blob: 49cdaa19e5b9b800f9c23b8770f28e4cb33c8fbf [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 */
245 vma->vm_end = STACK_TOP_MAX;
246 vma->vm_start = vma->vm_end - PAGE_SIZE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700247 vma->vm_flags = VM_STACK_FLAGS;
Coly Li3ed75eb2007-10-18 23:39:15 -0700248 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Rik van Riel5beb4932010-03-05 13:42:07 -0800249 INIT_LIST_HEAD(&vma->anon_vma_chain);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700250 err = insert_vm_struct(mm, vma);
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800251 if (err)
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700252 goto err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700253
254 mm->stack_vm = mm->total_vm = 1;
255 up_write(&mm->mmap_sem);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700256 bprm->p = vma->vm_end - sizeof(void *);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700257 return 0;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700258err:
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800259 up_write(&mm->mmap_sem);
260 bprm->vma = NULL;
261 kmem_cache_free(vm_area_cachep, vma);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700262 return err;
263}
264
265static bool valid_arg_len(struct linux_binprm *bprm, long len)
266{
267 return len <= MAX_ARG_STRLEN;
268}
269
270#else
271
272static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
273 int write)
274{
275 struct page *page;
276
277 page = bprm->page[pos / PAGE_SIZE];
278 if (!page && write) {
279 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
280 if (!page)
281 return NULL;
282 bprm->page[pos / PAGE_SIZE] = page;
283 }
284
285 return page;
286}
287
288static void put_arg_page(struct page *page)
289{
290}
291
292static void free_arg_page(struct linux_binprm *bprm, int i)
293{
294 if (bprm->page[i]) {
295 __free_page(bprm->page[i]);
296 bprm->page[i] = NULL;
297 }
298}
299
300static void free_arg_pages(struct linux_binprm *bprm)
301{
302 int i;
303
304 for (i = 0; i < MAX_ARG_PAGES; i++)
305 free_arg_page(bprm, i);
306}
307
308static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
309 struct page *page)
310{
311}
312
313static int __bprm_mm_init(struct linux_binprm *bprm)
314{
315 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
316 return 0;
317}
318
319static bool valid_arg_len(struct linux_binprm *bprm, long len)
320{
321 return len <= bprm->p;
322}
323
324#endif /* CONFIG_MMU */
325
326/*
327 * Create a new mm_struct and populate it with a temporary stack
328 * vm_area_struct. We don't have enough context at this point to set the stack
329 * flags, permissions, and offset, so we use temporary values. We'll update
330 * them later in setup_arg_pages().
331 */
332int bprm_mm_init(struct linux_binprm *bprm)
333{
334 int err;
335 struct mm_struct *mm = NULL;
336
337 bprm->mm = mm = mm_alloc();
338 err = -ENOMEM;
339 if (!mm)
340 goto err;
341
342 err = init_new_context(current, mm);
343 if (err)
344 goto err;
345
346 err = __bprm_mm_init(bprm);
347 if (err)
348 goto err;
349
350 return 0;
351
352err:
353 if (mm) {
354 bprm->mm = NULL;
355 mmdrop(mm);
356 }
357
358 return err;
359}
360
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361/*
362 * count() counts the number of strings in array ARGV.
363 */
364static int count(char __user * __user * argv, int max)
365{
366 int i = 0;
367
368 if (argv != NULL) {
369 for (;;) {
370 char __user * p;
371
372 if (get_user(p, argv))
373 return -EFAULT;
374 if (!p)
375 break;
376 argv++;
Jason Baron362e6662008-10-15 22:01:52 -0700377 if (i++ >= max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 return -E2BIG;
379 cond_resched();
380 }
381 }
382 return i;
383}
384
385/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700386 * 'copy_strings()' copies argument/environment strings from the old
387 * processes's memory to the new process's stack. The call to get_user_pages()
388 * ensures the destination page is created and not swapped out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 */
Adrian Bunk75c96f82005-05-05 16:16:09 -0700390static int copy_strings(int argc, char __user * __user * argv,
391 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392{
393 struct page *kmapped_page = NULL;
394 char *kaddr = NULL;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700395 unsigned long kpos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 int ret;
397
398 while (argc-- > 0) {
399 char __user *str;
400 int len;
401 unsigned long pos;
402
403 if (get_user(str, argv+argc) ||
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700404 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 ret = -EFAULT;
406 goto out;
407 }
408
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700409 if (!valid_arg_len(bprm, len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 ret = -E2BIG;
411 goto out;
412 }
413
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700414 /* We're going to work our way backwords. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 pos = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700416 str += len;
417 bprm->p -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
419 while (len > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 int offset, bytes_to_copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
422 offset = pos % PAGE_SIZE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700423 if (offset == 0)
424 offset = PAGE_SIZE;
425
426 bytes_to_copy = offset;
427 if (bytes_to_copy > len)
428 bytes_to_copy = len;
429
430 offset -= bytes_to_copy;
431 pos -= bytes_to_copy;
432 str -= bytes_to_copy;
433 len -= bytes_to_copy;
434
435 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
436 struct page *page;
437
438 page = get_arg_page(bprm, pos, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 if (!page) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700440 ret = -E2BIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 goto out;
442 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700444 if (kmapped_page) {
445 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700447 put_arg_page(kmapped_page);
448 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 kmapped_page = page;
450 kaddr = kmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700451 kpos = pos & PAGE_MASK;
452 flush_arg_page(bprm, kpos, kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700454 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 ret = -EFAULT;
456 goto out;
457 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 }
459 }
460 ret = 0;
461out:
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700462 if (kmapped_page) {
463 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700465 put_arg_page(kmapped_page);
466 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 return ret;
468}
469
470/*
471 * Like copy_strings, but get argv and its values from kernel memory.
472 */
473int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
474{
475 int r;
476 mm_segment_t oldfs = get_fs();
477 set_fs(KERNEL_DS);
478 r = copy_strings(argc, (char __user * __user *)argv, bprm);
479 set_fs(oldfs);
480 return r;
481}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482EXPORT_SYMBOL(copy_strings_kernel);
483
484#ifdef CONFIG_MMU
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700485
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700487 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
488 * the binfmt code determines where the new stack should reside, we shift it to
489 * its final location. The process proceeds as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 *
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700491 * 1) Use shift to calculate the new vma endpoints.
492 * 2) Extend vma to cover both the old and new ranges. This ensures the
493 * arguments passed to subsequent functions are consistent.
494 * 3) Move vma's page tables to the new range.
495 * 4) Free up any cleared pgd range.
496 * 5) Shrink the vma to cover only the new range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700498static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
500 struct mm_struct *mm = vma->vm_mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700501 unsigned long old_start = vma->vm_start;
502 unsigned long old_end = vma->vm_end;
503 unsigned long length = old_end - old_start;
504 unsigned long new_start = old_start - shift;
505 unsigned long new_end = old_end - shift;
506 struct mmu_gather *tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700508 BUG_ON(new_start > new_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700510 /*
511 * ensure there are no vmas between where we want to go
512 * and where we are
513 */
514 if (vma != find_vma(mm, new_start))
515 return -EFAULT;
516
517 /*
518 * cover the whole range: [new_start, old_end)
519 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800520 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
521 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700522
523 /*
524 * move the page tables downwards, on failure we rely on
525 * process cleanup to remove whatever mess we made.
526 */
527 if (length != move_page_tables(vma, old_start,
528 vma, new_start, length))
529 return -ENOMEM;
530
531 lru_add_drain();
532 tlb = tlb_gather_mmu(mm, 0);
533 if (new_end > old_start) {
534 /*
535 * when the old and new regions overlap clear from new_end.
