blob: 36254645b7cc4d81f7c73b2940767fe06de1642c [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>
Hugh Dickinsba92a432008-07-25 01:45:43 -070031#include <linux/swap.h>
Neil Horman74aadce2007-10-16 23:26:35 -070032#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/init.h>
Hugh Dickinsca5b1722008-07-28 15:46:18 -070034#include <linux/pagemap.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020035#include <linux/perf_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/highmem.h>
37#include <linux/spinlock.h>
38#include <linux/key.h>
39#include <linux/personality.h>
40#include <linux/binfmts.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/utsname.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080042#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/module.h>
44#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <linux/mount.h>
46#include <linux/security.h>
47#include <linux/syscalls.h>
Jay Lan8f0ab512006-09-30 23:28:59 -070048#include <linux/tsacct_kern.h>
Matt Helsley9f460802005-11-07 00:59:16 -080049#include <linux/cn_proc.h>
Al Viro473ae302006-04-26 14:04:08 -040050#include <linux/audit.h>
Roland McGrath6341c392008-07-25 19:45:44 -070051#include <linux/tracehook.h>
Johannes Berg5f4123b2008-07-09 10:28:40 +020052#include <linux/kmod.h>
Eric Paris6110e3a2008-12-17 13:53:20 -050053#include <linux/fsnotify.h>
Al Viro5ad4e532009-03-29 19:50:06 -040054#include <linux/fs_struct.h>
Neil Horman61be2282009-09-23 15:56:58 -070055#include <linux/pipe_fs_i.h>
Ying Han3d5992d2010-10-26 14:21:23 -070056#include <linux/oom.h>
Oleg Nesterov0e028462011-03-06 18:02:54 +010057#include <linux/compat.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
Xiaotian Feng1b0d3002010-10-27 15:34:08 -070069struct core_name {
70 char *corename;
71 int used, size;
72};
73static atomic_t call_count = ATOMIC_INIT(1);
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075/* The maximal length of core_pattern is also specified in sysctl.c */
76
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070077static LIST_HEAD(formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078static DEFINE_RWLOCK(binfmt_lock);
79
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070080int __register_binfmt(struct linux_binfmt * fmt, int insert)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081{
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 if (!fmt)
83 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 write_lock(&binfmt_lock);
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070085 insert ? list_add(&fmt->lh, &formats) :
86 list_add_tail(&fmt->lh, &formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 write_unlock(&binfmt_lock);
88 return 0;
89}
90
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070091EXPORT_SYMBOL(__register_binfmt);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Alexey Dobriyanf6b450d2007-10-16 23:26:04 -070093void unregister_binfmt(struct linux_binfmt * fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094{
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 write_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070096 list_del(&fmt->lh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 write_unlock(&binfmt_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098}
99
100EXPORT_SYMBOL(unregister_binfmt);
101
102static inline void put_binfmt(struct linux_binfmt * fmt)
103{
104 module_put(fmt->module);
105}
106
107/*
108 * Note that a shared library must be both readable and executable due to
109 * security reasons.
110 *
111 * Also note that we take the address to load from from the file itself.
112 */
Heiko Carstens1e7bfb22009-01-14 14:14:29 +0100113SYSCALL_DEFINE1(uselib, const char __user *, library)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114{
Al Viro964bd182008-07-26 03:33:14 -0400115 struct file *file;
Al Viro964bd182008-07-26 03:33:14 -0400116 char *tmp = getname(library);
117 int error = PTR_ERR(tmp);
Al Viro47c805d2011-02-23 17:44:09 -0500118 static const struct open_flags uselib_flags = {
119 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
120 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
121 .intent = LOOKUP_OPEN
122 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
Al Viro6e8341a2009-04-06 11:16:22 -0400124 if (IS_ERR(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 goto out;
126
Al Viro47c805d2011-02-23 17:44:09 -0500127 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
Al Viro6e8341a2009-04-06 11:16:22 -0400128 putname(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 error = PTR_ERR(file);
130 if (IS_ERR(file))
131 goto out;
132
Al Viro6e8341a2009-04-06 11:16:22 -0400133 error = -EINVAL;
134 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
135 goto exit;
136
137 error = -EACCES;
138 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
139 goto exit;
140
Eric Paris2a12a9d2009-12-17 21:24:21 -0500141 fsnotify_open(file);
Eric Paris6110e3a2008-12-17 13:53:20 -0500142
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 error = -ENOEXEC;
144 if(file->f_op) {
145 struct linux_binfmt * fmt;
146
147 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -0700148 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 if (!fmt->load_shlib)
150 continue;
151 if (!try_module_get(fmt->module))
152 continue;
153 read_unlock(&binfmt_lock);
154 error = fmt->load_shlib(file);
155 read_lock(&binfmt_lock);
156 put_binfmt(fmt);
157 if (error != -ENOEXEC)
158 break;
159 }
160 read_unlock(&binfmt_lock);
161 }
Al Viro6e8341a2009-04-06 11:16:22 -0400162exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 fput(file);
164out:
165 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166}
167
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700168#ifdef CONFIG_MMU
Oleg Nesterovae6b5852011-03-06 18:03:11 +0100169/*
170 * The nascent bprm->mm is not visible until exec_mmap() but it can
171 * use a lot of memory, account these pages in current->mm temporary
172 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
173 * change the counter back via acct_arg_size(0).
174 */
Oleg Nesterov0e028462011-03-06 18:02:54 +0100175static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100176{
177 struct mm_struct *mm = current->mm;
178 long diff = (long)(pages - bprm->vma_pages);
179
180 if (!mm || !diff)
181 return;
182
183 bprm->vma_pages = pages;
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100184 add_mm_counter(mm, MM_ANONPAGES, diff);
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100185}
186
Oleg Nesterov0e028462011-03-06 18:02:54 +0100187static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700188 int write)
189{
190 struct page *page;
191 int ret;
192
193#ifdef CONFIG_STACK_GROWSUP
194 if (write) {
Michal Hockod05f3162011-05-24 17:11:44 -0700195 ret = expand_downwards(bprm->vma, pos);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700196 if (ret < 0)
197 return NULL;
198 }
199#endif
200 ret = get_user_pages(current, bprm->mm, pos,
201 1, write, 1, &page, NULL);
202 if (ret <= 0)
203 return NULL;
204
205 if (write) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700206 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800207 struct rlimit *rlim;
208
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100209 acct_arg_size(bprm, size / PAGE_SIZE);
210
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800211 /*
212 * We've historically supported up to 32 pages (ARG_MAX)
213 * of argument strings even with small stacks
214 */
215 if (size <= ARG_MAX)
216 return page;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700217
218 /*
219 * Limit to 1/4-th the stack size for the argv+env strings.
220 * This ensures that:
221 * - the remaining binfmt code will not run out of stack space,
222 * - the program will have a reasonable amount of stack left
223 * to work from.
224 */
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800225 rlim = current->signal->rlim;
Jiri Slabyd554ed892010-03-05 13:42:42 -0800226 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700227 put_page(page);
228 return NULL;
229 }
230 }
231
232 return page;
233}
234
235static void put_arg_page(struct page *page)
236{
237 put_page(page);
238}
239
240static void free_arg_page(struct linux_binprm *bprm, int i)
241{
242}
243
244static void free_arg_pages(struct linux_binprm *bprm)
245{
246}
247
248static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
249 struct page *page)
250{
251 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
252}
253
254static int __bprm_mm_init(struct linux_binprm *bprm)
255{
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800256 int err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700257 struct vm_area_struct *vma = NULL;
258 struct mm_struct *mm = bprm->mm;
259
260 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
261 if (!vma)
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800262 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700263
264 down_write(&mm->mmap_sem);
265 vma->vm_mm = mm;
266
267 /*
268 * Place the stack at the largest stack address the architecture
269 * supports. Later, we'll move this to an appropriate place. We don't
270 * use STACK_TOP because that can depend on attributes which aren't
271 * configured yet.
272 */
Michal Hockoaacb3d12011-07-26 16:08:40 -0700273 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700274 vma->vm_end = STACK_TOP_MAX;
275 vma->vm_start = vma->vm_end - PAGE_SIZE;
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700276 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
Coly Li3ed75eb2007-10-18 23:39:15 -0700277 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Rik van Riel5beb4932010-03-05 13:42:07 -0800278 INIT_LIST_HEAD(&vma->anon_vma_chain);
Tavis Ormandy462e635e2010-12-09 15:29:42 +0100279
280 err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
281 if (err)
282 goto err;
283
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700284 err = insert_vm_struct(mm, vma);
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800285 if (err)
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700286 goto err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700287
288 mm->stack_vm = mm->total_vm = 1;
289 up_write(&mm->mmap_sem);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700290 bprm->p = vma->vm_end - sizeof(void *);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700291 return 0;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700292err:
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800293 up_write(&mm->mmap_sem);
294 bprm->vma = NULL;
295 kmem_cache_free(vm_area_cachep, vma);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700296 return err;
297}
298
299static bool valid_arg_len(struct linux_binprm *bprm, long len)
300{
301 return len <= MAX_ARG_STRLEN;
302}
303
304#else
305
Oleg Nesterov0e028462011-03-06 18:02:54 +0100306static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100307{
308}
309
Oleg Nesterov0e028462011-03-06 18:02:54 +0100310static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700311 int write)
312{
313 struct page *page;
314
315 page = bprm->page[pos / PAGE_SIZE];
316 if (!page && write) {
317 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
318 if (!page)
319 return NULL;
320 bprm->page[pos / PAGE_SIZE] = page;
321 }
322
323 return page;
324}
325
326static void put_arg_page(struct page *page)
327{
328}
329
330static void free_arg_page(struct linux_binprm *bprm, int i)
331{
332 if (bprm->page[i]) {
333 __free_page(bprm->page[i]);
334 bprm->page[i] = NULL;
335 }
336}
337
338static void free_arg_pages(struct linux_binprm *bprm)
339{
340 int i;
341
342 for (i = 0; i < MAX_ARG_PAGES; i++)
343 free_arg_page(bprm, i);
344}
345
346static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
347 struct page *page)
348{
349}
350
351static int __bprm_mm_init(struct linux_binprm *bprm)
352{
353 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
354 return 0;
355}
356
357static bool valid_arg_len(struct linux_binprm *bprm, long len)
358{
359 return len <= bprm->p;
360}
361
362#endif /* CONFIG_MMU */
363
364/*
365 * Create a new mm_struct and populate it with a temporary stack
366 * vm_area_struct. We don't have enough context at this point to set the stack
367 * flags, permissions, and offset, so we use temporary values. We'll update
368 * them later in setup_arg_pages().
