blob: aeb135c7ff5c0c1a6f1a1dfe3b5f2018dfbd6731 [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>
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080062
63#include <trace/events/task.h>
David Howellsa6f76f22008-11-14 10:39:24 +110064#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Linus Torvalds1da177e2005-04-16 15:20:36 -070066int core_uses_pid;
Dan Aloni71ce92f2007-05-16 22:11:16 -070067char core_pattern[CORENAME_MAX_SIZE] = "core";
Neil Hormana2939802009-09-23 15:56:56 -070068unsigned int core_pipe_limit;
Alan Coxd6e71142005-06-23 00:09:43 -070069int suid_dumpable = 0;
70
Xiaotian Feng1b0d3002010-10-27 15:34:08 -070071struct core_name {
72 char *corename;
73 int used, size;
74};
75static atomic_t call_count = ATOMIC_INIT(1);
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/* The maximal length of core_pattern is also specified in sysctl.c */
78
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070079static LIST_HEAD(formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080static DEFINE_RWLOCK(binfmt_lock);
81
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070082int __register_binfmt(struct linux_binfmt * fmt, int insert)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083{
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 if (!fmt)
85 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 write_lock(&binfmt_lock);
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070087 insert ? list_add(&fmt->lh, &formats) :
88 list_add_tail(&fmt->lh, &formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 write_unlock(&binfmt_lock);
90 return 0;
91}
92
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070093EXPORT_SYMBOL(__register_binfmt);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
Alexey Dobriyanf6b450d2007-10-16 23:26:04 -070095void unregister_binfmt(struct linux_binfmt * fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096{
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 write_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070098 list_del(&fmt->lh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 write_unlock(&binfmt_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100}
101
102EXPORT_SYMBOL(unregister_binfmt);
103
104static inline void put_binfmt(struct linux_binfmt * fmt)
105{
106 module_put(fmt->module);
107}
108
109/*
110 * Note that a shared library must be both readable and executable due to
111 * security reasons.
112 *
113 * Also note that we take the address to load from from the file itself.
114 */
Heiko Carstens1e7bfb22009-01-14 14:14:29 +0100115SYSCALL_DEFINE1(uselib, const char __user *, library)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
Al Viro964bd182008-07-26 03:33:14 -0400117 struct file *file;
Al Viro964bd182008-07-26 03:33:14 -0400118 char *tmp = getname(library);
119 int error = PTR_ERR(tmp);
Al Viro47c805d2011-02-23 17:44:09 -0500120 static const struct open_flags uselib_flags = {
121 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
122 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
123 .intent = LOOKUP_OPEN
124 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
Al Viro6e8341a2009-04-06 11:16:22 -0400126 if (IS_ERR(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 goto out;
128
Al Viro47c805d2011-02-23 17:44:09 -0500129 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
Al Viro6e8341a2009-04-06 11:16:22 -0400130 putname(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 error = PTR_ERR(file);
132 if (IS_ERR(file))
133 goto out;
134
Al Viro6e8341a2009-04-06 11:16:22 -0400135 error = -EINVAL;
136 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
137 goto exit;
138
139 error = -EACCES;
140 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
141 goto exit;
142
Eric Paris2a12a9d2009-12-17 21:24:21 -0500143 fsnotify_open(file);
Eric Paris6110e3a2008-12-17 13:53:20 -0500144
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 error = -ENOEXEC;
146 if(file->f_op) {
147 struct linux_binfmt * fmt;
148
149 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -0700150 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 if (!fmt->load_shlib)
152 continue;
153 if (!try_module_get(fmt->module))
154 continue;
155 read_unlock(&binfmt_lock);
156 error = fmt->load_shlib(file);
157 read_lock(&binfmt_lock);
158 put_binfmt(fmt);
159 if (error != -ENOEXEC)
160 break;
161 }
162 read_unlock(&binfmt_lock);
163 }
Al Viro6e8341a2009-04-06 11:16:22 -0400164exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 fput(file);
166out:
167 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168}
169
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700170#ifdef CONFIG_MMU
Oleg Nesterovae6b5852011-03-06 18:03:11 +0100171/*
172 * The nascent bprm->mm is not visible until exec_mmap() but it can
173 * use a lot of memory, account these pages in current->mm temporary
174 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
175 * change the counter back via acct_arg_size(0).
176 */
Oleg Nesterov0e028462011-03-06 18:02:54 +0100177static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100178{
179 struct mm_struct *mm = current->mm;
180 long diff = (long)(pages - bprm->vma_pages);
181
182 if (!mm || !diff)
183 return;
184
185 bprm->vma_pages = pages;
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100186 add_mm_counter(mm, MM_ANONPAGES, diff);
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100187}
188
Oleg Nesterov0e028462011-03-06 18:02:54 +0100189static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700190 int write)
191{
192 struct page *page;
193 int ret;
194
195#ifdef CONFIG_STACK_GROWSUP
196 if (write) {
Michal Hockod05f3162011-05-24 17:11:44 -0700197 ret = expand_downwards(bprm->vma, pos);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700198 if (ret < 0)
199 return NULL;
200 }
201#endif
202 ret = get_user_pages(current, bprm->mm, pos,
203 1, write, 1, &page, NULL);
204 if (ret <= 0)
205 return NULL;
206
207 if (write) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700208 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800209 struct rlimit *rlim;
210
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100211 acct_arg_size(bprm, size / PAGE_SIZE);
212
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800213 /*
214 * We've historically supported up to 32 pages (ARG_MAX)
215 * of argument strings even with small stacks
216 */
217 if (size <= ARG_MAX)
218 return page;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700219
220 /*
221 * Limit to 1/4-th the stack size for the argv+env strings.
222 * This ensures that:
223 * - the remaining binfmt code will not run out of stack space,
224 * - the program will have a reasonable amount of stack left
225 * to work from.
226 */
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800227 rlim = current->signal->rlim;
Jiri Slabyd554ed892010-03-05 13:42:42 -0800228 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700229 put_page(page);
230 return NULL;
231 }
232 }
233
234 return page;
235}
236
237static void put_arg_page(struct page *page)
238{
239 put_page(page);
240}
241
242static void free_arg_page(struct linux_binprm *bprm, int i)
243{
244}
245
246static void free_arg_pages(struct linux_binprm *bprm)
247{
248}
249
250static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
251 struct page *page)
252{
253 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
254}
255
256static int __bprm_mm_init(struct linux_binprm *bprm)
257{
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800258 int err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700259 struct vm_area_struct *vma = NULL;
260 struct mm_struct *mm = bprm->mm;
261
262 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
263 if (!vma)
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800264 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700265
266 down_write(&mm->mmap_sem);
267 vma->vm_mm = mm;
268
269 /*
270 * Place the stack at the largest stack address the architecture
271 * supports. Later, we'll move this to an appropriate place. We don't
272 * use STACK_TOP because that can depend on attributes which aren't
273 * configured yet.
