blob: 9badbc0bfb1d23afd6cf64ace5ad1f1ae757d294 [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>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020036#include <linux/perf_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/highmem.h>
38#include <linux/spinlock.h>
39#include <linux/key.h>
40#include <linux/personality.h>
41#include <linux/binfmts.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/utsname.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080043#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <linux/module.h>
45#include <linux/namei.h>
46#include <linux/proc_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/mount.h>
48#include <linux/security.h>
49#include <linux/syscalls.h>
Jay Lan8f0ab512006-09-30 23:28:59 -070050#include <linux/tsacct_kern.h>
Matt Helsley9f460802005-11-07 00:59:16 -080051#include <linux/cn_proc.h>
Al Viro473ae302006-04-26 14:04:08 -040052#include <linux/audit.h>
Roland McGrath6341c392008-07-25 19:45:44 -070053#include <linux/tracehook.h>
Johannes Berg5f4123b2008-07-09 10:28:40 +020054#include <linux/kmod.h>
Eric Paris6110e3a2008-12-17 13:53:20 -050055#include <linux/fsnotify.h>
Al Viro5ad4e532009-03-29 19:50:06 -040056#include <linux/fs_struct.h>
Neil Horman61be2282009-09-23 15:56:58 -070057#include <linux/pipe_fs_i.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59#include <asm/uaccess.h>
60#include <asm/mmu_context.h>
Ollie Wildb6a2fea2007-07-19 01:48:16 -070061#include <asm/tlb.h>
David Howellsa6f76f22008-11-14 10:39:24 +110062#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
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";
Neil Hormana2939802009-09-23 15:56:56 -070066unsigned int core_pipe_limit;
Alan Coxd6e71142005-06-23 00:09:43 -070067int suid_dumpable = 0;
68
Linus Torvalds1da177e2005-04-16 15:20:36 -070069/* The maximal length of core_pattern is also specified in sysctl.c */
70
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070071static LIST_HEAD(formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072static DEFINE_RWLOCK(binfmt_lock);
73
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070074int __register_binfmt(struct linux_binfmt * fmt, int insert)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 if (!fmt)
77 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 write_lock(&binfmt_lock);
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070079 insert ? list_add(&fmt->lh, &formats) :
80 list_add_tail(&fmt->lh, &formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 write_unlock(&binfmt_lock);
82 return 0;
83}
84
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070085EXPORT_SYMBOL(__register_binfmt);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Alexey Dobriyanf6b450d2007-10-16 23:26:04 -070087void unregister_binfmt(struct linux_binfmt * fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 write_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070090 list_del(&fmt->lh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 write_unlock(&binfmt_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092}
93
94EXPORT_SYMBOL(unregister_binfmt);
95
96static inline void put_binfmt(struct linux_binfmt * fmt)
97{
98 module_put(fmt->module);
99}
100
101/*
102 * Note that a shared library must be both readable and executable due to
103 * security reasons.
104 *
105 * Also note that we take the address to load from from the file itself.
106 */
Heiko Carstens1e7bfb22009-01-14 14:14:29 +0100107SYSCALL_DEFINE1(uselib, const char __user *, library)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108{
Al Viro964bd182008-07-26 03:33:14 -0400109 struct file *file;
Al Viro964bd182008-07-26 03:33:14 -0400110 char *tmp = getname(library);
111 int error = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
Al Viro6e8341a2009-04-06 11:16:22 -0400113 if (IS_ERR(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 goto out;
115
Al Viro6e8341a2009-04-06 11:16:22 -0400116 file = do_filp_open(AT_FDCWD, tmp,
117 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
118 MAY_READ | MAY_EXEC | MAY_OPEN);
119 putname(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 error = PTR_ERR(file);
121 if (IS_ERR(file))
122 goto out;
123
Al Viro6e8341a2009-04-06 11:16:22 -0400124 error = -EINVAL;
125 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
126 goto exit;
127
128 error = -EACCES;
129 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
130 goto exit;
131
Eric Paris6110e3a2008-12-17 13:53:20 -0500132 fsnotify_open(file->f_path.dentry);
133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 error = -ENOEXEC;
135 if(file->f_op) {
136 struct linux_binfmt * fmt;
137
138 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -0700139 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 if (!fmt->load_shlib)
141 continue;
142 if (!try_module_get(fmt->module))
143 continue;
144 read_unlock(&binfmt_lock);
145 error = fmt->load_shlib(file);
146 read_lock(&binfmt_lock);
147 put_binfmt(fmt);
148 if (error != -ENOEXEC)
149 break;
150 }
151 read_unlock(&binfmt_lock);
152 }
Al Viro6e8341a2009-04-06 11:16:22 -0400153exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 fput(file);
155out:
156 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157}
158
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700159#ifdef CONFIG_MMU
160
161static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
162 int write)
163{
164 struct page *page;
165 int ret;
166
167#ifdef CONFIG_STACK_GROWSUP
168 if (write) {
169 ret = expand_stack_downwards(bprm->vma, pos);
170 if (ret < 0)
171 return NULL;
172 }
173#endif
174 ret = get_user_pages(current, bprm->mm, pos,
175 1, write, 1, &page, NULL);
176 if (ret <= 0)
177 return NULL;
178
179 if (write) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700180 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800181 struct rlimit *rlim;
182
183 /*
184 * We've historically supported up to 32 pages (ARG_MAX)
185 * of argument strings even with small stacks
186 */
187 if (size <= ARG_MAX)
188 return page;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700189
190 /*
191 * Limit to 1/4-th the stack size for the argv+env strings.
192 * This ensures that:
193 * - the remaining binfmt code will not run out of stack space,
194 * - the program will have a reasonable amount of stack left
195 * to work from.
196 */
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800197 rlim = current->signal->rlim;
Jiri Slabyd554ed892010-03-05 13:42:42 -0800198 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700199 put_page(page);
200 return NULL;
201 }
202 }
203
204 return page;
205}
206
207static void put_arg_page(struct page *page)
208{
209 put_page(page);
210}
211
212static void free_arg_page(struct linux_binprm *bprm, int i)
213{
214}
215
216static void free_arg_pages(struct linux_binprm *bprm)
217{
218}
219
220static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
221 struct page *page)
222{
223 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
224}
225
226static int __bprm_mm_init(struct linux_binprm *bprm)
227{
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800228 int err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700229 struct vm_area_struct *vma = NULL;
230 struct mm_struct *mm = bprm->mm;
231
232 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
233 if (!vma)
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800234 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700235
236 down_write(&mm->mmap_sem);
237 vma->vm_mm = mm;
238
239 /*
240 * Place the stack at the largest stack address the architecture
241 * supports. Later, we'll move this to an appropriate place. We don't
242 * use STACK_TOP because that can depend on attributes which aren't
243 * configured yet.
244 */
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700245 BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700246 vma->vm_end = STACK_TOP_MAX;
247 vma->vm_start = vma->vm_end - PAGE_SIZE;
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700248 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
Coly Li3ed75eb2007-10-18 23:39:15 -0700249 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Rik van Riel5beb4932010-03-05 13:42:07 -0800250 INIT_LIST_HEAD(&vma->anon_vma_chain);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700251 err = insert_vm_struct(mm, vma);
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800252 if (err)
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700253 goto err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700254
255 mm->stack_vm = mm->total_vm = 1;
256 up_write(&mm->mmap_sem);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700257 bprm->p = vma->vm_end - sizeof(void *);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700258 return 0;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700259err:
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800260 up_write(&mm->mmap_sem);
261 bprm->vma = NULL;
262 kmem_cache_free(vm_area_cachep, vma);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700263 return err;
264}
265
266static bool valid_arg_len(struct linux_binprm *bprm, long len)
267{
268 return len <= MAX_ARG_STRLEN;
269}
270
271#else
272
273static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
274 int write)
275{
276 struct page *page;
277
278 page = bprm->page[pos / PAGE_SIZE];
279 if (!page && write) {
280 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
281 if (!page)
282 return NULL;
283 bprm->page[pos / PAGE_SIZE] = page;
284 }
285
286 return page;
287}
288
289static void put_arg_page(struct page *page)
290{
291}
292
293static void free_arg_page(struct linux_binprm *bprm, int i)
294{
295 if (bprm->page[i]) {
296 __free_page(bprm->page[i]);
297 bprm->page[i] = NULL;
298 }
299}
300
301static void free_arg_pages(struct linux_binprm *bprm)
302{
303 int i;
304
305 for (i = 0; i < MAX_ARG_PAGES; i++)
306 free_arg_page(bprm, i);
307}
308
309static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
310 struct page *page)
311{
312}
313
314static int __bprm_mm_init(struct linux_binprm *bprm)
315{
316 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
317 return 0;
318}
319
320static bool valid_arg_len(struct linux_binprm *bprm, long len)
321{
322 return len <= bprm->p;
323}
324
325#endif /* CONFIG_MMU */
326
327/*
328 * Create a new mm_struct and populate it with a temporary stack
329 * vm_area_struct. We don't have enough context at this point to set the stack
330 * flags, permissions, and offset, so we use temporary values. We'll update
331 * them later in setup_arg_pages().
