blob: b4e5b8a9216a29b4531a701b3d76247b6ee5471f [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55#include <asm/uaccess.h>
56#include <asm/mmu_context.h>
Ollie Wildb6a2fea2007-07-19 01:48:16 -070057#include <asm/tlb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
David Woodhouse702773b2008-06-16 12:11:54 +010059#ifdef __alpha__
60/* for /sbin/loader handling in search_binary_handler() */
61#include <linux/a.out.h>
62#endif
63
Linus Torvalds1da177e2005-04-16 15:20:36 -070064int core_uses_pid;
Dan Aloni71ce92f2007-05-16 22:11:16 -070065char core_pattern[CORENAME_MAX_SIZE] = "core";
Alan Coxd6e71142005-06-23 00:09:43 -070066int suid_dumpable = 0;
67
Linus Torvalds1da177e2005-04-16 15:20:36 -070068/* The maximal length of core_pattern is also specified in sysctl.c */
69
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070070static LIST_HEAD(formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071static DEFINE_RWLOCK(binfmt_lock);
72
73int register_binfmt(struct linux_binfmt * fmt)
74{
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 if (!fmt)
76 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 write_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070078 list_add(&fmt->lh, &formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 write_unlock(&binfmt_lock);
80 return 0;
81}
82
83EXPORT_SYMBOL(register_binfmt);
84
Alexey Dobriyanf6b450d2007-10-16 23:26:04 -070085void unregister_binfmt(struct linux_binfmt * fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086{
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 write_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070088 list_del(&fmt->lh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 write_unlock(&binfmt_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090}
91
92EXPORT_SYMBOL(unregister_binfmt);
93
94static inline void put_binfmt(struct linux_binfmt * fmt)
95{
96 module_put(fmt->module);
97}
98
99/*
100 * Note that a shared library must be both readable and executable due to
101 * security reasons.
102 *
103 * Also note that we take the address to load from from the file itself.
104 */
105asmlinkage long sys_uselib(const char __user * library)
106{
Al Viro964bd182008-07-26 03:33:14 -0400107 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 struct nameidata nd;
Al Viro964bd182008-07-26 03:33:14 -0400109 char *tmp = getname(library);
110 int error = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
Al Viro964bd182008-07-26 03:33:14 -0400112 if (!IS_ERR(tmp)) {
113 error = path_lookup_open(AT_FDCWD, tmp,
114 LOOKUP_FOLLOW, &nd,
115 FMODE_READ|FMODE_EXEC);
116 putname(tmp);
117 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 if (error)
119 goto out;
120
121 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -0800122 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 goto exit;
124
Al Viro30524472008-07-22 00:02:33 -0400125 error = -EACCES;
126 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
127 goto exit;
128
Al Virob77b0642008-07-17 09:37:02 -0400129 error = vfs_permission(&nd, MAY_READ | MAY_EXEC | MAY_OPEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 if (error)
131 goto exit;
132
Andi Kleenabe8be32008-02-08 04:20:23 -0800133 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 error = PTR_ERR(file);
135 if (IS_ERR(file))
136 goto out;
137
138 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 Hellwige56b6a52008-05-19 07:53:34 +0200682 err = vfs_permission(&nd, MAY_EXEC | MAY_OPEN);
683 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
690 err = deny_write_access(file);
691 if (err) {
692 fput(file);
693 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200696 return file;
697
698 out_path_put:
699 release_open_intent(&nd);
700 path_put(&nd.path);
701 out:
702 return ERR_PTR(err);
703}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704EXPORT_SYMBOL(open_exec);
705
706int kernel_read(struct file *file, unsigned long offset,
707 char *addr, unsigned long count)
708{
709 mm_segment_t old_fs;
710 loff_t pos = offset;
711 int result;
712
713 old_fs = get_fs();
714 set_fs(get_ds());
715 /* The cast to a user pointer is valid due to the set_fs() */
716 result = vfs_read(file, (void __user *)addr, count, &pos);
717 set_fs(old_fs);
718 return result;
719}
720
721EXPORT_SYMBOL(kernel_read);
722
723static int exec_mmap(struct mm_struct *mm)
724{
725 struct task_struct *tsk;
726 struct mm_struct * old_mm, *active_mm;
727
728 /* Notify parent that we're no longer interested in the old VM */
729 tsk = current;
730 old_mm = current->mm;
731 mm_release(tsk, old_mm);
732
733 if (old_mm) {
734 /*
735 * Make sure that if there is a core dump in progress
736 * for the old mm, we get out and die instead of going
737 * through with the exec. We must hold mmap_sem around
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700738 * checking core_state and changing tsk->mm.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 */
740 down_read(&old_mm->mmap_sem);
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700741 if (unlikely(old_mm->core_state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 up_read(&old_mm->mmap_sem);
743 return -EINTR;
744 }
745 }
746 task_lock(tsk);
747 active_mm = tsk->active_mm;
748 tsk->mm = mm;
749 tsk->active_mm = mm;
750 activate_mm(active_mm, mm);
751 task_unlock(tsk);
752 arch_pick_mmap_layout(mm);
753 if (old_mm) {
754 up_read(&old_mm->mmap_sem);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +0200755 BUG_ON(active_mm != old_mm);
Balbir Singh31a78f22008-09-28 23:09:31 +0100756 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 mmput(old_mm);
758 return 0;
759 }
760 mmdrop(active_mm);
761 return 0;
762}
763
764/*
765 * This function makes sure the current process has its own signal table,
766 * so that flush_signal_handlers can later reset the handlers without
767 * disturbing other processes. (Other processes might share the signal
768 * table via the CLONE_SIGHAND option to clone().)