274 */
Michal Hockoaacb3d12011-07-26 16:08:40 -0700275 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700276 vma->vm_end = STACK_TOP_MAX;
277 vma->vm_start = vma->vm_end - PAGE_SIZE;
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700278 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
Coly Li3ed75eb2007-10-18 23:39:15 -0700279 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Rik van Riel5beb4932010-03-05 13:42:07 -0800280 INIT_LIST_HEAD(&vma->anon_vma_chain);
Tavis Ormandy462e635e2010-12-09 15:29:42 +0100281
282 err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
283 if (err)
284 goto err;
285
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700286 err = insert_vm_struct(mm, vma);
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800287 if (err)
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700288 goto err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700289
290 mm->stack_vm = mm->total_vm = 1;
291 up_write(&mm->mmap_sem);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700292 bprm->p = vma->vm_end - sizeof(void *);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700293 return 0;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700294err:
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800295 up_write(&mm->mmap_sem);
296 bprm->vma = NULL;
297 kmem_cache_free(vm_area_cachep, vma);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700298 return err;
299}
300
301static bool valid_arg_len(struct linux_binprm *bprm, long len)
302{
303 return len <= MAX_ARG_STRLEN;
304}
305
306#else
307
Oleg Nesterov0e028462011-03-06 18:02:54 +0100308static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100309{
310}
311
Oleg Nesterov0e028462011-03-06 18:02:54 +0100312static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700313 int write)
314{
315 struct page *page;
316
317 page = bprm->page[pos / PAGE_SIZE];
318 if (!page && write) {
319 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
320 if (!page)
321 return NULL;
322 bprm->page[pos / PAGE_SIZE] = page;
323 }
324
325 return page;
326}
327
328static void put_arg_page(struct page *page)
329{
330}
331
332static void free_arg_page(struct linux_binprm *bprm, int i)
333{
334 if (bprm->page[i]) {
335 __free_page(bprm->page[i]);
336 bprm->page[i] = NULL;
337 }
338}
339
340static void free_arg_pages(struct linux_binprm *bprm)
341{
342 int i;
343
344 for (i = 0; i < MAX_ARG_PAGES; i++)
345 free_arg_page(bprm, i);
346}
347
348static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
349 struct page *page)
350{
351}
352
353static int __bprm_mm_init(struct linux_binprm *bprm)
354{
355 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
356 return 0;
357}
358
359static bool valid_arg_len(struct linux_binprm *bprm, long len)
360{
361 return len <= bprm->p;
362}
363
364#endif /* CONFIG_MMU */
365
366/*
367 * Create a new mm_struct and populate it with a temporary stack
368 * vm_area_struct. We don't have enough context at this point to set the stack
369 * flags, permissions, and offset, so we use temporary values. We'll update
370 * them later in setup_arg_pages().
371 */
372int bprm_mm_init(struct linux_binprm *bprm)
373{
374 int err;
375 struct mm_struct *mm = NULL;
376
377 bprm->mm = mm = mm_alloc();
378 err = -ENOMEM;
379 if (!mm)
380 goto err;
381
382 err = init_new_context(current, mm);
383 if (err)
384 goto err;
385
386 err = __bprm_mm_init(bprm);
387 if (err)
388 goto err;
389
390 return 0;
391
392err:
393 if (mm) {
394 bprm->mm = NULL;
395 mmdrop(mm);
396 }
397
398 return err;
399}
400
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100401struct user_arg_ptr {
Oleg Nesterov0e028462011-03-06 18:02:54 +0100402#ifdef CONFIG_COMPAT
403 bool is_compat;
404#endif
405 union {
406 const char __user *const __user *native;
407#ifdef CONFIG_COMPAT
408 compat_uptr_t __user *compat;
409#endif
410 } ptr;
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100411};
412
413static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100414{
Oleg Nesterov0e028462011-03-06 18:02:54 +0100415 const char __user *native;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100416
Oleg Nesterov0e028462011-03-06 18:02:54 +0100417#ifdef CONFIG_COMPAT
418 if (unlikely(argv.is_compat)) {
419 compat_uptr_t compat;
420
421 if (get_user(compat, argv.ptr.compat + nr))
422 return ERR_PTR(-EFAULT);
423
424 return compat_ptr(compat);
425 }
426#endif
427
428 if (get_user(native, argv.ptr.native + nr))
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100429 return ERR_PTR(-EFAULT);
430
Oleg Nesterov0e028462011-03-06 18:02:54 +0100431 return native;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100432}
433
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434/*
435 * count() counts the number of strings in array ARGV.
436 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100437static int count(struct user_arg_ptr argv, int max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438{
439 int i = 0;
440
Oleg Nesterov0e028462011-03-06 18:02:54 +0100441 if (argv.ptr.native != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 for (;;) {
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100443 const char __user *p = get_user_arg_ptr(argv, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 if (!p)
446 break;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100447
448 if (IS_ERR(p))
449 return -EFAULT;
450
Jason Baron362e6662008-10-15 22:01:52 -0700451 if (i++ >= max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 return -E2BIG;
Roland McGrath9aea5a62010-09-07 19:37:06 -0700453
454 if (fatal_signal_pending(current))
455 return -ERESTARTNOHAND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 cond_resched();
457 }
458 }
459 return i;
460}
461
462/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700463 * 'copy_strings()' copies argument/environment strings from the old
464 * processes's memory to the new process's stack. The call to get_user_pages()
465 * ensures the destination page is created and not swapped out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100467static int copy_strings(int argc, struct user_arg_ptr argv,
Adrian Bunk75c96f82005-05-05 16:16:09 -0700468 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
470 struct page *kmapped_page = NULL;
471 char *kaddr = NULL;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700472 unsigned long kpos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 int ret;
474
475 while (argc-- > 0) {
David Howellsd7627462010-08-17 23:52:56 +0100476 const char __user *str;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 int len;
478 unsigned long pos;
479
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100480 ret = -EFAULT;
481 str = get_user_arg_ptr(argv, argc);
482 if (IS_ERR(str))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100485 len = strnlen_user(str, MAX_ARG_STRLEN);
486 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 goto out;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100488
489 ret = -E2BIG;
490 if (!valid_arg_len(bprm, len))
491 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700493 /* We're going to work our way backwords. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 pos = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700495 str += len;
496 bprm->p -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
498 while (len > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 int offset, bytes_to_copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
Roland McGrath9aea5a62010-09-07 19:37:06 -0700501 if (fatal_signal_pending(current)) {
502 ret = -ERESTARTNOHAND;
503 goto out;
504 }
Roland McGrath7993bc12010-09-07 19:36:28 -0700505 cond_resched();
506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 offset = pos % PAGE_SIZE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700508 if (offset == 0)
509 offset = PAGE_SIZE;
510
511 bytes_to_copy = offset;
512 if (bytes_to_copy > len)
513 bytes_to_copy = len;
514
515 offset -= bytes_to_copy;
516 pos -= bytes_to_copy;
517 str -= bytes_to_copy;
518 len -= bytes_to_copy;
519
520 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
521 struct page *page;
522
523 page = get_arg_page(bprm, pos, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 if (!page) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700525 ret = -E2BIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 goto out;
527 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700529 if (kmapped_page) {
530 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700532 put_arg_page(kmapped_page);
533 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 kmapped_page = page;
535 kaddr = kmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700536 kpos = pos & PAGE_MASK;
537 flush_arg_page(bprm, kpos, kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700539 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 ret = -EFAULT;
541 goto out;
542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 }
544 }
545 ret = 0;
546out:
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700547 if (kmapped_page) {
548 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700550 put_arg_page(kmapped_page);
551 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 return ret;
553}
554
555/*
556 * Like copy_strings, but get argv and its values from kernel memory.