332 */
333int bprm_mm_init(struct linux_binprm *bprm)
334{
335 int err;
336 struct mm_struct *mm = NULL;
337
338 bprm->mm = mm = mm_alloc();
339 err = -ENOMEM;
340 if (!mm)
341 goto err;
342
343 err = init_new_context(current, mm);
344 if (err)
345 goto err;
346
347 err = __bprm_mm_init(bprm);
348 if (err)
349 goto err;
350
351 return 0;
352
353err:
354 if (mm) {
355 bprm->mm = NULL;
356 mmdrop(mm);
357 }
358
359 return err;
360}
361
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362/*
363 * count() counts the number of strings in array ARGV.
364 */
365static int count(char __user * __user * argv, int max)
366{
367 int i = 0;
368
369 if (argv != NULL) {
370 for (;;) {
371 char __user * p;
372
373 if (get_user(p, argv))
374 return -EFAULT;
375 if (!p)
376 break;
377 argv++;
Jason Baron362e6662008-10-15 22:01:52 -0700378 if (i++ >= max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 return -E2BIG;
380 cond_resched();
381 }
382 }
383 return i;
384}
385
386/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700387 * 'copy_strings()' copies argument/environment strings from the old
388 * processes's memory to the new process's stack. The call to get_user_pages()
389 * ensures the destination page is created and not swapped out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 */
Adrian Bunk75c96f82005-05-05 16:16:09 -0700391static int copy_strings(int argc, char __user * __user * argv,
392 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
394 struct page *kmapped_page = NULL;
395 char *kaddr = NULL;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700396 unsigned long kpos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 int ret;
398
399 while (argc-- > 0) {
400 char __user *str;
401 int len;
402 unsigned long pos;
403
404 if (get_user(str, argv+argc) ||
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700405 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 ret = -EFAULT;
407 goto out;
408 }
409
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700410 if (!valid_arg_len(bprm, len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 ret = -E2BIG;
412 goto out;
413 }
414
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700415 /* We're going to work our way backwords. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 pos = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700417 str += len;
418 bprm->p -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
420 while (len > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 int offset, bytes_to_copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422
423 offset = pos % PAGE_SIZE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700424 if (offset == 0)
425 offset = PAGE_SIZE;
426
427 bytes_to_copy = offset;
428 if (bytes_to_copy > len)
429 bytes_to_copy = len;
430
431 offset -= bytes_to_copy;
432 pos -= bytes_to_copy;
433 str -= bytes_to_copy;
434 len -= bytes_to_copy;
435
436 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
437 struct page *page;
438
439 page = get_arg_page(bprm, pos, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 if (!page) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700441 ret = -E2BIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 goto out;
443 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700445 if (kmapped_page) {
446 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700448 put_arg_page(kmapped_page);
449 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 kmapped_page = page;
451 kaddr = kmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700452 kpos = pos & PAGE_MASK;
453 flush_arg_page(bprm, kpos, kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700455 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 ret = -EFAULT;
457 goto out;
458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
460 }
461 ret = 0;
462out:
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700463 if (kmapped_page) {
464 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700466 put_arg_page(kmapped_page);
467 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 return ret;
469}
470
471/*
472 * Like copy_strings, but get argv and its values from kernel memory.
473 */
474int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
475{
476 int r;
477 mm_segment_t oldfs = get_fs();
478 set_fs(KERNEL_DS);
479 r = copy_strings(argc, (char __user * __user *)argv, bprm);
480 set_fs(oldfs);
481 return r;
482}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483EXPORT_SYMBOL(copy_strings_kernel);
484
485#ifdef CONFIG_MMU
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700486
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700488 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
489 * the binfmt code determines where the new stack should reside, we shift it to
490 * its final location. The process proceeds as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 *
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700492 * 1) Use shift to calculate the new vma endpoints.
493 * 2) Extend vma to cover both the old and new ranges. This ensures the
494 * arguments passed to subsequent functions are consistent.
495 * 3) Move vma's page tables to the new range.
496 * 4) Free up any cleared pgd range.
497 * 5) Shrink the vma to cover only the new range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700499static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
501 struct mm_struct *mm = vma->vm_mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700502 unsigned long old_start = vma->vm_start;
503 unsigned long old_end = vma->vm_end;
504 unsigned long length = old_end - old_start;
505 unsigned long new_start = old_start - shift;
506 unsigned long new_end = old_end - shift;
507 struct mmu_gather *tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700509 BUG_ON(new_start > new_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700511 /*
512 * ensure there are no vmas between where we want to go
513 * and where we are
514 */
515 if (vma != find_vma(mm, new_start))
516 return -EFAULT;
517
518 /*
519 * cover the whole range: [new_start, old_end)
520 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800521 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
522 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700523
524 /*
525 * move the page tables downwards, on failure we rely on
526 * process cleanup to remove whatever mess we made.
527 */
528 if (length != move_page_tables(vma, old_start,
529 vma, new_start, length))
530 return -ENOMEM;
531
532 lru_add_drain();
533 tlb = tlb_gather_mmu(mm, 0);
534 if (new_end > old_start) {
535 /*
536 * when the old and new regions overlap clear from new_end.
537 */
Jan Beulich42b77722008-07-23 21:27:10 -0700538 free_pgd_range(tlb, new_end, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700539 vma->vm_next ? vma->vm_next->vm_start : 0);
540 } else {
541 /*
542 * otherwise, clean from old_start; this is done to not touch
543 * the address space in [new_end, old_start) some architectures
544 * have constraints on va-space that make this illegal (IA64) -
545 * for the others its just a little faster.
546 */
Jan Beulich42b77722008-07-23 21:27:10 -0700547 free_pgd_range(tlb, old_start, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700548 vma->vm_next ? vma->vm_next->vm_start : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700550 tlb_finish_mmu(tlb, new_end, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700552 /*
Rik van Riel5beb4932010-03-05 13:42:07 -0800553 * Shrink the vma to just the new range. Always succeeds.
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700554 */
555 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
556
557 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558}
559
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700560/*
561 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
562 * the stack is optionally relocated, and some extra space is added.
563 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564int setup_arg_pages(struct linux_binprm *bprm,
565 unsigned long stack_top,
566 int executable_stack)
567{
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700568 unsigned long ret;
569 unsigned long stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 struct mm_struct *mm = current->mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700571 struct vm_area_struct *vma = bprm->vma;
572 struct vm_area_struct *prev = NULL;
573 unsigned long vm_flags;
574 unsigned long stack_base;
Michael Neuling803bf5e2010-02-10 13:56:42 -0800575 unsigned long stack_size;
576 unsigned long stack_expand;
577 unsigned long rlim_stack;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
579#ifdef CONFIG_STACK_GROWSUP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 /* Limit stack size to 1GB */
Jiri Slabyd554ed892010-03-05 13:42:42 -0800581 stack_base = rlimit_max(RLIMIT_STACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 if (stack_base > (1 << 30))
583 stack_base = 1 << 30;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700585 /* Make sure we didn't let the argument array grow too large. */
586 if (vma->vm_end - vma->vm_start > stack_base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 return -ENOMEM;
588
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700589 stack_base = PAGE_ALIGN(stack_top - stack_base);
590
591 stack_shift = vma->vm_start - stack_base;
592 mm->arg_start = bprm->p - stack_shift;
593 bprm->p = vma->vm_end - stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594#else
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700595 stack_top = arch_align_stack(stack_top);
596 stack_top = PAGE_ALIGN(stack_top);
597 stack_shift = vma->vm_end - stack_top;
598
599 bprm->p -= stack_shift;
600 mm->arg_start = bprm->p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601#endif
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700602
603 if (bprm->loader)
604 bprm->loader -= stack_shift;
605 bprm->exec -= stack_shift;
606
607 down_write(&mm->mmap_sem);
Hugh Dickins96a8e132008-07-10 21:19:20 +0100608 vm_flags = VM_STACK_FLAGS;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700609
610 /*
611 * Adjust stack execute permissions; explicitly enable for
612 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
613 * (arch default) otherwise.