769 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800770static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771{
772 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700773 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 spinlock_t *lock = &oldsighand->siglock;
775 int count;
776
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700777 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 goto no_thread_group;
779
780 /*
781 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800784 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 /*
786 * Another group action in progress, just
787 * return so that the signal is processed.
788 */
789 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 return -EAGAIN;
791 }
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800792 sig->group_exit_task = tsk;
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700793 zap_other_threads(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
Oleg Nesterovfea9d172008-02-08 04:19:19 -0800795 /* Account for the thread group leader hanging around: */
796 count = thread_group_leader(tsk) ? 1 : 2;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700797 sig->notify_count = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 while (atomic_read(&sig->count) > count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 __set_current_state(TASK_UNINTERRUPTIBLE);
800 spin_unlock_irq(lock);
801 schedule();
802 spin_lock_irq(lock);
803 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 spin_unlock_irq(lock);
805
806 /*
807 * At this point all other threads have exited, all we have to
808 * do is to wait for the thread group leader to become inactive,
809 * and to assume its PID:
810 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700811 if (!thread_group_leader(tsk)) {
Oleg Nesterov81879262008-12-01 14:18:16 -0800812 struct task_struct *leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700813
Oleg Nesterov2800d8d2008-04-30 00:53:12 -0700814 sig->notify_count = -1; /* for exit_notify() */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700815 for (;;) {
816 write_lock_irq(&tasklist_lock);
817 if (likely(leader->exit_state))
818 break;
819 __set_current_state(TASK_UNINTERRUPTIBLE);
820 write_unlock_irq(&tasklist_lock);
821 schedule();
822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823
Roland McGrathf5e90282006-04-10 22:54:16 -0700824 /*
825 * The only record we have of the real-time age of a
826 * process, regardless of execs it's done, is start_time.
827 * All the past CPU time is accumulated in signal_struct
828 * from sister threads now dead. But in this non-leader
829 * exec, nothing survives from the original leader thread,
830 * whose birth marks the true age of this process now.
831 * When we take on its identity by switching to its PID, we
832 * also take its birthdate (always earlier than our own).
833 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700834 tsk->start_time = leader->start_time;
Roland McGrathf5e90282006-04-10 22:54:16 -0700835
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700836 BUG_ON(!same_thread_group(leader, tsk));
837 BUG_ON(has_group_leader_pid(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 /*
839 * An exec() starts a new thread group with the
840 * TGID of the previous thread group. Rehash the
841 * two threads with a switched PID, and release
842 * the former thread group leader:
843 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800844
845 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700846 * The old leader becomes a thread of the this thread group.
847 * Note: The old leader also uses this pid until release_task
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800848 * is called. Odd but simple and correct.
849 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700850 detach_pid(tsk, PIDTYPE_PID);
851 tsk->pid = leader->pid;
Sukadev Bhattiprolu3743ca02007-10-18 23:39:51 -0700852 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700853 transfer_pid(leader, tsk, PIDTYPE_PGID);
854 transfer_pid(leader, tsk, PIDTYPE_SID);
855 list_replace_rcu(&leader->tasks, &tsk->tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700857 tsk->group_leader = tsk;
858 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -0600859
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700860 tsk->exit_signal = SIGCHLD;
Oleg Nesterov962b5642005-11-23 13:37:43 -0800861
862 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
863 leader->exit_state = EXIT_DEAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 write_unlock_irq(&tasklist_lock);
Oleg Nesterov81879262008-12-01 14:18:16 -0800865
866 release_task(leader);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800867 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700869 sig->group_exit_task = NULL;
870 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
872no_thread_group:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 exit_itimers(sig);
Oleg Nesterovcbaffba2008-05-26 20:55:42 +0400874 flush_itimer_signals();
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300875
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700876 if (atomic_read(&oldsighand->count) != 1) {
877 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700879 * This ->sighand is shared with the CLONE_SIGHAND
880 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700882 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
883 if (!newsighand)
884 return -ENOMEM;
885
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 atomic_set(&newsighand->count, 1);
887 memcpy(newsighand->action, oldsighand->action,
888 sizeof(newsighand->action));
889
890 write_lock_irq(&tasklist_lock);
891 spin_lock(&oldsighand->siglock);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700892 rcu_assign_pointer(tsk->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 spin_unlock(&oldsighand->siglock);
894 write_unlock_irq(&tasklist_lock);
895
Davide Libenzifba2afa2007-05-10 22:23:13 -0700896 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 }
898
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700899 BUG_ON(!thread_group_leader(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 return 0;
901}
Oleg Nesterov0840a902007-10-16 23:27:22 -0700902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903/*
904 * These functions flushes out all traces of the currently running executable
905 * so that a new one can be started
906 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800907static void flush_old_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908{
909 long j = -1;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700910 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
912 spin_lock(&files->file_lock);
913 for (;;) {
914 unsigned long set, i;
915
916 j++;
917 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700918 fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800919 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700921 set = fdt->close_on_exec->fds_bits[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 if (!set)
923 continue;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700924 fdt->close_on_exec->fds_bits[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 spin_unlock(&files->file_lock);
926 for ( ; set ; i++,set >>= 1) {
927 if (set & 1) {
928 sys_close(i);
929 }
930 }
931 spin_lock(&files->file_lock);
932
933 }
934 spin_unlock(&files->file_lock);
935}
936
Andrew Morton59714d62008-02-04 22:27:21 -0800937char *get_task_comm(char *buf, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938{
939 /* buf must be at least sizeof(tsk->comm) in size */
940 task_lock(tsk);
941 strncpy(buf, tsk->comm, sizeof(tsk->comm));
942 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -0800943 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944}
945
946void set_task_comm(struct task_struct *tsk, char *buf)
947{
948 task_lock(tsk);
949 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
950 task_unlock(tsk);
951}
952
953int flush_old_exec(struct linux_binprm * bprm)
954{
955 char * name;
956 int i, ch, retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 char tcomm[sizeof(current->comm)];
958
959 /*
960 * Make sure we have a private signal table and that
961 * we are unassociated from the previous thread group.