557 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100558int copy_strings_kernel(int argc, const char *const *__argv,
David Howellsd7627462010-08-17 23:52:56 +0100559 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
561 int r;
562 mm_segment_t oldfs = get_fs();
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100563 struct user_arg_ptr argv = {
Oleg Nesterov0e028462011-03-06 18:02:54 +0100564 .ptr.native = (const char __user *const __user *)__argv,
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100565 };
566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 set_fs(KERNEL_DS);
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100568 r = copy_strings(argc, argv, bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 set_fs(oldfs);
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 return r;
572}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573EXPORT_SYMBOL(copy_strings_kernel);
574
575#ifdef CONFIG_MMU
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700576
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700578 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
579 * the binfmt code determines where the new stack should reside, we shift it to
580 * its final location. The process proceeds as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 *
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700582 * 1) Use shift to calculate the new vma endpoints.
583 * 2) Extend vma to cover both the old and new ranges. This ensures the
584 * arguments passed to subsequent functions are consistent.
585 * 3) Move vma's page tables to the new range.
586 * 4) Free up any cleared pgd range.
587 * 5) Shrink the vma to cover only the new range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700589static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590{
591 struct mm_struct *mm = vma->vm_mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700592 unsigned long old_start = vma->vm_start;
593 unsigned long old_end = vma->vm_end;
594 unsigned long length = old_end - old_start;
595 unsigned long new_start = old_start - shift;
596 unsigned long new_end = old_end - shift;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700597 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700599 BUG_ON(new_start > new_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700601 /*
602 * ensure there are no vmas between where we want to go
603 * and where we are
604 */
605 if (vma != find_vma(mm, new_start))
606 return -EFAULT;
607
608 /*
609 * cover the whole range: [new_start, old_end)
610 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800611 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
612 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700613
614 /*
615 * move the page tables downwards, on failure we rely on
616 * process cleanup to remove whatever mess we made.
617 */
618 if (length != move_page_tables(vma, old_start,
619 vma, new_start, length))
620 return -ENOMEM;
621
622 lru_add_drain();
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700623 tlb_gather_mmu(&tlb, mm, 0);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700624 if (new_end > old_start) {
625 /*
626 * when the old and new regions overlap clear from new_end.
627 */
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700628 free_pgd_range(&tlb, new_end, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700629 vma->vm_next ? vma->vm_next->vm_start : 0);
630 } else {
631 /*
632 * otherwise, clean from old_start; this is done to not touch
633 * the address space in [new_end, old_start) some architectures
634 * have constraints on va-space that make this illegal (IA64) -
635 * for the others its just a little faster.
636 */
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700637 free_pgd_range(&tlb, old_start, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700638 vma->vm_next ? vma->vm_next->vm_start : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 }
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700640 tlb_finish_mmu(&tlb, new_end, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700642 /*
Rik van Riel5beb4932010-03-05 13:42:07 -0800643 * Shrink the vma to just the new range. Always succeeds.
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700644 */
645 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
646
647 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648}
649
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700650/*
651 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
652 * the stack is optionally relocated, and some extra space is added.
653 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654int setup_arg_pages(struct linux_binprm *bprm,
655 unsigned long stack_top,
656 int executable_stack)
657{
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700658 unsigned long ret;
659 unsigned long stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 struct mm_struct *mm = current->mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700661 struct vm_area_struct *vma = bprm->vma;
662 struct vm_area_struct *prev = NULL;
663 unsigned long vm_flags;
664 unsigned long stack_base;
Michael Neuling803bf5e2010-02-10 13:56:42 -0800665 unsigned long stack_size;
666 unsigned long stack_expand;
667 unsigned long rlim_stack;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
669#ifdef CONFIG_STACK_GROWSUP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 /* Limit stack size to 1GB */
Jiri Slabyd554ed892010-03-05 13:42:42 -0800671 stack_base = rlimit_max(RLIMIT_STACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 if (stack_base > (1 << 30))
673 stack_base = 1 << 30;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700675 /* Make sure we didn't let the argument array grow too large. */
676 if (vma->vm_end - vma->vm_start > stack_base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 return -ENOMEM;
678
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700679 stack_base = PAGE_ALIGN(stack_top - stack_base);
680
681 stack_shift = vma->vm_start - stack_base;
682 mm->arg_start = bprm->p - stack_shift;
683 bprm->p = vma->vm_end - stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684#else
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700685 stack_top = arch_align_stack(stack_top);
686 stack_top = PAGE_ALIGN(stack_top);
Roland McGrath1b528182010-09-07 19:35:49 -0700687
688 if (unlikely(stack_top < mmap_min_addr) ||
689 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
690 return -ENOMEM;
691
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700692 stack_shift = vma->vm_end - stack_top;
693
694 bprm->p -= stack_shift;
695 mm->arg_start = bprm->p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696#endif
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700697
698 if (bprm->loader)
699 bprm->loader -= stack_shift;
700 bprm->exec -= stack_shift;
701
702 down_write(&mm->mmap_sem);
Hugh Dickins96a8e132008-07-10 21:19:20 +0100703 vm_flags = VM_STACK_FLAGS;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700704
705 /*
706 * Adjust stack execute permissions; explicitly enable for
707 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
708 * (arch default) otherwise.
709 */
710 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
711 vm_flags |= VM_EXEC;
712 else if (executable_stack == EXSTACK_DISABLE_X)
713 vm_flags &= ~VM_EXEC;
714 vm_flags |= mm->def_flags;
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700715 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700716
717 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
718 vm_flags);
719 if (ret)
720 goto out_unlock;
721 BUG_ON(prev != vma);
722
723 /* Move stack pages down in memory. */
724 if (stack_shift) {
725 ret = shift_arg_pages(vma, stack_shift);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800726 if (ret)
727 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
729
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700730 /* mprotect_fixup is overkill to remove the temporary stack flags */
731 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
732
Michael Neuling5ef097d2010-03-05 13:42:57 -0800733 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
Michael Neuling803bf5e2010-02-10 13:56:42 -0800734 stack_size = vma->vm_end - vma->vm_start;
735 /*
736 * Align this down to a page boundary as expand_stack
737 * will align it up.
