Howells86a264a2008-11-14 10:39:18 +11001429 "%d", cred->gid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 if (rc > out_end - out_ptr)
1431 goto out;
1432 out_ptr += rc;
1433 break;
1434 /* signal that caused the coredump */
1435 case 's':
1436 rc = snprintf(out_ptr, out_end - out_ptr,
1437 "%ld", signr);
1438 if (rc > out_end - out_ptr)
1439 goto out;
1440 out_ptr += rc;
1441 break;
1442 /* UNIX time of coredump */
1443 case 't': {
1444 struct timeval tv;
1445 do_gettimeofday(&tv);
1446 rc = snprintf(out_ptr, out_end - out_ptr,
1447 "%lu", tv.tv_sec);
1448 if (rc > out_end - out_ptr)
1449 goto out;
1450 out_ptr += rc;
1451 break;
1452 }
1453 /* hostname */
1454 case 'h':
1455 down_read(&uts_sem);
1456 rc = snprintf(out_ptr, out_end - out_ptr,
Serge E. Hallyne9ff3992006-10-02 02:18:11 -07001457 "%s", utsname()->nodename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 up_read(&uts_sem);
1459 if (rc > out_end - out_ptr)
1460 goto out;
1461 out_ptr += rc;
1462 break;
1463 /* executable */
1464 case 'e':
1465 rc = snprintf(out_ptr, out_end - out_ptr,
1466 "%s", current->comm);
1467 if (rc > out_end - out_ptr)
1468 goto out;
1469 out_ptr += rc;
1470 break;
Neil Horman74aadce2007-10-16 23:26:35 -07001471 /* core limit size */
1472 case 'c':
1473 rc = snprintf(out_ptr, out_end - out_ptr,
1474 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1475 if (rc > out_end - out_ptr)
1476 goto out;
1477 out_ptr += rc;
1478 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 default:
1480 break;
1481 }
1482 ++pat_ptr;
1483 }
1484 }
1485 /* Backward compatibility with core_uses_pid:
1486 *
1487 * If core_pattern does not include a %p (as is the default)
1488 * and core_uses_pid is set, then .%pid will be appended to
Alan Coxc4bbafd2007-04-16 22:53:13 -07001489 * the filename. Do not do this for piped commands. */
Oleg Nesterov64093242008-10-18 20:28:22 -07001490 if (!ispipe && !pid_in_pattern && core_uses_pid) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 rc = snprintf(out_ptr, out_end - out_ptr,
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001492 ".%d", task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 if (rc > out_end - out_ptr)
1494 goto out;
1495 out_ptr += rc;
1496 }
Alan Coxc4bbafd2007-04-16 22:53:13 -07001497out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 *out_ptr = 0;
Alan Coxc4bbafd2007-04-16 22:53:13 -07001499 return ispipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500}
1501
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001502static int zap_process(struct task_struct *start)
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001503{
1504 struct task_struct *t;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001505 int nr = 0;
Oleg Nesterov281de332006-06-26 00:26:06 -07001506
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001507 start->signal->flags = SIGNAL_GROUP_EXIT;
1508 start->signal->group_stop_count = 0;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001509
1510 t = start;
1511 do {
1512 if (t != current && t->mm) {
Oleg Nesterov281de332006-06-26 00:26:06 -07001513 sigaddset(&t->pending.signal, SIGKILL);
1514 signal_wake_up(t, 1);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001515 nr++;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001516 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001517 } while_each_thread(start, t);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001518
1519 return nr;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001520}
1521
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001522static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001523 struct core_state *core_state, int exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
1525 struct task_struct *g, *p;
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001526 unsigned long flags;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001527 int nr = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001529 spin_lock_irq(&tsk->sighand->siglock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001530 if (!signal_group_exit(tsk->signal)) {
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001531 mm->core_state = core_state;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001532 tsk->signal->group_exit_code = exit_code;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001533 nr = zap_process(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 }
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001535 spin_unlock_irq(&tsk->sighand->siglock);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001536 if (unlikely(nr < 0))
1537 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001539 if (atomic_read(&mm->mm_users) == nr + 1)
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001540 goto done;
Oleg Nesterove4901f92008-07-25 01:47:31 -07001541 /*
1542 * We should find and kill all tasks which use this mm, and we should
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001543 * count them correctly into ->nr_threads. We don't take tasklist
Oleg Nesterove4901f92008-07-25 01:47:31 -07001544 * lock, but this is safe wrt:
1545 *
1546 * fork:
1547 * None of sub-threads can fork after zap_process(leader). All
1548 * processes which were created before this point should be
1549 * visible to zap_threads() because copy_process() adds the new
1550 * process to the tail of init_task.tasks list, and lock/unlock
1551 * of ->siglock provides a memory barrier.
