blob: da80612a35f42e9bc1bb2d09fe79090e44b3808b [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);
Ying Han3d5992d2010-10-26 14:21:23 -0700844 if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
845 atomic_dec(&old_mm->oom_disable_count);
846 atomic_inc(&tsk->mm->oom_disable_count);
847 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 task_unlock(tsk);
849 arch_pick_mmap_layout(mm);
850 if (old_mm) {
851 up_read(&old_mm->mmap_sem);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +0200852 BUG_ON(active_mm != old_mm);
Balbir Singh31a78f22008-09-28 23:09:31 +0100853 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 mmput(old_mm);
855 return 0;
856 }
857 mmdrop(active_mm);
858 return 0;
859}
860
861/*
862 * This function makes sure the current process has its own signal table,
863 * so that flush_signal_handlers can later reset the handlers without
864 * disturbing other processes. (Other processes might share the signal
865 * table via the CLONE_SIGHAND option to clone().)
866 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800867static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868{
869 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700870 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 spinlock_t *lock = &oldsighand->siglock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700873 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 goto no_thread_group;
875
876 /*
877 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800880 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 /*
882 * Another group action in progress, just
883 * return so that the signal is processed.
884 */
885 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 return -EAGAIN;
887 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
Oleg Nesterovd3441932010-05-26 14:43:11 -0700889 sig->group_exit_task = tsk;
890 sig->notify_count = zap_other_threads(tsk);
891 if (!thread_group_leader(tsk))
892 sig->notify_count--;
893
894 while (sig->notify_count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 __set_current_state(TASK_UNINTERRUPTIBLE);
896 spin_unlock_irq(lock);
897 schedule();
898 spin_lock_irq(lock);
899 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 spin_unlock_irq(lock);
901
902 /*
903 * At this point all other threads have exited, all we have to
904 * do is to wait for the thread group leader to become inactive,
905 * and to assume its PID:
906 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700907 if (!thread_group_leader(tsk)) {
Oleg Nesterov81879262008-12-01 14:18:16 -0800908 struct task_struct *leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700909
Oleg Nesterov2800d8d2008-04-30 00:53:12 -0700910 sig->notify_count = -1; /* for exit_notify() */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700911 for (;;) {
912 write_lock_irq(&tasklist_lock);
913 if (likely(leader->exit_state))
914 break;
915 __set_current_state(TASK_UNINTERRUPTIBLE);
916 write_unlock_irq(&tasklist_lock);
917 schedule();
918 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Roland McGrathf5e90282006-04-10 22:54:16 -0700920 /*
921 * The only record we have of the real-time age of a
922 * process, regardless of execs it's done, is start_time.
923 * All the past CPU time is accumulated in signal_struct
924 * from sister threads now dead. But in this non-leader
925 * exec, nothing survives from the original leader thread,
926 * whose birth marks the true age of this process now.
927 * When we take on its identity by switching to its PID, we
928 * also take its birthdate (always earlier than our own).
929 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700930 tsk->start_time = leader->start_time;
Roland McGrathf5e90282006-04-10 22:54:16 -0700931
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700932 BUG_ON(!same_thread_group(leader, tsk));
933 BUG_ON(has_group_leader_pid(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 /*
935 * An exec() starts a new thread group with the
936 * TGID of the previous thread group. Rehash the
937 * two threads with a switched PID, and release
938 * the former thread group leader:
939 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800940
941 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700942 * The old leader becomes a thread of the this thread group.
943 * Note: The old leader also uses this pid until release_task
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800944 * is called. Odd but simple and correct.
945 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700946 detach_pid(tsk, PIDTYPE_PID);
947 tsk->pid = leader->pid;
Sukadev Bhattiprolu3743ca02007-10-18 23:39:51 -0700948 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700949 transfer_pid(leader, tsk, PIDTYPE_PGID);
950 transfer_pid(leader, tsk, PIDTYPE_SID);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800951
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700952 list_replace_rcu(&leader->tasks, &tsk->tasks);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800953 list_replace_init(&leader->sibling, &tsk->sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700955 tsk->group_leader = tsk;
956 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -0600957
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700958 tsk->exit_signal = SIGCHLD;
Oleg Nesterov087806b2011-06-22 23:10:26 +0200959 leader->exit_signal = -1;
Oleg Nesterov962b5642005-11-23 13:37:43 -0800960
961 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
962 leader->exit_state = EXIT_DEAD;
Oleg Nesteroveac1b5e2011-07-21 20:00:43 +0200963
964 /*
965 * We are going to release_task()->ptrace_unlink() silently,
966 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
967 * the tracer wont't block again waiting for this thread.
968 */
969 if (unlikely(leader->ptrace))
970 __wake_up_parent(leader, leader->parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 write_unlock_irq(&tasklist_lock);
Oleg Nesterov81879262008-12-01 14:18:16 -0800972
973 release_task(leader);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800974 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700976 sig->group_exit_task = NULL;
977 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
979no_thread_group:
Jiri Pirko1f102062009-09-22 16:44:10 -0700980 if (current->mm)
981 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
982
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 exit_itimers(sig);
Oleg Nesterovcbaffba2008-05-26 20:55:42 +0400984 flush_itimer_signals();
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300985
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700986 if (atomic_read(&oldsighand->count) != 1) {
987 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700989 * This ->sighand is shared with the CLONE_SIGHAND
990 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700992 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
993 if (!newsighand)
994 return -ENOMEM;
995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 atomic_set(&newsighand->count, 1);
997 memcpy(newsighand->action, oldsighand->action,
998 sizeof(newsighand->action));
999
1000 write_lock_irq(&tasklist_lock);
1001 spin_lock(&oldsighand->siglock);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001002 rcu_assign_pointer(tsk->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 spin_unlock(&oldsighand->siglock);
1004 write_unlock_irq(&tasklist_lock);
1005
Davide Libenzifba2afa2007-05-10 22:23:13 -07001006 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 }
1008
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001009 BUG_ON(!thread_group_leader(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 return 0;
1011}
Oleg Nesterov0840a902007-10-16 23:27:22 -07001012
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013/*
1014 * These functions flushes out all traces of the currently running executable
1015 * so that a new one can be started
1016 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001017static void flush_old_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018{
1019 long j = -1;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001020 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
1022 spin_lock(&files->file_lock);
1023 for (;;) {
1024 unsigned long set, i;
1025
1026 j++;
1027 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001028 fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -08001029 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001031 set = fdt->close_on_exec->fds_bits[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 if (!set)
1033 continue;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001034 fdt->close_on_exec->fds_bits[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 spin_unlock(&files->file_lock);
1036 for ( ; set ; i++,set >>= 1) {
1037 if (set & 1) {
1038 sys_close(i);
1039 }
1040 }
1041 spin_lock(&files->file_lock);
1042
1043 }
1044 spin_unlock(&files->file_lock);
1045}
1046
Andrew Morton59714d62008-02-04 22:27:21 -08001047char *get_task_comm(char *buf, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048{
1049 /* buf must be at least sizeof(tsk->comm) in size */
1050 task_lock(tsk);
1051 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1052 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -08001053 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054}
J Freyensee7d74f492011-05-06 16:56:47 -07001055EXPORT_SYMBOL_GPL(get_task_comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
1057void set_task_comm(struct task_struct *tsk, char *buf)
1058{
1059 task_lock(tsk);
john stultz4614a696b2009-12-14 18:00:05 -08001060
1061 /*
1062 * Threads may access current->comm without holding
1063 * the task lock, so write the string carefully.