738 */
739 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700740#ifdef CONFIG_STACK_GROWSUP
Michael Neuling803bf5e2010-02-10 13:56:42 -0800741 if (stack_size + stack_expand > rlim_stack)
742 stack_base = vma->vm_start + rlim_stack;
743 else
744 stack_base = vma->vm_end + stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700745#else
Michael Neuling803bf5e2010-02-10 13:56:42 -0800746 if (stack_size + stack_expand > rlim_stack)
747 stack_base = vma->vm_end - rlim_stack;
748 else
749 stack_base = vma->vm_start - stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700750#endif
Eric B Munson3af9e852010-05-18 15:30:49 +0100751 current->mm->start_stack = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700752 ret = expand_stack(vma, stack_base);
753 if (ret)
754 ret = -EFAULT;
755
756out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 up_write(&mm->mmap_sem);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800758 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760EXPORT_SYMBOL(setup_arg_pages);
761
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762#endif /* CONFIG_MMU */
763
764struct file *open_exec(const char *name)
765{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 struct file *file;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200767 int err;
Al Viro47c805d2011-02-23 17:44:09 -0500768 static const struct open_flags open_exec_flags = {
769 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
770 .acc_mode = MAY_EXEC | MAY_OPEN,
771 .intent = LOOKUP_OPEN
772 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Al Viro47c805d2011-02-23 17:44:09 -0500774 file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
Al Viro6e8341a2009-04-06 11:16:22 -0400775 if (IS_ERR(file))
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200776 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200778 err = -EACCES;
Al Viro6e8341a2009-04-06 11:16:22 -0400779 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
780 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200781
Al Viro6e8341a2009-04-06 11:16:22 -0400782 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
783 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200784
Eric Paris2a12a9d2009-12-17 21:24:21 -0500785 fsnotify_open(file);
Eric Paris6110e3a2008-12-17 13:53:20 -0500786
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200787 err = deny_write_access(file);
Al Viro6e8341a2009-04-06 11:16:22 -0400788 if (err)
789 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Al Viro6e8341a2009-04-06 11:16:22 -0400791out:
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200792 return file;
793
Al Viro6e8341a2009-04-06 11:16:22 -0400794exit:
795 fput(file);
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200796 return ERR_PTR(err);
797}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798EXPORT_SYMBOL(open_exec);
799
Mimi Zohar6777d772009-08-21 14:32:48 -0400800int kernel_read(struct file *file, loff_t offset,
801 char *addr, unsigned long count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
803 mm_segment_t old_fs;
804 loff_t pos = offset;
805 int result;
806
807 old_fs = get_fs();
808 set_fs(get_ds());
809 /* The cast to a user pointer is valid due to the set_fs() */
810 result = vfs_read(file, (void __user *)addr, count, &pos);
811 set_fs(old_fs);
812 return result;
813}
814
815EXPORT_SYMBOL(kernel_read);
816
817static int exec_mmap(struct mm_struct *mm)
818{
819 struct task_struct *tsk;
820 struct mm_struct * old_mm, *active_mm;
821
822 /* Notify parent that we're no longer interested in the old VM */
823 tsk = current;
824 old_mm = current->mm;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800825 sync_mm_rss(tsk, old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 mm_release(tsk, old_mm);
827
828 if (old_mm) {
829 /*
830 * Make sure that if there is a core dump in progress
831 * for the old mm, we get out and die instead of going
832 * through with the exec. We must hold mmap_sem around
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700833 * checking core_state and changing tsk->mm.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 */
835 down_read(&old_mm->mmap_sem);
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700836 if (unlikely(old_mm->core_state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 up_read(&old_mm->mmap_sem);
838 return -EINTR;
839 }
840 }
841 task_lock(tsk);
842 active_mm = tsk->active_mm;
843 tsk->mm = mm;
844 tsk->active_mm = mm;
845 activate_mm(active_mm, mm);
846 task_unlock(tsk);
847 arch_pick_mmap_layout(mm);
848 if (old_mm) {
849 up_read(&old_mm->mmap_sem);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +0200850 BUG_ON(active_mm != old_mm);
Balbir Singh31a78f22008-09-28 23:09:31 +0100851 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 mmput(old_mm);
853 return 0;
854 }
855 mmdrop(active_mm);
856 return 0;
857}
858
859/*
860 * This function makes sure the current process has its own signal table,
861 * so that flush_signal_handlers can later reset the handlers without
862 * disturbing other processes. (Other processes might share the signal
863 * table via the CLONE_SIGHAND option to clone().)
864 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800865static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866{
867 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700868 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 spinlock_t *lock = &oldsighand->siglock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700871 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 goto no_thread_group;
873
874 /*
875 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800878 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 /*
880 * Another group action in progress, just
881 * return so that the signal is processed.
882 */
883 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 return -EAGAIN;
885 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
Oleg Nesterovd3441932010-05-26 14:43:11 -0700887 sig->group_exit_task = tsk;
888 sig->notify_count = zap_other_threads(tsk);
889 if (!thread_group_leader(tsk))
890 sig->notify_count--;
891
892 while (sig->notify_count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 __set_current_state(TASK_UNINTERRUPTIBLE);
894 spin_unlock_irq(lock);
895 schedule();
896 spin_lock_irq(lock);
897 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 spin_unlock_irq(lock);
899
900 /*
901 * At this point all other threads have exited, all we have to
902 * do is to wait for the thread group leader to become inactive,
903 * and to assume its PID:
904 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700905 if (!thread_group_leader(tsk)) {
Oleg Nesterov81879262008-12-01 14:18:16 -0800906 struct task_struct *leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700907
Oleg Nesterov2800d8d2008-04-30 00:53:12 -0700908 sig->notify_count = -1; /* for exit_notify() */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700909 for (;;) {
910 write_lock_irq(&tasklist_lock);
911 if (likely(leader->exit_state))
912 break;
913 __set_current_state(TASK_UNINTERRUPTIBLE);
914 write_unlock_irq(&tasklist_lock);
915 schedule();
916 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Roland McGrathf5e90282006-04-10 22:54:16 -0700918 /*
919 * The only record we have of the real-time age of a
920 * process, regardless of execs it's done, is start_time.
921 * All the past CPU time is accumulated in signal_struct
922 * from sister threads now dead. But in this non-leader
923 * exec, nothing survives from the original leader thread,
924 * whose birth marks the true age of this process now.
925 * When we take on its identity by switching to its PID, we
926 * also take its birthdate (always earlier than our own).
927 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700928 tsk->start_time = leader->start_time;
Roland McGrathf5e90282006-04-10 22:54:16 -0700929
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700930 BUG_ON(!same_thread_group(leader, tsk));
931 BUG_ON(has_group_leader_pid(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 /*
933 * An exec() starts a new thread group with the
934 * TGID of the previous thread group. Rehash the
935 * two threads with a switched PID, and release
936 * the former thread group leader:
937 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800938
939 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700940 * The old leader becomes a thread of the this thread group.
941 * Note: The old leader also uses this pid until release_task
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800942 * is called. Odd but simple and correct.
943 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700944 detach_pid(tsk, PIDTYPE_PID);
945 tsk->pid = leader->pid;
Sukadev Bhattiprolu3743ca02007-10-18 23:39:51 -0700946 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700947 transfer_pid(leader, tsk, PIDTYPE_PGID);
948 transfer_pid(leader, tsk, PIDTYPE_SID);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800949
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700950 list_replace_rcu(&leader->tasks, &tsk->tasks);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800951 list_replace_init(&leader->sibling, &tsk->sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700953 tsk->group_leader = tsk;
954 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -0600955
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700956 tsk->exit_signal = SIGCHLD;
Oleg Nesterov087806b2011-06-22 23:10:26 +0200957 leader->exit_signal = -1;
Oleg Nesterov962b5642005-11-23 13:37:43 -0800958
959 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
960 leader->exit_state = EXIT_DEAD;
Oleg Nesteroveac1b5e2011-07-21 20:00:43 +0200961
962 /*
963 * We are going to release_task()->ptrace_unlink() silently,
964 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
965 * the tracer wont't block again waiting for this thread.