536 */
Jan Beulich42b77722008-07-23 21:27:10 -0700537 free_pgd_range(tlb, new_end, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700538 vma->vm_next ? vma->vm_next->vm_start : 0);
539 } else {
540 /*
541 * otherwise, clean from old_start; this is done to not touch
542 * the address space in [new_end, old_start) some architectures
543 * have constraints on va-space that make this illegal (IA64) -
544 * for the others its just a little faster.
545 */
Jan Beulich42b77722008-07-23 21:27:10 -0700546 free_pgd_range(tlb, old_start, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700547 vma->vm_next ? vma->vm_next->vm_start : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700549 tlb_finish_mmu(tlb, new_end, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700551 /*
Rik van Riel5beb4932010-03-05 13:42:07 -0800552 * Shrink the vma to just the new range. Always succeeds.
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700553 */
554 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
555
556 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557}
558
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700559/*
560 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
561 * the stack is optionally relocated, and some extra space is added.
562 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563int setup_arg_pages(struct linux_binprm *bprm,
564 unsigned long stack_top,
565 int executable_stack)
566{
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700567 unsigned long ret;
568 unsigned long stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 struct mm_struct *mm = current->mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700570 struct vm_area_struct *vma = bprm->vma;
571 struct vm_area_struct *prev = NULL;
572 unsigned long vm_flags;
573 unsigned long stack_base;
Michael Neuling803bf5e2010-02-10 13:56:42 -0800574 unsigned long stack_size;
575 unsigned long stack_expand;
576 unsigned long rlim_stack;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578#ifdef CONFIG_STACK_GROWSUP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 /* Limit stack size to 1GB */
Jiri Slabyd554ed892010-03-05 13:42:42 -0800580 stack_base = rlimit_max(RLIMIT_STACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 if (stack_base > (1 << 30))
582 stack_base = 1 << 30;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700584 /* Make sure we didn't let the argument array grow too large. */
585 if (vma->vm_end - vma->vm_start > stack_base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 return -ENOMEM;
587
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700588 stack_base = PAGE_ALIGN(stack_top - stack_base);
589
590 stack_shift = vma->vm_start - stack_base;
591 mm->arg_start = bprm->p - stack_shift;
592 bprm->p = vma->vm_end - stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593#else
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700594 stack_top = arch_align_stack(stack_top);
595 stack_top = PAGE_ALIGN(stack_top);
596 stack_shift = vma->vm_end - stack_top;
597
598 bprm->p -= stack_shift;
599 mm->arg_start = bprm->p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600#endif
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700601
602 if (bprm->loader)
603 bprm->loader -= stack_shift;
604 bprm->exec -= stack_shift;
605
606 down_write(&mm->mmap_sem);
Hugh Dickins96a8e132008-07-10 21:19:20 +0100607 vm_flags = VM_STACK_FLAGS;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700608
609 /*
610 * Adjust stack execute permissions; explicitly enable for
611 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
612 * (arch default) otherwise.
613 */
614 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
615 vm_flags |= VM_EXEC;
616 else if (executable_stack == EXSTACK_DISABLE_X)
617 vm_flags &= ~VM_EXEC;
618 vm_flags |= mm->def_flags;
619
620 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
621 vm_flags);
622 if (ret)
623 goto out_unlock;
624 BUG_ON(prev != vma);
625
626 /* Move stack pages down in memory. */
627 if (stack_shift) {
628 ret = shift_arg_pages(vma, stack_shift);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800629 if (ret)
630 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 }
632
Michael Neuling5ef097d2010-03-05 13:42:57 -0800633 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
Michael Neuling803bf5e2010-02-10 13:56:42 -0800634 stack_size = vma->vm_end - vma->vm_start;
635 /*
636 * Align this down to a page boundary as expand_stack
637 * will align it up.
638 */
639 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700640#ifdef CONFIG_STACK_GROWSUP
Michael Neuling803bf5e2010-02-10 13:56:42 -0800641 if (stack_size + stack_expand > rlim_stack)
642 stack_base = vma->vm_start + rlim_stack;
643 else
644 stack_base = vma->vm_end + stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700645#else
Michael Neuling803bf5e2010-02-10 13:56:42 -0800646 if (stack_size + stack_expand > rlim_stack)
647 stack_base = vma->vm_end - rlim_stack;
648 else
649 stack_base = vma->vm_start - stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700650#endif
651 ret = expand_stack(vma, stack_base);
652 if (ret)
653 ret = -EFAULT;
654
655out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 up_write(&mm->mmap_sem);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800657 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659EXPORT_SYMBOL(setup_arg_pages);
660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661#endif /* CONFIG_MMU */
662
663struct file *open_exec(const char *name)
664{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 struct file *file;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200666 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Al Viro6e8341a2009-04-06 11:16:22 -0400668 file = do_filp_open(AT_FDCWD, name,
669 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
670 MAY_EXEC | MAY_OPEN);
671 if (IS_ERR(file))
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200672 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200674 err = -EACCES;
Al Viro6e8341a2009-04-06 11:16:22 -0400675 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
676 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200677
Al Viro6e8341a2009-04-06 11:16:22 -0400678 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
679 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200680
Eric Paris6110e3a2008-12-17 13:53:20 -0500681 fsnotify_open(file->f_path.dentry);
682
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200683 err = deny_write_access(file);
Al Viro6e8341a2009-04-06 11:16:22 -0400684 if (err)
685 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
Al Viro6e8341a2009-04-06 11:16:22 -0400687out:
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200688 return file;
689
Al Viro6e8341a2009-04-06 11:16:22 -0400690exit:
691 fput(file);
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200692 return ERR_PTR(err);
693}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694EXPORT_SYMBOL(open_exec);
695
Mimi Zohar6777d772009-08-21 14:32:48 -0400696int kernel_read(struct file *file, loff_t offset,
697 char *addr, unsigned long count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
699 mm_segment_t old_fs;
700 loff_t pos = offset;
701 int result;
702
703 old_fs = get_fs();
704 set_fs(get_ds());
705 /* The cast to a user pointer is valid due to the set_fs() */
706 result = vfs_read(file, (void __user *)addr, count, &pos);
707 set_fs(old_fs);
708 return result;
709}
710
711EXPORT_SYMBOL(kernel_read);
712
713static int exec_mmap(struct mm_struct *mm)
714{
715 struct task_struct *tsk;
716 struct mm_struct * old_mm, *active_mm;
717
718 /* Notify parent that we're no longer interested in the old VM */
719 tsk = current;
720 old_mm = current->mm;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800721 sync_mm_rss(tsk, old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 mm_release(tsk, old_mm);
723
724 if (old_mm) {
725 /*
726 * Make sure that if there is a core dump in progress
727 * for the old mm, we get out and die instead of going
728 * through with the exec. We must hold mmap_sem around
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700729 * checking core_state and changing tsk->mm.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 */
731 down_read(&old_mm->mmap_sem);
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700732 if (unlikely(old_mm->core_state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 up_read(&old_mm->mmap_sem);
734 return -EINTR;
735 }
736 }
737 task_lock(tsk);
738 active_mm = tsk->active_mm;
739 tsk->mm = mm;
740 tsk->active_mm = mm;
741 activate_mm(active_mm, mm);
742 task_unlock(tsk);
743 arch_pick_mmap_layout(mm);
744 if (old_mm) {
745 up_read(&old_mm->mmap_sem);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +0200746 BUG_ON(active_mm != old_mm);
Balbir Singh31a78f22008-09-28 23:09:31 +0100747 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 mmput(old_mm);
749 return 0;
750 }
751 mmdrop(active_mm);
752 return 0;
753}
754
755/*
756 * This function makes sure the current process has its own signal table,
757 * so that flush_signal_handlers can later reset the handlers without
758 * disturbing other processes. (Other processes might share the signal
759 * table via the CLONE_SIGHAND option to clone().)