614 */
615 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
616 vm_flags |= VM_EXEC;
617 else if (executable_stack == EXSTACK_DISABLE_X)
618 vm_flags &= ~VM_EXEC;
619 vm_flags |= mm->def_flags;
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700620 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700621
622 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
623 vm_flags);
624 if (ret)
625 goto out_unlock;
626 BUG_ON(prev != vma);
627
628 /* Move stack pages down in memory. */
629 if (stack_shift) {
630 ret = shift_arg_pages(vma, stack_shift);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800631 if (ret)
632 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 }
634
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700635 /* mprotect_fixup is overkill to remove the temporary stack flags */
636 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
637
Michael Neuling5ef097d2010-03-05 13:42:57 -0800638 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
Michael Neuling803bf5e2010-02-10 13:56:42 -0800639 stack_size = vma->vm_end - vma->vm_start;
640 /*
641 * Align this down to a page boundary as expand_stack
642 * will align it up.
643 */
644 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700645#ifdef CONFIG_STACK_GROWSUP
Michael Neuling803bf5e2010-02-10 13:56:42 -0800646 if (stack_size + stack_expand > rlim_stack)
647 stack_base = vma->vm_start + rlim_stack;
648 else
649 stack_base = vma->vm_end + stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700650#else
Michael Neuling803bf5e2010-02-10 13:56:42 -0800651 if (stack_size + stack_expand > rlim_stack)
652 stack_base = vma->vm_end - rlim_stack;
653 else
654 stack_base = vma->vm_start - stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700655#endif
656 ret = expand_stack(vma, stack_base);
657 if (ret)
658 ret = -EFAULT;
659
660out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 up_write(&mm->mmap_sem);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800662 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664EXPORT_SYMBOL(setup_arg_pages);
665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666#endif /* CONFIG_MMU */
667
668struct file *open_exec(const char *name)
669{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 struct file *file;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200671 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
Al Viro6e8341a2009-04-06 11:16:22 -0400673 file = do_filp_open(AT_FDCWD, name,
674 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
675 MAY_EXEC | MAY_OPEN);
676 if (IS_ERR(file))
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200677 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200679 err = -EACCES;
Al Viro6e8341a2009-04-06 11:16:22 -0400680 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
681 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200682
Al Viro6e8341a2009-04-06 11:16:22 -0400683 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
684 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200685
Eric Paris6110e3a2008-12-17 13:53:20 -0500686 fsnotify_open(file->f_path.dentry);
687
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200688 err = deny_write_access(file);
Al Viro6e8341a2009-04-06 11:16:22 -0400689 if (err)
690 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
Al Viro6e8341a2009-04-06 11:16:22 -0400692out:
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200693 return file;
694
Al Viro6e8341a2009-04-06 11:16:22 -0400695exit:
696 fput(file);
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200697 return ERR_PTR(err);
698}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699EXPORT_SYMBOL(open_exec);
700
Mimi Zohar6777d772009-08-21 14:32:48 -0400701int kernel_read(struct file *file, loff_t offset,
702 char *addr, unsigned long count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703{
704 mm_segment_t old_fs;
705 loff_t pos = offset;
706 int result;
707
708 old_fs = get_fs();
709 set_fs(get_ds());
710 /* The cast to a user pointer is valid due to the set_fs() */
711 result = vfs_read(file, (void __user *)addr, count, &pos);
712 set_fs(old_fs);
713 return result;
714}
715
716EXPORT_SYMBOL(kernel_read);
717
718static int exec_mmap(struct mm_struct *mm)
719{
720 struct task_struct *tsk;
721 struct mm_struct * old_mm, *active_mm;
722
723 /* Notify parent that we're no longer interested in the old VM */
724 tsk = current;
725 old_mm = current->mm;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800726 sync_mm_rss(tsk, old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 mm_release(tsk, old_mm);
728
729 if (old_mm) {
730 /*
731 * Make sure that if there is a core dump in progress
732 * for the old mm, we get out and die instead of going
733 * through with the exec. We must hold mmap_sem around
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700734 * checking core_state and changing tsk->mm.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 */
736 down_read(&old_mm->mmap_sem);
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700737 if (unlikely(old_mm->core_state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 up_read(&old_mm->mmap_sem);
739 return -EINTR;
740 }
741 }
742 task_lock(tsk);
743 active_mm = tsk->active_mm;
744 tsk->mm = mm;
745 tsk->active_mm = mm;
746 activate_mm(active_mm, mm);
747 task_unlock(tsk);
748 arch_pick_mmap_layout(mm);
749 if (old_mm) {
750 up_read(&old_mm->mmap_sem);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +0200751 BUG_ON(active_mm != old_mm);
Balbir Singh31a78f22008-09-28 23:09:31 +0100752 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 mmput(old_mm);
754 return 0;
755 }
756 mmdrop(active_mm);
757 return 0;
758}
759
760/*
761 * This function makes sure the current process has its own signal table,
762 * so that flush_signal_handlers can later reset the handlers without
763 * disturbing other processes. (Other processes might share the signal
764 * table via the CLONE_SIGHAND option to clone().)
765 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800766static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767{
768 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700769 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 spinlock_t *lock = &oldsighand->siglock;
771 int count;
772
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700773 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 goto no_thread_group;
775
776 /*
777 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800780 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 /*
782 * Another group action in progress, just
783 * return so that the signal is processed.
784 */
785 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 return -EAGAIN;
787 }
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800788 sig->group_exit_task = tsk;
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700789 zap_other_threads(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Oleg Nesterovfea9d172008-02-08 04:19:19 -0800791 /* Account for the thread group leader hanging around: */
792 count = thread_group_leader(tsk) ? 1 : 2;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700793 sig->notify_count = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 while (atomic_read(&sig->count) > count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 __set_current_state(TASK_UNINTERRUPTIBLE);
796 spin_unlock_irq(lock);
797 schedule();
798 spin_lock_irq(lock);
799 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 spin_unlock_irq(lock);
801
802 /*
803 * At this point all other threads have exited, all we have to
804 * do is to wait for the thread group leader to become inactive,
805 * and to assume its PID:
806 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700807 if (!thread_group_leader(tsk)) {
Oleg Nesterov81879262008-12-01 14:18:16 -0800808 struct task_struct *leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700809
Oleg Nesterov2800d8d2008-04-30 00:53:12 -0700810 sig->notify_count = -1; /* for exit_notify() */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700811 for (;;) {
812 write_lock_irq(&tasklist_lock);
813 if (likely(leader->exit_state))
814 break;
815 __set_current_state(TASK_UNINTERRUPTIBLE);
816 write_unlock_irq(&tasklist_lock);
817 schedule();
818 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Roland McGrathf5e90282006-04-10 22:54:16 -0700820 /*
821 * The only record we have of the real-time age of a
822 * process, regardless of execs it's done, is start_time.
823 * All the past CPU time is accumulated in signal_struct
824 * from sister threads now dead. But in this non-leader
825 * exec, nothing survives from the original leader thread,
826 * whose birth marks the true age of this process now.
827 * When we take on its identity by switching to its PID, we
828 * also take its birthdate (always earlier than our own).
829 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700830 tsk->start_time = leader->start_time;
Roland McGrathf5e90282006-04-10 22:54:16 -0700831
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700832 BUG_ON(!same_thread_group(leader, tsk));
833 BUG_ON(has_group_leader_pid(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 /*
835 * An exec() starts a new thread group with the
836 * TGID of the previous thread group. Rehash the
837 * two threads with a switched PID, and release
838 * the former thread group leader:
839 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800840
841 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700842 * The old leader becomes a thread of the this thread group.
843 * Note: The old leader also uses this pid until release_task
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800844 * is called. Odd but simple and correct.