962 */
963 retval = de_thread(current);
964 if (retval)
965 goto out;
966
Matt Helsley925d1c42008-04-29 01:01:36 -0700967 set_mm_exe_file(bprm->mm, bprm->file);
968
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 * Release all of the old mmap stuff
971 */
972 retval = exec_mmap(bprm->mm);
973 if (retval)
Al Virofd8328b2008-04-22 05:11:59 -0400974 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
976 bprm->mm = NULL; /* We're using it now */
977
978 /* This is the point of no return */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 current->sas_ss_sp = current->sas_ss_size = 0;
980
981 if (current->euid == current->uid && current->egid == current->gid)
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -0700982 set_dumpable(current->mm, 1);
Alan Coxd6e71142005-06-23 00:09:43 -0700983 else
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -0700984 set_dumpable(current->mm, suid_dumpable);
Alan Coxd6e71142005-06-23 00:09:43 -0700985
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 name = bprm->filename;
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -0700987
988 /* Copies the binary name from after last slash */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 for (i=0; (ch = *(name++)) != '\0';) {
990 if (ch == '/')
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -0700991 i = 0; /* overwrite what we wrote */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 else
993 if (i < (sizeof(tcomm) - 1))
994 tcomm[i++] = ch;
995 }
996 tcomm[i] = '\0';
997 set_task_comm(current, tcomm);
998
999 current->flags &= ~PF_RANDOMIZE;
1000 flush_thread();
1001
Benjamin Herrenschmidt0551fbd2006-02-28 16:59:19 -08001002 /* Set the new mm task size. We have to do that late because it may
1003 * depend on TIF_32BIT which is only updated in flush_thread() on
1004 * some architectures like powerpc
1005 */
1006 current->mm->task_size = TASK_SIZE;
1007
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001008 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1009 suid_keys(current);
1010 set_dumpable(current->mm, suid_dumpable);
1011 current->pdeath_signal = 0;
1012 } else if (file_permission(bprm->file, MAY_READ) ||
1013 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 suid_keys(current);
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001015 set_dumpable(current->mm, suid_dumpable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 }
1017
1018 /* An exec changes our domain. We are no longer part of the thread
1019 group */
1020
1021 current->self_exec_id++;
1022
1023 flush_signal_handlers(current, 0);
1024 flush_old_files(current->files);
1025
1026 return 0;
1027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028out:
1029 return retval;
1030}
1031
1032EXPORT_SYMBOL(flush_old_exec);
1033
1034/*
1035 * Fill the binprm structure from the inode.
1036 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1037 */
1038int prepare_binprm(struct linux_binprm *bprm)
1039{
1040 int mode;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001041 struct inode * inode = bprm->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 int retval;
1043
1044 mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 if (bprm->file->f_op == NULL)
1046 return -EACCES;
1047
1048 bprm->e_uid = current->euid;
1049 bprm->e_gid = current->egid;
1050
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001051 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 /* Set-uid? */
1053 if (mode & S_ISUID) {
1054 current->personality &= ~PER_CLEAR_ON_SETID;
1055 bprm->e_uid = inode->i_uid;
1056 }
1057
1058 /* Set-gid? */
1059 /*
1060 * If setgid is set but no group execute bit then this
1061 * is a candidate for mandatory locking, not a setgid
1062 * executable.
1063 */
1064 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1065 current->personality &= ~PER_CLEAR_ON_SETID;
1066 bprm->e_gid = inode->i_gid;
1067 }
1068 }
1069
1070 /* fill in binprm security blob */
1071 retval = security_bprm_set(bprm);
1072 if (retval)
1073 return retval;
1074
1075 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1076 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1077}
1078
1079EXPORT_SYMBOL(prepare_binprm);
1080
Arjan van de Ven858119e2006-01-14 13:20:43 -08001081static int unsafe_exec(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082{
Roland McGrath6341c392008-07-25 19:45:44 -07001083 int unsafe = tracehook_unsafe_exec(p);
1084
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 if (atomic_read(&p->fs->count) > 1 ||
1086 atomic_read(&p->files->count) > 1 ||
1087 atomic_read(&p->sighand->count) > 1)
1088 unsafe |= LSM_UNSAFE_SHARE;
1089
1090 return unsafe;
1091}
1092
1093void compute_creds(struct linux_binprm *bprm)
1094{
1095 int unsafe;
1096
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001097 if (bprm->e_uid != current->uid) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 suid_keys(current);
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001099 current->pdeath_signal = 0;
1100 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 exec_keys(current);
1102
1103 task_lock(current);
1104 unsafe = unsafe_exec(current);
1105 security_bprm_apply_creds(bprm, unsafe);
1106 task_unlock(current);
1107 security_bprm_post_apply_creds(bprm);
1108}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109EXPORT_SYMBOL(compute_creds);
1110
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001111/*
1112 * Arguments are '\0' separated strings found at the location bprm->p
1113 * points to; chop off the first by relocating brpm->p to right after
1114 * the first '\0' encountered.