760 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800761static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762{
763 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700764 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 spinlock_t *lock = &oldsighand->siglock;
766 int count;
767
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700768 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 goto no_thread_group;
770
771 /*
772 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800775 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 /*
777 * Another group action in progress, just
778 * return so that the signal is processed.
779 */
780 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 return -EAGAIN;
782 }
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800783 sig->group_exit_task = tsk;
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700784 zap_other_threads(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
Oleg Nesterovfea9d172008-02-08 04:19:19 -0800786 /* Account for the thread group leader hanging around: */
787 count = thread_group_leader(tsk) ? 1 : 2;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700788 sig->notify_count = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 while (atomic_read(&sig->count) > count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 __set_current_state(TASK_UNINTERRUPTIBLE);
791 spin_unlock_irq(lock);
792 schedule();
793 spin_lock_irq(lock);
794 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 spin_unlock_irq(lock);
796
797 /*
798 * At this point all other threads have exited, all we have to
799 * do is to wait for the thread group leader to become inactive,
800 * and to assume its PID:
801 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700802 if (!thread_group_leader(tsk)) {
Oleg Nesterov81879262008-12-01 14:18:16 -0800803 struct task_struct *leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700804
Oleg Nesterov2800d8d2008-04-30 00:53:12 -0700805 sig->notify_count = -1; /* for exit_notify() */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700806 for (;;) {
807 write_lock_irq(&tasklist_lock);
808 if (likely(leader->exit_state))
809 break;
810 __set_current_state(TASK_UNINTERRUPTIBLE);
811 write_unlock_irq(&tasklist_lock);
812 schedule();
813 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
Roland McGrathf5e90282006-04-10 22:54:16 -0700815 /*
816 * The only record we have of the real-time age of a
817 * process, regardless of execs it's done, is start_time.
818 * All the past CPU time is accumulated in signal_struct
819 * from sister threads now dead. But in this non-leader
820 * exec, nothing survives from the original leader thread,
821 * whose birth marks the true age of this process now.
822 * When we take on its identity by switching to its PID, we
823 * also take its birthdate (always earlier than our own).
824 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700825 tsk->start_time = leader->start_time;
Roland McGrathf5e90282006-04-10 22:54:16 -0700826
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700827 BUG_ON(!same_thread_group(leader, tsk));
828 BUG_ON(has_group_leader_pid(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 /*
830 * An exec() starts a new thread group with the
831 * TGID of the previous thread group. Rehash the
832 * two threads with a switched PID, and release
833 * the former thread group leader:
834 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800835
836 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700837 * The old leader becomes a thread of the this thread group.
838 * Note: The old leader also uses this pid until release_task
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800839 * is called. Odd but simple and correct.
840 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700841 detach_pid(tsk, PIDTYPE_PID);
842 tsk->pid = leader->pid;
Sukadev Bhattiprolu3743ca02007-10-18 23:39:51 -0700843 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700844 transfer_pid(leader, tsk, PIDTYPE_PGID);
845 transfer_pid(leader, tsk, PIDTYPE_SID);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800846
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700847 list_replace_rcu(&leader->tasks, &tsk->tasks);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800848 list_replace_init(&leader->sibling, &tsk->sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700850 tsk->group_leader = tsk;
851 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -0600852
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700853 tsk->exit_signal = SIGCHLD;
Oleg Nesterov962b5642005-11-23 13:37:43 -0800854
855 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
856 leader->exit_state = EXIT_DEAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 write_unlock_irq(&tasklist_lock);
Oleg Nesterov81879262008-12-01 14:18:16 -0800858
859 release_task(leader);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800860 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700862 sig->group_exit_task = NULL;
863 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
865no_thread_group:
Jiri Pirko1f102062009-09-22 16:44:10 -0700866 if (current->mm)
867 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
868
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 exit_itimers(sig);
Oleg Nesterovcbaffba2008-05-26 20:55:42 +0400870 flush_itimer_signals();
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300871
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700872 if (atomic_read(&oldsighand->count) != 1) {
873 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700875 * This ->sighand is shared with the CLONE_SIGHAND
876 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700878 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
879 if (!newsighand)
880 return -ENOMEM;
881
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 atomic_set(&newsighand->count, 1);
883 memcpy(newsighand->action, oldsighand->action,
884 sizeof(newsighand->action));
885
886 write_lock_irq(&tasklist_lock);
887 spin_lock(&oldsighand->siglock);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700888 rcu_assign_pointer(tsk->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 spin_unlock(&oldsighand->siglock);
890 write_unlock_irq(&tasklist_lock);
891
Davide Libenzifba2afa2007-05-10 22:23:13 -0700892 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 }
894
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700895 BUG_ON(!thread_group_leader(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 return 0;
897}
Oleg Nesterov0840a902007-10-16 23:27:22 -0700898
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899/*
900 * These functions flushes out all traces of the currently running executable
901 * so that a new one can be started
902 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800903static void flush_old_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904{
905 long j = -1;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700906 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
908 spin_lock(&files->file_lock);
909 for (;;) {
910 unsigned long set, i;
911
912 j++;
913 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700914 fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800915 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700917 set = fdt->close_on_exec->fds_bits[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 if (!set)
919 continue;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700920 fdt->close_on_exec->fds_bits[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 spin_unlock(&files->file_lock);
922 for ( ; set ; i++,set >>= 1) {
923 if (set & 1) {
924 sys_close(i);
925 }
926 }
927 spin_lock(&files->file_lock);
928
929 }
930 spin_unlock(&files->file_lock);
931}
932
Andrew Morton59714d62008-02-04 22:27:21 -0800933char *get_task_comm(char *buf, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934{
935 /* buf must be at least sizeof(tsk->comm) in size */
936 task_lock(tsk);
937 strncpy(buf, tsk->comm, sizeof(tsk->comm));
938 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -0800939 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940}
941
942void set_task_comm(struct task_struct *tsk, char *buf)
943{
944 task_lock(tsk);
john stultz4614a696b2009-12-14 18:00:05 -0800945
946 /*
947 * Threads may access current->comm without holding
948 * the task lock, so write the string carefully.
949 * Readers without a lock may see incomplete new
950 * names but are safe from non-terminating string reads.
951 */
952 memset(tsk->comm, 0, TASK_COMM_LEN);
953 wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
955 task_unlock(tsk);
Ingo Molnarcdd6c482009-09-21 12:02:48 +0200956 perf_event_comm(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957}
958
959int flush_old_exec(struct linux_binprm * bprm)
960{
Linus Torvalds221af7f2010-01-28 22:14:42 -0800961 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
963 /*
964 * Make sure we have a private signal table and that
965 * we are unassociated from the previous thread group.