845 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700846 detach_pid(tsk, PIDTYPE_PID);
847 tsk->pid = leader->pid;
Sukadev Bhattiprolu3743ca02007-10-18 23:39:51 -0700848 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700849 transfer_pid(leader, tsk, PIDTYPE_PGID);
850 transfer_pid(leader, tsk, PIDTYPE_SID);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800851
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700852 list_replace_rcu(&leader->tasks, &tsk->tasks);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800853 list_replace_init(&leader->sibling, &tsk->sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700855 tsk->group_leader = tsk;
856 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -0600857
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700858 tsk->exit_signal = SIGCHLD;
Oleg Nesterov962b5642005-11-23 13:37:43 -0800859
860 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
861 leader->exit_state = EXIT_DEAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 write_unlock_irq(&tasklist_lock);
Oleg Nesterov81879262008-12-01 14:18:16 -0800863
864 release_task(leader);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800865 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700867 sig->group_exit_task = NULL;
868 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
870no_thread_group:
Jiri Pirko1f102062009-09-22 16:44:10 -0700871 if (current->mm)
872 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
873
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 exit_itimers(sig);
Oleg Nesterovcbaffba2008-05-26 20:55:42 +0400875 flush_itimer_signals();
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300876
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700877 if (atomic_read(&oldsighand->count) != 1) {
878 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700880 * This ->sighand is shared with the CLONE_SIGHAND
881 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700883 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
884 if (!newsighand)
885 return -ENOMEM;
886
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 atomic_set(&newsighand->count, 1);
888 memcpy(newsighand->action, oldsighand->action,
889 sizeof(newsighand->action));
890
891 write_lock_irq(&tasklist_lock);
892 spin_lock(&oldsighand->siglock);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700893 rcu_assign_pointer(tsk->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 spin_unlock(&oldsighand->siglock);
895 write_unlock_irq(&tasklist_lock);
896
Davide Libenzifba2afa2007-05-10 22:23:13 -0700897 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 }
899
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700900 BUG_ON(!thread_group_leader(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 return 0;
902}
Oleg Nesterov0840a902007-10-16 23:27:22 -0700903
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904/*
905 * These functions flushes out all traces of the currently running executable
906 * so that a new one can be started
907 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800908static void flush_old_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909{
910 long j = -1;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700911 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
913 spin_lock(&files->file_lock);
914 for (;;) {
915 unsigned long set, i;
916
917 j++;
918 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700919 fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800920 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700922 set = fdt->close_on_exec->fds_bits[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 if (!set)
924 continue;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700925 fdt->close_on_exec->fds_bits[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 spin_unlock(&files->file_lock);
927 for ( ; set ; i++,set >>= 1) {
928 if (set & 1) {
929 sys_close(i);
930 }
931 }
932 spin_lock(&files->file_lock);
933
934 }
935 spin_unlock(&files->file_lock);
936}
937
Andrew Morton59714d62008-02-04 22:27:21 -0800938char *get_task_comm(char *buf, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939{
940 /* buf must be at least sizeof(tsk->comm) in size */
941 task_lock(tsk);
942 strncpy(buf, tsk->comm, sizeof(tsk->comm));
943 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -0800944 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945}
946
947void set_task_comm(struct task_struct *tsk, char *buf)
948{
949 task_lock(tsk);
john stultz4614a696b2009-12-14 18:00:05 -0800950
951 /*
952 * Threads may access current->comm without holding
953 * the task lock, so write the string carefully.
954 * Readers without a lock may see incomplete new
955 * names but are safe from non-terminating string reads.
956 */
957 memset(tsk->comm, 0, TASK_COMM_LEN);
958 wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
960 task_unlock(tsk);
Ingo Molnarcdd6c482009-09-21 12:02:48 +0200961 perf_event_comm(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962}
963
964int flush_old_exec(struct linux_binprm * bprm)
965{
Linus Torvalds221af7f2010-01-28 22:14:42 -0800966 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967
968 /*
969 * Make sure we have a private signal table and that
970 * we are unassociated from the previous thread group.
971 */
972 retval = de_thread(current);
973 if (retval)
974 goto out;
975
Matt Helsley925d1c42008-04-29 01:01:36 -0700976 set_mm_exe_file(bprm->mm, bprm->file);
977
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 * Release all of the old mmap stuff
980 */
981 retval = exec_mmap(bprm->mm);
982 if (retval)
Al Virofd8328b2008-04-22 05:11:59 -0400983 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
985 bprm->mm = NULL; /* We're using it now */
Linus Torvalds7ab02af2010-02-02 12:37:44 -0800986
987 current->flags &= ~PF_RANDOMIZE;
988 flush_thread();
989 current->personality &= ~bprm->per_clear;
990
Linus Torvalds221af7f2010-01-28 22:14:42 -0800991 return 0;
992
993out:
994 return retval;
995}
996EXPORT_SYMBOL(flush_old_exec);
997
998void setup_new_exec(struct linux_binprm * bprm)
999{
1000 int i, ch;
1001 char * name;
1002 char tcomm[sizeof(current->comm)];
1003
1004 arch_pick_mmap_layout(current->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
1006 /* This is the point of no return */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 current->sas_ss_sp = current->sas_ss_size = 0;
1008
David Howellsda9592e2008-11-14 10:39:05 +11001009 if (current_euid() == current_uid() && current_egid() == current_gid())
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001010 set_dumpable(current->mm, 1);
Alan Coxd6e71142005-06-23 00:09:43 -07001011 else
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001012 set_dumpable(current->mm, suid_dumpable);
Alan Coxd6e71142005-06-23 00:09:43 -07001013
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 name = bprm->filename;
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -07001015
1016 /* Copies the binary name from after last slash */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 for (i=0; (ch = *(name++)) != '\0';) {
1018 if (ch == '/')
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -07001019 i = 0; /* overwrite what we wrote */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 else
1021 if (i < (sizeof(tcomm) - 1))
1022 tcomm[i++] = ch;
1023 }
1024 tcomm[i] = '\0';
1025 set_task_comm(current, tcomm);
1026
Benjamin Herrenschmidt0551fbd2006-02-28 16:59:19 -08001027 /* Set the new mm task size. We have to do that late because it may
1028 * depend on TIF_32BIT which is only updated in flush_thread() on
1029 * some architectures like powerpc
1030 */
1031 current->mm->task_size = TASK_SIZE;
1032
David Howellsa6f76f22008-11-14 10:39:24 +11001033 /* install the new credentials */
1034 if (bprm->cred->uid != current_euid() ||
1035 bprm->cred->gid != current_egid()) {
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001036 current->pdeath_signal = 0;
1037 } else if (file_permission(bprm->file, MAY_READ) ||
David Howellsa6f76f22008-11-14 10:39:24 +11001038 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001039 set_dumpable(current->mm, suid_dumpable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 }
1041
Ingo Molnarf65cb452008-12-16 13:40:44 +01001042 /*
1043 * Flush performance counters when crossing a
1044 * security domain:
1045 */
1046 if (!get_dumpable(current->mm))
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001047 perf_event_exit_task(current);
Ingo Molnarf65cb452008-12-16 13:40:44 +01001048
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 /* An exec changes our domain. We are no longer part of the thread
1050 group */
1051
1052 current->self_exec_id++;
1053
1054 flush_signal_handlers(current, 0);
1055 flush_old_files(current->files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056}
Linus Torvalds221af7f2010-01-28 22:14:42 -08001057EXPORT_SYMBOL(setup_new_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
David Howellsa6f76f22008-11-14 10:39:24 +11001059/*
Oleg Nesterova2a84742009-09-05 11:17:13 -07001060 * Prepare credentials and lock ->cred_guard_mutex.
1061 * install_exec_creds() commits the new creds and drops the lock.
1062 * Or, if exec fails before, free_bprm() should release ->cred and
1063 * and unlock.
1064 */
1065int prepare_bprm_creds(struct linux_binprm *bprm)
1066{
1067 if (mutex_lock_interruptible(&current->cred_guard_mutex))
1068 return -ERESTARTNOINTR;
1069
1070 bprm->cred = prepare_exec_creds();
1071 if (likely(bprm->cred))
1072 return 0;
1073
1074 mutex_unlock(&current->cred_guard_mutex);
1075 return -ENOMEM;
1076}
1077
1078void free_bprm(struct linux_binprm *bprm)
1079{
1080 free_arg_pages(bprm);
1081 if (bprm->cred) {
1082 mutex_unlock(&current->cred_guard_mutex);
1083 abort_creds(bprm->cred);
1084 }
1085 kfree(bprm);
1086}
1087
1088/*
David Howellsa6f76f22008-11-14 10:39:24 +11001089 * install the new credentials for this executable
1090 */
1091void install_exec_creds(struct linux_binprm *bprm)
1092{
1093 security_bprm_committing_creds(bprm);
1094
1095 commit_creds(bprm->cred);
1096 bprm->cred = NULL;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001097 /*
1098 * cred_guard_mutex must be held at least to this point to prevent
David Howellsa6f76f22008-11-14 10:39:24 +11001099 * ptrace_attach() from altering our determination of the task's
Oleg Nesterova2a84742009-09-05 11:17:13 -07001100 * credentials; any time after this it may be unlocked.