1064 * Readers without a lock may see incomplete new
1065 * names but are safe from non-terminating string reads.
1066 */
1067 memset(tsk->comm, 0, TASK_COMM_LEN);
1068 wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1070 task_unlock(tsk);
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001071 perf_event_comm(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072}
1073
1074int flush_old_exec(struct linux_binprm * bprm)
1075{
Linus Torvalds221af7f2010-01-28 22:14:42 -08001076 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
1078 /*
1079 * Make sure we have a private signal table and that
1080 * we are unassociated from the previous thread group.
1081 */
1082 retval = de_thread(current);
1083 if (retval)
1084 goto out;
1085
Matt Helsley925d1c42008-04-29 01:01:36 -07001086 set_mm_exe_file(bprm->mm, bprm->file);
1087
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 * Release all of the old mmap stuff
1090 */
Oleg Nesterov3c77f842010-11-30 20:55:34 +01001091 acct_arg_size(bprm, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 retval = exec_mmap(bprm->mm);
1093 if (retval)
Al Virofd8328b2008-04-22 05:11:59 -04001094 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
1096 bprm->mm = NULL; /* We're using it now */
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001097
Mathias Krausedac853a2011-06-09 20:05:18 +02001098 set_fs(USER_DS);
KOSAKI Motohiro98391cf2010-10-27 15:34:16 -07001099 current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001100 flush_thread();
1101 current->personality &= ~bprm->per_clear;
1102
Linus Torvalds221af7f2010-01-28 22:14:42 -08001103 return 0;
1104
1105out:
1106 return retval;
1107}
1108EXPORT_SYMBOL(flush_old_exec);
1109
Al Viro1b5d7832011-06-19 12:49:47 -04001110void would_dump(struct linux_binprm *bprm, struct file *file)
1111{
1112 if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1113 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1114}
1115EXPORT_SYMBOL(would_dump);
1116
Linus Torvalds221af7f2010-01-28 22:14:42 -08001117void setup_new_exec(struct linux_binprm * bprm)
1118{
1119 int i, ch;
David Howellsd7627462010-08-17 23:52:56 +01001120 const char *name;
Linus Torvalds221af7f2010-01-28 22:14:42 -08001121 char tcomm[sizeof(current->comm)];
1122
1123 arch_pick_mmap_layout(current->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
1125 /* This is the point of no return */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 current->sas_ss_sp = current->sas_ss_size = 0;
1127
David Howellsda9592e2008-11-14 10:39:05 +11001128 if (current_euid() == current_uid() && current_egid() == current_gid())
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001129 set_dumpable(current->mm, 1);
Alan Coxd6e71142005-06-23 00:09:43 -07001130 else
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001131 set_dumpable(current->mm, suid_dumpable);
Alan Coxd6e71142005-06-23 00:09:43 -07001132
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 name = bprm->filename;
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -07001134
1135 /* Copies the binary name from after last slash */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 for (i=0; (ch = *(name++)) != '\0';) {
1137 if (ch == '/')
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -07001138 i = 0; /* overwrite what we wrote */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 else
1140 if (i < (sizeof(tcomm) - 1))
1141 tcomm[i++] = ch;
1142 }
1143 tcomm[i] = '\0';
1144 set_task_comm(current, tcomm);
1145
Benjamin Herrenschmidt0551fbd2006-02-28 16:59:19 -08001146 /* Set the new mm task size. We have to do that late because it may
1147 * depend on TIF_32BIT which is only updated in flush_thread() on
1148 * some architectures like powerpc
1149 */
1150 current->mm->task_size = TASK_SIZE;
1151
David Howellsa6f76f22008-11-14 10:39:24 +11001152 /* install the new credentials */
1153 if (bprm->cred->uid != current_euid() ||
1154 bprm->cred->gid != current_egid()) {
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001155 current->pdeath_signal = 0;
Al Viro1b5d7832011-06-19 12:49:47 -04001156 } else {
1157 would_dump(bprm, bprm->file);
1158 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1159 set_dumpable(current->mm, suid_dumpable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 }
1161
Ingo Molnarf65cb452008-12-16 13:40:44 +01001162 /*
1163 * Flush performance counters when crossing a
1164 * security domain:
1165 */
1166 if (!get_dumpable(current->mm))
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001167 perf_event_exit_task(current);
Ingo Molnarf65cb452008-12-16 13:40:44 +01001168
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 /* An exec changes our domain. We are no longer part of the thread
1170 group */
1171
1172 current->self_exec_id++;
1173
1174 flush_signal_handlers(current, 0);
1175 flush_old_files(current->files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176}
Linus Torvalds221af7f2010-01-28 22:14:42 -08001177EXPORT_SYMBOL(setup_new_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
David Howellsa6f76f22008-11-14 10:39:24 +11001179/*
Oleg Nesterova2a84742009-09-05 11:17:13 -07001180 * Prepare credentials and lock ->cred_guard_mutex.
1181 * install_exec_creds() commits the new creds and drops the lock.
1182 * Or, if exec fails before, free_bprm() should release ->cred and
1183 * and unlock.
1184 */
1185int prepare_bprm_creds(struct linux_binprm *bprm)
1186{
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001187 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
Oleg Nesterova2a84742009-09-05 11:17:13 -07001188 return -ERESTARTNOINTR;
1189
1190 bprm->cred = prepare_exec_creds();
1191 if (likely(bprm->cred))
1192 return 0;
1193
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001194 mutex_unlock(&current->signal->cred_guard_mutex);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001195 return -ENOMEM;
1196}
1197
1198void free_bprm(struct linux_binprm *bprm)
1199{
1200 free_arg_pages(bprm);
1201 if (bprm->cred) {
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001202 mutex_unlock(&current->signal->cred_guard_mutex);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001203 abort_creds(bprm->cred);
1204 }
1205 kfree(bprm);
1206}
1207
1208/*
David Howellsa6f76f22008-11-14 10:39:24 +11001209 * install the new credentials for this executable
1210 */
1211void install_exec_creds(struct linux_binprm *bprm)
1212{
1213 security_bprm_committing_creds(bprm);
1214
1215 commit_creds(bprm->cred);
1216 bprm->cred = NULL;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001217 /*
1218 * cred_guard_mutex must be held at least to this point to prevent
David Howellsa6f76f22008-11-14 10:39:24 +11001219 * ptrace_attach() from altering our determination of the task's
Oleg Nesterova2a84742009-09-05 11:17:13 -07001220 * credentials; any time after this it may be unlocked.