966 */
967 if (unlikely(leader->ptrace))
968 __wake_up_parent(leader, leader->parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 write_unlock_irq(&tasklist_lock);
Oleg Nesterov81879262008-12-01 14:18:16 -0800970
971 release_task(leader);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800972 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700974 sig->group_exit_task = NULL;
975 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
977no_thread_group:
Jiri Pirko1f102062009-09-22 16:44:10 -0700978 if (current->mm)
979 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 exit_itimers(sig);
Oleg Nesterovcbaffba2008-05-26 20:55:42 +0400982 flush_itimer_signals();
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300983
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700984 if (atomic_read(&oldsighand->count) != 1) {
985 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700987 * This ->sighand is shared with the CLONE_SIGHAND
988 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700990 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
991 if (!newsighand)
992 return -ENOMEM;
993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 atomic_set(&newsighand->count, 1);
995 memcpy(newsighand->action, oldsighand->action,
996 sizeof(newsighand->action));
997
998 write_lock_irq(&tasklist_lock);
999 spin_lock(&oldsighand->siglock);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001000 rcu_assign_pointer(tsk->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 spin_unlock(&oldsighand->siglock);
1002 write_unlock_irq(&tasklist_lock);
1003
Davide Libenzifba2afa2007-05-10 22:23:13 -07001004 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 }
1006
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001007 BUG_ON(!thread_group_leader(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 return 0;
1009}
Oleg Nesterov0840a902007-10-16 23:27:22 -07001010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011/*
1012 * These functions flushes out all traces of the currently running executable
1013 * so that a new one can be started
1014 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001015static void flush_old_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016{
1017 long j = -1;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001018 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
1020 spin_lock(&files->file_lock);
1021 for (;;) {
1022 unsigned long set, i;
1023
1024 j++;
1025 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001026 fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -08001027 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001029 set = fdt->close_on_exec->fds_bits[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 if (!set)
1031 continue;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07001032 fdt->close_on_exec->fds_bits[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 spin_unlock(&files->file_lock);
1034 for ( ; set ; i++,set >>= 1) {
1035 if (set & 1) {
1036 sys_close(i);
1037 }
1038 }
1039 spin_lock(&files->file_lock);
1040
1041 }
1042 spin_unlock(&files->file_lock);
1043}
1044
Andrew Morton59714d62008-02-04 22:27:21 -08001045char *get_task_comm(char *buf, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046{
1047 /* buf must be at least sizeof(tsk->comm) in size */
1048 task_lock(tsk);
1049 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1050 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -08001051 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052}
J Freyensee7d74f492011-05-06 16:56:47 -07001053EXPORT_SYMBOL_GPL(get_task_comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
1055void set_task_comm(struct task_struct *tsk, char *buf)
1056{
1057 task_lock(tsk);
john stultz4614a696b2009-12-14 18:00:05 -08001058
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -08001059 trace_task_rename(tsk, buf);
1060
john stultz4614a696b2009-12-14 18:00:05 -08001061 /*
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 Virof47ec3f2011-11-21 21:15:42 -05001232static int 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;
Vasiliy Kulikov72fa5992011-08-08 19:02:04 +04001462 const struct cred *cred = current_cred();
1463
1464 /*
1465 * We move the actual failure in case of RLIMIT_NPROC excess from
1466 * set*uid() to execve() because too many poorly written programs
1467 * don't check setuid() return code. Here we additionally recheck
1468 * whether NPROC limit is still exceeded.
1469 */
1470 if ((current->flags & PF_NPROC_EXCEEDED) &&
1471 atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1472 retval = -EAGAIN;
1473 goto out_ret;
1474 }
1475
1476 /* We're below the limit (still or again), so we don't want to make
1477 * further execve() calls fail. */
1478 current->flags &= ~PF_NPROC_EXCEEDED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
Al Viro3b125382008-04-22 05:31:30 -04001480 retval = unshare_files(&displaced);
Al Virofd8328b2008-04-22 05:11:59 -04001481 if (retval)
1482 goto out_ret;
1483
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 retval = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001485 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 if (!bprm)
Al Virofd8328b2008-04-22 05:11:59 -04001487 goto out_files;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
Oleg Nesterova2a84742009-09-05 11:17:13 -07001489 retval = prepare_bprm_creds(bprm);
1490 if (retval)
David Howellsa6f76f22008-11-14 10:39:24 +11001491 goto out_free;
Al Viro498052b2009-03-30 07:20:30 -04001492
1493 retval = check_unsafe_exec(bprm);
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001494 if (retval < 0)
Oleg Nesterova2a84742009-09-05 11:17:13 -07001495 goto out_free;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001496 clear_in_exec = retval;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001497 current->in_execve = 1;
David Howellsa6f76f22008-11-14 10:39:24 +11001498
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 file = open_exec(filename);
1500 retval = PTR_ERR(file);
1501 if (IS_ERR(file))
Al Viro498052b2009-03-30 07:20:30 -04001502 goto out_unmark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
1504 sched_exec();
1505
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 bprm->file = file;
1507 bprm->filename = filename;
1508 bprm->interp = filename;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001509
1510 retval = bprm_mm_init(bprm);
1511 if (retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 goto out_file;
1513
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001514 bprm->argc = count(argv, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 if ((retval = bprm->argc) < 0)
David Howellsa6f76f22008-11-14 10:39:24 +11001516 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001518 bprm->envc = count(envp, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 if ((retval = bprm->envc) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 goto out;
1521
1522 retval = prepare_binprm(bprm);
1523 if (retval < 0)
1524 goto out;
1525
1526 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1527 if (retval < 0)
1528 goto out;
1529
1530 bprm->exec = bprm->p;
1531 retval = copy_strings(bprm->envc, envp, bprm);
1532 if (retval < 0)
1533 goto out;
1534
1535 retval = copy_strings(bprm->argc, argv, bprm);
1536 if (retval < 0)
1537 goto out;
1538
1539 retval = search_binary_handler(bprm,regs);
David Howellsa6f76f22008-11-14 10:39:24 +11001540 if (retval < 0)
1541 goto out;
1542
1543 /* execve succeeded */
Al Viro498052b2009-03-30 07:20:30 -04001544 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001545 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001546 acct_update_integrals(current);
1547 free_bprm(bprm);
1548 if (displaced)
1549 put_files_struct(displaced);
1550 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551
1552out:
Oleg Nesterov3c77f842010-11-30 20:55:34 +01001553 if (bprm->mm) {
1554 acct_arg_size(bprm, 0);
1555 mmput(bprm->mm);
1556 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
1558out_file:
1559 if (bprm->file) {
1560 allow_write_access(bprm->file);
1561 fput(bprm->file);
1562 }
David Howellsa6f76f22008-11-14 10:39:24 +11001563
Al Viro498052b2009-03-30 07:20:30 -04001564out_unmark:
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001565 if (clear_in_exec)
1566 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001567 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001568
1569out_free:
Al Viro08a6fac2008-05-10 16:38:25 -04001570 free_bprm(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
Al Virofd8328b2008-04-22 05:11:59 -04001572out_files:
Al Viro3b125382008-04-22 05:31:30 -04001573 if (displaced)
1574 reset_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575out_ret:
1576 return retval;
1577}
1578
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001579int do_execve(const char *filename,
1580 const char __user *const __user *__argv,
1581 const char __user *const __user *__envp,
1582 struct pt_regs *regs)
1583{
Oleg Nesterov0e028462011-03-06 18:02:54 +01001584 struct user_arg_ptr argv = { .ptr.native = __argv };
1585 struct user_arg_ptr envp = { .ptr.native = __envp };
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001586 return do_execve_common(filename, argv, envp, regs);
1587}
1588
Oleg Nesterov0e028462011-03-06 18:02:54 +01001589#ifdef CONFIG_COMPAT
1590int compat_do_execve(char *filename,
1591 compat_uptr_t __user *__argv,
1592 compat_uptr_t __user *__envp,
1593 struct pt_regs *regs)
1594{
1595 struct user_arg_ptr argv = {
1596 .is_compat = true,
1597 .ptr.compat = __argv,
1598 };
1599 struct user_arg_ptr envp = {
1600 .is_compat = true,
1601 .ptr.compat = __envp,
1602 };
1603 return do_execve_common(filename, argv, envp, regs);
1604}
1605#endif
1606
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001607void set_binfmt(struct linux_binfmt *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608{
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001609 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001611 if (mm->binfmt)
1612 module_put(mm->binfmt->module);
1613
1614 mm->binfmt = new;
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001615 if (new)
1616 __module_get(new->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617}
1618
1619EXPORT_SYMBOL(set_binfmt);
1620
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001621static int expand_corename(struct core_name *cn)
1622{
1623 char *old_corename = cn->corename;
1624
1625 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1626 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1627
1628 if (!cn->corename) {
1629 kfree(old_corename);
1630 return -ENOMEM;
1631 }
1632
1633 return 0;
1634}
1635
1636static int cn_printf(struct core_name *cn, const char *fmt, ...)