369 */
370int bprm_mm_init(struct linux_binprm *bprm)
371{
372 int err;
373 struct mm_struct *mm = NULL;
374
375 bprm->mm = mm = mm_alloc();
376 err = -ENOMEM;
377 if (!mm)
378 goto err;
379
380 err = init_new_context(current, mm);
381 if (err)
382 goto err;
383
384 err = __bprm_mm_init(bprm);
385 if (err)
386 goto err;
387
388 return 0;
389
390err:
391 if (mm) {
392 bprm->mm = NULL;
393 mmdrop(mm);
394 }
395
396 return err;
397}
398
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100399struct user_arg_ptr {
Oleg Nesterov0e028462011-03-06 18:02:54 +0100400#ifdef CONFIG_COMPAT
401 bool is_compat;
402#endif
403 union {
404 const char __user *const __user *native;
405#ifdef CONFIG_COMPAT
406 compat_uptr_t __user *compat;
407#endif
408 } ptr;
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100409};
410
411static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100412{
Oleg Nesterov0e028462011-03-06 18:02:54 +0100413 const char __user *native;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100414
Oleg Nesterov0e028462011-03-06 18:02:54 +0100415#ifdef CONFIG_COMPAT
416 if (unlikely(argv.is_compat)) {
417 compat_uptr_t compat;
418
419 if (get_user(compat, argv.ptr.compat + nr))
420 return ERR_PTR(-EFAULT);
421
422 return compat_ptr(compat);
423 }
424#endif
425
426 if (get_user(native, argv.ptr.native + nr))
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100427 return ERR_PTR(-EFAULT);
428
Oleg Nesterov0e028462011-03-06 18:02:54 +0100429 return native;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100430}
431
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432/*
433 * count() counts the number of strings in array ARGV.
434 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100435static int count(struct user_arg_ptr argv, int max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436{
437 int i = 0;
438
Oleg Nesterov0e028462011-03-06 18:02:54 +0100439 if (argv.ptr.native != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 for (;;) {
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100441 const char __user *p = get_user_arg_ptr(argv, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 if (!p)
444 break;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100445
446 if (IS_ERR(p))
447 return -EFAULT;
448
Jason Baron362e6662008-10-15 22:01:52 -0700449 if (i++ >= max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 return -E2BIG;
Roland McGrath9aea5a62010-09-07 19:37:06 -0700451
452 if (fatal_signal_pending(current))
453 return -ERESTARTNOHAND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 cond_resched();
455 }
456 }
457 return i;
458}
459
460/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700461 * 'copy_strings()' copies argument/environment strings from the old
462 * processes's memory to the new process's stack. The call to get_user_pages()
463 * ensures the destination page is created and not swapped out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100465static int copy_strings(int argc, struct user_arg_ptr argv,
Adrian Bunk75c96f82005-05-05 16:16:09 -0700466 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
468 struct page *kmapped_page = NULL;
469 char *kaddr = NULL;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700470 unsigned long kpos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 int ret;
472
473 while (argc-- > 0) {
David Howellsd7627462010-08-17 23:52:56 +0100474 const char __user *str;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 int len;
476 unsigned long pos;
477
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100478 ret = -EFAULT;
479 str = get_user_arg_ptr(argv, argc);
480 if (IS_ERR(str))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100483 len = strnlen_user(str, MAX_ARG_STRLEN);
484 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 goto out;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100486
487 ret = -E2BIG;
488 if (!valid_arg_len(bprm, len))
489 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700491 /* We're going to work our way backwords. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 pos = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700493 str += len;
494 bprm->p -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496 while (len > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 int offset, bytes_to_copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Roland McGrath9aea5a62010-09-07 19:37:06 -0700499 if (fatal_signal_pending(current)) {
500 ret = -ERESTARTNOHAND;
501 goto out;
502 }
Roland McGrath7993bc12010-09-07 19:36:28 -0700503 cond_resched();
504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 offset = pos % PAGE_SIZE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700506 if (offset == 0)
507 offset = PAGE_SIZE;
508
509 bytes_to_copy = offset;
510 if (bytes_to_copy > len)
511 bytes_to_copy = len;
512
513 offset -= bytes_to_copy;
514 pos -= bytes_to_copy;
515 str -= bytes_to_copy;
516 len -= bytes_to_copy;
517
518 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
519 struct page *page;
520
521 page = get_arg_page(bprm, pos, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 if (!page) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700523 ret = -E2BIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 goto out;
525 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700527 if (kmapped_page) {
528 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700530 put_arg_page(kmapped_page);
531 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 kmapped_page = page;
533 kaddr = kmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700534 kpos = pos & PAGE_MASK;
535 flush_arg_page(bprm, kpos, kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700537 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 ret = -EFAULT;
539 goto out;
540 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 }
542 }
543 ret = 0;
544out:
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700545 if (kmapped_page) {
546 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700548 put_arg_page(kmapped_page);
549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 return ret;
551}
552
553/*
554 * Like copy_strings, but get argv and its values from kernel memory.
555 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100556int copy_strings_kernel(int argc, const char *const *__argv,
David Howellsd7627462010-08-17 23:52:56 +0100557 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558{
559 int r;
560 mm_segment_t oldfs = get_fs();
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100561 struct user_arg_ptr argv = {
Oleg Nesterov0e028462011-03-06 18:02:54 +0100562 .ptr.native = (const char __user *const __user *)__argv,
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100563 };
564
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 set_fs(KERNEL_DS);
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100566 r = copy_strings(argc, argv, bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 set_fs(oldfs);
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 return r;
570}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571EXPORT_SYMBOL(copy_strings_kernel);
572
573#ifdef CONFIG_MMU
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700576 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
577 * the binfmt code determines where the new stack should reside, we shift it to
578 * its final location. The process proceeds as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 *
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700580 * 1) Use shift to calculate the new vma endpoints.
581 * 2) Extend vma to cover both the old and new ranges. This ensures the
582 * arguments passed to subsequent functions are consistent.
583 * 3) Move vma's page tables to the new range.
584 * 4) Free up any cleared pgd range.
585 * 5) Shrink the vma to cover only the new range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700587static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588{
589 struct mm_struct *mm = vma->vm_mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700590 unsigned long old_start = vma->vm_start;
591 unsigned long old_end = vma->vm_end;
592 unsigned long length = old_end - old_start;
593 unsigned long new_start = old_start - shift;
594 unsigned long new_end = old_end - shift;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700595 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700597 BUG_ON(new_start > new_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700599 /*
600 * ensure there are no vmas between where we want to go
601 * and where we are
602 */
603 if (vma != find_vma(mm, new_start))
604 return -EFAULT;
605
606 /*
607 * cover the whole range: [new_start, old_end)
608 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800609 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
610 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700611
612 /*
613 * move the page tables downwards, on failure we rely on
614 * process cleanup to remove whatever mess we made.
615 */
616 if (length != move_page_tables(vma, old_start,
617 vma, new_start, length))
618 return -ENOMEM;
619
620 lru_add_drain();
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700621 tlb_gather_mmu(&tlb, mm, 0);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700622 if (new_end > old_start) {
623 /*
624 * when the old and new regions overlap clear from new_end.
625 */
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700626 free_pgd_range(&tlb, new_end, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700627 vma->vm_next ? vma->vm_next->vm_start : 0);
628 } else {
629 /*
630 * otherwise, clean from old_start; this is done to not touch
631 * the address space in [new_end, old_start) some architectures
632 * have constraints on va-space that make this illegal (IA64) -
633 * for the others its just a little faster.
634 */
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700635 free_pgd_range(&tlb, old_start, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700636 vma->vm_next ? vma->vm_next->vm_start : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 }
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700638 tlb_finish_mmu(&tlb, new_end, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700640 /*
Rik van Riel5beb4932010-03-05 13:42:07 -0800641 * Shrink the vma to just the new range. Always succeeds.
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700642 */
643 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
644
645 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
647
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700648/*
649 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
650 * the stack is optionally relocated, and some extra space is added.
651 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652int setup_arg_pages(struct linux_binprm *bprm,
653 unsigned long stack_top,
654 int executable_stack)
655{
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700656 unsigned long ret;
657 unsigned long stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 struct mm_struct *mm = current->mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700659 struct vm_area_struct *vma = bprm->vma;
660 struct vm_area_struct *prev = NULL;
661 unsigned long vm_flags;
662 unsigned long stack_base;
Michael Neuling803bf5e2010-02-10 13:56:42 -0800663 unsigned long stack_size;
664 unsigned long stack_expand;
665 unsigned long rlim_stack;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666
667#ifdef CONFIG_STACK_GROWSUP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 /* Limit stack size to 1GB */
Jiri Slabyd554ed892010-03-05 13:42:42 -0800669 stack_base = rlimit_max(RLIMIT_STACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 if (stack_base > (1 << 30))
671 stack_base = 1 << 30;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700673 /* Make sure we didn't let the argument array grow too large. */
674 if (vma->vm_end - vma->vm_start > stack_base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 return -ENOMEM;
676
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700677 stack_base = PAGE_ALIGN(stack_top - stack_base);
678
679 stack_shift = vma->vm_start - stack_base;
680 mm->arg_start = bprm->p - stack_shift;
681 bprm->p = vma->vm_end - stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682#else
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700683 stack_top = arch_align_stack(stack_top);
684 stack_top = PAGE_ALIGN(stack_top);
Roland McGrath1b528182010-09-07 19:35:49 -0700685
686 if (unlikely(stack_top < mmap_min_addr) ||
687 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
688 return -ENOMEM;
689
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700690 stack_shift = vma->vm_end - stack_top;
691
692 bprm->p -= stack_shift;
693 mm->arg_start = bprm->p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694#endif
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700695
696 if (bprm->loader)
697 bprm->loader -= stack_shift;
698 bprm->exec -= stack_shift;
699
700 down_write(&mm->mmap_sem);
Hugh Dickins96a8e132008-07-10 21:19:20 +0100701 vm_flags = VM_STACK_FLAGS;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700702
703 /*
704 * Adjust stack execute permissions; explicitly enable for
705 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
706 * (arch default) otherwise.