966 */
967 retval = de_thread(current);
968 if (retval)
969 goto out;
970
Matt Helsley925d1c42008-04-29 01:01:36 -0700971 set_mm_exe_file(bprm->mm, bprm->file);
972
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 * Release all of the old mmap stuff
975 */
976 retval = exec_mmap(bprm->mm);
977 if (retval)
Al Virofd8328b2008-04-22 05:11:59 -0400978 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
980 bprm->mm = NULL; /* We're using it now */
Linus Torvalds7ab02af2010-02-02 12:37:44 -0800981
982 current->flags &= ~PF_RANDOMIZE;
983 flush_thread();
984 current->personality &= ~bprm->per_clear;
985
Linus Torvalds221af7f2010-01-28 22:14:42 -0800986 return 0;
987
988out:
989 return retval;
990}
991EXPORT_SYMBOL(flush_old_exec);
992
993void setup_new_exec(struct linux_binprm * bprm)
994{
995 int i, ch;
996 char * name;
997 char tcomm[sizeof(current->comm)];
998
999 arch_pick_mmap_layout(current->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
1001 /* This is the point of no return */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 current->sas_ss_sp = current->sas_ss_size = 0;
1003
David Howellsda9592e2008-11-14 10:39:05 +11001004 if (current_euid() == current_uid() && current_egid() == current_gid())
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001005 set_dumpable(current->mm, 1);
Alan Coxd6e71142005-06-23 00:09:43 -07001006 else
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001007 set_dumpable(current->mm, suid_dumpable);
Alan Coxd6e71142005-06-23 00:09:43 -07001008
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 name = bprm->filename;
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -07001010
1011 /* Copies the binary name from after last slash */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 for (i=0; (ch = *(name++)) != '\0';) {
1013 if (ch == '/')
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -07001014 i = 0; /* overwrite what we wrote */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 else
1016 if (i < (sizeof(tcomm) - 1))
1017 tcomm[i++] = ch;
1018 }
1019 tcomm[i] = '\0';
1020 set_task_comm(current, tcomm);
1021
Benjamin Herrenschmidt0551fbd2006-02-28 16:59:19 -08001022 /* Set the new mm task size. We have to do that late because it may
1023 * depend on TIF_32BIT which is only updated in flush_thread() on
1024 * some architectures like powerpc
1025 */
1026 current->mm->task_size = TASK_SIZE;
1027
David Howellsa6f76f22008-11-14 10:39:24 +11001028 /* install the new credentials */
1029 if (bprm->cred->uid != current_euid() ||
1030 bprm->cred->gid != current_egid()) {
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001031 current->pdeath_signal = 0;
1032 } else if (file_permission(bprm->file, MAY_READ) ||
David Howellsa6f76f22008-11-14 10:39:24 +11001033 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001034 set_dumpable(current->mm, suid_dumpable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 }
1036
Ingo Molnarf65cb452008-12-16 13:40:44 +01001037 /*
1038 * Flush performance counters when crossing a
1039 * security domain:
1040 */
1041 if (!get_dumpable(current->mm))
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001042 perf_event_exit_task(current);
Ingo Molnarf65cb452008-12-16 13:40:44 +01001043
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 /* An exec changes our domain. We are no longer part of the thread
1045 group */
1046
1047 current->self_exec_id++;
1048
1049 flush_signal_handlers(current, 0);
1050 flush_old_files(current->files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051}
Linus Torvalds221af7f2010-01-28 22:14:42 -08001052EXPORT_SYMBOL(setup_new_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
David Howellsa6f76f22008-11-14 10:39:24 +11001054/*
Oleg Nesterova2a84742009-09-05 11:17:13 -07001055 * Prepare credentials and lock ->cred_guard_mutex.
1056 * install_exec_creds() commits the new creds and drops the lock.
1057 * Or, if exec fails before, free_bprm() should release ->cred and
1058 * and unlock.
1059 */
1060int prepare_bprm_creds(struct linux_binprm *bprm)
1061{
1062 if (mutex_lock_interruptible(&current->cred_guard_mutex))
1063 return -ERESTARTNOINTR;
1064
1065 bprm->cred = prepare_exec_creds();
1066 if (likely(bprm->cred))
1067 return 0;
1068
1069 mutex_unlock(&current->cred_guard_mutex);
1070 return -ENOMEM;
1071}
1072
1073void free_bprm(struct linux_binprm *bprm)
1074{
1075 free_arg_pages(bprm);
1076 if (bprm->cred) {
1077 mutex_unlock(&current->cred_guard_mutex);
1078 abort_creds(bprm->cred);
1079 }
1080 kfree(bprm);
1081}
1082
1083/*
David Howellsa6f76f22008-11-14 10:39:24 +11001084 * install the new credentials for this executable
1085 */
1086void install_exec_creds(struct linux_binprm *bprm)
1087{
1088 security_bprm_committing_creds(bprm);
1089
1090 commit_creds(bprm->cred);
1091 bprm->cred = NULL;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001092 /*
1093 * cred_guard_mutex must be held at least to this point to prevent
David Howellsa6f76f22008-11-14 10:39:24 +11001094 * ptrace_attach() from altering our determination of the task's
Oleg Nesterova2a84742009-09-05 11:17:13 -07001095 * credentials; any time after this it may be unlocked.
1096 */
David Howellsa6f76f22008-11-14 10:39:24 +11001097 security_bprm_committed_creds(bprm);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001098 mutex_unlock(&current->cred_guard_mutex);
David Howellsa6f76f22008-11-14 10:39:24 +11001099}
1100EXPORT_SYMBOL(install_exec_creds);
1101
1102/*
1103 * determine how safe it is to execute the proposed program
David Howells5e751e92009-05-08 13:55:22 +01001104 * - the caller must hold current->cred_guard_mutex to protect against
David Howellsa6f76f22008-11-14 10:39:24 +11001105 * PTRACE_ATTACH
1106 */
Al Viro498052b2009-03-30 07:20:30 -04001107int check_unsafe_exec(struct linux_binprm *bprm)
David Howellsa6f76f22008-11-14 10:39:24 +11001108{
David Howells0bf2f3a2009-02-06 11:45:46 +00001109 struct task_struct *p = current, *t;
Al Virof1191b52009-03-30 07:35:18 -04001110 unsigned n_fs;
Al Viro498052b2009-03-30 07:20:30 -04001111 int res = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001112
1113 bprm->unsafe = tracehook_unsafe_exec(p);
1114
David Howells0bf2f3a2009-02-06 11:45:46 +00001115 n_fs = 1;
Al Viro498052b2009-03-30 07:20:30 -04001116 write_lock(&p->fs->lock);
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001117 rcu_read_lock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001118 for (t = next_thread(p); t != p; t = next_thread(t)) {
1119 if (t->fs == p->fs)
1120 n_fs++;
David Howells0bf2f3a2009-02-06 11:45:46 +00001121 }
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001122 rcu_read_unlock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001123
Al Virof1191b52009-03-30 07:35:18 -04001124 if (p->fs->users > n_fs) {
David Howellsa6f76f22008-11-14 10:39:24 +11001125 bprm->unsafe |= LSM_UNSAFE_SHARE;
Al Viro498052b2009-03-30 07:20:30 -04001126 } else {
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001127 res = -EAGAIN;
1128 if (!p->fs->in_exec) {
1129 p->fs->in_exec = 1;
1130 res = 1;
1131 }
Al Viro498052b2009-03-30 07:20:30 -04001132 }
Al Viro498052b2009-03-30 07:20:30 -04001133 write_unlock(&p->fs->lock);
1134
1135 return res;
David Howellsa6f76f22008-11-14 10:39:24 +11001136}
1137
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138/*
1139 * Fill the binprm structure from the inode.
1140 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
David Howellsa6f76f22008-11-14 10:39:24 +11001141 *
1142 * This may be called multiple times for binary chains (scripts for example).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 */
1144int prepare_binprm(struct linux_binprm *bprm)
1145{
David Howellsa6f76f22008-11-14 10:39:24 +11001146 umode_t mode;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001147 struct inode * inode = bprm->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 int retval;
1149
1150 mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 if (bprm->file->f_op == NULL)
1152 return -EACCES;
1153
David Howellsa6f76f22008-11-14 10:39:24 +11001154 /* clear any previous set[ug]id data from a previous binary */
1155 bprm->cred->euid = current_euid();
1156 bprm->cred->egid = current_egid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
David Howellsa6f76f22008-11-14 10:39:24 +11001158 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 /* Set-uid? */
1160 if (mode & S_ISUID) {
David Howellsa6f76f22008-11-14 10:39:24 +11001161 bprm->per_clear |= PER_CLEAR_ON_SETID;
1162 bprm->cred->euid = inode->i_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 }
1164
1165 /* Set-gid? */
1166 /*
1167 * If setgid is set but no group execute bit then this
1168 * is a candidate for mandatory locking, not a setgid
1169 * executable.