1115 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001116int remove_arg_zero(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117{
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001118 int ret = 0;
1119 unsigned long offset;
1120 char *kaddr;
1121 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001123 if (!bprm->argc)
1124 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001126 do {
1127 offset = bprm->p & ~PAGE_MASK;
1128 page = get_arg_page(bprm, bprm->p, 0);
1129 if (!page) {
1130 ret = -EFAULT;
1131 goto out;
1132 }
1133 kaddr = kmap_atomic(page, KM_USER0);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001134
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001135 for (; offset < PAGE_SIZE && kaddr[offset];
1136 offset++, bprm->p++)
1137 ;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001138
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001139 kunmap_atomic(kaddr, KM_USER0);
1140 put_arg_page(page);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001141
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001142 if (offset == PAGE_SIZE)
1143 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1144 } while (offset == PAGE_SIZE);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001145
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001146 bprm->p++;
1147 bprm->argc--;
1148 ret = 0;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001149
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001150out:
1151 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153EXPORT_SYMBOL(remove_arg_zero);
1154
1155/*
1156 * cycle the list of binary formats handler, until one recognizes the image
1157 */
1158int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1159{
Roland McGrath85f33462008-12-09 19:36:38 -08001160 unsigned int depth = bprm->recursion_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 int try,retval;
1162 struct linux_binfmt *fmt;
David Woodhouse702773b2008-06-16 12:11:54 +01001163#ifdef __alpha__
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 /* handle /sbin/loader.. */
1165 {
1166 struct exec * eh = (struct exec *) bprm->buf;
1167
1168 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1169 (eh->fh.f_flags & 0x3000) == 0x3000)
1170 {
1171 struct file * file;
1172 unsigned long loader;
1173
1174 allow_write_access(bprm->file);
1175 fput(bprm->file);
1176 bprm->file = NULL;
1177
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001178 loader = bprm->vma->vm_end - sizeof(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179
1180 file = open_exec("/sbin/loader");
1181 retval = PTR_ERR(file);
1182 if (IS_ERR(file))
1183 return retval;
1184
1185 /* Remember if the application is TASO. */
Kirill A. Shutemov53112482008-10-15 22:02:37 -07001186 bprm->taso = eh->ah.entry < 0x100000000UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
1188 bprm->file = file;
1189 bprm->loader = loader;
1190 retval = prepare_binprm(bprm);
1191 if (retval<0)
1192 return retval;
1193 /* should call search_binary_handler recursively here,
1194 but it does not matter */
1195 }
1196 }
1197#endif
1198 retval = security_bprm_check(bprm);
1199 if (retval)
1200 return retval;
1201
1202 /* kernel module loader fixup */
1203 /* so we don't try to load run modprobe in kernel space. */
1204 set_fs(USER_DS);
Al Viro473ae302006-04-26 14:04:08 -04001205
1206 retval = audit_bprm(bprm);
1207 if (retval)
1208 return retval;
1209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 retval = -ENOENT;
1211 for (try=0; try<2; try++) {
1212 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -07001213 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1215 if (!fn)
1216 continue;
1217 if (!try_module_get(fmt->module))
1218 continue;
1219 read_unlock(&binfmt_lock);
1220 retval = fn(bprm, regs);
Roland McGrath85f33462008-12-09 19:36:38 -08001221 /*
1222 * Restore the depth counter to its starting value
1223 * in this call, so we don't have to rely on every
1224 * load_binary function to restore it on return.
1225 */
1226 bprm->recursion_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 if (retval >= 0) {
Roland McGrath85f33462008-12-09 19:36:38 -08001228 if (depth == 0)
1229 tracehook_report_exec(fmt, bprm, regs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 put_binfmt(fmt);
1231 allow_write_access(bprm->file);
1232 if (bprm->file)
1233 fput(bprm->file);
1234 bprm->file = NULL;
1235 current->did_exec = 1;
Matt Helsley9f460802005-11-07 00:59:16 -08001236 proc_exec_connector(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 return retval;
1238 }
1239 read_lock(&binfmt_lock);
1240 put_binfmt(fmt);
1241 if (retval != -ENOEXEC || bprm->mm == NULL)
1242 break;
1243 if (!bprm->file) {
1244 read_unlock(&binfmt_lock);
1245 return retval;
1246 }
1247 }
1248 read_unlock(&binfmt_lock);
1249 if (retval != -ENOEXEC || bprm->mm == NULL) {
1250 break;
Johannes Berg5f4123b2008-07-09 10:28:40 +02001251#ifdef CONFIG_MODULES
1252 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1254 if (printable(bprm->buf[0]) &&
1255 printable(bprm->buf[1]) &&
1256 printable(bprm->buf[2]) &&
1257 printable(bprm->buf[3]))
1258 break; /* -ENOEXEC */
1259 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1260#endif
1261 }
1262 }
1263 return retval;
1264}
1265
1266EXPORT_SYMBOL(search_binary_handler);
1267
Al Viro08a6fac2008-05-10 16:38:25 -04001268void free_bprm(struct linux_binprm *bprm)
1269{
1270 free_arg_pages(bprm);
1271 kfree(bprm);
1272}
1273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274/*
1275 * sys_execve() executes a new program.