1637{
1638 char *cur;
1639 int need;
1640 int ret;
1641 va_list arg;
1642
1643 va_start(arg, fmt);
1644 need = vsnprintf(NULL, 0, fmt, arg);
1645 va_end(arg);
1646
1647 if (likely(need < cn->size - cn->used - 1))
1648 goto out_printf;
1649
1650 ret = expand_corename(cn);
1651 if (ret)
1652 goto expand_fail;
1653
1654out_printf:
1655 cur = cn->corename + cn->used;
1656 va_start(arg, fmt);
1657 vsnprintf(cur, need + 1, fmt, arg);
1658 va_end(arg);
1659 cn->used += need;
1660 return 0;
1661
1662expand_fail:
1663 return ret;
1664}
1665
Jiri Slaby2c563732011-07-26 16:08:33 -07001666static void cn_escape(char *str)
1667{
1668 for (; *str; str++)
1669 if (*str == '/')
1670 *str = '!';
1671}
1672
Jiri Slaby57cc0832011-05-26 16:25:46 -07001673static int cn_print_exe_file(struct core_name *cn)
1674{
1675 struct file *exe_file;
Jiri Slaby2c563732011-07-26 16:08:33 -07001676 char *pathbuf, *path;
Jiri Slaby57cc0832011-05-26 16:25:46 -07001677 int ret;
1678
1679 exe_file = get_mm_exe_file(current->mm);
Jiri Slaby2c563732011-07-26 16:08:33 -07001680 if (!exe_file) {
1681 char *commstart = cn->corename + cn->used;
1682 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1683 cn_escape(commstart);
1684 return ret;
1685 }
Jiri Slaby57cc0832011-05-26 16:25:46 -07001686
1687 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1688 if (!pathbuf) {
1689 ret = -ENOMEM;
1690 goto put_exe_file;
1691 }
1692
1693 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1694 if (IS_ERR(path)) {
1695 ret = PTR_ERR(path);
1696 goto free_buf;
1697 }
1698
Jiri Slaby2c563732011-07-26 16:08:33 -07001699 cn_escape(path);
Jiri Slaby57cc0832011-05-26 16:25:46 -07001700
1701 ret = cn_printf(cn, "%s", path);
1702
1703free_buf:
1704 kfree(pathbuf);
1705put_exe_file:
1706 fput(exe_file);
1707 return ret;
1708}
1709
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710/* format_corename will inspect the pattern parameter, and output a
1711 * name into corename, which must have space for at least
1712 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1713 */
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001714static int format_corename(struct core_name *cn, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715{
David Howells86a264a2008-11-14 10:39:18 +11001716 const struct cred *cred = current_cred();
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001717 const char *pat_ptr = core_pattern;
1718 int ispipe = (*pat_ptr == '|');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 int pid_in_pattern = 0;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001720 int err = 0;
1721
1722 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1723 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1724 cn->used = 0;
1725
1726 if (!cn->corename)
1727 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
1729 /* Repeat as long as we have more pattern to process and more output
1730 space */
1731 while (*pat_ptr) {
1732 if (*pat_ptr != '%') {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001733 if (*pat_ptr == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 goto out;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001735 err = cn_printf(cn, "%c", *pat_ptr++);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 } else {
1737 switch (*++pat_ptr) {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001738 /* single % at the end, drop that */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 case 0:
1740 goto out;
1741 /* Double percent, output one percent */
1742 case '%':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001743 err = cn_printf(cn, "%c", '%');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 break;
1745 /* pid */
1746 case 'p':
1747 pid_in_pattern = 1;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001748 err = cn_printf(cn, "%d",
1749 task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 break;
1751 /* uid */
1752 case 'u':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001753 err = cn_printf(cn, "%d", cred->uid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 break;
1755 /* gid */
1756 case 'g':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001757 err = cn_printf(cn, "%d", cred->gid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 break;
1759 /* signal that caused the coredump */
1760 case 's':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001761 err = cn_printf(cn, "%ld", signr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 break;
1763 /* UNIX time of coredump */
1764 case 't': {
1765 struct timeval tv;
1766 do_gettimeofday(&tv);
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001767 err = cn_printf(cn, "%lu", tv.tv_sec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 break;
1769 }
1770 /* hostname */
Jiri Slaby2c563732011-07-26 16:08:33 -07001771 case 'h': {
1772 char *namestart = cn->corename + cn->used;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 down_read(&uts_sem);
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001774 err = cn_printf(cn, "%s",
1775 utsname()->nodename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 up_read(&uts_sem);
Jiri Slaby2c563732011-07-26 16:08:33 -07001777 cn_escape(namestart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 break;
Jiri Slaby2c563732011-07-26 16:08:33 -07001779 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 /* executable */
Jiri Slaby2c563732011-07-26 16:08:33 -07001781 case 'e': {
1782 char *commstart = cn->corename + cn->used;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001783 err = cn_printf(cn, "%s", current->comm);
Jiri Slaby2c563732011-07-26 16:08:33 -07001784 cn_escape(commstart);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 break;
Jiri Slaby2c563732011-07-26 16:08:33 -07001786 }
Jiri Slaby57cc0832011-05-26 16:25:46 -07001787 case 'E':
1788 err = cn_print_exe_file(cn);
1789 break;
Neil Horman74aadce2007-10-16 23:26:35 -07001790 /* core limit size */
1791 case 'c':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001792 err = cn_printf(cn, "%lu",
1793 rlimit(RLIMIT_CORE));
Neil Horman74aadce2007-10-16 23:26:35 -07001794 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 default:
1796 break;
1797 }
1798 ++pat_ptr;
1799 }
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001800
1801 if (err)
1802 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 }
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001804
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 /* Backward compatibility with core_uses_pid:
1806 *
1807 * If core_pattern does not include a %p (as is the default)
1808 * and core_uses_pid is set, then .%pid will be appended to
Alan Coxc4bbafd2007-04-16 22:53:13 -07001809 * the filename. Do not do this for piped commands. */
Oleg Nesterov64093242008-10-18 20:28:22 -07001810 if (!ispipe && !pid_in_pattern && core_uses_pid) {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001811 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1812 if (err)
1813 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 }
Alan Coxc4bbafd2007-04-16 22:53:13 -07001815out:
Alan Coxc4bbafd2007-04-16 22:53:13 -07001816 return ispipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817}
1818
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001819static int zap_process(struct task_struct *start, int exit_code)
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001820{
1821 struct task_struct *t;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001822 int nr = 0;
Oleg Nesterov281de332006-06-26 00:26:06 -07001823
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001824 start->signal->flags = SIGNAL_GROUP_EXIT;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001825 start->signal->group_exit_code = exit_code;