707 */
708 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
709 vm_flags |= VM_EXEC;
710 else if (executable_stack == EXSTACK_DISABLE_X)
711 vm_flags &= ~VM_EXEC;
712 vm_flags |= mm->def_flags;
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700713 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700714
715 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
716 vm_flags);
717 if (ret)
718 goto out_unlock;
719 BUG_ON(prev != vma);
720
721 /* Move stack pages down in memory. */
722 if (stack_shift) {
723 ret = shift_arg_pages(vma, stack_shift);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800724 if (ret)
725 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 }
727
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700728 /* mprotect_fixup is overkill to remove the temporary stack flags */
729 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
730
Michael Neuling5ef097d2010-03-05 13:42:57 -0800731 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
Michael Neuling803bf5e2010-02-10 13:56:42 -0800732 stack_size = vma->vm_end - vma->vm_start;
733 /*
734 * Align this down to a page boundary as expand_stack
735 * will align it up.
736 */
737 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700738#ifdef CONFIG_STACK_GROWSUP
Michael Neuling803bf5e2010-02-10 13:56:42 -0800739 if (stack_size + stack_expand > rlim_stack)
740 stack_base = vma->vm_start + rlim_stack;
741 else
742 stack_base = vma->vm_end + stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700743#else
Michael Neuling803bf5e2010-02-10 13:56:42 -0800744 if (stack_size + stack_expand > rlim_stack)
745 stack_base = vma->vm_end - rlim_stack;
746 else
747 stack_base = vma->vm_start - stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700748#endif
Eric B Munson3af9e852010-05-18 15:30:49 +0100749 current->mm->start_stack = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700750 ret = expand_stack(vma, stack_base);
751 if (ret)
752 ret = -EFAULT;
753
754out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 up_write(&mm->mmap_sem);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800756 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758EXPORT_SYMBOL(setup_arg_pages);
759
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760#endif /* CONFIG_MMU */
761
762struct file *open_exec(const char *name)
763{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 struct file *file;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200765 int err;
Al Viro47c805d2011-02-23 17:44:09 -0500766 static const struct open_flags open_exec_flags = {
767 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
768 .acc_mode = MAY_EXEC | MAY_OPEN,
769 .intent = LOOKUP_OPEN
770 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
Al Viro47c805d2011-02-23 17:44:09 -0500772 file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
Al Viro6e8341a2009-04-06 11:16:22 -0400773 if (IS_ERR(file))
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200774 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200776 err = -EACCES;
Al Viro6e8341a2009-04-06 11:16:22 -0400777 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
778 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200779
Al Viro6e8341a2009-04-06 11:16:22 -0400780 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
781 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200782
Eric Paris2a12a9d2009-12-17 21:24:21 -0500783 fsnotify_open(file);
Eric Paris6110e3a2008-12-17 13:53:20 -0500784
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200785 err = deny_write_access(file);
Al Viro6e8341a2009-04-06 11:16:22 -0400786 if (err)
787 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
Al Viro6e8341a2009-04-06 11:16:22 -0400789out:
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200790 return file;
791
Al Viro6e8341a2009-04-06 11:16:22 -0400792exit:
793 fput(file);
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200794 return ERR_PTR(err);
795}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796EXPORT_SYMBOL(open_exec);
797
Mimi Zohar6777d772009-08-21 14:32:48 -0400798int kernel_read(struct file *file, loff_t offset,
799 char *addr, unsigned long count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800{
801 mm_segment_t old_fs;
802 loff_t pos = offset;
803 int result;
804
805 old_fs = get_fs();
806 set_fs(get_ds());
807 /* The cast to a user pointer is valid due to the set_fs() */
808 result = vfs_read(file, (void __user *)addr, count, &pos);
809 set_fs(old_fs);
810 return result;
811}
812
813EXPORT_SYMBOL(kernel_read);
814
815static int exec_mmap(struct mm_struct *mm)
816{
817 struct task_struct *tsk;
818 struct mm_struct * old_mm, *active_mm;
819
820 /* Notify parent that we're no longer interested in the old VM */
821 tsk = current;
822 old_mm = current->mm;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800823 sync_mm_rss(tsk, old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 mm_release(tsk, old_mm);
825
826 if (old_mm) {
827 /*
828 * Make sure that if there is a core dump in progress
829 * for the old mm, we get out and die instead of going
830 * through with the exec. We must hold mmap_sem around
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700831 * checking core_state and changing tsk->mm.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 */
833 down_read(&old_mm->mmap_sem);
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700834 if (unlikely(old_mm->core_state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 up_read(&old_mm->mmap_sem);
836 return -EINTR;
837 }
838 }
839 task_lock(tsk);
840 active_mm = tsk->active_mm;
841 tsk->mm = mm;
842 tsk->active_mm = mm;
843 activate_mm(active_mm, mm);
844 task_unlock(tsk);
845 arch_pick_mmap_layout(mm);
846 if (old_mm) {
847 up_read(&old_mm->mmap_sem);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +0200848 BUG_ON(active_mm != old_mm);
Balbir Singh31a78f22008-09-28 23:09:31 +0100849 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 mmput(old_mm);
851 return 0;
852 }
853 mmdrop(active_mm);
854 return 0;
855}
856
857/*
858 * This function makes sure the current process has its own signal table,
859 * so that flush_signal_handlers can later reset the handlers without
860 * disturbing other processes. (Other processes might share the signal
861 * table via the CLONE_SIGHAND option to clone().)
862 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800863static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864{
865 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700866 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 spinlock_t *lock = &oldsighand->siglock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700869 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 goto no_thread_group;
871
872 /*
873 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800876 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 /*
878 * Another group action in progress, just
879 * return so that the signal is processed.
880 */
881 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 return -EAGAIN;
883 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Oleg Nesterovd3441932010-05-26 14:43:11 -0700885 sig->group_exit_task = tsk;
886 sig->notify_count = zap_other_threads(tsk);
887 if (!thread_group_leader(tsk))
888 sig->notify_count--;
889
890 while (sig->notify_count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 __set_current_state(TASK_UNINTERRUPTIBLE);
892 spin_unlock_irq(lock);
893 schedule();
894 spin_lock_irq(lock);
895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 spin_unlock_irq(lock);
897
898 /*
899 * At this point all other threads have exited, all we have to
900 * do is to wait for the thread group leader to become inactive,
901 * and to assume its PID:
902 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700903 if (!thread_group_leader(tsk)) {
Oleg Nesterov81879262008-12-01 14:18:16 -0800904 struct task_struct *leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700905
Oleg Nesterov2800d8d2008-04-30 00:53:12 -0700906 sig->notify_count = -1; /* for exit_notify() */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700907 for (;;) {
908 write_lock_irq(&tasklist_lock);
909 if (likely(leader->exit_state))
910 break;
911 __set_current_state(TASK_UNINTERRUPTIBLE);
912 write_unlock_irq(&tasklist_lock);
913 schedule();
914 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
Roland McGrathf5e90282006-04-10 22:54:16 -0700916 /*
917 * The only record we have of the real-time age of a
918 * process, regardless of execs it's done, is start_time.
919 * All the past CPU time is accumulated in signal_struct
920 * from sister threads now dead. But in this non-leader
921 * exec, nothing survives from the original leader thread,
922 * whose birth marks the true age of this process now.
923 * When we take on its identity by switching to its PID, we
924 * also take its birthdate (always earlier than our own).
925 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700926 tsk->start_time = leader->start_time;
Roland McGrathf5e90282006-04-10 22:54:16 -0700927
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700928 BUG_ON(!same_thread_group(leader, tsk));
929 BUG_ON(has_group_leader_pid(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 /*
931 * An exec() starts a new thread group with the
932 * TGID of the previous thread group. Rehash the
933 * two threads with a switched PID, and release
934 * the former thread group leader:
935 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800936
937 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700938 * The old leader becomes a thread of the this thread group.
939 * Note: The old leader also uses this pid until release_task
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800940 * is called. Odd but simple and correct.
941 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700942 detach_pid(tsk, PIDTYPE_PID);
943 tsk->pid = leader->pid;
Sukadev Bhattiprolu3743ca02007-10-18 23:39:51 -0700944 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700945 transfer_pid(leader, tsk, PIDTYPE_PGID);
946 transfer_pid(leader, tsk, PIDTYPE_SID);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800947
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700948 list_replace_rcu(&leader->tasks, &tsk->tasks);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800949 list_replace_init(&leader->sibling, &tsk->sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700951 tsk->group_leader = tsk;
952 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -0600953
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700954 tsk->exit_signal = SIGCHLD;
Oleg Nesterov087806b2011-06-22 23:10:26 +0200955 leader->exit_signal = -1;
Oleg Nesterov962b5642005-11-23 13:37:43 -0800956
957 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
958 leader->exit_state = EXIT_DEAD;
Oleg Nesteroveac1b5e2011-07-21 20:00:43 +0200959
960 /*
961 * We are going to release_task()->ptrace_unlink() silently,
962 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
963 * the tracer wont't block again waiting for this thread.