1276 */
1277int do_execve(char * filename,
1278 char __user *__user *argv,
1279 char __user *__user *envp,
1280 struct pt_regs * regs)
1281{
1282 struct linux_binprm *bprm;
1283 struct file *file;
Al Viro3b125382008-04-22 05:31:30 -04001284 struct files_struct *displaced;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
Al Viro3b125382008-04-22 05:31:30 -04001287 retval = unshare_files(&displaced);
Al Virofd8328b2008-04-22 05:11:59 -04001288 if (retval)
1289 goto out_ret;
1290
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 retval = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001292 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 if (!bprm)
Al Virofd8328b2008-04-22 05:11:59 -04001294 goto out_files;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
1296 file = open_exec(filename);
1297 retval = PTR_ERR(file);
1298 if (IS_ERR(file))
1299 goto out_kfree;
1300
1301 sched_exec();
1302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 bprm->file = file;
1304 bprm->filename = filename;
1305 bprm->interp = filename;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001306
1307 retval = bprm_mm_init(bprm);
1308 if (retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 goto out_file;
1310
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001311 bprm->argc = count(argv, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 if ((retval = bprm->argc) < 0)
1313 goto out_mm;
1314
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001315 bprm->envc = count(envp, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 if ((retval = bprm->envc) < 0)
1317 goto out_mm;
1318
1319 retval = security_bprm_alloc(bprm);
1320 if (retval)
1321 goto out;
1322
1323 retval = prepare_binprm(bprm);
1324 if (retval < 0)
1325 goto out;
1326
1327 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1328 if (retval < 0)
1329 goto out;
1330
1331 bprm->exec = bprm->p;
1332 retval = copy_strings(bprm->envc, envp, bprm);
1333 if (retval < 0)
1334 goto out;
1335
1336 retval = copy_strings(bprm->argc, argv, bprm);
1337 if (retval < 0)
1338 goto out;
1339
Oleg Nesterov7b34e422008-07-25 01:47:37 -07001340 current->flags &= ~PF_KTHREAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 retval = search_binary_handler(bprm,regs);
1342 if (retval >= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 /* execve success */
1344 security_bprm_free(bprm);
1345 acct_update_integrals(current);
Al Viro08a6fac2008-05-10 16:38:25 -04001346 free_bprm(bprm);
Al Viro3b125382008-04-22 05:31:30 -04001347 if (displaced)
1348 put_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 return retval;
1350 }
1351
1352out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 if (bprm->security)
1354 security_bprm_free(bprm);
1355
1356out_mm:
1357 if (bprm->mm)
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001358 mmput (bprm->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
1360out_file:
1361 if (bprm->file) {
1362 allow_write_access(bprm->file);
1363 fput(bprm->file);
1364 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365out_kfree:
Al Viro08a6fac2008-05-10 16:38:25 -04001366 free_bprm(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Al Virofd8328b2008-04-22 05:11:59 -04001368out_files:
Al Viro3b125382008-04-22 05:31:30 -04001369 if (displaced)
1370 reset_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371out_ret:
1372 return retval;
1373}
1374
1375int set_binfmt(struct linux_binfmt *new)
1376{
1377 struct linux_binfmt *old = current->binfmt;
1378
1379 if (new) {
1380 if (!try_module_get(new->module))
1381 return -1;
1382 }
1383 current->binfmt = new;
1384 if (old)
1385 module_put(old->module);
1386 return 0;
1387}
1388
1389EXPORT_SYMBOL(set_binfmt);
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391/* format_corename will inspect the pattern parameter, and output a
1392 * name into corename, which must have space for at least
1393 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1394 */
Oleg Nesterov64093242008-10-18 20:28:22 -07001395static int format_corename(char *corename, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396{
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001397 const char *pat_ptr = core_pattern;
1398 int ispipe = (*pat_ptr == '|');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 char *out_ptr = corename;
1400 char *const out_end = corename + CORENAME_MAX_SIZE;
1401 int rc;
1402 int pid_in_pattern = 0;
1403
1404 /* Repeat as long as we have more pattern to process and more output
1405 space */
1406 while (*pat_ptr) {
1407 if (*pat_ptr != '%') {
1408 if (out_ptr == out_end)
1409 goto out;
1410 *out_ptr++ = *pat_ptr++;
1411 } else {
1412 switch (*++pat_ptr) {
1413 case 0:
1414 goto out;
1415 /* Double percent, output one percent */
1416 case '%':
1417 if (out_ptr == out_end)
1418 goto out;
1419 *out_ptr++ = '%';
1420 break;
1421 /* pid */
1422 case 'p':
1423 pid_in_pattern = 1;
1424 rc = snprintf(out_ptr, out_end - out_ptr,
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001425 "%d", task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 if (rc > out_end - out_ptr)
1427 goto out;
1428 out_ptr += rc;
1429 break;
1430 /* uid */
1431 case 'u':
1432 rc = snprintf(out_ptr, out_end - out_ptr,
1433 "%d", current->uid);
1434 if (rc > out_end - out_ptr)
1435 goto out;
1436 out_ptr += rc;
1437 break;
1438 /* gid */
1439 case 'g':
1440 rc = snprintf(out_ptr, out_end - out_ptr,