1221 */
David Howellsa6f76f22008-11-14 10:39:24 +11001222 security_bprm_committed_creds(bprm);
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001223 mutex_unlock(&current->signal->cred_guard_mutex);
David Howellsa6f76f22008-11-14 10:39:24 +11001224}
1225EXPORT_SYMBOL(install_exec_creds);
1226
1227/*
1228 * determine how safe it is to execute the proposed program
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001229 * - the caller must hold ->cred_guard_mutex to protect against
David Howellsa6f76f22008-11-14 10:39:24 +11001230 * PTRACE_ATTACH
1231 */
Al Viro498052b2009-03-30 07:20:30 -04001232int check_unsafe_exec(struct linux_binprm *bprm)
David Howellsa6f76f22008-11-14 10:39:24 +11001233{
David Howells0bf2f3a2009-02-06 11:45:46 +00001234 struct task_struct *p = current, *t;
Al Virof1191b52009-03-30 07:35:18 -04001235 unsigned n_fs;
Al Viro498052b2009-03-30 07:20:30 -04001236 int res = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001237
Tejun Heo4b9d33e2011-06-17 16:50:38 +02001238 if (p->ptrace) {
1239 if (p->ptrace & PT_PTRACE_CAP)
1240 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1241 else
1242 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1243 }
David Howellsa6f76f22008-11-14 10:39:24 +11001244
David Howells0bf2f3a2009-02-06 11:45:46 +00001245 n_fs = 1;
Nick Piggin2a4419b2010-08-18 04:37:33 +10001246 spin_lock(&p->fs->lock);
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001247 rcu_read_lock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001248 for (t = next_thread(p); t != p; t = next_thread(t)) {
1249 if (t->fs == p->fs)
1250 n_fs++;
David Howells0bf2f3a2009-02-06 11:45:46 +00001251 }
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001252 rcu_read_unlock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001253
Al Virof1191b52009-03-30 07:35:18 -04001254 if (p->fs->users > n_fs) {
David Howellsa6f76f22008-11-14 10:39:24 +11001255 bprm->unsafe |= LSM_UNSAFE_SHARE;
Al Viro498052b2009-03-30 07:20:30 -04001256 } else {
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001257 res = -EAGAIN;
1258 if (!p->fs->in_exec) {
1259 p->fs->in_exec = 1;
1260 res = 1;
1261 }
Al Viro498052b2009-03-30 07:20:30 -04001262 }
Nick Piggin2a4419b2010-08-18 04:37:33 +10001263 spin_unlock(&p->fs->lock);
Al Viro498052b2009-03-30 07:20:30 -04001264
1265 return res;
David Howellsa6f76f22008-11-14 10:39:24 +11001266}
1267
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268/*
1269 * Fill the binprm structure from the inode.
1270 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
David Howellsa6f76f22008-11-14 10:39:24 +11001271 *
1272 * This may be called multiple times for binary chains (scripts for example).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 */
1274int prepare_binprm(struct linux_binprm *bprm)
1275{
David Howellsa6f76f22008-11-14 10:39:24 +11001276 umode_t mode;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001277 struct inode * inode = bprm->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 int retval;
1279
1280 mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 if (bprm->file->f_op == NULL)
1282 return -EACCES;
1283
David Howellsa6f76f22008-11-14 10:39:24 +11001284 /* clear any previous set[ug]id data from a previous binary */
1285 bprm->cred->euid = current_euid();
1286 bprm->cred->egid = current_egid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
David Howellsa6f76f22008-11-14 10:39:24 +11001288 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 /* Set-uid? */
1290 if (mode & S_ISUID) {
David Howellsa6f76f22008-11-14 10:39:24 +11001291 bprm->per_clear |= PER_CLEAR_ON_SETID;
1292 bprm->cred->euid = inode->i_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 }
1294
1295 /* Set-gid? */
1296 /*
1297 * If setgid is set but no group execute bit then this
1298 * is a candidate for mandatory locking, not a setgid
1299 * executable.
1300 */
1301 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
David Howellsa6f76f22008-11-14 10:39:24 +11001302 bprm->per_clear |= PER_CLEAR_ON_SETID;
1303 bprm->cred->egid = inode->i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 }
1305 }
1306
1307 /* fill in binprm security blob */
David Howellsa6f76f22008-11-14 10:39:24 +11001308 retval = security_bprm_set_creds(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 if (retval)
1310 return retval;
David Howellsa6f76f22008-11-14 10:39:24 +11001311 bprm->cred_prepared = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
David Howellsa6f76f22008-11-14 10:39:24 +11001313 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1314 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315}
1316
1317EXPORT_SYMBOL(prepare_binprm);
1318
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001319/*
1320 * Arguments are '\0' separated strings found at the location bprm->p
1321 * points to; chop off the first by relocating brpm->p to right after
1322 * the first '\0' encountered.
1323 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001324int remove_arg_zero(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325{
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001326 int ret = 0;
1327 unsigned long offset;
1328 char *kaddr;
1329 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001331 if (!bprm->argc)
1332 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001334 do {
1335 offset = bprm->p & ~PAGE_MASK;
1336 page = get_arg_page(bprm, bprm->p, 0);
1337 if (!page) {
1338 ret = -EFAULT;
1339 goto out;
1340 }
1341 kaddr = kmap_atomic(page, KM_USER0);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001342
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001343 for (; offset < PAGE_SIZE && kaddr[offset];
1344 offset++, bprm->p++)
1345 ;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001346
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001347 kunmap_atomic(kaddr, KM_USER0);
1348 put_arg_page(page);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001349
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001350 if (offset == PAGE_SIZE)
1351 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1352 } while (offset == PAGE_SIZE);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001353
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001354 bprm->p++;
1355 bprm->argc--;
1356 ret = 0;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001357
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001358out:
1359 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361EXPORT_SYMBOL(remove_arg_zero);
1362
1363/*
1364 * cycle the list of binary formats handler, until one recognizes the image
1365 */
1366int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1367{
Roland McGrath85f33462008-12-09 19:36:38 -08001368 unsigned int depth = bprm->recursion_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 int try,retval;
1370 struct linux_binfmt *fmt;
Denys Vlasenkobb188d72011-06-29 04:13:39 +02001371 pid_t old_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 retval = security_bprm_check(bprm);
1374 if (retval)
1375 return retval;
1376
Al Viro473ae302006-04-26 14:04:08 -04001377 retval = audit_bprm(bprm);
1378 if (retval)
1379 return retval;
1380
Denys Vlasenkobb188d72011-06-29 04:13:39 +02001381 /* Need to fetch pid before load_binary changes it */
1382 rcu_read_lock();
1383 old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1384 rcu_read_unlock();
1385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 retval = -ENOENT;
1387 for (try=0; try<2; try++) {
1388 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -07001389 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1391 if (!fn)
1392 continue;
1393 if (!try_module_get(fmt->module))
1394 continue;
1395 read_unlock(&binfmt_lock);
1396 retval = fn(bprm, regs);
Roland McGrath85f33462008-12-09 19:36:38 -08001397 /*
1398 * Restore the depth counter to its starting value
1399 * in this call, so we don't have to rely on every
1400 * load_binary function to restore it on return.