1101 */
David Howellsa6f76f22008-11-14 10:39:24 +11001102 security_bprm_committed_creds(bprm);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001103 mutex_unlock(&current->cred_guard_mutex);
David Howellsa6f76f22008-11-14 10:39:24 +11001104}
1105EXPORT_SYMBOL(install_exec_creds);
1106
1107/*
1108 * determine how safe it is to execute the proposed program
David Howells5e751e92009-05-08 13:55:22 +01001109 * - the caller must hold current->cred_guard_mutex to protect against
David Howellsa6f76f22008-11-14 10:39:24 +11001110 * PTRACE_ATTACH
1111 */
Al Viro498052b2009-03-30 07:20:30 -04001112int check_unsafe_exec(struct linux_binprm *bprm)
David Howellsa6f76f22008-11-14 10:39:24 +11001113{
David Howells0bf2f3a2009-02-06 11:45:46 +00001114 struct task_struct *p = current, *t;
Al Virof1191b52009-03-30 07:35:18 -04001115 unsigned n_fs;
Al Viro498052b2009-03-30 07:20:30 -04001116 int res = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001117
1118 bprm->unsafe = tracehook_unsafe_exec(p);
1119
David Howells0bf2f3a2009-02-06 11:45:46 +00001120 n_fs = 1;
Al Viro498052b2009-03-30 07:20:30 -04001121 write_lock(&p->fs->lock);
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001122 rcu_read_lock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001123 for (t = next_thread(p); t != p; t = next_thread(t)) {
1124 if (t->fs == p->fs)
1125 n_fs++;
David Howells0bf2f3a2009-02-06 11:45:46 +00001126 }
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001127 rcu_read_unlock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001128
Al Virof1191b52009-03-30 07:35:18 -04001129 if (p->fs->users > n_fs) {
David Howellsa6f76f22008-11-14 10:39:24 +11001130 bprm->unsafe |= LSM_UNSAFE_SHARE;
Al Viro498052b2009-03-30 07:20:30 -04001131 } else {
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001132 res = -EAGAIN;
1133 if (!p->fs->in_exec) {
1134 p->fs->in_exec = 1;
1135 res = 1;
1136 }
Al Viro498052b2009-03-30 07:20:30 -04001137 }
Al Viro498052b2009-03-30 07:20:30 -04001138 write_unlock(&p->fs->lock);
1139
1140 return res;
David Howellsa6f76f22008-11-14 10:39:24 +11001141}
1142
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143/*
1144 * Fill the binprm structure from the inode.
1145 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
David Howellsa6f76f22008-11-14 10:39:24 +11001146 *
1147 * This may be called multiple times for binary chains (scripts for example).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 */
1149int prepare_binprm(struct linux_binprm *bprm)
1150{
David Howellsa6f76f22008-11-14 10:39:24 +11001151 umode_t mode;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001152 struct inode * inode = bprm->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 int retval;
1154
1155 mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 if (bprm->file->f_op == NULL)
1157 return -EACCES;
1158
David Howellsa6f76f22008-11-14 10:39:24 +11001159 /* clear any previous set[ug]id data from a previous binary */
1160 bprm->cred->euid = current_euid();
1161 bprm->cred->egid = current_egid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
David Howellsa6f76f22008-11-14 10:39:24 +11001163 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 /* Set-uid? */
1165 if (mode & S_ISUID) {
David Howellsa6f76f22008-11-14 10:39:24 +11001166 bprm->per_clear |= PER_CLEAR_ON_SETID;
1167 bprm->cred->euid = inode->i_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 }
1169
1170 /* Set-gid? */
1171 /*
1172 * If setgid is set but no group execute bit then this
1173 * is a candidate for mandatory locking, not a setgid
1174 * executable.
1175 */
1176 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
David Howellsa6f76f22008-11-14 10:39:24 +11001177 bprm->per_clear |= PER_CLEAR_ON_SETID;
1178 bprm->cred->egid = inode->i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 }
1180 }
1181
1182 /* fill in binprm security blob */
David Howellsa6f76f22008-11-14 10:39:24 +11001183 retval = security_bprm_set_creds(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 if (retval)
1185 return retval;
David Howellsa6f76f22008-11-14 10:39:24 +11001186 bprm->cred_prepared = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
David Howellsa6f76f22008-11-14 10:39:24 +11001188 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1189 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190}
1191
1192EXPORT_SYMBOL(prepare_binprm);
1193
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001194/*
1195 * Arguments are '\0' separated strings found at the location bprm->p
1196 * points to; chop off the first by relocating brpm->p to right after
1197 * the first '\0' encountered.
1198 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001199int remove_arg_zero(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200{
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001201 int ret = 0;
1202 unsigned long offset;
1203 char *kaddr;
1204 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001206 if (!bprm->argc)
1207 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001209 do {
1210 offset = bprm->p & ~PAGE_MASK;
1211 page = get_arg_page(bprm, bprm->p, 0);
1212 if (!page) {
1213 ret = -EFAULT;
1214 goto out;
1215 }
1216 kaddr = kmap_atomic(page, KM_USER0);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001217
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001218 for (; offset < PAGE_SIZE && kaddr[offset];
1219 offset++, bprm->p++)
1220 ;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001221
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001222 kunmap_atomic(kaddr, KM_USER0);
1223 put_arg_page(page);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001224
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001225 if (offset == PAGE_SIZE)
1226 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1227 } while (offset == PAGE_SIZE);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001228
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001229 bprm->p++;
1230 bprm->argc--;
1231 ret = 0;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001232
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001233out:
1234 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236EXPORT_SYMBOL(remove_arg_zero);
1237
1238/*
1239 * cycle the list of binary formats handler, until one recognizes the image
1240 */
1241int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1242{
Roland McGrath85f33462008-12-09 19:36:38 -08001243 unsigned int depth = bprm->recursion_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 int try,retval;
1245 struct linux_binfmt *fmt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 retval = security_bprm_check(bprm);
1248 if (retval)
1249 return retval;
1250
1251 /* kernel module loader fixup */
1252 /* so we don't try to load run modprobe in kernel space. */
1253 set_fs(USER_DS);
Al Viro473ae302006-04-26 14:04:08 -04001254
1255 retval = audit_bprm(bprm);
1256 if (retval)
1257 return retval;
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 retval = -ENOENT;
1260 for (try=0; try<2; try++) {
1261 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -07001262 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1264 if (!fn)
1265 continue;
1266 if (!try_module_get(fmt->module))
1267 continue;
1268 read_unlock(&binfmt_lock);
1269 retval = fn(bprm, regs);
Roland McGrath85f33462008-12-09 19:36:38 -08001270 /*
1271 * Restore the depth counter to its starting value
1272 * in this call, so we don't have to rely on every
1273 * load_binary function to restore it on return.
1274 */
1275 bprm->recursion_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 if (retval >= 0) {
Roland McGrath85f33462008-12-09 19:36:38 -08001277 if (depth == 0)
1278 tracehook_report_exec(fmt, bprm, regs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 put_binfmt(fmt);
1280 allow_write_access(bprm->file);
1281 if (bprm->file)
1282 fput(bprm->file);
1283 bprm->file = NULL;
1284 current->did_exec = 1;
Matt Helsley9f460802005-11-07 00:59:16 -08001285 proc_exec_connector(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 return retval;
1287 }
1288 read_lock(&binfmt_lock);
1289 put_binfmt(fmt);
1290 if (retval != -ENOEXEC || bprm->mm == NULL)
1291 break;
1292 if (!bprm->file) {
1293 read_unlock(&binfmt_lock);
1294 return retval;
1295 }
1296 }
1297 read_unlock(&binfmt_lock);
1298 if (retval != -ENOEXEC || bprm->mm == NULL) {
1299 break;
Johannes Berg5f4123b2008-07-09 10:28:40 +02001300#ifdef CONFIG_MODULES
1301 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1303 if (printable(bprm->buf[0]) &&
1304 printable(bprm->buf[1]) &&
1305 printable(bprm->buf[2]) &&
1306 printable(bprm->buf[3]))
1307 break; /* -ENOEXEC */
1308 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1309#endif
1310 }
1311 }
1312 return retval;
1313}
1314
1315EXPORT_SYMBOL(search_binary_handler);
1316
1317/*
1318 * sys_execve() executes a new program.