1441 "%d", current->gid);
1442 if (rc > out_end - out_ptr)
1443 goto out;
1444 out_ptr += rc;
1445 break;
1446 /* signal that caused the coredump */
1447 case 's':
1448 rc = snprintf(out_ptr, out_end - out_ptr,
1449 "%ld", signr);
1450 if (rc > out_end - out_ptr)
1451 goto out;
1452 out_ptr += rc;
1453 break;
1454 /* UNIX time of coredump */
1455 case 't': {
1456 struct timeval tv;
1457 do_gettimeofday(&tv);
1458 rc = snprintf(out_ptr, out_end - out_ptr,
1459 "%lu", tv.tv_sec);
1460 if (rc > out_end - out_ptr)
1461 goto out;
1462 out_ptr += rc;
1463 break;
1464 }
1465 /* hostname */
1466 case 'h':
1467 down_read(&uts_sem);
1468 rc = snprintf(out_ptr, out_end - out_ptr,
Serge E. Hallyne9ff3992006-10-02 02:18:11 -07001469 "%s", utsname()->nodename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 up_read(&uts_sem);
1471 if (rc > out_end - out_ptr)
1472 goto out;
1473 out_ptr += rc;
1474 break;
1475 /* executable */
1476 case 'e':
1477 rc = snprintf(out_ptr, out_end - out_ptr,
1478 "%s", current->comm);
1479 if (rc > out_end - out_ptr)
1480 goto out;
1481 out_ptr += rc;
1482 break;
Neil Horman74aadce2007-10-16 23:26:35 -07001483 /* core limit size */
1484 case 'c':
1485 rc = snprintf(out_ptr, out_end - out_ptr,
1486 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1487 if (rc > out_end - out_ptr)
1488 goto out;
1489 out_ptr += rc;
1490 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 default:
1492 break;
1493 }
1494 ++pat_ptr;
1495 }
1496 }
1497 /* Backward compatibility with core_uses_pid:
1498 *
1499 * If core_pattern does not include a %p (as is the default)
1500 * and core_uses_pid is set, then .%pid will be appended to
Alan Coxc4bbafd2007-04-16 22:53:13 -07001501 * the filename. Do not do this for piped commands. */
Oleg Nesterov64093242008-10-18 20:28:22 -07001502 if (!ispipe && !pid_in_pattern && core_uses_pid) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 rc = snprintf(out_ptr, out_end - out_ptr,
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001504 ".%d", task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 if (rc > out_end - out_ptr)
1506 goto out;
1507 out_ptr += rc;
1508 }
Alan Coxc4bbafd2007-04-16 22:53:13 -07001509out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 *out_ptr = 0;
Alan Coxc4bbafd2007-04-16 22:53:13 -07001511 return ispipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512}
1513
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001514static int zap_process(struct task_struct *start)
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001515{
1516 struct task_struct *t;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001517 int nr = 0;
Oleg Nesterov281de332006-06-26 00:26:06 -07001518
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001519 start->signal->flags = SIGNAL_GROUP_EXIT;
1520 start->signal->group_stop_count = 0;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001521
1522 t = start;
1523 do {
1524 if (t != current && t->mm) {
Oleg Nesterov281de332006-06-26 00:26:06 -07001525 sigaddset(&t->pending.signal, SIGKILL);
1526 signal_wake_up(t, 1);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001527 nr++;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001528 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001529 } while_each_thread(start, t);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001530
1531 return nr;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001532}
1533
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001534static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001535 struct core_state *core_state, int exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536{
1537 struct task_struct *g, *p;
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001538 unsigned long flags;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001539 int nr = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001541 spin_lock_irq(&tsk->sighand->siglock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001542 if (!signal_group_exit(tsk->signal)) {
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001543 mm->core_state = core_state;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001544 tsk->signal->group_exit_code = exit_code;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001545 nr = zap_process(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 }
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001547 spin_unlock_irq(&tsk->sighand->siglock);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001548 if (unlikely(nr < 0))
1549 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001551 if (atomic_read(&mm->mm_users) == nr + 1)
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001552 goto done;
Oleg Nesterove4901f92008-07-25 01:47:31 -07001553 /*
1554 * We should find and kill all tasks which use this mm, and we should
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001555 * count them correctly into ->nr_threads. We don't take tasklist
Oleg Nesterove4901f92008-07-25 01:47:31 -07001556 * lock, but this is safe wrt:
1557 *
1558 * fork:
1559 * None of sub-threads can fork after zap_process(leader). All
1560 * processes which were created before this point should be
1561 * visible to zap_threads() because copy_process() adds the new
1562 * process to the tail of init_task.tasks list, and lock/unlock
1563 * of ->siglock provides a memory barrier.