1401 */
1402 bprm->recursion_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 if (retval >= 0) {
Roland McGrath85f33462008-12-09 19:36:38 -08001404 if (depth == 0)
Denys Vlasenkobb188d72011-06-29 04:13:39 +02001405 ptrace_event(PTRACE_EVENT_EXEC,
1406 old_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 put_binfmt(fmt);
1408 allow_write_access(bprm->file);
1409 if (bprm->file)
1410 fput(bprm->file);
1411 bprm->file = NULL;
1412 current->did_exec = 1;
Matt Helsley9f460802005-11-07 00:59:16 -08001413 proc_exec_connector(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 return retval;
1415 }
1416 read_lock(&binfmt_lock);
1417 put_binfmt(fmt);
1418 if (retval != -ENOEXEC || bprm->mm == NULL)
1419 break;
1420 if (!bprm->file) {
1421 read_unlock(&binfmt_lock);
1422 return retval;
1423 }
1424 }
1425 read_unlock(&binfmt_lock);
Tetsuo Handab4edf8b2011-07-26 16:08:42 -07001426#ifdef CONFIG_MODULES
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 if (retval != -ENOEXEC || bprm->mm == NULL) {
1428 break;
Johannes Berg5f4123b2008-07-09 10:28:40 +02001429 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1431 if (printable(bprm->buf[0]) &&
1432 printable(bprm->buf[1]) &&
1433 printable(bprm->buf[2]) &&
1434 printable(bprm->buf[3]))
1435 break; /* -ENOEXEC */
Tetsuo Handa91219352011-07-26 16:08:41 -07001436 if (try)
1437 break; /* -ENOEXEC */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 }
Tetsuo Handab4edf8b2011-07-26 16:08:42 -07001440#else
1441 break;
1442#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 }
1444 return retval;
1445}
1446
1447EXPORT_SYMBOL(search_binary_handler);
1448
1449/*
1450 * sys_execve() executes a new program.
1451 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001452static int do_execve_common(const char *filename,
1453 struct user_arg_ptr argv,
1454 struct user_arg_ptr envp,
1455 struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456{
1457 struct linux_binprm *bprm;
1458 struct file *file;
Al Viro3b125382008-04-22 05:31:30 -04001459 struct files_struct *displaced;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001460 bool clear_in_exec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Al Viro3b125382008-04-22 05:31:30 -04001463 retval = unshare_files(&displaced);
Al Virofd8328b2008-04-22 05:11:59 -04001464 if (retval)
1465 goto out_ret;
1466
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 retval = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001468 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 if (!bprm)
Al Virofd8328b2008-04-22 05:11:59 -04001470 goto out_files;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
Oleg Nesterova2a84742009-09-05 11:17:13 -07001472 retval = prepare_bprm_creds(bprm);
1473 if (retval)
David Howellsa6f76f22008-11-14 10:39:24 +11001474 goto out_free;
Al Viro498052b2009-03-30 07:20:30 -04001475
1476 retval = check_unsafe_exec(bprm);
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001477 if (retval < 0)
Oleg Nesterova2a84742009-09-05 11:17:13 -07001478 goto out_free;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001479 clear_in_exec = retval;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001480 current->in_execve = 1;
David Howellsa6f76f22008-11-14 10:39:24 +11001481
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 file = open_exec(filename);
1483 retval = PTR_ERR(file);
1484 if (IS_ERR(file))
Al Viro498052b2009-03-30 07:20:30 -04001485 goto out_unmark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486
1487 sched_exec();
1488
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 bprm->file = file;
1490 bprm->filename = filename;
1491 bprm->interp = filename;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001492
1493 retval = bprm_mm_init(bprm);
1494 if (retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 goto out_file;
1496
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001497 bprm->argc = count(argv, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 if ((retval = bprm->argc) < 0)
David Howellsa6f76f22008-11-14 10:39:24 +11001499 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001501 bprm->envc = count(envp, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 if ((retval = bprm->envc) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 goto out;
1504
1505 retval = prepare_binprm(bprm);
1506 if (retval < 0)
1507 goto out;
1508
1509 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1510 if (retval < 0)
1511 goto out;
1512
1513 bprm->exec = bprm->p;
1514 retval = copy_strings(bprm->envc, envp, bprm);
1515 if (retval < 0)
1516 goto out;
1517
1518 retval = copy_strings(bprm->argc, argv, bprm);
1519 if (retval < 0)
1520 goto out;
1521
1522 retval = search_binary_handler(bprm,regs);
David Howellsa6f76f22008-11-14 10:39:24 +11001523 if (retval < 0)
1524 goto out;
1525
1526 /* execve succeeded */
Al Viro498052b2009-03-30 07:20:30 -04001527 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001528 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001529 acct_update_integrals(current);
1530 free_bprm(bprm);
1531 if (displaced)
1532 put_files_struct(displaced);
1533 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
1535out:
Oleg Nesterov3c77f842010-11-30 20:55:34 +01001536 if (bprm->mm) {
1537 acct_arg_size(bprm, 0);
1538 mmput(bprm->mm);
1539 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
1541out_file:
1542 if (bprm->file) {
1543 allow_write_access(bprm->file);
1544 fput(bprm->file);
1545 }
David Howellsa6f76f22008-11-14 10:39:24 +11001546
Al Viro498052b2009-03-30 07:20:30 -04001547out_unmark:
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001548 if (clear_in_exec)
1549 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001550 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001551
1552out_free:
Al Viro08a6fac2008-05-10 16:38:25 -04001553 free_bprm(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
Al Virofd8328b2008-04-22 05:11:59 -04001555out_files:
Al Viro3b125382008-04-22 05:31:30 -04001556 if (displaced)
1557 reset_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558out_ret:
1559 return retval;
1560}
1561
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001562int do_execve(const char *filename,
1563 const char __user *const __user *__argv,
1564 const char __user *const __user *__envp,
1565 struct pt_regs *regs)
1566{
Oleg Nesterov0e028462011-03-06 18:02:54 +01001567 struct user_arg_ptr argv = { .ptr.native = __argv };
1568 struct user_arg_ptr envp = { .ptr.native = __envp };
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001569 return do_execve_common(filename, argv, envp, regs);
1570}
1571
Oleg Nesterov0e028462011-03-06 18:02:54 +01001572#ifdef CONFIG_COMPAT
1573int compat_do_execve(char *filename,
1574 compat_uptr_t __user *__argv,
1575 compat_uptr_t __user *__envp,
1576 struct pt_regs *regs)
1577{
1578 struct user_arg_ptr argv = {
1579 .is_compat = true,
1580 .ptr.compat = __argv,
1581 };
1582 struct user_arg_ptr envp = {
1583 .is_compat = true,
1584 .ptr.compat = __envp,
1585 };
1586 return do_execve_common(filename, argv, envp, regs);
1587}
1588#endif
1589
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001590void set_binfmt(struct linux_binfmt *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591{
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001592 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001594 if (mm->binfmt)
1595 module_put(mm->binfmt->module);
1596
1597 mm->binfmt = new;
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001598 if (new)
1599 __module_get(new->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600}
1601
1602EXPORT_SYMBOL(set_binfmt);
1603
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001604static int expand_corename(struct core_name *cn)
1605{
1606 char *old_corename = cn->corename;
1607
1608 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1609 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1610
1611 if (!cn->corename) {
1612 kfree(old_corename);
1613 return -ENOMEM;
1614 }
1615
1616 return 0;
1617}
1618
1619static int cn_printf(struct core_name *cn, const char *fmt, ...)