1170 */
1171 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
David Howellsa6f76f22008-11-14 10:39:24 +11001172 bprm->per_clear |= PER_CLEAR_ON_SETID;
1173 bprm->cred->egid = inode->i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 }
1175 }
1176
1177 /* fill in binprm security blob */
David Howellsa6f76f22008-11-14 10:39:24 +11001178 retval = security_bprm_set_creds(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 if (retval)
1180 return retval;
David Howellsa6f76f22008-11-14 10:39:24 +11001181 bprm->cred_prepared = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
David Howellsa6f76f22008-11-14 10:39:24 +11001183 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1184 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185}
1186
1187EXPORT_SYMBOL(prepare_binprm);
1188
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001189/*
1190 * Arguments are '\0' separated strings found at the location bprm->p
1191 * points to; chop off the first by relocating brpm->p to right after
1192 * the first '\0' encountered.
1193 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001194int remove_arg_zero(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195{
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001196 int ret = 0;
1197 unsigned long offset;
1198 char *kaddr;
1199 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001201 if (!bprm->argc)
1202 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001204 do {
1205 offset = bprm->p & ~PAGE_MASK;
1206 page = get_arg_page(bprm, bprm->p, 0);
1207 if (!page) {
1208 ret = -EFAULT;
1209 goto out;
1210 }
1211 kaddr = kmap_atomic(page, KM_USER0);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001212
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001213 for (; offset < PAGE_SIZE && kaddr[offset];
1214 offset++, bprm->p++)
1215 ;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001216
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001217 kunmap_atomic(kaddr, KM_USER0);
1218 put_arg_page(page);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001219
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001220 if (offset == PAGE_SIZE)
1221 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1222 } while (offset == PAGE_SIZE);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001223
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001224 bprm->p++;
1225 bprm->argc--;
1226 ret = 0;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001227
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001228out:
1229 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231EXPORT_SYMBOL(remove_arg_zero);
1232
1233/*
1234 * cycle the list of binary formats handler, until one recognizes the image
1235 */
1236int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1237{
Roland McGrath85f33462008-12-09 19:36:38 -08001238 unsigned int depth = bprm->recursion_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 int try,retval;
1240 struct linux_binfmt *fmt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 retval = security_bprm_check(bprm);
1243 if (retval)
1244 return retval;
1245
1246 /* kernel module loader fixup */
1247 /* so we don't try to load run modprobe in kernel space. */
1248 set_fs(USER_DS);
Al Viro473ae302006-04-26 14:04:08 -04001249
1250 retval = audit_bprm(bprm);
1251 if (retval)
1252 return retval;
1253
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 retval = -ENOENT;
1255 for (try=0; try<2; try++) {
1256 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -07001257 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1259 if (!fn)
1260 continue;
1261 if (!try_module_get(fmt->module))
1262 continue;
1263 read_unlock(&binfmt_lock);
1264 retval = fn(bprm, regs);
Roland McGrath85f33462008-12-09 19:36:38 -08001265 /*
1266 * Restore the depth counter to its starting value
1267 * in this call, so we don't have to rely on every
1268 * load_binary function to restore it on return.
1269 */
1270 bprm->recursion_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 if (retval >= 0) {
Roland McGrath85f33462008-12-09 19:36:38 -08001272 if (depth == 0)
1273 tracehook_report_exec(fmt, bprm, regs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 put_binfmt(fmt);
1275 allow_write_access(bprm->file);
1276 if (bprm->file)
1277 fput(bprm->file);
1278 bprm->file = NULL;
1279 current->did_exec = 1;
Matt Helsley9f460802005-11-07 00:59:16 -08001280 proc_exec_connector(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 return retval;
1282 }
1283 read_lock(&binfmt_lock);
1284 put_binfmt(fmt);
1285 if (retval != -ENOEXEC || bprm->mm == NULL)
1286 break;
1287 if (!bprm->file) {
1288 read_unlock(&binfmt_lock);
1289 return retval;
1290 }
1291 }
1292 read_unlock(&binfmt_lock);
1293 if (retval != -ENOEXEC || bprm->mm == NULL) {
1294 break;
Johannes Berg5f4123b2008-07-09 10:28:40 +02001295#ifdef CONFIG_MODULES
1296 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1298 if (printable(bprm->buf[0]) &&
1299 printable(bprm->buf[1]) &&
1300 printable(bprm->buf[2]) &&
1301 printable(bprm->buf[3]))
1302 break; /* -ENOEXEC */
1303 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1304#endif
1305 }
1306 }
1307 return retval;
1308}
1309
1310EXPORT_SYMBOL(search_binary_handler);
1311
1312/*
1313 * sys_execve() executes a new program.
1314 */
1315int do_execve(char * filename,
1316 char __user *__user *argv,
1317 char __user *__user *envp,
1318 struct pt_regs * regs)
1319{
1320 struct linux_binprm *bprm;
1321 struct file *file;
Al Viro3b125382008-04-22 05:31:30 -04001322 struct files_struct *displaced;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001323 bool clear_in_exec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
Al Viro3b125382008-04-22 05:31:30 -04001326 retval = unshare_files(&displaced);
Al Virofd8328b2008-04-22 05:11:59 -04001327 if (retval)
1328 goto out_ret;
1329
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 retval = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001331 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 if (!bprm)
Al Virofd8328b2008-04-22 05:11:59 -04001333 goto out_files;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
Oleg Nesterova2a84742009-09-05 11:17:13 -07001335 retval = prepare_bprm_creds(bprm);
1336 if (retval)
David Howellsa6f76f22008-11-14 10:39:24 +11001337 goto out_free;
Al Viro498052b2009-03-30 07:20:30 -04001338
1339 retval = check_unsafe_exec(bprm);
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001340 if (retval < 0)
Oleg Nesterova2a84742009-09-05 11:17:13 -07001341 goto out_free;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001342 clear_in_exec = retval;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001343 current->in_execve = 1;
David Howellsa6f76f22008-11-14 10:39:24 +11001344
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 file = open_exec(filename);
1346 retval = PTR_ERR(file);
1347 if (IS_ERR(file))
Al Viro498052b2009-03-30 07:20:30 -04001348 goto out_unmark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
1350 sched_exec();
1351
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 bprm->file = file;
1353 bprm->filename = filename;
1354 bprm->interp = filename;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001355
1356 retval = bprm_mm_init(bprm);
1357 if (retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 goto out_file;
1359
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001360 bprm->argc = count(argv, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 if ((retval = bprm->argc) < 0)
David Howellsa6f76f22008-11-14 10:39:24 +11001362 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001364 bprm->envc = count(envp, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 if ((retval = bprm->envc) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 goto out;
1367
1368 retval = prepare_binprm(bprm);
1369 if (retval < 0)
1370 goto out;
1371
1372 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1373 if (retval < 0)
1374 goto out;
1375
1376 bprm->exec = bprm->p;
1377 retval = copy_strings(bprm->envc, envp, bprm);
1378 if (retval < 0)
1379 goto out;
1380
1381 retval = copy_strings(bprm->argc, argv, bprm);
1382 if (retval < 0)
1383 goto out;
1384
Oleg Nesterov7b34e422008-07-25 01:47:37 -07001385 current->flags &= ~PF_KTHREAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 retval = search_binary_handler(bprm,regs);
David Howellsa6f76f22008-11-14 10:39:24 +11001387 if (retval < 0)
1388 goto out;
1389
Stefani Seiboldd899bf72009-09-22 16:45:40 -07001390 current->stack_start = current->mm->start_stack;
1391
David Howellsa6f76f22008-11-14 10:39:24 +11001392 /* execve succeeded */
Al Viro498052b2009-03-30 07:20:30 -04001393 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001394 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001395 acct_update_integrals(current);
1396 free_bprm(bprm);
1397 if (displaced)
1398 put_files_struct(displaced);
1399 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