964 */
965 if (unlikely(leader->ptrace))
966 __wake_up_parent(leader, leader->parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 write_unlock_irq(&tasklist_lock);
Oleg Nesterov81879262008-12-01 14:18:16 -0800968
969 release_task(leader);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700972 sig->group_exit_task = NULL;
973 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
975no_thread_group:
Jiri Pirko1f102062009-09-22 16:44:10 -0700976 if (current->mm)
977 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 exit_itimers(sig);
Oleg Nesterovcbaffba2008-05-26 20:55:42 +0400980 flush_itimer_signals();
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300981
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700982 if (atomic_read(&oldsighand->count) != 1) {
983 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700985 * This ->sighand is shared with the CLONE_SIGHAND
986 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700988 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
989 if (!newsighand)
990 return -ENOMEM;
991
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 atomic_set(&newsighand->count, 1);
993 memcpy(newsighand->action, oldsighand->action,
994 sizeof(newsighand->action));
995
996 write_lock_irq(&tasklist_lock);
997 spin_lock(&oldsighand->siglock);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700998 rcu_assign_pointer(tsk->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 spin_unlock(&oldsighand->siglock);
1000 write_unlock_irq(&tasklist_lock);
1001
Davide Libenzifba2afa2007-05-10 22:23:13 -07001002 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 }
1004
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001005 BUG_ON(!thread_group_leader(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 return 0;
1007}
Oleg Nesterov0840a902007-10-16 23:27:22 -07001008
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009/*
1010 * These functions flushes out all traces of the currently running executable
1011 * so that a new one can be started
1012 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001013static void flush_old_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014{
1015 long j = -1;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001016 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
1018 spin_lock(&files->file_lock);
1019 for (;;) {
1020 unsigned long set, i;
1021
1022 j++;
1023 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001024 fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -08001025 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001027 set = fdt->close_on_exec->fds_bits[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 if (!set)
1029 continue;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001030 fdt->close_on_exec->fds_bits[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 spin_unlock(&files->file_lock);
1032 for ( ; set ; i++,set >>= 1) {
1033 if (set & 1) {
1034 sys_close(i);
1035 }
1036 }
1037 spin_lock(&files->file_lock);
1038
1039 }
1040 spin_unlock(&files->file_lock);
1041}
1042
Andrew Morton59714d62008-02-04 22:27:21 -08001043char *get_task_comm(char *buf, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044{
1045 /* buf must be at least sizeof(tsk->comm) in size */
1046 task_lock(tsk);
1047 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1048 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -08001049 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050}
J Freyensee7d74f492011-05-06 16:56:47 -07001051EXPORT_SYMBOL_GPL(get_task_comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
1053void set_task_comm(struct task_struct *tsk, char *buf)
1054{
1055 task_lock(tsk);
john stultz4614a696b2009-12-14 18:00:05 -08001056
1057 /*
1058 * Threads may access current->comm without holding
1059 * the task lock, so write the string carefully.
1060 * Readers without a lock may see incomplete new
1061 * names but are safe from non-terminating string reads.
1062 */
1063 memset(tsk->comm, 0, TASK_COMM_LEN);
1064 wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1066 task_unlock(tsk);
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001067 perf_event_comm(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068}
1069
1070int flush_old_exec(struct linux_binprm * bprm)
1071{
Linus Torvalds221af7f2010-01-28 22:14:42 -08001072 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
1074 /*
1075 * Make sure we have a private signal table and that
1076 * we are unassociated from the previous thread group.
1077 */
1078 retval = de_thread(current);
1079 if (retval)
1080 goto out;
1081
Matt Helsley925d1c42008-04-29 01:01:36 -07001082 set_mm_exe_file(bprm->mm, bprm->file);
1083
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 * Release all of the old mmap stuff
1086 */
Oleg Nesterov3c77f842010-11-30 20:55:34 +01001087 acct_arg_size(bprm, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 retval = exec_mmap(bprm->mm);
1089 if (retval)
Al Virofd8328b2008-04-22 05:11:59 -04001090 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
1092 bprm->mm = NULL; /* We're using it now */
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001093
Mathias Krausedac853a2011-06-09 20:05:18 +02001094 set_fs(USER_DS);
KOSAKI Motohiro98391cf2010-10-27 15:34:16 -07001095 current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001096 flush_thread();
1097 current->personality &= ~bprm->per_clear;
1098
Linus Torvalds221af7f2010-01-28 22:14:42 -08001099 return 0;
1100
1101out:
1102 return retval;
1103}
1104EXPORT_SYMBOL(flush_old_exec);
1105
Al Viro1b5d7832011-06-19 12:49:47 -04001106void would_dump(struct linux_binprm *bprm, struct file *file)
1107{
1108 if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1109 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1110}
1111EXPORT_SYMBOL(would_dump);
1112
Linus Torvalds221af7f2010-01-28 22:14:42 -08001113void setup_new_exec(struct linux_binprm * bprm)
1114{
1115 int i, ch;
David Howellsd7627462010-08-17 23:52:56 +01001116 const char *name;
Linus Torvalds221af7f2010-01-28 22:14:42 -08001117 char tcomm[sizeof(current->comm)];
1118
1119 arch_pick_mmap_layout(current->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
1121 /* This is the point of no return */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 current->sas_ss_sp = current->sas_ss_size = 0;
1123
David Howellsda9592e2008-11-14 10:39:05 +11001124 if (current_euid() == current_uid() && current_egid() == current_gid())
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001125 set_dumpable(current->mm, 1);
Alan Coxd6e71142005-06-23 00:09:43 -07001126 else
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001127 set_dumpable(current->mm, suid_dumpable);
Alan Coxd6e71142005-06-23 00:09:43 -07001128
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 name = bprm->filename;
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -07001130
1131 /* Copies the binary name from after last slash */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 for (i=0; (ch = *(name++)) != '\0';) {
1133 if (ch == '/')
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -07001134 i = 0; /* overwrite what we wrote */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 else
1136 if (i < (sizeof(tcomm) - 1))
1137 tcomm[i++] = ch;
1138 }
1139 tcomm[i] = '\0';
1140 set_task_comm(current, tcomm);
1141
Benjamin Herrenschmidt0551fbd2006-02-28 16:59:19 -08001142 /* Set the new mm task size. We have to do that late because it may
1143 * depend on TIF_32BIT which is only updated in flush_thread() on
1144 * some architectures like powerpc
1145 */
1146 current->mm->task_size = TASK_SIZE;
1147
David Howellsa6f76f22008-11-14 10:39:24 +11001148 /* install the new credentials */
1149 if (bprm->cred->uid != current_euid() ||
1150 bprm->cred->gid != current_egid()) {
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001151 current->pdeath_signal = 0;
Al Viro1b5d7832011-06-19 12:49:47 -04001152 } else {
1153 would_dump(bprm, bprm->file);
1154 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1155 set_dumpable(current->mm, suid_dumpable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 }
1157
Ingo Molnarf65cb452008-12-16 13:40:44 +01001158 /*
1159 * Flush performance counters when crossing a
1160 * security domain:
1161 */
1162 if (!get_dumpable(current->mm))
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001163 perf_event_exit_task(current);
Ingo Molnarf65cb452008-12-16 13:40:44 +01001164
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 /* An exec changes our domain. We are no longer part of the thread
1166 group */
1167
1168 current->self_exec_id++;
1169
1170 flush_signal_handlers(current, 0);
1171 flush_old_files(current->files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172}
Linus Torvalds221af7f2010-01-28 22:14:42 -08001173EXPORT_SYMBOL(setup_new_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
David Howellsa6f76f22008-11-14 10:39:24 +11001175/*
Oleg Nesterova2a84742009-09-05 11:17:13 -07001176 * Prepare credentials and lock ->cred_guard_mutex.
1177 * install_exec_creds() commits the new creds and drops the lock.
1178 * Or, if exec fails before, free_bprm() should release ->cred and
1179 * and unlock.
1180 */
1181int prepare_bprm_creds(struct linux_binprm *bprm)
1182{
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001183 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
Oleg Nesterova2a84742009-09-05 11:17:13 -07001184 return -ERESTARTNOINTR;
1185
1186 bprm->cred = prepare_exec_creds();
1187 if (likely(bprm->cred))
1188 return 0;
1189
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001190 mutex_unlock(&current->signal->cred_guard_mutex);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001191 return -ENOMEM;
1192}
1193
1194void free_bprm(struct linux_binprm *bprm)
1195{
1196 free_arg_pages(bprm);
1197 if (bprm->cred) {
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001198 mutex_unlock(&current->signal->cred_guard_mutex);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001199 abort_creds(bprm->cred);
1200 }
1201 kfree(bprm);
1202}
1203
1204/*
David Howellsa6f76f22008-11-14 10:39:24 +11001205 * install the new credentials for this executable
1206 */
1207void install_exec_creds(struct linux_binprm *bprm)
1208{
1209 security_bprm_committing_creds(bprm);
1210
1211 commit_creds(bprm->cred);
1212 bprm->cred = NULL;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001213 /*
1214 * cred_guard_mutex must be held at least to this point to prevent
David Howellsa6f76f22008-11-14 10:39:24 +11001215 * ptrace_attach() from altering our determination of the task's
Oleg Nesterova2a84742009-09-05 11:17:13 -07001216 * credentials; any time after this it may be unlocked.