1564 *
1565 * do_exit:
1566 * The caller holds mm->mmap_sem. This means that the task which
1567 * uses this mm can't pass exit_mm(), so it can't exit or clear
1568 * its ->mm.
1569 *
1570 * de_thread:
1571 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1572 * we must see either old or new leader, this does not matter.
1573 * However, it can change p->sighand, so lock_task_sighand(p)
1574 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1575 * it can't fail.
1576 *
1577 * Note also that "g" can be the old leader with ->mm == NULL
1578 * and already unhashed and thus removed from ->thread_group.
1579 * This is OK, __unhash_process()->list_del_rcu() does not
1580 * clear the ->next pointer, we will find the new leader via
1581 * next_thread().
1582 */
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001583 rcu_read_lock();
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001584 for_each_process(g) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001585 if (g == tsk->group_leader)
1586 continue;
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001587 if (g->flags & PF_KTHREAD)
1588 continue;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001589 p = g;
1590 do {
1591 if (p->mm) {
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001592 if (unlikely(p->mm == mm)) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001593 lock_task_sighand(p, &flags);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001594 nr += zap_process(p);
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001595 unlock_task_sighand(p, &flags);
1596 }
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001597 break;
1598 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001599 } while_each_thread(g, p);
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001600 }
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001601 rcu_read_unlock();
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001602done:
Oleg Nesterovc5f1cc82008-07-25 01:47:42 -07001603 atomic_set(&core_state->nr_threads, nr);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001604 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605}
1606
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001607static int coredump_wait(int exit_code, struct core_state *core_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608{
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001609 struct task_struct *tsk = current;
1610 struct mm_struct *mm = tsk->mm;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001611 struct completion *vfork_done;
Oleg Nesterov2384f552005-10-30 15:02:47 -08001612 int core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001614 init_completion(&core_state->startup);
Oleg Nesterovb564daf2008-07-25 01:47:44 -07001615 core_state->dumper.task = tsk;
1616 core_state->dumper.next = NULL;
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001617 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001618 up_write(&mm->mmap_sem);
1619
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001620 if (unlikely(core_waiters < 0))
1621 goto fail;
1622
1623 /*
1624 * Make sure nobody is waiting for us to release the VM,
1625 * otherwise we can deadlock when we wait on each other
1626 */
1627 vfork_done = tsk->vfork_done;
1628 if (vfork_done) {
1629 tsk->vfork_done = NULL;
1630 complete(vfork_done);
1631 }
1632
Oleg Nesterov2384f552005-10-30 15:02:47 -08001633 if (core_waiters)
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001634 wait_for_completion(&core_state->startup);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001635fail:
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001636 return core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637}
1638
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001639static void coredump_finish(struct mm_struct *mm)
1640{
1641 struct core_thread *curr, *next;
1642 struct task_struct *task;
1643
1644 next = mm->core_state->dumper.next;
1645 while ((curr = next) != NULL) {
1646 next = curr->next;
1647 task = curr->task;
1648 /*
1649 * see exit_mm(), curr->task must not see
1650 * ->task == NULL before we read ->next.
1651 */
1652 smp_mb();
1653 curr->task = NULL;
1654 wake_up_process(task);
1655 }
1656
1657 mm->core_state = NULL;
1658}
1659
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001660/*
1661 * set_dumpable converts traditional three-value dumpable to two flags and
1662 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1663 * these bits are not changed atomically. So get_dumpable can observe the
1664 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1665 * return either old dumpable or new one by paying attention to the order of
1666 * modifying the bits.
1667 *
1668 * dumpable | mm->flags (binary)
1669 * old new | initial interim final
1670 * ---------+-----------------------
1671 * 0 1 | 00 01 01
1672 * 0 2 | 00 10(*) 11
1673 * 1 0 | 01 00 00
1674 * 1 2 | 01 11 11
1675 * 2 0 | 11 10(*) 00
1676 * 2 1 | 11 11 01
1677 *
1678 * (*) get_dumpable regards interim value of 10 as 11.
1679 */
1680void set_dumpable(struct mm_struct *mm, int value)
1681{
1682 switch (value) {
1683 case 0:
1684 clear_bit(MMF_DUMPABLE, &mm->flags);
1685 smp_wmb();
1686 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1687 break;
1688 case 1:
1689 set_bit(MMF_DUMPABLE, &mm->flags);
1690 smp_wmb();
1691 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1692 break;
1693 case 2:
1694 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1695 smp_wmb();
1696 set_bit(MMF_DUMPABLE, &mm->flags);
1697 break;
1698 }
1699}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001700
1701int get_dumpable(struct mm_struct *mm)
1702{
1703 int ret;
1704
1705 ret = mm->flags & 0x3;
1706 return (ret >= 2) ? 2 : ret;
1707}
1708
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1710{
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001711 struct core_state core_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 char corename[CORENAME_MAX_SIZE + 1];
1713 struct mm_struct *mm = current->mm;
1714 struct linux_binfmt * binfmt;
1715 struct inode * inode;
1716 struct file * file;
1717 int retval = 0;
Alan Coxd6e71142005-06-23 00:09:43 -07001718 int fsuid = current->fsuid;
1719 int flag = 0;
Andi Kleend025c9d2006-09-30 23:29:28 -07001720 int ispipe = 0;
Neil Horman7dc0b222007-10-16 23:26:34 -07001721 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
Neil Horman74aadce2007-10-16 23:26:35 -07001722 char **helper_argv = NULL;
1723 int helper_argc = 0;
1724 char *delimit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725
Steve Grubb0a4ff8c2007-04-19 10:28:21 -04001726 audit_core_dumps(signr);
1727
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 binfmt = current->binfmt;
1729 if (!binfmt || !binfmt->core_dump)
1730 goto fail;
1731 down_write(&mm->mmap_sem);
Roland McGrath00ec99d2007-11-11 19:13:43 -08001732 /*
1733 * If another thread got here first, or we are not dumpable, bail out.