1401out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 if (bprm->mm)
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001403 mmput (bprm->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
1405out_file:
1406 if (bprm->file) {
1407 allow_write_access(bprm->file);
1408 fput(bprm->file);
1409 }
David Howellsa6f76f22008-11-14 10:39:24 +11001410
Al Viro498052b2009-03-30 07:20:30 -04001411out_unmark:
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001412 if (clear_in_exec)
1413 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001414 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001415
1416out_free:
Al Viro08a6fac2008-05-10 16:38:25 -04001417 free_bprm(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Al Virofd8328b2008-04-22 05:11:59 -04001419out_files:
Al Viro3b125382008-04-22 05:31:30 -04001420 if (displaced)
1421 reset_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422out_ret:
1423 return retval;
1424}
1425
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001426void set_binfmt(struct linux_binfmt *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427{
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001428 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001430 if (mm->binfmt)
1431 module_put(mm->binfmt->module);
1432
1433 mm->binfmt = new;
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001434 if (new)
1435 __module_get(new->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436}
1437
1438EXPORT_SYMBOL(set_binfmt);
1439
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440/* format_corename will inspect the pattern parameter, and output a
1441 * name into corename, which must have space for at least
1442 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1443 */
Oleg Nesterov64093242008-10-18 20:28:22 -07001444static int format_corename(char *corename, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445{
David Howells86a264a2008-11-14 10:39:18 +11001446 const struct cred *cred = current_cred();
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001447 const char *pat_ptr = core_pattern;
1448 int ispipe = (*pat_ptr == '|');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 char *out_ptr = corename;
1450 char *const out_end = corename + CORENAME_MAX_SIZE;
1451 int rc;
1452 int pid_in_pattern = 0;
1453
1454 /* Repeat as long as we have more pattern to process and more output
1455 space */
1456 while (*pat_ptr) {
1457 if (*pat_ptr != '%') {
1458 if (out_ptr == out_end)
1459 goto out;
1460 *out_ptr++ = *pat_ptr++;
1461 } else {
1462 switch (*++pat_ptr) {
1463 case 0:
1464 goto out;
1465 /* Double percent, output one percent */
1466 case '%':
1467 if (out_ptr == out_end)
1468 goto out;
1469 *out_ptr++ = '%';
1470 break;
1471 /* pid */
1472 case 'p':
1473 pid_in_pattern = 1;
1474 rc = snprintf(out_ptr, out_end - out_ptr,
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001475 "%d", task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 if (rc > out_end - out_ptr)
1477 goto out;
1478 out_ptr += rc;
1479 break;
1480 /* uid */
1481 case 'u':
1482 rc = snprintf(out_ptr, out_end - out_ptr,
David Howells86a264a2008-11-14 10:39:18 +11001483 "%d", cred->uid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 if (rc > out_end - out_ptr)
1485 goto out;
1486 out_ptr += rc;
1487 break;
1488 /* gid */
1489 case 'g':
1490 rc = snprintf(out_ptr, out_end - out_ptr,
David Howells86a264a2008-11-14 10:39:18 +11001491 "%d", cred->gid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 if (rc > out_end - out_ptr)
1493 goto out;
1494 out_ptr += rc;
1495 break;
1496 /* signal that caused the coredump */
1497 case 's':
1498 rc = snprintf(out_ptr, out_end - out_ptr,
1499 "%ld", signr);
1500 if (rc > out_end - out_ptr)
1501 goto out;
1502 out_ptr += rc;
1503 break;
1504 /* UNIX time of coredump */
1505 case 't': {
1506 struct timeval tv;
1507 do_gettimeofday(&tv);
1508 rc = snprintf(out_ptr, out_end - out_ptr,
1509 "%lu", tv.tv_sec);
1510 if (rc > out_end - out_ptr)
1511 goto out;
1512 out_ptr += rc;
1513 break;
1514 }
1515 /* hostname */
1516 case 'h':
1517 down_read(&uts_sem);
1518 rc = snprintf(out_ptr, out_end - out_ptr,
Serge E. Hallyne9ff3992006-10-02 02:18:11 -07001519 "%s", utsname()->nodename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 up_read(&uts_sem);
1521 if (rc > out_end - out_ptr)
1522 goto out;
1523 out_ptr += rc;
1524 break;
1525 /* executable */
1526 case 'e':
1527 rc = snprintf(out_ptr, out_end - out_ptr,
1528 "%s", current->comm);
1529 if (rc > out_end - out_ptr)
1530 goto out;
1531 out_ptr += rc;
1532 break;
Neil Horman74aadce2007-10-16 23:26:35 -07001533 /* core limit size */
1534 case 'c':
1535 rc = snprintf(out_ptr, out_end - out_ptr,
Jiri Slabyd554ed892010-03-05 13:42:42 -08001536 "%lu", rlimit(RLIMIT_CORE));
Neil Horman74aadce2007-10-16 23:26:35 -07001537 if (rc > out_end - out_ptr)
1538 goto out;
1539 out_ptr += rc;
1540 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 default:
1542 break;
1543 }
1544 ++pat_ptr;
1545 }
1546 }
1547 /* Backward compatibility with core_uses_pid:
1548 *
1549 * If core_pattern does not include a %p (as is the default)
1550 * and core_uses_pid is set, then .%pid will be appended to
Alan Coxc4bbafd2007-04-16 22:53:13 -07001551 * the filename. Do not do this for piped commands. */
Oleg Nesterov64093242008-10-18 20:28:22 -07001552 if (!ispipe && !pid_in_pattern && core_uses_pid) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 rc = snprintf(out_ptr, out_end - out_ptr,
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001554 ".%d", task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 if (rc > out_end - out_ptr)
1556 goto out;
1557 out_ptr += rc;
1558 }
Alan Coxc4bbafd2007-04-16 22:53:13 -07001559out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 *out_ptr = 0;
Alan Coxc4bbafd2007-04-16 22:53:13 -07001561 return ispipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562}
1563
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001564static int zap_process(struct task_struct *start, int exit_code)
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001565{
1566 struct task_struct *t;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001567 int nr = 0;
Oleg Nesterov281de332006-06-26 00:26:06 -07001568
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001569 start->signal->flags = SIGNAL_GROUP_EXIT;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001570 start->signal->group_exit_code = exit_code;
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001571 start->signal->group_stop_count = 0;
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001572
1573 t = start;
1574 do {
1575 if (t != current && t->mm) {
Oleg Nesterov281de332006-06-26 00:26:06 -07001576 sigaddset(&t->pending.signal, SIGKILL);
1577 signal_wake_up(t, 1);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001578 nr++;
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001579 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001580 } while_each_thread(start, t);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001581
1582 return nr;
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001583}
1584
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001585static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001586 struct core_state *core_state, int exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587{
1588 struct task_struct *g, *p;
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001589 unsigned long flags;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001590 int nr = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001592 spin_lock_irq(&tsk->sighand->siglock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001593 if (!signal_group_exit(tsk->signal)) {
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001594 mm->core_state = core_state;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001595 nr = zap_process(tsk, exit_code);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 }
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001597 spin_unlock_irq(&tsk->sighand->siglock);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001598 if (unlikely(nr < 0))
1599 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001601 if (atomic_read(&mm->mm_users) == nr + 1)
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001602 goto done;
Oleg Nesterove4901f92008-07-25 01:47:31 -07001603 /*
1604 * We should find and kill all tasks which use this mm, and we should
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001605 * count them correctly into ->nr_threads. We don't take tasklist
Oleg Nesterove4901f92008-07-25 01:47:31 -07001606 * lock, but this is safe wrt:
1607 *
1608 * fork:
1609 * None of sub-threads can fork after zap_process(leader). All
1610 * processes which were created before this point should be
1611 * visible to zap_threads() because copy_process() adds the new
1612 * process to the tail of init_task.tasks list, and lock/unlock
1613 * of ->siglock provides a memory barrier.