1552 *
1553 * do_exit:
1554 * The caller holds mm->mmap_sem. This means that the task which
1555 * uses this mm can't pass exit_mm(), so it can't exit or clear
1556 * its ->mm.
1557 *
1558 * de_thread:
1559 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1560 * we must see either old or new leader, this does not matter.
1561 * However, it can change p->sighand, so lock_task_sighand(p)
1562 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1563 * it can't fail.
1564 *
1565 * Note also that "g" can be the old leader with ->mm == NULL
1566 * and already unhashed and thus removed from ->thread_group.
1567 * This is OK, __unhash_process()->list_del_rcu() does not
1568 * clear the ->next pointer, we will find the new leader via
1569 * next_thread().
1570 */
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001571 rcu_read_lock();
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001572 for_each_process(g) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001573 if (g == tsk->group_leader)
1574 continue;
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001575 if (g->flags & PF_KTHREAD)
1576 continue;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001577 p = g;
1578 do {
1579 if (p->mm) {
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001580 if (unlikely(p->mm == mm)) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001581 lock_task_sighand(p, &flags);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001582 nr += zap_process(p);
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001583 unlock_task_sighand(p, &flags);
1584 }
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001585 break;
1586 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001587 } while_each_thread(g, p);
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001588 }
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001589 rcu_read_unlock();
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001590done:
Oleg Nesterovc5f1cc82008-07-25 01:47:42 -07001591 atomic_set(&core_state->nr_threads, nr);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001592 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593}
1594
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001595static int coredump_wait(int exit_code, struct core_state *core_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596{
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001597 struct task_struct *tsk = current;
1598 struct mm_struct *mm = tsk->mm;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001599 struct completion *vfork_done;
Oleg Nesterov2384f552005-10-30 15:02:47 -08001600 int core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001602 init_completion(&core_state->startup);
Oleg Nesterovb564daf2008-07-25 01:47:44 -07001603 core_state->dumper.task = tsk;
1604 core_state->dumper.next = NULL;
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001605 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001606 up_write(&mm->mmap_sem);
1607
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001608 if (unlikely(core_waiters < 0))
1609 goto fail;
1610
1611 /*
1612 * Make sure nobody is waiting for us to release the VM,
1613 * otherwise we can deadlock when we wait on each other
1614 */
1615 vfork_done = tsk->vfork_done;
1616 if (vfork_done) {
1617 tsk->vfork_done = NULL;
1618 complete(vfork_done);
1619 }
1620
Oleg Nesterov2384f552005-10-30 15:02:47 -08001621 if (core_waiters)
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001622 wait_for_completion(&core_state->startup);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001623fail:
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001624 return core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625}
1626
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001627static void coredump_finish(struct mm_struct *mm)
1628{
1629 struct core_thread *curr, *next;
1630 struct task_struct *task;
1631
1632 next = mm->core_state->dumper.next;
1633 while ((curr = next) != NULL) {
1634 next = curr->next;
1635 task = curr->task;
1636 /*
1637 * see exit_mm(), curr->task must not see
1638 * ->task == NULL before we read ->next.
1639 */
1640 smp_mb();
1641 curr->task = NULL;
1642 wake_up_process(task);
1643 }
1644
1645 mm->core_state = NULL;
1646}
1647
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001648/*
1649 * set_dumpable converts traditional three-value dumpable to two flags and
1650 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1651 * these bits are not changed atomically. So get_dumpable can observe the
1652 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1653 * return either old dumpable or new one by paying attention to the order of
1654 * modifying the bits.