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001826 start->signal->group_stop_count = 0;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001827
1828 t = start;
1829 do {
Tejun Heo6dfca322011-06-02 11:14:00 +02001830 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001831 if (t != current && t->mm) {
Oleg Nesterov281de332006-06-26 00:26:06 -07001832 sigaddset(&t->pending.signal, SIGKILL);
1833 signal_wake_up(t, 1);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001834 nr++;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001835 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001836 } while_each_thread(start, t);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001837
1838 return nr;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001839}
1840
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001841static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001842 struct core_state *core_state, int exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843{
1844 struct task_struct *g, *p;
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001845 unsigned long flags;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001846 int nr = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001848 spin_lock_irq(&tsk->sighand->siglock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001849 if (!signal_group_exit(tsk->signal)) {
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001850 mm->core_state = core_state;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001851 nr = zap_process(tsk, exit_code);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 }
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001853 spin_unlock_irq(&tsk->sighand->siglock);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001854 if (unlikely(nr < 0))
1855 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001857 if (atomic_read(&mm->mm_users) == nr + 1)
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001858 goto done;
Oleg Nesterove4901f92008-07-25 01:47:31 -07001859 /*
1860 * We should find and kill all tasks which use this mm, and we should
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001861 * count them correctly into ->nr_threads. We don't take tasklist
Oleg Nesterove4901f92008-07-25 01:47:31 -07001862 * lock, but this is safe wrt:
1863 *
1864 * fork:
1865 * None of sub-threads can fork after zap_process(leader). All
1866 * processes which were created before this point should be
1867 * visible to zap_threads() because copy_process() adds the new
1868 * process to the tail of init_task.tasks list, and lock/unlock
1869 * of ->siglock provides a memory barrier.
1870 *
1871 * do_exit:
1872 * The caller holds mm->mmap_sem. This means that the task which
1873 * uses this mm can't pass exit_mm(), so it can't exit or clear
1874 * its ->mm.
1875 *
1876 * de_thread:
1877 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1878 * we must see either old or new leader, this does not matter.
1879 * However, it can change p->sighand, so lock_task_sighand(p)
1880 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1881 * it can't fail.
1882 *
1883 * Note also that "g" can be the old leader with ->mm == NULL
1884 * and already unhashed and thus removed from ->thread_group.
1885 * This is OK, __unhash_process()->list_del_rcu() does not
1886 * clear the ->next pointer, we will find the new leader via
1887 * next_thread().
1888 */
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001889 rcu_read_lock();
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001890 for_each_process(g) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001891 if (g == tsk->group_leader)
1892 continue;
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001893 if (g->flags & PF_KTHREAD)
1894 continue;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001895 p = g;
1896 do {
1897 if (p->mm) {
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001898 if (unlikely(p->mm == mm)) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001899 lock_task_sighand(p, &flags);
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001900 nr += zap_process(p, exit_code);
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001901 unlock_task_sighand(p, &flags);
1902 }
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001903 break;
1904 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001905 } while_each_thread(g, p);
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001906 }
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001907 rcu_read_unlock();
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001908done:
Oleg Nesterovc5f1cc82008-07-25 01:47:42 -07001909 atomic_set(&core_state->nr_threads, nr);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001910 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911}
1912
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001913static int coredump_wait(int exit_code, struct core_state *core_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914{
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001915 struct task_struct *tsk = current;
1916 struct mm_struct *mm = tsk->mm;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001917 struct completion *vfork_done;
Oleg Nesterov269b005a2010-05-26 14:43:08 -07001918 int core_waiters = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001920 init_completion(&core_state->startup);
Oleg Nesterovb564daf2008-07-25 01:47:44 -07001921 core_state->dumper.task = tsk;
1922 core_state->dumper.next = NULL;
Oleg Nesterov269b005a2010-05-26 14:43:08 -07001923
1924 down_write(&mm->mmap_sem);
1925 if (!mm->core_state)
1926 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001927 up_write(&mm->mmap_sem);
1928
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001929 if (unlikely(core_waiters < 0))
1930 goto fail;
1931
1932 /*
1933 * Make sure nobody is waiting for us to release the VM,
1934 * otherwise we can deadlock when we wait on each other
1935 */
1936 vfork_done = tsk->vfork_done;
1937 if (vfork_done) {
1938 tsk->vfork_done = NULL;
1939 complete(vfork_done);
1940 }
1941
Oleg Nesterov2384f552005-10-30 15:02:47 -08001942 if (core_waiters)
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001943 wait_for_completion(&core_state->startup);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001944fail:
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001945 return core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946}
1947
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001948static void coredump_finish(struct mm_struct *mm)
1949{
1950 struct core_thread *curr, *next;
1951 struct task_struct *task;
1952
1953 next = mm->core_state->dumper.next;
1954 while ((curr = next) != NULL) {
1955 next = curr->next;
1956 task = curr->task;
1957 /*
1958 * see exit_mm(), curr->task must not see
1959 * ->task == NULL before we read ->next.
1960 */
1961 smp_mb();
1962 curr->task = NULL;
1963 wake_up_process(task);
1964 }
1965
1966 mm->core_state = NULL;
1967}
1968
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001969/*
1970 * set_dumpable converts traditional three-value dumpable to two flags and
1971 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1972 * these bits are not changed atomically. So get_dumpable can observe the
1973 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1974 * return either old dumpable or new one by paying attention to the order of
1975 * modifying the bits.
1976 *
1977 * dumpable | mm->flags (binary)
1978 * old new | initial interim final
1979 * ---------+-----------------------
1980 * 0 1 | 00 01 01
1981 * 0 2 | 00 10(*) 11
1982 * 1 0 | 01 00 00
1983 * 1 2 | 01 11 11
1984 * 2 0 | 11 10(*) 00
1985 * 2 1 | 11 11 01
1986 *
1987 * (*) get_dumpable regards interim value of 10 as 11.