1614 *
1615 * do_exit:
1616 * The caller holds mm->mmap_sem. This means that the task which
1617 * uses this mm can't pass exit_mm(), so it can't exit or clear
1618 * its ->mm.
1619 *
1620 * de_thread:
1621 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1622 * we must see either old or new leader, this does not matter.
1623 * However, it can change p->sighand, so lock_task_sighand(p)
1624 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1625 * it can't fail.
1626 *
1627 * Note also that "g" can be the old leader with ->mm == NULL
1628 * and already unhashed and thus removed from ->thread_group.
1629 * This is OK, __unhash_process()->list_del_rcu() does not
1630 * clear the ->next pointer, we will find the new leader via
1631 * next_thread().
1632 */
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001633 rcu_read_lock();
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001634 for_each_process(g) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001635 if (g == tsk->group_leader)
1636 continue;
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001637 if (g->flags & PF_KTHREAD)
1638 continue;
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001639 p = g;
1640 do {
1641 if (p->mm) {
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001642 if (unlikely(p->mm == mm)) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001643 lock_task_sighand(p, &flags);
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001644 nr += zap_process(p, exit_code);
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001645 unlock_task_sighand(p, &flags);
1646 }
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001647 break;
1648 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001649 } while_each_thread(g, p);
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001650 }
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001651 rcu_read_unlock();
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001652done:
Oleg Nesterovc5f1cc82008-07-25 01:47:42 -07001653 atomic_set(&core_state->nr_threads, nr);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001654 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655}
1656
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001657static int coredump_wait(int exit_code, struct core_state *core_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658{
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001659 struct task_struct *tsk = current;
1660 struct mm_struct *mm = tsk->mm;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001661 struct completion *vfork_done;
Oleg Nesterov2384f552005-10-30 15:02:47 -08001662 int core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001664 init_completion(&core_state->startup);
Oleg Nesterovb564daf2008-07-25 01:47:44 -07001665 core_state->dumper.task = tsk;
1666 core_state->dumper.next = NULL;
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001667 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001668 up_write(&mm->mmap_sem);
1669
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001670 if (unlikely(core_waiters < 0))
1671 goto fail;
1672
1673 /*
1674 * Make sure nobody is waiting for us to release the VM,
1675 * otherwise we can deadlock when we wait on each other
1676 */
1677 vfork_done = tsk->vfork_done;
1678 if (vfork_done) {
1679 tsk->vfork_done = NULL;
1680 complete(vfork_done);
1681 }
1682
Oleg Nesterov2384f552005-10-30 15:02:47 -08001683 if (core_waiters)
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001684 wait_for_completion(&core_state->startup);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001685fail:
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001686 return core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687}
1688
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001689static void coredump_finish(struct mm_struct *mm)
1690{
1691 struct core_thread *curr, *next;
1692 struct task_struct *task;
1693
1694 next = mm->core_state->dumper.next;
1695 while ((curr = next) != NULL) {
1696 next = curr->next;
1697 task = curr->task;
1698 /*
1699 * see exit_mm(), curr->task must not see
1700 * ->task == NULL before we read ->next.
1701 */
1702 smp_mb();
1703 curr->task = NULL;
1704 wake_up_process(task);
1705 }
1706
1707 mm->core_state = NULL;
1708}
1709
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001710/*
1711 * set_dumpable converts traditional three-value dumpable to two flags and
1712 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1713 * these bits are not changed atomically. So get_dumpable can observe the
1714 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1715 * return either old dumpable or new one by paying attention to the order of
1716 * modifying the bits.
1717 *
1718 * dumpable | mm->flags (binary)
1719 * old new | initial interim final
1720 * ---------+-----------------------
1721 * 0 1 | 00 01 01
1722 * 0 2 | 00 10(*) 11
1723 * 1 0 | 01 00 00
1724 * 1 2 | 01 11 11
1725 * 2 0 | 11 10(*) 00
1726 * 2 1 | 11 11 01
1727 *
1728 * (*) get_dumpable regards interim value of 10 as 11.
1729 */
1730void set_dumpable(struct mm_struct *mm, int value)
1731{
1732 switch (value) {
1733 case 0:
1734 clear_bit(MMF_DUMPABLE, &mm->flags);
1735 smp_wmb();
1736 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1737 break;
1738 case 1:
1739 set_bit(MMF_DUMPABLE, &mm->flags);
1740 smp_wmb();
1741 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1742 break;
1743 case 2:
1744 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1745 smp_wmb();
1746 set_bit(MMF_DUMPABLE, &mm->flags);
1747 break;
1748 }
1749}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001750
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001751static int __get_dumpable(unsigned long mm_flags)
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001752{
1753 int ret;
1754
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001755 ret = mm_flags & MMF_DUMPABLE_MASK;
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001756 return (ret >= 2) ? 2 : ret;
1757}
1758
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001759int get_dumpable(struct mm_struct *mm)
1760{
1761 return __get_dumpable(mm->flags);
1762}
1763
Neil Horman61be2282009-09-23 15:56:58 -07001764static void wait_for_dump_helpers(struct file *file)
1765{
1766 struct pipe_inode_info *pipe;
1767
1768 pipe = file->f_path.dentry->d_inode->i_pipe;
1769
1770 pipe_lock(pipe);
1771 pipe->readers++;
1772 pipe->writers--;
1773
1774 while ((pipe->readers > 1) && (!signal_pending(current))) {
1775 wake_up_interruptible_sync(&pipe->wait);
1776 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1777 pipe_wait(pipe);
1778 }
1779
1780 pipe->readers--;
1781 pipe->writers++;
1782 pipe_unlock(pipe);
1783
1784}
1785
1786
WANG Cong8cd3ac32009-01-06 14:42:48 -08001787void do_coredump(long signr, int exit_code, struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788{
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001789 struct core_state core_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 char corename[CORENAME_MAX_SIZE + 1];
1791 struct mm_struct *mm = current->mm;
1792 struct linux_binfmt * binfmt;
1793 struct inode * inode;
David Howellsd84f4f92008-11-14 10:39:23 +11001794 const struct cred *old_cred;
1795 struct cred *cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 int retval = 0;
Alan Coxd6e71142005-06-23 00:09:43 -07001797 int flag = 0;
Andi Kleend025c9d2006-09-30 23:29:28 -07001798 int ispipe = 0;
Neil Horman74aadce2007-10-16 23:26:35 -07001799 char **helper_argv = NULL;
1800 int helper_argc = 0;
Neil Hormana2939802009-09-23 15:56:56 -07001801 int dump_count = 0;
1802 static atomic_t core_dump_count = ATOMIC_INIT(0);
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001803 struct coredump_params cprm = {
1804 .signr = signr,
1805 .regs = regs,
Jiri Slabyd554ed892010-03-05 13:42:42 -08001806 .limit = rlimit(RLIMIT_CORE),
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001807 /*
1808 * We must use the same mm->flags while dumping core to avoid
1809 * inconsistency of bit flags, since this flag is not protected
1810 * by any locks.