1734 */
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001735 if (mm->core_state || !get_dumpable(mm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 up_write(&mm->mmap_sem);
1737 goto fail;
1738 }
Alan Coxd6e71142005-06-23 00:09:43 -07001739
1740 /*
1741 * We cannot trust fsuid as being the "true" uid of the
1742 * process nor do we know its entire history. We only know it
1743 * was tainted so we dump it as root in mode 2.
1744 */
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001745 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
Alan Coxd6e71142005-06-23 00:09:43 -07001746 flag = O_EXCL; /* Stop rewrite attacks */
1747 current->fsuid = 0; /* Dump root private */
1748 }
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001749
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001750 retval = coredump_wait(exit_code, &core_state);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001751 if (retval < 0)
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001752 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753
1754 /*
1755 * Clear any false indication of pending signals that might
1756 * be seen by the filesystem code called to write the core file.
1757 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 clear_thread_flag(TIF_SIGPENDING);
1759
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 /*
1761 * lock_kernel() because format_corename() is controlled by sysctl, which
1762 * uses lock_kernel()
1763 */
1764 lock_kernel();
Oleg Nesterov64093242008-10-18 20:28:22 -07001765 ispipe = format_corename(corename, signr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 unlock_kernel();
Neil Horman7dc0b222007-10-16 23:26:34 -07001767 /*
1768 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1769 * to a pipe. Since we're not writing directly to the filesystem
1770 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1771 * created unless the pipe reader choses to write out the core file
1772 * at which point file size limits and permissions will be imposed
1773 * as it does with any other process
1774 */
Neil Horman74aadce2007-10-16 23:26:35 -07001775 if ((!ispipe) && (core_limit < binfmt->min_coredump))
Neil Horman7dc0b222007-10-16 23:26:34 -07001776 goto fail_unlock;
1777
Alan Coxc4bbafd2007-04-16 22:53:13 -07001778 if (ispipe) {
Neil Horman74aadce2007-10-16 23:26:35 -07001779 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1780 /* Terminate the string before the first option */
1781 delimit = strchr(corename, ' ');
1782 if (delimit)
1783 *delimit = '\0';
Neil Horman32321132007-10-16 23:26:36 -07001784 delimit = strrchr(helper_argv[0], '/');
1785 if (delimit)
1786 delimit++;
1787 else
1788 delimit = helper_argv[0];
1789 if (!strcmp(delimit, current->comm)) {
1790 printk(KERN_NOTICE "Recursive core dump detected, "
1791 "aborting\n");
1792 goto fail_unlock;
1793 }
1794
1795 core_limit = RLIM_INFINITY;
1796
Andi Kleend025c9d2006-09-30 23:29:28 -07001797 /* SIGPIPE can happen, but it's just never processed */
Neil Horman32321132007-10-16 23:26:36 -07001798 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1799 &file)) {
Andi Kleend025c9d2006-09-30 23:29:28 -07001800 printk(KERN_INFO "Core dump to %s pipe failed\n",
1801 corename);
1802 goto fail_unlock;
1803 }
Andi Kleend025c9d2006-09-30 23:29:28 -07001804 } else
1805 file = filp_open(corename,
Alexey Dobriyan6d4df672006-12-06 20:40:39 -08001806 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1807 0600);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 if (IS_ERR(file))
1809 goto fail_unlock;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001810 inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 if (inode->i_nlink > 1)
1812 goto close_fail; /* multiple links - don't dump */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001813 if (!ispipe && d_unhashed(file->f_path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 goto close_fail;
1815
Andi Kleend025c9d2006-09-30 23:29:28 -07001816 /* AK: actually i see no reason to not allow this for named pipes etc.,
1817 but keep the previous behaviour for now. */
1818 if (!ispipe && !S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 goto close_fail;
Ingo Molnarc46f7392007-11-28 13:59:18 +01001820 /*
1821 * Dont allow local users get cute and trick others to coredump
1822 * into their pre-created files:
1823 */
1824 if (inode->i_uid != current->fsuid)
1825 goto close_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 if (!file->f_op)
1827 goto close_fail;
1828 if (!file->f_op->write)
1829 goto close_fail;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001830 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 goto close_fail;
1832
Neil Horman7dc0b222007-10-16 23:26:34 -07001833 retval = binfmt->core_dump(signr, regs, file, core_limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834
1835 if (retval)
1836 current->signal->group_exit_code |= 0x80;
1837close_fail:
1838 filp_close(file, NULL);
1839fail_unlock:
Neil Horman74aadce2007-10-16 23:26:35 -07001840 if (helper_argv)
1841 argv_free(helper_argv);
1842
Alan Coxd6e71142005-06-23 00:09:43 -07001843 current->fsuid = fsuid;
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001844 coredump_finish(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845fail:
1846 return retval;
1847}