1655 *
1656 * dumpable | mm->flags (binary)
1657 * old new | initial interim final
1658 * ---------+-----------------------
1659 * 0 1 | 00 01 01
1660 * 0 2 | 00 10(*) 11
1661 * 1 0 | 01 00 00
1662 * 1 2 | 01 11 11
1663 * 2 0 | 11 10(*) 00
1664 * 2 1 | 11 11 01
1665 *
1666 * (*) get_dumpable regards interim value of 10 as 11.
1667 */
1668void set_dumpable(struct mm_struct *mm, int value)
1669{
1670 switch (value) {
1671 case 0:
1672 clear_bit(MMF_DUMPABLE, &mm->flags);
1673 smp_wmb();
1674 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1675 break;
1676 case 1:
1677 set_bit(MMF_DUMPABLE, &mm->flags);
1678 smp_wmb();
1679 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1680 break;
1681 case 2:
1682 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1683 smp_wmb();
1684 set_bit(MMF_DUMPABLE, &mm->flags);
1685 break;
1686 }
1687}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001688
1689int get_dumpable(struct mm_struct *mm)
1690{
1691 int ret;
1692
1693 ret = mm->flags & 0x3;
1694 return (ret >= 2) ? 2 : ret;
1695}
1696
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1698{
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001699 struct core_state core_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 char corename[CORENAME_MAX_SIZE + 1];
1701 struct mm_struct *mm = current->mm;
1702 struct linux_binfmt * binfmt;
1703 struct inode * inode;
1704 struct file * file;
David Howellsd84f4f92008-11-14 10:39:23 +11001705 const struct cred *old_cred;
1706 struct cred *cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 int retval = 0;
Alan Coxd6e71142005-06-23 00:09:43 -07001708 int flag = 0;
Andi Kleend025c9d2006-09-30 23:29:28 -07001709 int ispipe = 0;
Neil Horman7dc0b222007-10-16 23:26:34 -07001710 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
Neil Horman74aadce2007-10-16 23:26:35 -07001711 char **helper_argv = NULL;
1712 int helper_argc = 0;
1713 char *delimit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
Steve Grubb0a4ff8c2007-04-19 10:28:21 -04001715 audit_core_dumps(signr);
1716
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 binfmt = current->binfmt;
1718 if (!binfmt || !binfmt->core_dump)
1719 goto fail;
David Howellsd84f4f92008-11-14 10:39:23 +11001720
1721 cred = prepare_creds();
1722 if (!cred) {
1723 retval = -ENOMEM;
1724 goto fail;
1725 }
1726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 down_write(&mm->mmap_sem);
Roland McGrath00ec99d2007-11-11 19:13:43 -08001728 /*
1729 * If another thread got here first, or we are not dumpable, bail out.
1730 */
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001731 if (mm->core_state || !get_dumpable(mm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 up_write(&mm->mmap_sem);
David Howellsd84f4f92008-11-14 10:39:23 +11001733 put_cred(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 goto fail;
1735 }
Alan Coxd6e71142005-06-23 00:09:43 -07001736
1737 /*
1738 * We cannot trust fsuid as being the "true" uid of the
1739 * process nor do we know its entire history. We only know it
1740 * was tainted so we dump it as root in mode 2.
1741 */
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001742 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
Alan Coxd6e71142005-06-23 00:09:43 -07001743 flag = O_EXCL; /* Stop rewrite attacks */
David Howellsd84f4f92008-11-14 10:39:23 +11001744 cred->fsuid = 0; /* Dump root private */
Alan Coxd6e71142005-06-23 00:09:43 -07001745 }
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001746
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001747 retval = coredump_wait(exit_code, &core_state);
David Howellsd84f4f92008-11-14 10:39:23 +11001748 if (retval < 0) {
1749 put_cred(cred);
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001750 goto fail;
David Howellsd84f4f92008-11-14 10:39:23 +11001751 }
1752
1753 old_cred = override_creds(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754
1755 /*
1756 * Clear any false indication of pending signals that might
1757 * be seen by the filesystem code called to write the core file.