1620{
1621 char *cur;
1622 int need;
1623 int ret;
1624 va_list arg;
1625
1626 va_start(arg, fmt);
1627 need = vsnprintf(NULL, 0, fmt, arg);
1628 va_end(arg);
1629
1630 if (likely(need < cn->size - cn->used - 1))
1631 goto out_printf;
1632
1633 ret = expand_corename(cn);
1634 if (ret)
1635 goto expand_fail;
1636
1637out_printf:
1638 cur = cn->corename + cn->used;
1639 va_start(arg, fmt);
1640 vsnprintf(cur, need + 1, fmt, arg);
1641 va_end(arg);
1642 cn->used += need;
1643 return 0;
1644
1645expand_fail:
1646 return ret;
1647}
1648
Jiri Slaby2c563732011-07-26 16:08:33 -07001649static void cn_escape(char *str)
1650{
1651 for (; *str; str++)
1652 if (*str == '/')
1653 *str = '!';
1654}
1655
Jiri Slaby57cc0832011-05-26 16:25:46 -07001656static int cn_print_exe_file(struct core_name *cn)
1657{
1658 struct file *exe_file;
Jiri Slaby2c563732011-07-26 16:08:33 -07001659 char *pathbuf, *path;
Jiri Slaby57cc0832011-05-26 16:25:46 -07001660 int ret;
1661
1662 exe_file = get_mm_exe_file(current->mm);
Jiri Slaby2c563732011-07-26 16:08:33 -07001663 if (!exe_file) {
1664 char *commstart = cn->corename + cn->used;
1665 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1666 cn_escape(commstart);
1667 return ret;
1668 }
Jiri Slaby57cc0832011-05-26 16:25:46 -07001669
1670 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1671 if (!pathbuf) {
1672 ret = -ENOMEM;
1673 goto put_exe_file;
1674 }
1675
1676 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1677 if (IS_ERR(path)) {
1678 ret = PTR_ERR(path);
1679 goto free_buf;
1680 }
1681
Jiri Slaby2c563732011-07-26 16:08:33 -07001682 cn_escape(path);
Jiri Slaby57cc0832011-05-26 16:25:46 -07001683
1684 ret = cn_printf(cn, "%s", path);
1685
1686free_buf:
1687 kfree(pathbuf);
1688put_exe_file:
1689 fput(exe_file);
1690 return ret;
1691}
1692
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693/* format_corename will inspect the pattern parameter, and output a
1694 * name into corename, which must have space for at least
1695 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1696 */
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001697static int format_corename(struct core_name *cn, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698{
David Howells86a264a2008-11-14 10:39:18 +11001699 const struct cred *cred = current_cred();
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001700 const char *pat_ptr = core_pattern;
1701 int ispipe = (*pat_ptr == '|');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 int pid_in_pattern = 0;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001703 int err = 0;
1704
1705 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1706 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1707 cn->used = 0;
1708
1709 if (!cn->corename)
1710 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
1712 /* Repeat as long as we have more pattern to process and more output
1713 space */
1714 while (*pat_ptr) {
1715 if (*pat_ptr != '%') {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001716 if (*pat_ptr == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 goto out;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001718 err = cn_printf(cn, "%c", *pat_ptr++);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 } else {
1720 switch (*++pat_ptr) {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001721 /* single % at the end, drop that */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 case 0:
1723 goto out;
1724 /* Double percent, output one percent */
1725 case '%':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001726 err = cn_printf(cn, "%c", '%');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 break;
1728 /* pid */
1729 case 'p':
1730 pid_in_pattern = 1;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001731 err = cn_printf(cn, "%d",
1732 task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 break;
1734 /* uid */
1735 case 'u':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001736 err = cn_printf(cn, "%d", cred->uid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 break;
1738 /* gid */
1739 case 'g':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001740 err = cn_printf(cn, "%d", cred->gid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 break;
1742 /* signal that caused the coredump */
1743 case 's':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001744 err = cn_printf(cn, "%ld", signr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 break;
1746 /* UNIX time of coredump */
1747 case 't': {
1748 struct timeval tv;
1749 do_gettimeofday(&tv);
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001750 err = cn_printf(cn, "%lu", tv.tv_sec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 break;
1752 }
1753 /* hostname */
Jiri Slaby2c563732011-07-26 16:08:33 -07001754 case 'h': {
1755 char *namestart = cn->corename + cn->used;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 down_read(&uts_sem);
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001757 err = cn_printf(cn, "%s",
1758 utsname()->nodename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 up_read(&uts_sem);
Jiri Slaby2c563732011-07-26 16:08:33 -07001760 cn_escape(namestart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 break;
Jiri Slaby2c563732011-07-26 16:08:33 -07001762 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 /* executable */
Jiri Slaby2c563732011-07-26 16:08:33 -07001764 case 'e': {
1765 char *commstart = cn->corename + cn->used;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001766 err = cn_printf(cn, "%s", current->comm);
Jiri Slaby2c563732011-07-26 16:08:33 -07001767 cn_escape(commstart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 break;
Jiri Slaby2c563732011-07-26 16:08:33 -07001769 }
Jiri Slaby57cc0832011-05-26 16:25:46 -07001770 case 'E':
1771 err = cn_print_exe_file(cn);
1772 break;
Neil Horman74aadce2007-10-16 23:26:35 -07001773 /* core limit size */
1774 case 'c':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001775 err = cn_printf(cn, "%lu",
1776 rlimit(RLIMIT_CORE));
Neil Horman74aadce2007-10-16 23:26:35 -07001777 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 default:
1779 break;
1780 }
1781 ++pat_ptr;
1782 }
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001783
1784 if (err)
1785 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 }
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001787
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 /* Backward compatibility with core_uses_pid:
1789 *
1790 * If core_pattern does not include a %p (as is the default)
1791 * and core_uses_pid is set, then .%pid will be appended to
Alan Coxc4bbafd2007-04-16 22:53:13 -07001792 * the filename. Do not do this for piped commands. */
Oleg Nesterov64093242008-10-18 20:28:22 -07001793 if (!ispipe && !pid_in_pattern && core_uses_pid) {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001794 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1795 if (err)
1796 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 }
Alan Coxc4bbafd2007-04-16 22:53:13 -07001798out:
Alan Coxc4bbafd2007-04-16 22:53:13 -07001799 return ispipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800}
1801
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001802static int zap_process(struct task_struct *start, int exit_code)
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001803{
1804 struct task_struct *t;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001805 int nr = 0;
Oleg Nesterov281de332006-06-26 00:26:06 -07001806
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001807 start->signal->flags = SIGNAL_GROUP_EXIT;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001808 start->signal->group_exit_code = exit_code;