1217 */
David Howellsa6f76f22008-11-14 10:39:24 +11001218 security_bprm_committed_creds(bprm);
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001219 mutex_unlock(&current->signal->cred_guard_mutex);
David Howellsa6f76f22008-11-14 10:39:24 +11001220}
1221EXPORT_SYMBOL(install_exec_creds);
1222
1223/*
1224 * determine how safe it is to execute the proposed program
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001225 * - the caller must hold ->cred_guard_mutex to protect against
David Howellsa6f76f22008-11-14 10:39:24 +11001226 * PTRACE_ATTACH
1227 */
Al Viro498052b2009-03-30 07:20:30 -04001228int check_unsafe_exec(struct linux_binprm *bprm)
David Howellsa6f76f22008-11-14 10:39:24 +11001229{
David Howells0bf2f3a2009-02-06 11:45:46 +00001230 struct task_struct *p = current, *t;
Al Virof1191b52009-03-30 07:35:18 -04001231 unsigned n_fs;
Al Viro498052b2009-03-30 07:20:30 -04001232 int res = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001233
Tejun Heo4b9d33e2011-06-17 16:50:38 +02001234 if (p->ptrace) {
1235 if (p->ptrace & PT_PTRACE_CAP)
1236 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1237 else
1238 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1239 }
David Howellsa6f76f22008-11-14 10:39:24 +11001240
David Howells0bf2f3a2009-02-06 11:45:46 +00001241 n_fs = 1;
Nick Piggin2a4419b2010-08-18 04:37:33 +10001242 spin_lock(&p->fs->lock);
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001243 rcu_read_lock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001244 for (t = next_thread(p); t != p; t = next_thread(t)) {
1245 if (t->fs == p->fs)
1246 n_fs++;
David Howells0bf2f3a2009-02-06 11:45:46 +00001247 }
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001248 rcu_read_unlock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001249
Al Virof1191b52009-03-30 07:35:18 -04001250 if (p->fs->users > n_fs) {
David Howellsa6f76f22008-11-14 10:39:24 +11001251 bprm->unsafe |= LSM_UNSAFE_SHARE;
Al Viro498052b2009-03-30 07:20:30 -04001252 } else {
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001253 res = -EAGAIN;
1254 if (!p->fs->in_exec) {
1255 p->fs->in_exec = 1;
1256 res = 1;
1257 }
Al Viro498052b2009-03-30 07:20:30 -04001258 }
Nick Piggin2a4419b2010-08-18 04:37:33 +10001259 spin_unlock(&p->fs->lock);
Al Viro498052b2009-03-30 07:20:30 -04001260
1261 return res;
David Howellsa6f76f22008-11-14 10:39:24 +11001262}
1263
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264/*
1265 * Fill the binprm structure from the inode.
1266 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
David Howellsa6f76f22008-11-14 10:39:24 +11001267 *
1268 * This may be called multiple times for binary chains (scripts for example).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 */
1270int prepare_binprm(struct linux_binprm *bprm)
1271{
David Howellsa6f76f22008-11-14 10:39:24 +11001272 umode_t mode;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001273 struct inode * inode = bprm->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 int retval;
1275
1276 mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 if (bprm->file->f_op == NULL)
1278 return -EACCES;
1279
David Howellsa6f76f22008-11-14 10:39:24 +11001280 /* clear any previous set[ug]id data from a previous binary */
1281 bprm->cred->euid = current_euid();
1282 bprm->cred->egid = current_egid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
David Howellsa6f76f22008-11-14 10:39:24 +11001284 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 /* Set-uid? */
1286 if (mode & S_ISUID) {
David Howellsa6f76f22008-11-14 10:39:24 +11001287 bprm->per_clear |= PER_CLEAR_ON_SETID;
1288 bprm->cred->euid = inode->i_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 }
1290
1291 /* Set-gid? */
1292 /*
1293 * If setgid is set but no group execute bit then this
1294 * is a candidate for mandatory locking, not a setgid
1295 * executable.
1296 */
1297 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
David Howellsa6f76f22008-11-14 10:39:24 +11001298 bprm->per_clear |= PER_CLEAR_ON_SETID;
1299 bprm->cred->egid = inode->i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 }
1301 }
1302
1303 /* fill in binprm security blob */
David Howellsa6f76f22008-11-14 10:39:24 +11001304 retval = security_bprm_set_creds(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 if (retval)
1306 return retval;
David Howellsa6f76f22008-11-14 10:39:24 +11001307 bprm->cred_prepared = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
David Howellsa6f76f22008-11-14 10:39:24 +11001309 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1310 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311}
1312
1313EXPORT_SYMBOL(prepare_binprm);
1314
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001315/*
1316 * Arguments are '\0' separated strings found at the location bprm->p
1317 * points to; chop off the first by relocating brpm->p to right after
1318 * the first '\0' encountered.
1319 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001320int remove_arg_zero(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321{
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001322 int ret = 0;
1323 unsigned long offset;
1324 char *kaddr;
1325 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001327 if (!bprm->argc)
1328 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001330 do {
1331 offset = bprm->p & ~PAGE_MASK;
1332 page = get_arg_page(bprm, bprm->p, 0);
1333 if (!page) {
1334 ret = -EFAULT;
1335 goto out;
1336 }
1337 kaddr = kmap_atomic(page, KM_USER0);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001338
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001339 for (; offset < PAGE_SIZE && kaddr[offset];
1340 offset++, bprm->p++)
1341 ;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001342
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001343 kunmap_atomic(kaddr, KM_USER0);
1344 put_arg_page(page);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001345
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001346 if (offset == PAGE_SIZE)
1347 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1348 } while (offset == PAGE_SIZE);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001349
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001350 bprm->p++;
1351 bprm->argc--;
1352 ret = 0;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001353
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001354out:
1355 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357EXPORT_SYMBOL(remove_arg_zero);
1358
1359/*
1360 * cycle the list of binary formats handler, until one recognizes the image
1361 */
1362int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1363{
Roland McGrath85f33462008-12-09 19:36:38 -08001364 unsigned int depth = bprm->recursion_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 int try,retval;
1366 struct linux_binfmt *fmt;
Denys Vlasenkobb188d72011-06-29 04:13:39 +02001367 pid_t old_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 retval = security_bprm_check(bprm);
1370 if (retval)
1371 return retval;
1372
Al Viro473ae302006-04-26 14:04:08 -04001373 retval = audit_bprm(bprm);
1374 if (retval)
1375 return retval;
1376
Denys Vlasenkobb188d72011-06-29 04:13:39 +02001377 /* Need to fetch pid before load_binary changes it */
1378 rcu_read_lock();
1379 old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1380 rcu_read_unlock();
1381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 retval = -ENOENT;
1383 for (try=0; try<2; try++) {
1384 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -07001385 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1387 if (!fn)
1388 continue;
1389 if (!try_module_get(fmt->module))
1390 continue;
1391 read_unlock(&binfmt_lock);
1392 retval = fn(bprm, regs);
Roland McGrath85f33462008-12-09 19:36:38 -08001393 /*
1394 * Restore the depth counter to its starting value
1395 * in this call, so we don't have to rely on every
1396 * load_binary function to restore it on return.
1397 */
1398 bprm->recursion_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 if (retval >= 0) {
Roland McGrath85f33462008-12-09 19:36:38 -08001400 if (depth == 0)
Denys Vlasenkobb188d72011-06-29 04:13:39 +02001401 ptrace_event(PTRACE_EVENT_EXEC,
1402 old_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 put_binfmt(fmt);
1404 allow_write_access(bprm->file);
1405 if (bprm->file)
1406 fput(bprm->file);
1407 bprm->file = NULL;
1408 current->did_exec = 1;
Matt Helsley9f460802005-11-07 00:59:16 -08001409 proc_exec_connector(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 return retval;
1411 }
1412 read_lock(&binfmt_lock);
1413 put_binfmt(fmt);
1414 if (retval != -ENOEXEC || bprm->mm == NULL)
1415 break;
1416 if (!bprm->file) {
1417 read_unlock(&binfmt_lock);
1418 return retval;
1419 }
1420 }
1421 read_unlock(&binfmt_lock);
Tetsuo Handab4edf8b2011-07-26 16:08:42 -07001422#ifdef CONFIG_MODULES
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 if (retval != -ENOEXEC || bprm->mm == NULL) {
1424 break;
Johannes Berg5f4123b2008-07-09 10:28:40 +02001425 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1427 if (printable(bprm->buf[0]) &&
1428 printable(bprm->buf[1]) &&
1429 printable(bprm->buf[2]) &&
1430 printable(bprm->buf[3]))
1431 break; /* -ENOEXEC */
Tetsuo Handa91219352011-07-26 16:08:41 -07001432 if (try)
1433 break; /* -ENOEXEC */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 }
Tetsuo Handab4edf8b2011-07-26 16:08:42 -07001436#else
1437 break;
1438#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 }
1440 return retval;
1441}
1442
1443EXPORT_SYMBOL(search_binary_handler);
1444
1445/*
1446 * sys_execve() executes a new program.
1447 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001448static int do_execve_common(const char *filename,
1449 struct user_arg_ptr argv,
1450 struct user_arg_ptr envp,
1451 struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452{
1453 struct linux_binprm *bprm;
1454 struct file *file;
Al Viro3b125382008-04-22 05:31:30 -04001455 struct files_struct *displaced;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001456 bool clear_in_exec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 int retval;
Vasiliy Kulikov72fa5992011-08-08 19:02:04 +04001458 const struct cred *cred = current_cred();
1459
1460 /*
1461 * We move the actual failure in case of RLIMIT_NPROC excess from
1462 * set*uid() to execve() because too many poorly written programs
1463 * don't check setuid() return code. Here we additionally recheck
1464 * whether NPROC limit is still exceeded.