1758 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 clear_thread_flag(TIF_SIGPENDING);
1760
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 /*
1762 * lock_kernel() because format_corename() is controlled by sysctl, which
1763 * uses lock_kernel()
1764 */
1765 lock_kernel();
Oleg Nesterov64093242008-10-18 20:28:22 -07001766 ispipe = format_corename(corename, signr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 unlock_kernel();
Neil Horman7dc0b222007-10-16 23:26:34 -07001768 /*
1769 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1770 * to a pipe. Since we're not writing directly to the filesystem
1771 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1772 * created unless the pipe reader choses to write out the core file
1773 * at which point file size limits and permissions will be imposed
1774 * as it does with any other process
1775 */
Neil Horman74aadce2007-10-16 23:26:35 -07001776 if ((!ispipe) && (core_limit < binfmt->min_coredump))
Neil Horman7dc0b222007-10-16 23:26:34 -07001777 goto fail_unlock;
1778
Alan Coxc4bbafd2007-04-16 22:53:13 -07001779 if (ispipe) {
Neil Horman74aadce2007-10-16 23:26:35 -07001780 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1781 /* Terminate the string before the first option */
1782 delimit = strchr(corename, ' ');
1783 if (delimit)
1784 *delimit = '\0';
Neil Horman32321132007-10-16 23:26:36 -07001785 delimit = strrchr(helper_argv[0], '/');
1786 if (delimit)
1787 delimit++;
1788 else
1789 delimit = helper_argv[0];
1790 if (!strcmp(delimit, current->comm)) {
1791 printk(KERN_NOTICE "Recursive core dump detected, "
1792 "aborting\n");
1793 goto fail_unlock;
1794 }
1795
1796 core_limit = RLIM_INFINITY;
1797
Andi Kleend025c9d2006-09-30 23:29:28 -07001798 /* SIGPIPE can happen, but it's just never processed */
Neil Horman32321132007-10-16 23:26:36 -07001799 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1800 &file)) {
Andi Kleend025c9d2006-09-30 23:29:28 -07001801 printk(KERN_INFO "Core dump to %s pipe failed\n",
1802 corename);
1803 goto fail_unlock;
1804 }
Andi Kleend025c9d2006-09-30 23:29:28 -07001805 } else
1806 file = filp_open(corename,
Alexey Dobriyan6d4df672006-12-06 20:40:39 -08001807 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1808 0600);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 if (IS_ERR(file))
1810 goto fail_unlock;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001811 inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 if (inode->i_nlink > 1)
1813 goto close_fail; /* multiple links - don't dump */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001814 if (!ispipe && d_unhashed(file->f_path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 goto close_fail;
1816
Andi Kleend025c9d2006-09-30 23:29:28 -07001817 /* AK: actually i see no reason to not allow this for named pipes etc.,
1818 but keep the previous behaviour for now. */
1819 if (!ispipe && !S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 goto close_fail;
Ingo Molnarc46f7392007-11-28 13:59:18 +01001821 /*
1822 * Dont allow local users get cute and trick others to coredump
1823 * into their pre-created files:
1824 */
David Howellsda9592e2008-11-14 10:39:05 +11001825 if (inode->i_uid != current_fsuid())
Ingo Molnarc46f7392007-11-28 13:59:18 +01001826 goto close_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 if (!file->f_op)
1828 goto close_fail;
1829 if (!file->f_op->write)
1830 goto close_fail;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001831 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 goto close_fail;
1833
Neil Horman7dc0b222007-10-16 23:26:34 -07001834 retval = binfmt->core_dump(signr, regs, file, core_limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
1836 if (retval)
1837 current->signal->group_exit_code |= 0x80;
1838close_fail:
1839 filp_close(file, NULL);
1840fail_unlock:
Neil Horman74aadce2007-10-16 23:26:35 -07001841 if (helper_argv)
1842 argv_free(helper_argv);
1843
David Howellsd84f4f92008-11-14 10:39:23 +11001844 revert_creds(old_cred);
1845 put_cred(cred);
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001846 coredump_finish(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847fail:
1848 return retval;
1849}