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001809 start->signal->group_stop_count = 0;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001810
1811 t = start;
1812 do {
Tejun Heo6dfca322011-06-02 11:14:00 +02001813 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001814 if (t != current && t->mm) {
Oleg Nesterov281de332006-06-26 00:26:06 -07001815 sigaddset(&t->pending.signal, SIGKILL);
1816 signal_wake_up(t, 1);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001817 nr++;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001818 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001819 } while_each_thread(start, t);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001820
1821 return nr;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001822}
1823
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001824static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001825 struct core_state *core_state, int exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826{
1827 struct task_struct *g, *p;
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001828 unsigned long flags;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001829 int nr = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001831 spin_lock_irq(&tsk->sighand->siglock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001832 if (!signal_group_exit(tsk->signal)) {
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001833 mm->core_state = core_state;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001834 nr = zap_process(tsk, exit_code);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 }
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001836 spin_unlock_irq(&tsk->sighand->siglock);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001837 if (unlikely(nr < 0))
1838 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001840 if (atomic_read(&mm->mm_users) == nr + 1)
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001841 goto done;
Oleg Nesterove4901f92008-07-25 01:47:31 -07001842 /*
1843 * We should find and kill all tasks which use this mm, and we should
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001844 * count them correctly into ->nr_threads. We don't take tasklist
Oleg Nesterove4901f92008-07-25 01:47:31 -07001845 * lock, but this is safe wrt:
1846 *
1847 * fork:
1848 * None of sub-threads can fork after zap_process(leader). All
1849 * processes which were created before this point should be
1850 * visible to zap_threads() because copy_process() adds the new
1851 * process to the tail of init_task.tasks list, and lock/unlock
1852 * of ->siglock provides a memory barrier.
1853 *
1854 * do_exit:
1855 * The caller holds mm->mmap_sem. This means that the task which
1856 * uses this mm can't pass exit_mm(), so it can't exit or clear
1857 * its ->mm.
1858 *
1859 * de_thread:
1860 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1861 * we must see either old or new leader, this does not matter.
1862 * However, it can change p->sighand, so lock_task_sighand(p)
1863 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1864 * it can't fail.
1865 *
1866 * Note also that "g" can be the old leader with ->mm == NULL
1867 * and already unhashed and thus removed from ->thread_group.
1868 * This is OK, __unhash_process()->list_del_rcu() does not
1869 * clear the ->next pointer, we will find the new leader via
1870 * next_thread().
1871 */
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001872 rcu_read_lock();
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001873 for_each_process(g) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001874 if (g == tsk->group_leader)
1875 continue;
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001876 if (g->flags & PF_KTHREAD)
1877 continue;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001878 p = g;
1879 do {
1880 if (p->mm) {
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001881 if (unlikely(p->mm == mm)) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001882 lock_task_sighand(p, &flags);
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001883 nr += zap_process(p, exit_code);
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001884 unlock_task_sighand(p, &flags);
1885 }
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001886 break;
1887 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001888 } while_each_thread(g, p);
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001889 }
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001890 rcu_read_unlock();
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001891done:
Oleg Nesterovc5f1cc82008-07-25 01:47:42 -07001892 atomic_set(&core_state->nr_threads, nr);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001893 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894}
1895
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001896static int coredump_wait(int exit_code, struct core_state *core_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897{
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001898 struct task_struct *tsk = current;
1899 struct mm_struct *mm = tsk->mm;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001900 struct completion *vfork_done;
Oleg Nesterov269b0052010-05-26 14:43:08 -07001901 int core_waiters = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001903 init_completion(&core_state->startup);
Oleg Nesterovb564daf2008-07-25 01:47:44 -07001904 core_state->dumper.task = tsk;
1905 core_state->dumper.next = NULL;
Oleg Nesterov269b0052010-05-26 14:43:08 -07001906
1907 down_write(&mm->mmap_sem);
1908 if (!mm->core_state)
1909 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001910 up_write(&mm->mmap_sem);
1911
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001912 if (unlikely(core_waiters < 0))
1913 goto fail;
1914
1915 /*
1916 * Make sure nobody is waiting for us to release the VM,
1917 * otherwise we can deadlock when we wait on each other
1918 */
1919 vfork_done = tsk->vfork_done;
1920 if (vfork_done) {
1921 tsk->vfork_done = NULL;
1922 complete(vfork_done);
1923 }
1924
Oleg Nesterov2384f552005-10-30 15:02:47 -08001925 if (core_waiters)
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001926 wait_for_completion(&core_state->startup);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001927fail:
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001928 return core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929}
1930
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001931static void coredump_finish(struct mm_struct *mm)
1932{
1933 struct core_thread *curr, *next;
1934 struct task_struct *task;
1935
1936 next = mm->core_state->dumper.next;
1937 while ((curr = next) != NULL) {
1938 next = curr->next;
1939 task = curr->task;
1940 /*
1941 * see exit_mm(), curr->task must not see
1942 * ->task == NULL before we read ->next.
1943 */
1944 smp_mb();
1945 curr->task = NULL;
1946 wake_up_process(task);
1947 }
1948
1949 mm->core_state = NULL;
1950}
1951
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001952/*
1953 * set_dumpable converts traditional three-value dumpable to two flags and
1954 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1955 * these bits are not changed atomically. So get_dumpable can observe the
1956 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1957 * return either old dumpable or new one by paying attention to the order of
1958 * modifying the bits.
1959 *
1960 * dumpable | mm->flags (binary)
1961 * old new | initial interim final
1962 * ---------+-----------------------
1963 * 0 1 | 00 01 01
1964 * 0 2 | 00 10(*) 11
1965 * 1 0 | 01 00 00
1966 * 1 2 | 01 11 11
1967 * 2 0 | 11 10(*) 00
1968 * 2 1 | 11 11 01
1969 *
1970 * (*) get_dumpable regards interim value of 10 as 11.