1465 */
1466 if ((current->flags & PF_NPROC_EXCEEDED) &&
1467 atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1468 retval = -EAGAIN;
1469 goto out_ret;
1470 }
1471
1472 /* We're below the limit (still or again), so we don't want to make
1473 * further execve() calls fail. */
1474 current->flags &= ~PF_NPROC_EXCEEDED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
Al Viro3b125382008-04-22 05:31:30 -04001476 retval = unshare_files(&displaced);
Al Virofd8328b2008-04-22 05:11:59 -04001477 if (retval)
1478 goto out_ret;
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 retval = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001481 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 if (!bprm)
Al Virofd8328b2008-04-22 05:11:59 -04001483 goto out_files;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Oleg Nesterova2a84742009-09-05 11:17:13 -07001485 retval = prepare_bprm_creds(bprm);
1486 if (retval)
David Howellsa6f76f22008-11-14 10:39:24 +11001487 goto out_free;
Al Viro498052b2009-03-30 07:20:30 -04001488
1489 retval = check_unsafe_exec(bprm);
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001490 if (retval < 0)
Oleg Nesterova2a84742009-09-05 11:17:13 -07001491 goto out_free;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001492 clear_in_exec = retval;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001493 current->in_execve = 1;
David Howellsa6f76f22008-11-14 10:39:24 +11001494
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 file = open_exec(filename);
1496 retval = PTR_ERR(file);
1497 if (IS_ERR(file))
Al Viro498052b2009-03-30 07:20:30 -04001498 goto out_unmark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
1500 sched_exec();
1501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 bprm->file = file;
1503 bprm->filename = filename;
1504 bprm->interp = filename;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001505
1506 retval = bprm_mm_init(bprm);
1507 if (retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 goto out_file;
1509
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001510 bprm->argc = count(argv, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 if ((retval = bprm->argc) < 0)
David Howellsa6f76f22008-11-14 10:39:24 +11001512 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001514 bprm->envc = count(envp, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 if ((retval = bprm->envc) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 goto out;
1517
1518 retval = prepare_binprm(bprm);
1519 if (retval < 0)
1520 goto out;
1521
1522 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1523 if (retval < 0)
1524 goto out;
1525
1526 bprm->exec = bprm->p;
1527 retval = copy_strings(bprm->envc, envp, bprm);
1528 if (retval < 0)
1529 goto out;
1530
1531 retval = copy_strings(bprm->argc, argv, bprm);
1532 if (retval < 0)
1533 goto out;
1534
1535 retval = search_binary_handler(bprm,regs);
David Howellsa6f76f22008-11-14 10:39:24 +11001536 if (retval < 0)
1537 goto out;
1538
1539 /* execve succeeded */
Al Viro498052b2009-03-30 07:20:30 -04001540 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001541 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001542 acct_update_integrals(current);
1543 free_bprm(bprm);
1544 if (displaced)
1545 put_files_struct(displaced);
1546 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
1548out:
Oleg Nesterov3c77f842010-11-30 20:55:34 +01001549 if (bprm->mm) {
1550 acct_arg_size(bprm, 0);
1551 mmput(bprm->mm);
1552 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
1554out_file:
1555 if (bprm->file) {
1556 allow_write_access(bprm->file);
1557 fput(bprm->file);
1558 }
David Howellsa6f76f22008-11-14 10:39:24 +11001559
Al Viro498052b2009-03-30 07:20:30 -04001560out_unmark:
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001561 if (clear_in_exec)
1562 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001563 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001564
1565out_free:
Al Viro08a6fac2008-05-10 16:38:25 -04001566 free_bprm(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
Al Virofd8328b2008-04-22 05:11:59 -04001568out_files:
Al Viro3b125382008-04-22 05:31:30 -04001569 if (displaced)
1570 reset_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571out_ret:
1572 return retval;
1573}
1574
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001575int do_execve(const char *filename,
1576 const char __user *const __user *__argv,
1577 const char __user *const __user *__envp,
1578 struct pt_regs *regs)
1579{
Oleg Nesterov0e028462011-03-06 18:02:54 +01001580 struct user_arg_ptr argv = { .ptr.native = __argv };
1581 struct user_arg_ptr envp = { .ptr.native = __envp };
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001582 return do_execve_common(filename, argv, envp, regs);
1583}
1584
Oleg Nesterov0e028462011-03-06 18:02:54 +01001585#ifdef CONFIG_COMPAT
1586int compat_do_execve(char *filename,
1587 compat_uptr_t __user *__argv,
1588 compat_uptr_t __user *__envp,
1589 struct pt_regs *regs)
1590{
1591 struct user_arg_ptr argv = {
1592 .is_compat = true,
1593 .ptr.compat = __argv,
1594 };
1595 struct user_arg_ptr envp = {
1596 .is_compat = true,
1597 .ptr.compat = __envp,
1598 };
1599 return do_execve_common(filename, argv, envp, regs);
1600}
1601#endif
1602
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001603void set_binfmt(struct linux_binfmt *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604{
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001605 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001607 if (mm->binfmt)
1608 module_put(mm->binfmt->module);
1609
1610 mm->binfmt = new;
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001611 if (new)
1612 __module_get(new->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613}
1614
1615EXPORT_SYMBOL(set_binfmt);
1616
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001617static int expand_corename(struct core_name *cn)
1618{
1619 char *old_corename = cn->corename;
1620
1621 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1622 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1623
1624 if (!cn->corename) {
1625 kfree(old_corename);
1626 return -ENOMEM;
1627 }
1628
1629 return 0;
1630}
1631
1632static int cn_printf(struct core_name *cn, const char *fmt, ...)
1633{
1634 char *cur;
1635 int need;
1636 int ret;
1637 va_list arg;
1638
1639 va_start(arg, fmt);
1640 need = vsnprintf(NULL, 0, fmt, arg);
1641 va_end(arg);
1642
1643 if (likely(need < cn->size - cn->used - 1))
1644 goto out_printf;
1645
1646 ret = expand_corename(cn);
1647 if (ret)
1648 goto expand_fail;
1649
1650out_printf:
1651 cur = cn->corename + cn->used;
1652 va_start(arg, fmt);
1653 vsnprintf(cur, need + 1, fmt, arg);
1654 va_end(arg);
1655 cn->used += need;
1656 return 0;
1657
1658expand_fail:
1659 return ret;
1660}
1661
Jiri Slaby2c563732011-07-26 16:08:33 -07001662static void cn_escape(char *str)
1663{
1664 for (; *str; str++)
1665 if (*str == '/')
1666 *str = '!';
1667}
1668
Jiri Slaby57cc0832011-05-26 16:25:46 -07001669static int cn_print_exe_file(struct core_name *cn)
1670{
1671 struct file *exe_file;
Jiri Slaby2c563732011-07-26 16:08:33 -07001672 char *pathbuf, *path;
Jiri Slaby57cc0832011-05-26 16:25:46 -07001673 int ret;
1674
1675 exe_file = get_mm_exe_file(current->mm);
Jiri Slaby2c563732011-07-26 16:08:33 -07001676 if (!exe_file) {
1677 char *commstart = cn->corename + cn->used;
1678 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1679 cn_escape(commstart);
1680 return ret;
1681 }
Jiri Slaby57cc0832011-05-26 16:25:46 -07001682
1683 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1684 if (!pathbuf) {
1685 ret = -ENOMEM;
1686 goto put_exe_file;
1687 }
1688
1689 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1690 if (IS_ERR(path)) {
1691 ret = PTR_ERR(path);
1692 goto free_buf;
1693 }
1694
Jiri Slaby2c563732011-07-26 16:08:33 -07001695 cn_escape(path);
Jiri Slaby57cc0832011-05-26 16:25:46 -07001696
1697 ret = cn_printf(cn, "%s", path);
1698
1699free_buf:
1700 kfree(pathbuf);
1701put_exe_file:
1702 fput(exe_file);
1703 return ret;
1704}
1705
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706/* format_corename will inspect the pattern parameter, and output a
1707 * name into corename, which must have space for at least
1708 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1709 */
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001710static int format_corename(struct core_name *cn, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711{
David Howells86a264a2008-11-14 10:39:18 +11001712 const struct cred *cred = current_cred();
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001713 const char *pat_ptr = core_pattern;
1714 int ispipe = (*pat_ptr == '|');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 int pid_in_pattern = 0;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001716 int err = 0;
1717
1718 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1719 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1720 cn->used = 0;
1721
1722 if (!cn->corename)
1723 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
1725 /* Repeat as long as we have more pattern to process and more output
1726 space */
1727 while (*pat_ptr) {
1728 if (*pat_ptr != '%') {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001729 if (*pat_ptr == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 goto out;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001731 err = cn_printf(cn, "%c", *pat_ptr++);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 } else {
1733 switch (*++pat_ptr) {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001734 /* single % at the end, drop that */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 case 0:
1736 goto out;
1737 /* Double percent, output one percent */
1738 case '%':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001739 err = cn_printf(cn, "%c", '%');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 break;
1741 /* pid */
1742 case 'p':
1743 pid_in_pattern = 1;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001744 err = cn_printf(cn, "%d",
1745 task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 break;
1747 /* uid */
1748 case 'u':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001749 err = cn_printf(cn, "%d", cred->uid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 break;
1751 /* gid */
1752 case 'g':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001753 err = cn_printf(cn, "%d", cred->gid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 break;
1755 /* signal that caused the coredump */
1756 case 's':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001757 err = cn_printf(cn, "%ld", signr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 break;
1759 /* UNIX time of coredump */
1760 case 't': {
1761 struct timeval tv;
1762 do_gettimeofday(&tv);
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001763 err = cn_printf(cn, "%lu", tv.tv_sec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 break;
1765 }
1766 /* hostname */
Jiri Slaby2c563732011-07-26 16:08:33 -07001767 case 'h': {
1768 char *namestart = cn->corename + cn->used;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 down_read(&uts_sem);
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001770 err = cn_printf(cn, "%s",
1771 utsname()->nodename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 up_read(&uts_sem);
Jiri Slaby2c563732011-07-26 16:08:33 -07001773 cn_escape(namestart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 break;
Jiri Slaby2c563732011-07-26 16:08:33 -07001775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 /* executable */
Jiri Slaby2c563732011-07-26 16:08:33 -07001777 case 'e': {
1778 char *commstart = cn->corename + cn->used;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001779 err = cn_printf(cn, "%s", current->comm);
Jiri Slaby2c563732011-07-26 16:08:33 -07001780 cn_escape(commstart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 break;
Jiri Slaby2c563732011-07-26 16:08:33 -07001782 }
Jiri Slaby57cc0832011-05-26 16:25:46 -07001783 case 'E':
1784 err = cn_print_exe_file(cn);
1785 break;
Neil Horman74aadce2007-10-16 23:26:35 -07001786 /* core limit size */
1787 case 'c':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001788 err = cn_printf(cn, "%lu",
1789 rlimit(RLIMIT_CORE));
Neil Horman74aadce2007-10-16 23:26:35 -07001790 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 default:
1792 break;
1793 }
1794 ++pat_ptr;
1795 }
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001796
1797 if (err)
1798 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 }
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001800
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 /* Backward compatibility with core_uses_pid:
1802 *
1803 * If core_pattern does not include a %p (as is the default)
1804 * and core_uses_pid is set, then .%pid will be appended to
Alan Coxc4bbafd2007-04-16 22:53:13 -07001805 * the filename. Do not do this for piped commands. */
Oleg Nesterov64093242008-10-18 20:28:22 -07001806 if (!ispipe && !pid_in_pattern && core_uses_pid) {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001807 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1808 if (err)
1809 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 }
Alan Coxc4bbafd2007-04-16 22:53:13 -07001811out:
Alan Coxc4bbafd2007-04-16 22:53:13 -07001812 return ispipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813}
1814
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001815static int zap_process(struct task_struct *start, int exit_code)
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001816{
1817 struct task_struct *t;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001818 int nr = 0;
Oleg Nesterov281de332006-06-26 00:26:06 -07001819
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001820 start->signal->flags = SIGNAL_GROUP_EXIT;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001821 start->signal->group_exit_code = exit_code;
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001822 start->signal->group_stop_count = 0;
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001823
1824 t = start;
1825 do {
Tejun Heo6dfca322011-06-02 11:14:00 +02001826 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001827 if (t != current && t->mm) {
Oleg Nesterov281de332006-06-26 00:26:06 -07001828 sigaddset(&t->pending.signal, SIGKILL);
1829 signal_wake_up(t, 1);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001830 nr++;
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001831 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001832 } while_each_thread(start, t);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001833
1834 return nr;
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001835}
1836
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001837static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001838 struct core_state *core_state, int exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839{
1840 struct task_struct *g, *p;
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001841 unsigned long flags;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001842 int nr = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001844 spin_lock_irq(&tsk->sighand->siglock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001845 if (!signal_group_exit(tsk->signal)) {
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001846 mm->core_state = core_state;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001847 nr = zap_process(tsk, exit_code);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 }
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001849 spin_unlock_irq(&tsk->sighand->siglock);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001850 if (unlikely(nr < 0))
1851 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001853 if (atomic_read(&mm->mm_users) == nr + 1)
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001854 goto done;
Oleg Nesterove4901f92008-07-25 01:47:31 -07001855 /*
1856 * We should find and kill all tasks which use this mm, and we should
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001857 * count them correctly into ->nr_threads. We don't take tasklist
Oleg Nesterove4901f92008-07-25 01:47:31 -07001858 * lock, but this is safe wrt:
1859 *
1860 * fork:
1861 * None of sub-threads can fork after zap_process(leader). All
1862 * processes which were created before this point should be
1863 * visible to zap_threads() because copy_process() adds the new
1864 * process to the tail of init_task.tasks list, and lock/unlock
1865 * of ->siglock provides a memory barrier.