1319 */
1320int do_execve(char * filename,
1321 char __user *__user *argv,
1322 char __user *__user *envp,
1323 struct pt_regs * regs)
1324{
1325 struct linux_binprm *bprm;
1326 struct file *file;
Al Viro3b125382008-04-22 05:31:30 -04001327 struct files_struct *displaced;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001328 bool clear_in_exec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
Al Viro3b125382008-04-22 05:31:30 -04001331 retval = unshare_files(&displaced);
Al Virofd8328b2008-04-22 05:11:59 -04001332 if (retval)
1333 goto out_ret;
1334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 retval = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001336 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 if (!bprm)
Al Virofd8328b2008-04-22 05:11:59 -04001338 goto out_files;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
Oleg Nesterova2a84742009-09-05 11:17:13 -07001340 retval = prepare_bprm_creds(bprm);
1341 if (retval)
David Howellsa6f76f22008-11-14 10:39:24 +11001342 goto out_free;
Al Viro498052b2009-03-30 07:20:30 -04001343
1344 retval = check_unsafe_exec(bprm);
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001345 if (retval < 0)
Oleg Nesterova2a84742009-09-05 11:17:13 -07001346 goto out_free;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001347 clear_in_exec = retval;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001348 current->in_execve = 1;
David Howellsa6f76f22008-11-14 10:39:24 +11001349
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 file = open_exec(filename);
1351 retval = PTR_ERR(file);
1352 if (IS_ERR(file))
Al Viro498052b2009-03-30 07:20:30 -04001353 goto out_unmark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
1355 sched_exec();
1356
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 bprm->file = file;
1358 bprm->filename = filename;
1359 bprm->interp = filename;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001360
1361 retval = bprm_mm_init(bprm);
1362 if (retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 goto out_file;
1364
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001365 bprm->argc = count(argv, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 if ((retval = bprm->argc) < 0)
David Howellsa6f76f22008-11-14 10:39:24 +11001367 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001369 bprm->envc = count(envp, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 if ((retval = bprm->envc) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 goto out;
1372
1373 retval = prepare_binprm(bprm);
1374 if (retval < 0)
1375 goto out;
1376
1377 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1378 if (retval < 0)
1379 goto out;
1380
1381 bprm->exec = bprm->p;
1382 retval = copy_strings(bprm->envc, envp, bprm);
1383 if (retval < 0)
1384 goto out;
1385
1386 retval = copy_strings(bprm->argc, argv, bprm);
1387 if (retval < 0)
1388 goto out;
1389
Oleg Nesterov7b34e422008-07-25 01:47:37 -07001390 current->flags &= ~PF_KTHREAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 retval = search_binary_handler(bprm,regs);
David Howellsa6f76f22008-11-14 10:39:24 +11001392 if (retval < 0)
1393 goto out;
1394
1395 /* execve succeeded */
Al Viro498052b2009-03-30 07:20:30 -04001396 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001397 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001398 acct_update_integrals(current);
1399 free_bprm(bprm);
1400 if (displaced)
1401 put_files_struct(displaced);
1402 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
1404out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 if (bprm->mm)
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001406 mmput (bprm->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
1408out_file:
1409 if (bprm->file) {
1410 allow_write_access(bprm->file);
1411 fput(bprm->file);
1412 }
David Howellsa6f76f22008-11-14 10:39:24 +11001413
Al Viro498052b2009-03-30 07:20:30 -04001414out_unmark:
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001415 if (clear_in_exec)
1416 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001417 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001418
1419out_free:
Al Viro08a6fac2008-05-10 16:38:25 -04001420 free_bprm(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
Al Virofd8328b2008-04-22 05:11:59 -04001422out_files:
Al Viro3b125382008-04-22 05:31:30 -04001423 if (displaced)
1424 reset_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425out_ret:
1426 return retval;
1427}
1428
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001429void set_binfmt(struct linux_binfmt *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430{
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001431 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001433 if (mm->binfmt)
1434 module_put(mm->binfmt->module);
1435
1436 mm->binfmt = new;
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001437 if (new)
1438 __module_get(new->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439}
1440
1441EXPORT_SYMBOL(set_binfmt);
1442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443/* format_corename will inspect the pattern parameter, and output a
1444 * name into corename, which must have space for at least
1445 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1446 */
Oleg Nesterov64093242008-10-18 20:28:22 -07001447static int format_corename(char *corename, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448{
David Howells86a264a2008-11-14 10:39:18 +11001449 const struct cred *cred = current_cred();
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001450 const char *pat_ptr = core_pattern;
1451 int ispipe = (*pat_ptr == '|');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 char *out_ptr = corename;
1453 char *const out_end = corename + CORENAME_MAX_SIZE;
1454 int rc;
1455 int pid_in_pattern = 0;
1456
1457 /* Repeat as long as we have more pattern to process and more output
1458 space */
1459 while (*pat_ptr) {
1460 if (*pat_ptr != '%') {
1461 if (out_ptr == out_end)
1462 goto out;
1463 *out_ptr++ = *pat_ptr++;
1464 } else {
1465 switch (*++pat_ptr) {
1466 case 0:
1467 goto out;
1468 /* Double percent, output one percent */
1469 case '%':
1470 if (out_ptr == out_end)
1471 goto out;
1472 *out_ptr++ = '%';
1473 break;
1474 /* pid */
1475 case 'p':
1476 pid_in_pattern = 1;
1477 rc = snprintf(out_ptr, out_end - out_ptr,
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001478 "%d", task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 if (rc > out_end - out_ptr)
1480 goto out;
1481 out_ptr += rc;
1482 break;
1483 /* uid */
1484 case 'u':
1485 rc = snprintf(out_ptr, out_end - out_ptr,
David Howells86a264a2008-11-14 10:39:18 +11001486 "%d", cred->uid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 if (rc > out_end - out_ptr)
1488 goto out;
1489 out_ptr += rc;
1490 break;
1491 /* gid */
1492 case 'g':
1493 rc = snprintf(out_ptr, out_end - out_ptr,
David Howells86a264a2008-11-14 10:39:18 +11001494 "%d", cred->gid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 if (rc > out_end - out_ptr)
1496 goto out;
1497 out_ptr += rc;
1498 break;
1499 /* signal that caused the coredump */
1500 case 's':
1501 rc = snprintf(out_ptr, out_end - out_ptr,
1502 "%ld", signr);
1503 if (rc > out_end - out_ptr)
1504 goto out;
1505 out_ptr += rc;
1506 break;
1507 /* UNIX time of coredump */
1508 case 't': {
1509 struct timeval tv;
1510 do_gettimeofday(&tv);
1511 rc = snprintf(out_ptr, out_end - out_ptr,
1512 "%lu", tv.tv_sec);
1513 if (rc > out_end - out_ptr)
1514 goto out;
1515 out_ptr += rc;
1516 break;
1517 }
1518 /* hostname */
1519 case 'h':
1520 down_read(&uts_sem);
1521 rc = snprintf(out_ptr, out_end - out_ptr,
Serge E. Hallyne9ff3992006-10-02 02:18:11 -07001522 "%s", utsname()->nodename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 up_read(&uts_sem);
1524 if (rc > out_end - out_ptr)
1525 goto out;
1526 out_ptr += rc;
1527 break;
1528 /* executable */
1529 case 'e':
1530 rc = snprintf(out_ptr, out_end - out_ptr,
1531 "%s", current->comm);
1532 if (rc > out_end - out_ptr)
1533 goto out;
1534 out_ptr += rc;
1535 break;
Neil Horman74aadce2007-10-16 23:26:35 -07001536 /* core limit size */
1537 case 'c':
1538 rc = snprintf(out_ptr, out_end - out_ptr,
Jiri Slabyd554ed892010-03-05 13:42:42 -08001539 "%lu", rlimit(RLIMIT_CORE));
Neil Horman74aadce2007-10-16 23:26:35 -07001540 if (rc > out_end - out_ptr)
1541 goto out;
1542 out_ptr += rc;
1543 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 default:
1545 break;
1546 }
1547 ++pat_ptr;
1548 }
1549 }
1550 /* Backward compatibility with core_uses_pid:
1551 *
1552 * If core_pattern does not include a %p (as is the default)
1553 * and core_uses_pid is set, then .%pid will be appended to
Alan Coxc4bbafd2007-04-16 22:53:13 -07001554 * the filename. Do not do this for piped commands. */
Oleg Nesterov64093242008-10-18 20:28:22 -07001555 if (!ispipe && !pid_in_pattern && core_uses_pid) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 rc = snprintf(out_ptr, out_end - out_ptr,
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001557 ".%d", task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 if (rc > out_end - out_ptr)
1559 goto out;
1560 out_ptr += rc;
1561 }
Alan Coxc4bbafd2007-04-16 22:53:13 -07001562out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 *out_ptr = 0;
Alan Coxc4bbafd2007-04-16 22:53:13 -07001564 return ispipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565}
1566
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001567static int zap_process(struct task_struct *start, int exit_code)
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001568{
1569 struct task_struct *t;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001570 int nr = 0;
Oleg Nesterov281de332006-06-26 00:26:06 -07001571
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001572 start->signal->flags = SIGNAL_GROUP_EXIT;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001573 start->signal->group_exit_code = exit_code;
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001574 start->signal->group_stop_count = 0;
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001575
1576 t = start;
1577 do {
1578 if (t != current && t->mm) {
Oleg Nesterov281de332006-06-26 00:26:06 -07001579 sigaddset(&t->pending.signal, SIGKILL);
1580 signal_wake_up(t, 1);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001581 nr++;
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001582 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001583 } while_each_thread(start, t);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001584
1585 return nr;
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001586}
1587
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001588static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001589 struct core_state *core_state, int exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590{
1591 struct task_struct *g, *p;
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001592 unsigned long flags;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001593 int nr = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001595 spin_lock_irq(&tsk->sighand->siglock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001596 if (!signal_group_exit(tsk->signal)) {
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001597 mm->core_state = core_state;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001598 nr = zap_process(tsk, exit_code);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 }
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001600 spin_unlock_irq(&tsk->sighand->siglock);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001601 if (unlikely(nr < 0))
1602 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001604 if (atomic_read(&mm->mm_users) == nr + 1)
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001605 goto done;
Oleg Nesterove4901f92008-07-25 01:47:31 -07001606 /*
1607 * We should find and kill all tasks which use this mm, and we should
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001608 * count them correctly into ->nr_threads. We don't take tasklist
Oleg Nesterove4901f92008-07-25 01:47:31 -07001609 * lock, but this is safe wrt:
1610 *
1611 * fork:
1612 * None of sub-threads can fork after zap_process(leader). All
1613 * processes which were created before this point should be
1614 * visible to zap_threads() because copy_process() adds the new
1615 * process to the tail of init_task.tasks list, and lock/unlock
1616 * of ->siglock provides a memory barrier.