1971 */
1972void set_dumpable(struct mm_struct *mm, int value)
1973{
1974 switch (value) {
1975 case 0:
1976 clear_bit(MMF_DUMPABLE, &mm->flags);
1977 smp_wmb();
1978 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1979 break;
1980 case 1:
1981 set_bit(MMF_DUMPABLE, &mm->flags);
1982 smp_wmb();
1983 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1984 break;
1985 case 2:
1986 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1987 smp_wmb();
1988 set_bit(MMF_DUMPABLE, &mm->flags);
1989 break;
1990 }
1991}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001992
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001993static int __get_dumpable(unsigned long mm_flags)
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001994{
1995 int ret;
1996
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001997 ret = mm_flags & MMF_DUMPABLE_MASK;
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001998 return (ret >= 2) ? 2 : ret;
1999}
2000
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002001int get_dumpable(struct mm_struct *mm)
2002{
2003 return __get_dumpable(mm->flags);
2004}
2005
Neil Horman61be2282009-09-23 15:56:58 -07002006static void wait_for_dump_helpers(struct file *file)
2007{
2008 struct pipe_inode_info *pipe;
2009
2010 pipe = file->f_path.dentry->d_inode->i_pipe;
2011
2012 pipe_lock(pipe);
2013 pipe->readers++;
2014 pipe->writers--;
2015
2016 while ((pipe->readers > 1) && (!signal_pending(current))) {
2017 wake_up_interruptible_sync(&pipe->wait);
2018 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2019 pipe_wait(pipe);
2020 }
2021
2022 pipe->readers--;
2023 pipe->writers++;
2024 pipe_unlock(pipe);
2025
2026}
2027
2028
Neil Horman898b3742010-05-26 14:42:59 -07002029/*
Holger Hans Peter Freyther1bef8292011-02-24 17:46:49 +01002030 * umh_pipe_setup
Neil Horman898b3742010-05-26 14:42:59 -07002031 * helper function to customize the process used
2032 * to collect the core in userspace. Specifically
2033 * it sets up a pipe and installs it as fd 0 (stdin)
2034 * for the process. Returns 0 on success, or
2035 * PTR_ERR on failure.
2036 * Note that it also sets the core limit to 1. This
2037 * is a special value that we use to trap recursive
2038 * core dumps
2039 */
David Howells87966992011-06-17 11:25:59 +01002040static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
Neil Horman898b3742010-05-26 14:42:59 -07002041{
2042 struct file *rp, *wp;
2043 struct fdtable *fdt;
2044 struct coredump_params *cp = (struct coredump_params *)info->data;
2045 struct files_struct *cf = current->files;
2046
2047 wp = create_write_pipe(0);
2048 if (IS_ERR(wp))
2049 return PTR_ERR(wp);
2050
2051 rp = create_read_pipe(wp, 0);
2052 if (IS_ERR(rp)) {
2053 free_write_pipe(wp);
2054 return PTR_ERR(rp);
2055 }
2056
2057 cp->file = wp;
2058
2059 sys_close(0);
2060 fd_install(0, rp);
2061 spin_lock(&cf->file_lock);
2062 fdt = files_fdtable(cf);
2063 FD_SET(0, fdt->open_fds);
2064 FD_CLR(0, fdt->close_on_exec);
2065 spin_unlock(&cf->file_lock);
2066
2067 /* and disallow core files too */
2068 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2069
2070 return 0;
2071}
2072
WANG Cong8cd3ac32009-01-06 14:42:48 -08002073void do_coredump(long signr, int exit_code, struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074{
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07002075 struct core_state core_state;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002076 struct core_name cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 struct mm_struct *mm = current->mm;
2078 struct linux_binfmt * binfmt;
David Howellsd84f4f92008-11-14 10:39:23 +11002079 const struct cred *old_cred;
2080 struct cred *cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 int retval = 0;
Alan Coxd6e71142005-06-23 00:09:43 -07002082 int flag = 0;
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002083 int ispipe;
Neil Hormana2939802009-09-23 15:56:56 -07002084 static atomic_t core_dump_count = ATOMIC_INIT(0);
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002085 struct coredump_params cprm = {
2086 .signr = signr,
2087 .regs = regs,
Jiri Slabyd554ed892010-03-05 13:42:42 -08002088 .limit = rlimit(RLIMIT_CORE),
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002089 /*
2090 * We must use the same mm->flags while dumping core to avoid
2091 * inconsistency of bit flags, since this flag is not protected
2092 * by any locks.
2093 */
2094 .mm_flags = mm->flags,
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002095 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096
Steve Grubb0a4ff8c2007-04-19 10:28:21 -04002097 audit_core_dumps(signr);
2098
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07002099 binfmt = mm->binfmt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 if (!binfmt || !binfmt->core_dump)
2101 goto fail;
Oleg Nesterov269b0052010-05-26 14:43:08 -07002102 if (!__get_dumpable(cprm.mm_flags))
2103 goto fail;
David Howellsd84f4f92008-11-14 10:39:23 +11002104
2105 cred = prepare_creds();
Oleg Nesterov5e43aef2010-05-26 14:43:07 -07002106 if (!cred)
David Howellsd84f4f92008-11-14 10:39:23 +11002107 goto fail;
Alan Coxd6e71142005-06-23 00:09:43 -07002108 /*
2109 * We cannot trust fsuid as being the "true" uid of the
2110 * process nor do we know its entire history. We only know it
2111 * was tainted so we dump it as root in mode 2.
2112 */
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002113 if (__get_dumpable(cprm.mm_flags) == 2) {
2114 /* Setuid core dump mode */
Alan Coxd6e71142005-06-23 00:09:43 -07002115 flag = O_EXCL; /* Stop rewrite attacks */
David Howellsd84f4f92008-11-14 10:39:23 +11002116 cred->fsuid = 0; /* Dump root private */
Alan Coxd6e71142005-06-23 00:09:43 -07002117 }
Oleg Nesterov1291cf42005-10-30 15:02:54 -08002118
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07002119 retval = coredump_wait(exit_code, &core_state);
Oleg Nesterov5e43aef2010-05-26 14:43:07 -07002120 if (retval < 0)
2121 goto fail_creds;
David Howellsd84f4f92008-11-14 10:39:23 +11002122
2123 old_cred = override_creds(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
2125 /*
2126 * Clear any false indication of pending signals that might
2127 * be seen by the filesystem code called to write the core file.