1866 *
1867 * do_exit:
1868 * The caller holds mm->mmap_sem. This means that the task which
1869 * uses this mm can't pass exit_mm(), so it can't exit or clear
1870 * its ->mm.
1871 *
1872 * de_thread:
1873 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1874 * we must see either old or new leader, this does not matter.
1875 * However, it can change p->sighand, so lock_task_sighand(p)
1876 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1877 * it can't fail.
1878 *
1879 * Note also that "g" can be the old leader with ->mm == NULL
1880 * and already unhashed and thus removed from ->thread_group.
1881 * This is OK, __unhash_process()->list_del_rcu() does not
1882 * clear the ->next pointer, we will find the new leader via
1883 * next_thread().
1884 */
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001885 rcu_read_lock();
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001886 for_each_process(g) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001887 if (g == tsk->group_leader)
1888 continue;
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001889 if (g->flags & PF_KTHREAD)
1890 continue;
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001891 p = g;
1892 do {
1893 if (p->mm) {
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001894 if (unlikely(p->mm == mm)) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001895 lock_task_sighand(p, &flags);
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001896 nr += zap_process(p, exit_code);
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001897 unlock_task_sighand(p, &flags);
1898 }
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001899 break;
1900 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001901 } while_each_thread(g, p);
Oleg Nesterovaceecc042006-06-26 00:26:05 -07001902 }
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001903 rcu_read_unlock();
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001904done:
Oleg Nesterovc5f1cc82008-07-25 01:47:42 -07001905 atomic_set(&core_state->nr_threads, nr);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001906 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907}
1908
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001909static int coredump_wait(int exit_code, struct core_state *core_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910{
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001911 struct task_struct *tsk = current;
1912 struct mm_struct *mm = tsk->mm;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001913 struct completion *vfork_done;
Oleg Nesterov269b0052010-05-26 14:43:08 -07001914 int core_waiters = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001916 init_completion(&core_state->startup);
Oleg Nesterovb564daf2008-07-25 01:47:44 -07001917 core_state->dumper.task = tsk;
1918 core_state->dumper.next = NULL;
Oleg Nesterov269b0052010-05-26 14:43:08 -07001919
1920 down_write(&mm->mmap_sem);
1921 if (!mm->core_state)
1922 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001923 up_write(&mm->mmap_sem);
1924
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001925 if (unlikely(core_waiters < 0))
1926 goto fail;
1927
1928 /*
1929 * Make sure nobody is waiting for us to release the VM,
1930 * otherwise we can deadlock when we wait on each other
1931 */
1932 vfork_done = tsk->vfork_done;
1933 if (vfork_done) {
1934 tsk->vfork_done = NULL;
1935 complete(vfork_done);
1936 }
1937
Oleg Nesterov2384f552005-10-30 15:02:47 -08001938 if (core_waiters)
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001939 wait_for_completion(&core_state->startup);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001940fail:
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001941 return core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942}
1943
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001944static void coredump_finish(struct mm_struct *mm)
1945{
1946 struct core_thread *curr, *next;
1947 struct task_struct *task;
1948
1949 next = mm->core_state->dumper.next;
1950 while ((curr = next) != NULL) {
1951 next = curr->next;
1952 task = curr->task;
1953 /*
1954 * see exit_mm(), curr->task must not see
1955 * ->task == NULL before we read ->next.
1956 */
1957 smp_mb();
1958 curr->task = NULL;
1959 wake_up_process(task);
1960 }
1961
1962 mm->core_state = NULL;
1963}
1964
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001965/*
1966 * set_dumpable converts traditional three-value dumpable to two flags and
1967 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1968 * these bits are not changed atomically. So get_dumpable can observe the
1969 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1970 * return either old dumpable or new one by paying attention to the order of
1971 * modifying the bits.
1972 *
1973 * dumpable | mm->flags (binary)
1974 * old new | initial interim final
1975 * ---------+-----------------------
1976 * 0 1 | 00 01 01
1977 * 0 2 | 00 10(*) 11
1978 * 1 0 | 01 00 00
1979 * 1 2 | 01 11 11
1980 * 2 0 | 11 10(*) 00
1981 * 2 1 | 11 11 01
1982 *
1983 * (*) get_dumpable regards interim value of 10 as 11.
1984 */
1985void set_dumpable(struct mm_struct *mm, int value)
1986{
1987 switch (value) {
1988 case 0:
1989 clear_bit(MMF_DUMPABLE, &mm->flags);
1990 smp_wmb();
1991 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1992 break;
1993 case 1:
1994 set_bit(MMF_DUMPABLE, &mm->flags);
1995 smp_wmb();
1996 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1997 break;
1998 case 2:
1999 set_bit(MMF_DUMP_SECURELY, &mm->flags);
2000 smp_wmb();
2001 set_bit(MMF_DUMPABLE, &mm->flags);
2002 break;
2003 }
2004}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07002005
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002006static int __get_dumpable(unsigned long mm_flags)
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07002007{
2008 int ret;
2009
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002010 ret = mm_flags & MMF_DUMPABLE_MASK;
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07002011 return (ret >= 2) ? 2 : ret;
2012}
2013
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002014int get_dumpable(struct mm_struct *mm)
2015{
2016 return __get_dumpable(mm->flags);
2017}
2018
Neil Horman61be2282009-09-23 15:56:58 -07002019static void wait_for_dump_helpers(struct file *file)
2020{
2021 struct pipe_inode_info *pipe;
2022
2023 pipe = file->f_path.dentry->d_inode->i_pipe;
2024
2025 pipe_lock(pipe);
2026 pipe->readers++;
2027 pipe->writers--;
2028
2029 while ((pipe->readers > 1) && (!signal_pending(current))) {
2030 wake_up_interruptible_sync(&pipe->wait);
2031 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2032 pipe_wait(pipe);
2033 }
2034
2035 pipe->readers--;
2036 pipe->writers++;
2037 pipe_unlock(pipe);
2038
2039}
2040
2041
Neil Horman898b3742010-05-26 14:42:59 -07002042/*
Holger Hans Peter Freyther1bef8292011-02-24 17:46:49 +01002043 * umh_pipe_setup
Neil Horman898b3742010-05-26 14:42:59 -07002044 * helper function to customize the process used
2045 * to collect the core in userspace. Specifically
2046 * it sets up a pipe and installs it as fd 0 (stdin)
2047 * for the process. Returns 0 on success, or
2048 * PTR_ERR on failure.
2049 * Note that it also sets the core limit to 1. This
2050 * is a special value that we use to trap recursive
2051 * core dumps
2052 */
David Howells87966992011-06-17 11:25:59 +01002053static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
Neil Horman898b3742010-05-26 14:42:59 -07002054{
2055 struct file *rp, *wp;
2056 struct fdtable *fdt;
2057 struct coredump_params *cp = (struct coredump_params *)info->data;
2058 struct files_struct *cf = current->files;
2059
2060 wp = create_write_pipe(0);
2061 if (IS_ERR(wp))
2062 return PTR_ERR(wp);
2063
2064 rp = create_read_pipe(wp, 0);
2065 if (IS_ERR(rp)) {
2066 free_write_pipe(wp);
2067 return PTR_ERR(rp);
2068 }
2069
2070 cp->file = wp;
2071
2072 sys_close(0);
2073 fd_install(0, rp);
2074 spin_lock(&cf->file_lock);
2075 fdt = files_fdtable(cf);
2076 FD_SET(0, fdt->open_fds);
2077 FD_CLR(0, fdt->close_on_exec);
2078 spin_unlock(&cf->file_lock);
2079
2080 /* and disallow core files too */
2081 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2082
2083 return 0;
2084}
2085
WANG Cong8cd3ac32009-01-06 14:42:48 -08002086void do_coredump(long signr, int exit_code, struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087{
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07002088 struct core_state core_state;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002089 struct core_name cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 struct mm_struct *mm = current->mm;
2091 struct linux_binfmt * binfmt;
David Howellsd84f4f92008-11-14 10:39:23 +11002092 const struct cred *old_cred;
2093 struct cred *cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 int retval = 0;
Alan Coxd6e71142005-06-23 00:09:43 -07002095 int flag = 0;
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002096 int ispipe;
Neil Hormana2939802009-09-23 15:56:56 -07002097 static atomic_t core_dump_count = ATOMIC_INIT(0);
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002098 struct coredump_params cprm = {
2099 .signr = signr,
2100 .regs = regs,
Jiri Slabyd554ed892010-03-05 13:42:42 -08002101 .limit = rlimit(RLIMIT_CORE),
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002102 /*
2103 * We must use the same mm->flags while dumping core to avoid
2104 * inconsistency of bit flags, since this flag is not protected
2105 * by any locks.