1617 *
1618 * do_exit:
1619 * The caller holds mm->mmap_sem. This means that the task which
1620 * uses this mm can't pass exit_mm(), so it can't exit or clear
1621 * its ->mm.
1622 *
1623 * de_thread:
1624 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1625 * we must see either old or new leader, this does not matter.
1626 * However, it can change p->sighand, so lock_task_sighand(p)
1627 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1628 * it can't fail.
1629 *
1630 * Note also that "g" can be the old leader with ->mm == NULL
1631 * and already unhashed and thus removed from ->thread_group.
1632 * This is OK, __unhash_process()->list_del_rcu() does not
1633 * clear the ->next pointer, we will find the new leader via
1634 * next_thread().
1635 */
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001636 rcu_read_lock();
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001637 for_each_process(g) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001638 if (g == tsk->group_leader)
1639 continue;
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001640 if (g->flags & PF_KTHREAD)
1641 continue;
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001642 p = g;
1643 do {
1644 if (p->mm) {
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001645 if (unlikely(p->mm == mm)) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001646 lock_task_sighand(p, &flags);
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001647 nr += zap_process(p, exit_code);
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001648 unlock_task_sighand(p, &flags);
1649 }
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001650 break;
1651 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001652 } while_each_thread(g, p);
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001653 }
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001654 rcu_read_unlock();
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001655done:
Oleg Nesterovc5f1cc82008-07-25 01:47:42 -07001656 atomic_set(&core_state->nr_threads, nr);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001657 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658}
1659
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001660static int coredump_wait(int exit_code, struct core_state *core_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661{
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001662 struct task_struct *tsk = current;
1663 struct mm_struct *mm = tsk->mm;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001664 struct completion *vfork_done;
Oleg Nesterov2384f552005-10-30 15:02:47 -08001665 int core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001667 init_completion(&core_state->startup);
Oleg Nesterovb564daf2008-07-25 01:47:44 -07001668 core_state->dumper.task = tsk;
1669 core_state->dumper.next = NULL;
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001670 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001671 up_write(&mm->mmap_sem);
1672
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001673 if (unlikely(core_waiters < 0))
1674 goto fail;
1675
1676 /*
1677 * Make sure nobody is waiting for us to release the VM,
1678 * otherwise we can deadlock when we wait on each other
1679 */
1680 vfork_done = tsk->vfork_done;
1681 if (vfork_done) {
1682 tsk->vfork_done = NULL;
1683 complete(vfork_done);
1684 }
1685
Oleg Nesterov2384f552005-10-30 15:02:47 -08001686 if (core_waiters)
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001687 wait_for_completion(&core_state->startup);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001688fail:
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001689 return core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690}
1691
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001692static void coredump_finish(struct mm_struct *mm)
1693{
1694 struct core_thread *curr, *next;
1695 struct task_struct *task;
1696
1697 next = mm->core_state->dumper.next;
1698 while ((curr = next) != NULL) {
1699 next = curr->next;
1700 task = curr->task;
1701 /*
1702 * see exit_mm(), curr->task must not see
1703 * ->task == NULL before we read ->next.
1704 */
1705 smp_mb();
1706 curr->task = NULL;
1707 wake_up_process(task);
1708 }
1709
1710 mm->core_state = NULL;
1711}
1712
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001713/*
1714 * set_dumpable converts traditional three-value dumpable to two flags and
1715 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1716 * these bits are not changed atomically. So get_dumpable can observe the
1717 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1718 * return either old dumpable or new one by paying attention to the order of
1719 * modifying the bits.
1720 *
1721 * dumpable | mm->flags (binary)
1722 * old new | initial interim final
1723 * ---------+-----------------------
1724 * 0 1 | 00 01 01
1725 * 0 2 | 00 10(*) 11
1726 * 1 0 | 01 00 00
1727 * 1 2 | 01 11 11
1728 * 2 0 | 11 10(*) 00
1729 * 2 1 | 11 11 01
1730 *
1731 * (*) get_dumpable regards interim value of 10 as 11.
1732 */
1733void set_dumpable(struct mm_struct *mm, int value)
1734{
1735 switch (value) {
1736 case 0:
1737 clear_bit(MMF_DUMPABLE, &mm->flags);
1738 smp_wmb();
1739 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1740 break;
1741 case 1:
1742 set_bit(MMF_DUMPABLE, &mm->flags);
1743 smp_wmb();
1744 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1745 break;
1746 case 2:
1747 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1748 smp_wmb();
1749 set_bit(MMF_DUMPABLE, &mm->flags);
1750 break;
1751 }
1752}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001753
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001754static int __get_dumpable(unsigned long mm_flags)
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001755{
1756 int ret;
1757
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001758 ret = mm_flags & MMF_DUMPABLE_MASK;
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001759 return (ret >= 2) ? 2 : ret;
1760}
1761
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001762int get_dumpable(struct mm_struct *mm)
1763{
1764 return __get_dumpable(mm->flags);
1765}
1766
Neil Horman61be2282009-09-23 15:56:58 -07001767static void wait_for_dump_helpers(struct file *file)
1768{
1769 struct pipe_inode_info *pipe;
1770
1771 pipe = file->f_path.dentry->d_inode->i_pipe;
1772
1773 pipe_lock(pipe);
1774 pipe->readers++;
1775 pipe->writers--;
1776
1777 while ((pipe->readers > 1) && (!signal_pending(current))) {
1778 wake_up_interruptible_sync(&pipe->wait);
1779 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1780 pipe_wait(pipe);
1781 }
1782
1783 pipe->readers--;
1784 pipe->writers++;
1785 pipe_unlock(pipe);
1786
1787}
1788
1789
WANG Cong8cd3ac32009-01-06 14:42:48 -08001790void do_coredump(long signr, int exit_code, struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791{
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001792 struct core_state core_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 char corename[CORENAME_MAX_SIZE + 1];
1794 struct mm_struct *mm = current->mm;
1795 struct linux_binfmt * binfmt;
1796 struct inode * inode;
David Howellsd84f4f92008-11-14 10:39:23 +11001797 const struct cred *old_cred;
1798 struct cred *cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 int retval = 0;
Alan Coxd6e71142005-06-23 00:09:43 -07001800 int flag = 0;
Andi Kleend025c9d2006-09-30 23:29:28 -07001801 int ispipe = 0;
Neil Horman74aadce2007-10-16 23:26:35 -07001802 char **helper_argv = NULL;
1803 int helper_argc = 0;
Neil Hormana2939802009-09-23 15:56:56 -07001804 int dump_count = 0;
1805 static atomic_t core_dump_count = ATOMIC_INIT(0);
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001806 struct coredump_params cprm = {
1807 .signr = signr,
1808 .regs = regs,
Jiri Slabyd554ed892010-03-05 13:42:42 -08001809 .limit = rlimit(RLIMIT_CORE),
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001810 /*
1811 * We must use the same mm->flags while dumping core to avoid
1812 * inconsistency of bit flags, since this flag is not protected
1813 * by any locks.