2128 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 clear_thread_flag(TIF_SIGPENDING);
2130
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002131 ispipe = format_corename(&cn, signr);
2132
Alan Coxc4bbafd2007-04-16 22:53:13 -07002133 if (ispipe) {
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002134 int dump_count;
2135 char **helper_argv;
2136
Oleg Nesterov99b64562011-07-26 16:08:34 -07002137 if (ispipe < 0) {
2138 printk(KERN_WARNING "format_corename failed\n");
2139 printk(KERN_WARNING "Aborting core\n");
2140 goto fail_corename;
2141 }
2142
Neil Horman898b3742010-05-26 14:42:59 -07002143 if (cprm.limit == 1) {
Neil Horman725eae32009-09-23 15:56:54 -07002144 /*
2145 * Normally core limits are irrelevant to pipes, since
2146 * we're not writing to the file system, but we use
Neil Horman898b3742010-05-26 14:42:59 -07002147 * cprm.limit of 1 here as a speacial value. Any
2148 * non-1 limit gets set to RLIM_INFINITY below, but
Neil Horman725eae32009-09-23 15:56:54 -07002149 * a limit of 0 skips the dump. This is a consistent
2150 * way to catch recursive crashes. We can still crash
Neil Horman898b3742010-05-26 14:42:59 -07002151 * if the core_pattern binary sets RLIM_CORE = !1
Neil Horman725eae32009-09-23 15:56:54 -07002152 * but it runs as root, and can do lots of stupid things
2153 * Note that we use task_tgid_vnr here to grab the pid
2154 * of the process group leader. That way we get the
2155 * right pid if a thread in a multi-threaded
2156 * core_pattern process dies.
2157 */
2158 printk(KERN_WARNING
Neil Horman898b3742010-05-26 14:42:59 -07002159 "Process %d(%s) has RLIMIT_CORE set to 1\n",
Neil Horman725eae32009-09-23 15:56:54 -07002160 task_tgid_vnr(current), current->comm);
2161 printk(KERN_WARNING "Aborting core\n");
2162 goto fail_unlock;
2163 }
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002164 cprm.limit = RLIM_INFINITY;
Neil Horman725eae32009-09-23 15:56:54 -07002165
Neil Hormana2939802009-09-23 15:56:56 -07002166 dump_count = atomic_inc_return(&core_dump_count);
2167 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2168 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2169 task_tgid_vnr(current), current->comm);
2170 printk(KERN_WARNING "Skipping core dump\n");
2171 goto fail_dropcount;
2172 }
2173
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002174 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
Tetsuo Handa350eaf72009-01-06 14:41:11 -08002175 if (!helper_argv) {
2176 printk(KERN_WARNING "%s failed to allocate memory\n",
2177 __func__);
Neil Hormana2939802009-09-23 15:56:56 -07002178 goto fail_dropcount;
Tetsuo Handa350eaf72009-01-06 14:41:11 -08002179 }
Neil Horman32321132007-10-16 23:26:36 -07002180
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002181 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2182 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2183 NULL, &cprm);
2184 argv_free(helper_argv);
2185 if (retval) {
Andi Kleend025c9d2006-09-30 23:29:28 -07002186 printk(KERN_INFO "Core dump to %s pipe failed\n",
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002187 cn.corename);
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002188 goto close_fail;
Andi Kleend025c9d2006-09-30 23:29:28 -07002189 }
Oleg Nesterovc7135412010-05-26 14:43:05 -07002190 } else {
2191 struct inode *inode;
2192
2193 if (cprm.limit < binfmt->min_coredump)
2194 goto fail_unlock;
2195
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002196 cprm.file = filp_open(cn.corename,
Alexey Dobriyan6d4df672006-12-06 20:40:39 -08002197 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2198 0600);
Oleg Nesterovc7135412010-05-26 14:43:05 -07002199 if (IS_ERR(cprm.file))
2200 goto fail_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201
Oleg Nesterovc7135412010-05-26 14:43:05 -07002202 inode = cprm.file->f_path.dentry->d_inode;
2203 if (inode->i_nlink > 1)
2204 goto close_fail;
2205 if (d_unhashed(cprm.file->f_path.dentry))
2206 goto close_fail;
2207 /*
2208 * AK: actually i see no reason to not allow this for named
2209 * pipes etc, but keep the previous behaviour for now.
2210 */
2211 if (!S_ISREG(inode->i_mode))
2212 goto close_fail;
2213 /*
2214 * Dont allow local users get cute and trick others to coredump
2215 * into their pre-created files.
2216 */
2217 if (inode->i_uid != current_fsuid())
2218 goto close_fail;
2219 if (!cprm.file->f_op || !cprm.file->f_op->write)
2220 goto close_fail;
2221 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2222 goto close_fail;
2223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002225 retval = binfmt->core_dump(&cprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 if (retval)
2227 current->signal->group_exit_code |= 0x80;
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002228
Neil Horman61be2282009-09-23 15:56:58 -07002229 if (ispipe && core_pipe_limit)
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002230 wait_for_dump_helpers(cprm.file);
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002231close_fail:
2232 if (cprm.file)
2233 filp_close(cprm.file, NULL);
Neil Hormana2939802009-09-23 15:56:56 -07002234fail_dropcount:
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002235 if (ispipe)
Neil Hormana2939802009-09-23 15:56:56 -07002236 atomic_dec(&core_dump_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237fail_unlock:
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002238 kfree(cn.corename);
2239fail_corename:
Oleg Nesterova94e2d42008-07-25 01:47:46 -07002240 coredump_finish(mm);
Oleg Nesterov5e43aef2010-05-26 14:43:07 -07002241 revert_creds(old_cred);
2242fail_creds:
2243 put_cred(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244fail:
WANG Cong8cd3ac32009-01-06 14:42:48 -08002245 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246}
Linus Torvalds3aa0ce82010-10-14 14:32:06 -07002247
2248/*
2249 * Core dumping helper functions. These are the only things you should
2250 * do on a core-file: use only these functions to write out all the
2251 * necessary info.
2252 */
2253int dump_write(struct file *file, const void *addr, int nr)
2254{
2255 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2256}
Linus Torvalds8fd01d62010-10-14 19:15:28 -07002257EXPORT_SYMBOL(dump_write);
Linus Torvalds3aa0ce82010-10-14 14:32:06 -07002258
2259int dump_seek(struct file *file, loff_t off)
2260{
2261 int ret = 1;
2262
2263 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2264 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2265 return 0;
2266 } else {
2267 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2268
2269 if (!buf)
2270 return 0;
2271 while (off > 0) {
2272 unsigned long n = off;
2273
2274 if (n > PAGE_SIZE)
2275 n = PAGE_SIZE;
2276 if (!dump_write(file, buf, n)) {
2277 ret = 0;
2278 break;
2279 }
2280 off -= n;
2281 }
2282 free_page((unsigned long)buf);
2283 }
2284 return ret;
2285}
Linus Torvalds8fd01d62010-10-14 19:15:28 -07002286EXPORT_SYMBOL(dump_seek);