2106 */
2107 .mm_flags = mm->flags,
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002108 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109
Steve Grubb0a4ff8c2007-04-19 10:28:21 -04002110 audit_core_dumps(signr);
2111
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07002112 binfmt = mm->binfmt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 if (!binfmt || !binfmt->core_dump)
2114 goto fail;
Oleg Nesterov269b0052010-05-26 14:43:08 -07002115 if (!__get_dumpable(cprm.mm_flags))
2116 goto fail;
David Howellsd84f4f92008-11-14 10:39:23 +11002117
2118 cred = prepare_creds();
Oleg Nesterov5e43aef52010-05-26 14:43:07 -07002119 if (!cred)
David Howellsd84f4f92008-11-14 10:39:23 +11002120 goto fail;
Alan Coxd6e71142005-06-23 00:09:43 -07002121 /*
2122 * We cannot trust fsuid as being the "true" uid of the
2123 * process nor do we know its entire history. We only know it
2124 * was tainted so we dump it as root in mode 2.
2125 */
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002126 if (__get_dumpable(cprm.mm_flags) == 2) {
2127 /* Setuid core dump mode */
Alan Coxd6e71142005-06-23 00:09:43 -07002128 flag = O_EXCL; /* Stop rewrite attacks */
David Howellsd84f4f92008-11-14 10:39:23 +11002129 cred->fsuid = 0; /* Dump root private */
Alan Coxd6e71142005-06-23 00:09:43 -07002130 }
Oleg Nesterov1291cf42005-10-30 15:02:54 -08002131
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07002132 retval = coredump_wait(exit_code, &core_state);
Oleg Nesterov5e43aef52010-05-26 14:43:07 -07002133 if (retval < 0)
2134 goto fail_creds;
David Howellsd84f4f92008-11-14 10:39:23 +11002135
2136 old_cred = override_creds(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137
2138 /*
2139 * Clear any false indication of pending signals that might
2140 * be seen by the filesystem code called to write the core file.
2141 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 clear_thread_flag(TIF_SIGPENDING);
2143
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002144 ispipe = format_corename(&cn, signr);
2145
Alan Coxc4bbafd2007-04-16 22:53:13 -07002146 if (ispipe) {
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002147 int dump_count;
2148 char **helper_argv;
2149
Oleg Nesterov99b64562011-07-26 16:08:34 -07002150 if (ispipe < 0) {
2151 printk(KERN_WARNING "format_corename failed\n");
2152 printk(KERN_WARNING "Aborting core\n");
2153 goto fail_corename;
2154 }
2155
Neil Horman898b3742010-05-26 14:42:59 -07002156 if (cprm.limit == 1) {
Neil Horman725eae32009-09-23 15:56:54 -07002157 /*
2158 * Normally core limits are irrelevant to pipes, since
2159 * we're not writing to the file system, but we use
Neil Horman898b3742010-05-26 14:42:59 -07002160 * cprm.limit of 1 here as a speacial value. Any
2161 * non-1 limit gets set to RLIM_INFINITY below, but
Neil Horman725eae32009-09-23 15:56:54 -07002162 * a limit of 0 skips the dump. This is a consistent
2163 * way to catch recursive crashes. We can still crash
Neil Horman898b3742010-05-26 14:42:59 -07002164 * if the core_pattern binary sets RLIM_CORE = !1
Neil Horman725eae32009-09-23 15:56:54 -07002165 * but it runs as root, and can do lots of stupid things
2166 * Note that we use task_tgid_vnr here to grab the pid
2167 * of the process group leader. That way we get the
2168 * right pid if a thread in a multi-threaded
2169 * core_pattern process dies.
2170 */
2171 printk(KERN_WARNING
Neil Horman898b3742010-05-26 14:42:59 -07002172 "Process %d(%s) has RLIMIT_CORE set to 1\n",
Neil Horman725eae32009-09-23 15:56:54 -07002173 task_tgid_vnr(current), current->comm);
2174 printk(KERN_WARNING "Aborting core\n");
2175 goto fail_unlock;
2176 }
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002177 cprm.limit = RLIM_INFINITY;
Neil Horman725eae32009-09-23 15:56:54 -07002178
Neil Hormana2939802009-09-23 15:56:56 -07002179 dump_count = atomic_inc_return(&core_dump_count);
2180 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2181 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2182 task_tgid_vnr(current), current->comm);
2183 printk(KERN_WARNING "Skipping core dump\n");
2184 goto fail_dropcount;
2185 }
2186
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002187 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
Tetsuo Handa350eaf72009-01-06 14:41:11 -08002188 if (!helper_argv) {
2189 printk(KERN_WARNING "%s failed to allocate memory\n",
2190 __func__);
Neil Hormana2939802009-09-23 15:56:56 -07002191 goto fail_dropcount;
Tetsuo Handa350eaf72009-01-06 14:41:11 -08002192 }
Neil Horman32321132007-10-16 23:26:36 -07002193
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002194 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2195 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2196 NULL, &cprm);
2197 argv_free(helper_argv);
2198 if (retval) {
Andi Kleend025c9d2006-09-30 23:29:28 -07002199 printk(KERN_INFO "Core dump to %s pipe failed\n",
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002200 cn.corename);
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002201 goto close_fail;
Andi Kleend025c9d2006-09-30 23:29:28 -07002202 }
Oleg Nesterovc7135412010-05-26 14:43:05 -07002203 } else {
2204 struct inode *inode;
2205
2206 if (cprm.limit < binfmt->min_coredump)
2207 goto fail_unlock;
2208
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002209 cprm.file = filp_open(cn.corename,
Alexey Dobriyan6d4df672006-12-06 20:40:39 -08002210 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2211 0600);
Oleg Nesterovc7135412010-05-26 14:43:05 -07002212 if (IS_ERR(cprm.file))
2213 goto fail_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214
Oleg Nesterovc7135412010-05-26 14:43:05 -07002215 inode = cprm.file->f_path.dentry->d_inode;
2216 if (inode->i_nlink > 1)
2217 goto close_fail;
2218 if (d_unhashed(cprm.file->f_path.dentry))
2219 goto close_fail;
2220 /*
2221 * AK: actually i see no reason to not allow this for named
2222 * pipes etc, but keep the previous behaviour for now.
2223 */
2224 if (!S_ISREG(inode->i_mode))
2225 goto close_fail;
2226 /*
2227 * Dont allow local users get cute and trick others to coredump
2228 * into their pre-created files.
2229 */
2230 if (inode->i_uid != current_fsuid())
2231 goto close_fail;
2232 if (!cprm.file->f_op || !cprm.file->f_op->write)
2233 goto close_fail;
2234 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2235 goto close_fail;
2236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002238 retval = binfmt->core_dump(&cprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 if (retval)
2240 current->signal->group_exit_code |= 0x80;
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002241
Neil Horman61be2282009-09-23 15:56:58 -07002242 if (ispipe && core_pipe_limit)
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002243 wait_for_dump_helpers(cprm.file);
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002244close_fail:
2245 if (cprm.file)
2246 filp_close(cprm.file, NULL);
Neil Hormana2939802009-09-23 15:56:56 -07002247fail_dropcount:
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002248 if (ispipe)
Neil Hormana2939802009-09-23 15:56:56 -07002249 atomic_dec(&core_dump_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250fail_unlock:
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002251 kfree(cn.corename);
2252fail_corename:
Oleg Nesterova94e2d42008-07-25 01:47:46 -07002253 coredump_finish(mm);
Oleg Nesterov5e43aef52010-05-26 14:43:07 -07002254 revert_creds(old_cred);
2255fail_creds:
2256 put_cred(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257fail:
WANG Cong8cd3ac32009-01-06 14:42:48 -08002258 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259}
Linus Torvalds3aa0ce82010-10-14 14:32:06 -07002260
2261/*
2262 * Core dumping helper functions. These are the only things you should
2263 * do on a core-file: use only these functions to write out all the
2264 * necessary info.
2265 */
2266int dump_write(struct file *file, const void *addr, int nr)
2267{
2268 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2269}
Linus Torvalds8fd01d62010-10-14 19:15:28 -07002270EXPORT_SYMBOL(dump_write);
Linus Torvalds3aa0ce82010-10-14 14:32:06 -07002271
2272int dump_seek(struct file *file, loff_t off)
2273{
2274 int ret = 1;
2275
2276 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2277 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2278 return 0;
2279 } else {
2280 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2281
2282 if (!buf)
2283 return 0;
2284 while (off > 0) {
2285 unsigned long n = off;
2286
2287 if (n > PAGE_SIZE)
2288 n = PAGE_SIZE;
2289 if (!dump_write(file, buf, n)) {
2290 ret = 0;
2291 break;
2292 }
2293 off -= n;
2294 }
2295 free_page((unsigned long)buf);
2296 }
2297 return ret;
2298}
Linus Torvalds8fd01d62010-10-14 19:15:28 -07002299EXPORT_SYMBOL(dump_seek);