1811 */
1812 .mm_flags = mm->flags,
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001813 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814
Steve Grubb0a4ff8c2007-04-19 10:28:21 -04001815 audit_core_dumps(signr);
1816
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001817 binfmt = mm->binfmt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 if (!binfmt || !binfmt->core_dump)
1819 goto fail;
David Howellsd84f4f92008-11-14 10:39:23 +11001820
1821 cred = prepare_creds();
1822 if (!cred) {
1823 retval = -ENOMEM;
1824 goto fail;
1825 }
1826
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 down_write(&mm->mmap_sem);
Roland McGrath00ec99d2007-11-11 19:13:43 -08001828 /*
1829 * If another thread got here first, or we are not dumpable, bail out.
1830 */
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001831 if (mm->core_state || !__get_dumpable(cprm.mm_flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 up_write(&mm->mmap_sem);
David Howellsd84f4f92008-11-14 10:39:23 +11001833 put_cred(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 goto fail;
1835 }
Alan Coxd6e71142005-06-23 00:09:43 -07001836
1837 /*
1838 * We cannot trust fsuid as being the "true" uid of the
1839 * process nor do we know its entire history. We only know it
1840 * was tainted so we dump it as root in mode 2.
1841 */
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001842 if (__get_dumpable(cprm.mm_flags) == 2) {
1843 /* Setuid core dump mode */
Alan Coxd6e71142005-06-23 00:09:43 -07001844 flag = O_EXCL; /* Stop rewrite attacks */
David Howellsd84f4f92008-11-14 10:39:23 +11001845 cred->fsuid = 0; /* Dump root private */
Alan Coxd6e71142005-06-23 00:09:43 -07001846 }
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001847
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001848 retval = coredump_wait(exit_code, &core_state);
David Howellsd84f4f92008-11-14 10:39:23 +11001849 if (retval < 0) {
1850 put_cred(cred);
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001851 goto fail;
David Howellsd84f4f92008-11-14 10:39:23 +11001852 }
1853
1854 old_cred = override_creds(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
1856 /*
1857 * Clear any false indication of pending signals that might
1858 * be seen by the filesystem code called to write the core file.
1859 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 clear_thread_flag(TIF_SIGPENDING);
1861
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 /*
1863 * lock_kernel() because format_corename() is controlled by sysctl, which
1864 * uses lock_kernel()
1865 */
1866 lock_kernel();
Oleg Nesterov64093242008-10-18 20:28:22 -07001867 ispipe = format_corename(corename, signr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 unlock_kernel();
Neil Horman725eae32009-09-23 15:56:54 -07001869
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001870 if ((!ispipe) && (cprm.limit < binfmt->min_coredump))
Neil Horman7dc0b222007-10-16 23:26:34 -07001871 goto fail_unlock;
1872
Alan Coxc4bbafd2007-04-16 22:53:13 -07001873 if (ispipe) {
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001874 if (cprm.limit == 0) {
Neil Horman725eae32009-09-23 15:56:54 -07001875 /*
1876 * Normally core limits are irrelevant to pipes, since
1877 * we're not writing to the file system, but we use
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001878 * cprm.limit of 0 here as a speacial value. Any
Neil Horman725eae32009-09-23 15:56:54 -07001879 * non-zero limit gets set to RLIM_INFINITY below, but
1880 * a limit of 0 skips the dump. This is a consistent
1881 * way to catch recursive crashes. We can still crash
1882 * if the core_pattern binary sets RLIM_CORE = !0
1883 * but it runs as root, and can do lots of stupid things
1884 * Note that we use task_tgid_vnr here to grab the pid
1885 * of the process group leader. That way we get the
1886 * right pid if a thread in a multi-threaded
1887 * core_pattern process dies.
1888 */
1889 printk(KERN_WARNING
1890 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1891 task_tgid_vnr(current), current->comm);
1892 printk(KERN_WARNING "Aborting core\n");
1893 goto fail_unlock;
1894 }
1895
Neil Hormana2939802009-09-23 15:56:56 -07001896 dump_count = atomic_inc_return(&core_dump_count);
1897 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1898 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1899 task_tgid_vnr(current), current->comm);
1900 printk(KERN_WARNING "Skipping core dump\n");
1901 goto fail_dropcount;
1902 }
1903
Neil Horman74aadce2007-10-16 23:26:35 -07001904 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
Tetsuo Handa350eaf72009-01-06 14:41:11 -08001905 if (!helper_argv) {
1906 printk(KERN_WARNING "%s failed to allocate memory\n",
1907 __func__);
Neil Hormana2939802009-09-23 15:56:56 -07001908 goto fail_dropcount;
Tetsuo Handa350eaf72009-01-06 14:41:11 -08001909 }
Neil Horman32321132007-10-16 23:26:36 -07001910
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001911 cprm.limit = RLIM_INFINITY;
Neil Horman32321132007-10-16 23:26:36 -07001912
Andi Kleend025c9d2006-09-30 23:29:28 -07001913 /* SIGPIPE can happen, but it's just never processed */
Neil Hormana2939802009-09-23 15:56:56 -07001914 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001915 &cprm.file)) {
Andi Kleend025c9d2006-09-30 23:29:28 -07001916 printk(KERN_INFO "Core dump to %s pipe failed\n",
1917 corename);
Neil Hormana2939802009-09-23 15:56:56 -07001918 goto fail_dropcount;
Andi Kleend025c9d2006-09-30 23:29:28 -07001919 }
Andi Kleend025c9d2006-09-30 23:29:28 -07001920 } else
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001921 cprm.file = filp_open(corename,
Alexey Dobriyan6d4df672006-12-06 20:40:39 -08001922 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1923 0600);
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001924 if (IS_ERR(cprm.file))
Neil Hormana2939802009-09-23 15:56:56 -07001925 goto fail_dropcount;
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001926 inode = cprm.file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 if (inode->i_nlink > 1)
1928 goto close_fail; /* multiple links - don't dump */
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001929 if (!ispipe && d_unhashed(cprm.file->f_path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 goto close_fail;
1931
Andi Kleend025c9d2006-09-30 23:29:28 -07001932 /* AK: actually i see no reason to not allow this for named pipes etc.,
1933 but keep the previous behaviour for now. */
1934 if (!ispipe && !S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 goto close_fail;
Ingo Molnarc46f7392007-11-28 13:59:18 +01001936 /*
1937 * Dont allow local users get cute and trick others to coredump
1938 * into their pre-created files:
Neil Horman76595f72010-03-05 13:44:16 -08001939 * Note, this is not relevant for pipes
Ingo Molnarc46f7392007-11-28 13:59:18 +01001940 */
Neil Horman76595f72010-03-05 13:44:16 -08001941 if (!ispipe && (inode->i_uid != current_fsuid()))
Ingo Molnarc46f7392007-11-28 13:59:18 +01001942 goto close_fail;
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001943 if (!cprm.file->f_op)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 goto close_fail;
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001945 if (!cprm.file->f_op->write)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 goto close_fail;
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001947 if (!ispipe &&
1948 do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 goto close_fail;
1950
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001951 retval = binfmt->core_dump(&cprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
1953 if (retval)
1954 current->signal->group_exit_code |= 0x80;
1955close_fail:
Neil Horman61be2282009-09-23 15:56:58 -07001956 if (ispipe && core_pipe_limit)
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001957 wait_for_dump_helpers(cprm.file);
1958 filp_close(cprm.file, NULL);
Neil Hormana2939802009-09-23 15:56:56 -07001959fail_dropcount:
1960 if (dump_count)
1961 atomic_dec(&core_dump_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962fail_unlock:
Neil Horman74aadce2007-10-16 23:26:35 -07001963 if (helper_argv)
1964 argv_free(helper_argv);
1965
David Howellsd84f4f92008-11-14 10:39:23 +11001966 revert_creds(old_cred);
1967 put_cred(cred);
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001968 coredump_finish(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969fail:
WANG Cong8cd3ac32009-01-06 14:42:48 -08001970 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971}