1814 */
1815 .mm_flags = mm->flags,
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001816 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817
Steve Grubb0a4ff8c2007-04-19 10:28:21 -04001818 audit_core_dumps(signr);
1819
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001820 binfmt = mm->binfmt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 if (!binfmt || !binfmt->core_dump)
1822 goto fail;
David Howellsd84f4f92008-11-14 10:39:23 +11001823
1824 cred = prepare_creds();
1825 if (!cred) {
1826 retval = -ENOMEM;
1827 goto fail;
1828 }
1829
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 down_write(&mm->mmap_sem);
Roland McGrath00ec99d2007-11-11 19:13:43 -08001831 /*
1832 * If another thread got here first, or we are not dumpable, bail out.
1833 */
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001834 if (mm->core_state || !__get_dumpable(cprm.mm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 up_write(&mm->mmap_sem);
David Howellsd84f4f92008-11-14 10:39:23 +11001836 put_cred(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 goto fail;
1838 }
Alan Coxd6e71142005-06-23 00:09:43 -07001839
1840 /*
1841 * We cannot trust fsuid as being the "true" uid of the
1842 * process nor do we know its entire history. We only know it
1843 * was tainted so we dump it as root in mode 2.
1844 */
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001845 if (__get_dumpable(cprm.mm_flags) == 2) {
1846 /* Setuid core dump mode */
Alan Coxd6e71142005-06-23 00:09:43 -07001847 flag = O_EXCL; /* Stop rewrite attacks */
David Howellsd84f4f92008-11-14 10:39:23 +11001848 cred->fsuid = 0; /* Dump root private */
Alan Coxd6e71142005-06-23 00:09:43 -07001849 }
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001850
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001851 retval = coredump_wait(exit_code, &core_state);
David Howellsd84f4f92008-11-14 10:39:23 +11001852 if (retval < 0) {
1853 put_cred(cred);
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001854 goto fail;
David Howellsd84f4f92008-11-14 10:39:23 +11001855 }
1856
1857 old_cred = override_creds(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
1859 /*
1860 * Clear any false indication of pending signals that might
1861 * be seen by the filesystem code called to write the core file.
1862 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 clear_thread_flag(TIF_SIGPENDING);
1864
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 /*
1866 * lock_kernel() because format_corename() is controlled by sysctl, which
1867 * uses lock_kernel()
1868 */
1869 lock_kernel();
Oleg Nesterov64093242008-10-18 20:28:22 -07001870 ispipe = format_corename(corename, signr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 unlock_kernel();
Neil Horman725eae32009-09-23 15:56:54 -07001872
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001873 if ((!ispipe) && (cprm.limit < binfmt->min_coredump))
Neil Horman7dc0b222007-10-16 23:26:34 -07001874 goto fail_unlock;
1875
Alan Coxc4bbafd2007-04-16 22:53:13 -07001876 if (ispipe) {
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001877 if (cprm.limit == 0) {
Neil Horman725eae32009-09-23 15:56:54 -07001878 /*
1879 * Normally core limits are irrelevant to pipes, since
1880 * we're not writing to the file system, but we use
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001881 * cprm.limit of 0 here as a speacial value. Any
Neil Horman725eae32009-09-23 15:56:54 -07001882 * non-zero limit gets set to RLIM_INFINITY below, but
1883 * a limit of 0 skips the dump. This is a consistent
1884 * way to catch recursive crashes. We can still crash
1885 * if the core_pattern binary sets RLIM_CORE = !0
1886 * but it runs as root, and can do lots of stupid things
1887 * Note that we use task_tgid_vnr here to grab the pid
1888 * of the process group leader. That way we get the
1889 * right pid if a thread in a multi-threaded
1890 * core_pattern process dies.
1891 */
1892 printk(KERN_WARNING
1893 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1894 task_tgid_vnr(current), current->comm);
1895 printk(KERN_WARNING "Aborting core\n");
1896 goto fail_unlock;
1897 }
1898
Neil Hormana2939802009-09-23 15:56:56 -07001899 dump_count = atomic_inc_return(&core_dump_count);
1900 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1901 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1902 task_tgid_vnr(current), current->comm);
1903 printk(KERN_WARNING "Skipping core dump\n");
1904 goto fail_dropcount;
1905 }
1906
Neil Horman74aadce2007-10-16 23:26:35 -07001907 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
Tetsuo Handa350eaf72009-01-06 14:41:11 -08001908 if (!helper_argv) {
1909 printk(KERN_WARNING "%s failed to allocate memory\n",
1910 __func__);
Neil Hormana2939802009-09-23 15:56:56 -07001911 goto fail_dropcount;
Tetsuo Handa350eaf72009-01-06 14:41:11 -08001912 }
Neil Horman32321132007-10-16 23:26:36 -07001913
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001914 cprm.limit = RLIM_INFINITY;
Neil Horman32321132007-10-16 23:26:36 -07001915
Andi Kleend025c9d2006-09-30 23:29:28 -07001916 /* SIGPIPE can happen, but it's just never processed */
Neil Hormana2939802009-09-23 15:56:56 -07001917 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001918 &cprm.file)) {
Andi Kleend025c9d2006-09-30 23:29:28 -07001919 printk(KERN_INFO "Core dump to %s pipe failed\n",
1920 corename);
Neil Hormana2939802009-09-23 15:56:56 -07001921 goto fail_dropcount;
Andi Kleend025c9d2006-09-30 23:29:28 -07001922 }
Andi Kleend025c9d2006-09-30 23:29:28 -07001923 } else
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001924 cprm.file = filp_open(corename,
Alexey Dobriyan6d4df672006-12-06 20:40:39 -08001925 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1926 0600);
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001927 if (IS_ERR(cprm.file))
Neil Hormana2939802009-09-23 15:56:56 -07001928 goto fail_dropcount;
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001929 inode = cprm.file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 if (inode->i_nlink > 1)
1931 goto close_fail; /* multiple links - don't dump */
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001932 if (!ispipe && d_unhashed(cprm.file->f_path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 goto close_fail;
1934
Andi Kleend025c9d2006-09-30 23:29:28 -07001935 /* AK: actually i see no reason to not allow this for named pipes etc.,
1936 but keep the previous behaviour for now. */
1937 if (!ispipe && !S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 goto close_fail;
Ingo Molnarc46f7392007-11-28 13:59:18 +01001939 /*
1940 * Dont allow local users get cute and trick others to coredump
1941 * into their pre-created files:
Neil Horman76595f72010-03-05 13:44:16 -08001942 * Note, this is not relevant for pipes
Ingo Molnarc46f7392007-11-28 13:59:18 +01001943 */
Neil Horman76595f72010-03-05 13:44:16 -08001944 if (!ispipe && (inode->i_uid != current_fsuid()))
Ingo Molnarc46f7392007-11-28 13:59:18 +01001945 goto close_fail;
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001946 if (!cprm.file->f_op)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 goto close_fail;
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001948 if (!cprm.file->f_op->write)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 goto close_fail;
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001950 if (!ispipe &&
1951 do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 goto close_fail;
1953
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001954 retval = binfmt->core_dump(&cprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
1956 if (retval)
1957 current->signal->group_exit_code |= 0x80;
1958close_fail:
Neil Horman61be2282009-09-23 15:56:58 -07001959 if (ispipe && core_pipe_limit)
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001960 wait_for_dump_helpers(cprm.file);
1961 filp_close(cprm.file, NULL);
Neil Hormana2939802009-09-23 15:56:56 -07001962fail_dropcount:
1963 if (dump_count)
1964 atomic_dec(&core_dump_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965fail_unlock:
Neil Horman74aadce2007-10-16 23:26:35 -07001966 if (helper_argv)
1967 argv_free(helper_argv);
1968
David Howellsd84f4f92008-11-14 10:39:23 +11001969 revert_creds(old_cred);
1970 put_cred(cred);
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001971 coredump_finish(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972fail:
WANG Cong8cd3ac32009-01-06 14:42:48 -08001973 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974}