1988 */
1989void set_dumpable(struct mm_struct *mm, int value)
1990{
1991 switch (value) {
1992 case 0:
1993 clear_bit(MMF_DUMPABLE, &mm->flags);
1994 smp_wmb();
1995 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1996 break;
1997 case 1:
1998 set_bit(MMF_DUMPABLE, &mm->flags);
1999 smp_wmb();
2000 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2001 break;
2002 case 2:
2003 set_bit(MMF_DUMP_SECURELY, &mm->flags);
2004 smp_wmb();
2005 set_bit(MMF_DUMPABLE, &mm->flags);
2006 break;
2007 }
2008}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07002009
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002010static int __get_dumpable(unsigned long mm_flags)
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07002011{
2012 int ret;
2013
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002014 ret = mm_flags & MMF_DUMPABLE_MASK;
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07002015 return (ret >= 2) ? 2 : ret;
2016}
2017
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002018int get_dumpable(struct mm_struct *mm)
2019{
2020 return __get_dumpable(mm->flags);
2021}
2022
Neil Horman61be2282009-09-23 15:56:58 -07002023static void wait_for_dump_helpers(struct file *file)
2024{
2025 struct pipe_inode_info *pipe;
2026
2027 pipe = file->f_path.dentry->d_inode->i_pipe;
2028
2029 pipe_lock(pipe);
2030 pipe->readers++;
2031 pipe->writers--;
2032
2033 while ((pipe->readers > 1) && (!signal_pending(current))) {
2034 wake_up_interruptible_sync(&pipe->wait);
2035 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2036 pipe_wait(pipe);
2037 }
2038
2039 pipe->readers--;
2040 pipe->writers++;
2041 pipe_unlock(pipe);
2042
2043}
2044
2045
Neil Horman898b3742010-05-26 14:42:59 -07002046/*
Holger Hans Peter Freyther1bef8292011-02-24 17:46:49 +01002047 * umh_pipe_setup
Neil Horman898b3742010-05-26 14:42:59 -07002048 * helper function to customize the process used
2049 * to collect the core in userspace. Specifically
2050 * it sets up a pipe and installs it as fd 0 (stdin)
2051 * for the process. Returns 0 on success, or
2052 * PTR_ERR on failure.
2053 * Note that it also sets the core limit to 1. This
2054 * is a special value that we use to trap recursive
2055 * core dumps
2056 */
David Howells87966992011-06-17 11:25:59 +01002057static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
Neil Horman898b3742010-05-26 14:42:59 -07002058{
2059 struct file *rp, *wp;
2060 struct fdtable *fdt;
2061 struct coredump_params *cp = (struct coredump_params *)info->data;
2062 struct files_struct *cf = current->files;
2063
2064 wp = create_write_pipe(0);
2065 if (IS_ERR(wp))
2066 return PTR_ERR(wp);
2067
2068 rp = create_read_pipe(wp, 0);
2069 if (IS_ERR(rp)) {
2070 free_write_pipe(wp);
2071 return PTR_ERR(rp);
2072 }
2073
2074 cp->file = wp;
2075
2076 sys_close(0);
2077 fd_install(0, rp);
2078 spin_lock(&cf->file_lock);
2079 fdt = files_fdtable(cf);
2080 FD_SET(0, fdt->open_fds);
2081 FD_CLR(0, fdt->close_on_exec);
2082 spin_unlock(&cf->file_lock);
2083
2084 /* and disallow core files too */
2085 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2086
2087 return 0;
2088}
2089
WANG Cong8cd3ac32009-01-06 14:42:48 -08002090void do_coredump(long signr, int exit_code, struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091{
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07002092 struct core_state core_state;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002093 struct core_name cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 struct mm_struct *mm = current->mm;
2095 struct linux_binfmt * binfmt;
David Howellsd84f4f92008-11-14 10:39:23 +11002096 const struct cred *old_cred;
2097 struct cred *cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 int retval = 0;
Alan Coxd6e71142005-06-23 00:09:43 -07002099 int flag = 0;
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002100 int ispipe;
Neil Hormana2939802009-09-23 15:56:56 -07002101 static atomic_t core_dump_count = ATOMIC_INIT(0);
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002102 struct coredump_params cprm = {
2103 .signr = signr,
2104 .regs = regs,
Jiri Slabyd554ed892010-03-05 13:42:42 -08002105 .limit = rlimit(RLIMIT_CORE),
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002106 /*
2107 * We must use the same mm->flags while dumping core to avoid
2108 * inconsistency of bit flags, since this flag is not protected
2109 * by any locks.
2110 */
2111 .mm_flags = mm->flags,
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002112 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
Steve Grubb0a4ff8c2007-04-19 10:28:21 -04002114 audit_core_dumps(signr);
2115
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07002116 binfmt = mm->binfmt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 if (!binfmt || !binfmt->core_dump)
2118 goto fail;
Oleg Nesterov269b005a2010-05-26 14:43:08 -07002119 if (!__get_dumpable(cprm.mm_flags))
2120 goto fail;
David Howellsd84f4f92008-11-14 10:39:23 +11002121
2122 cred = prepare_creds();
Oleg Nesterov5e43aef52010-05-26 14:43:07 -07002123 if (!cred)
David Howellsd84f4f92008-11-14 10:39:23 +11002124 goto fail;
Alan Coxd6e71142005-06-23 00:09:43 -07002125 /*
2126 * We cannot trust fsuid as being the "true" uid of the
2127 * process nor do we know its entire history. We only know it
2128 * was tainted so we dump it as root in mode 2.
2129 */
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002130 if (__get_dumpable(cprm.mm_flags) == 2) {
2131 /* Setuid core dump mode */
Alan Coxd6e71142005-06-23 00:09:43 -07002132 flag = O_EXCL; /* Stop rewrite attacks */
David Howellsd84f4f92008-11-14 10:39:23 +11002133 cred->fsuid = 0; /* Dump root private */
Alan Coxd6e71142005-06-23 00:09:43 -07002134 }
Oleg Nesterov1291cf42005-10-30 15:02:54 -08002135
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07002136 retval = coredump_wait(exit_code, &core_state);
Oleg Nesterov5e43aef52010-05-26 14:43:07 -07002137 if (retval < 0)
2138 goto fail_creds;
David Howellsd84f4f92008-11-14 10:39:23 +11002139
2140 old_cred = override_creds(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141
2142 /*
2143 * Clear any false indication of pending signals that might
2144 * be seen by the filesystem code called to write the core file.
2145 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 clear_thread_flag(TIF_SIGPENDING);
2147
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002148 ispipe = format_corename(&cn, signr);
2149
Alan Coxc4bbafd2007-04-16 22:53:13 -07002150 if (ispipe) {
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002151 int dump_count;
2152 char **helper_argv;
2153
Oleg Nesterov99b64562011-07-26 16:08:34 -07002154 if (ispipe < 0) {
2155 printk(KERN_WARNING "format_corename failed\n");
2156 printk(KERN_WARNING "Aborting core\n");
2157 goto fail_corename;
2158 }
2159
Neil Horman898b3742010-05-26 14:42:59 -07002160 if (cprm.limit == 1) {
Neil Horman725eae32009-09-23 15:56:54 -07002161 /*
2162 * Normally core limits are irrelevant to pipes, since
2163 * we're not writing to the file system, but we use
Neil Horman898b3742010-05-26 14:42:59 -07002164 * cprm.limit of 1 here as a speacial value. Any
2165 * non-1 limit gets set to RLIM_INFINITY below, but
Neil Horman725eae32009-09-23 15:56:54 -07002166 * a limit of 0 skips the dump. This is a consistent
2167 * way to catch recursive crashes. We can still crash
Neil Horman898b3742010-05-26 14:42:59 -07002168 * if the core_pattern binary sets RLIM_CORE = !1
Neil Horman725eae32009-09-23 15:56:54 -07002169 * but it runs as root, and can do lots of stupid things
2170 * Note that we use task_tgid_vnr here to grab the pid
2171 * of the process group leader. That way we get the
2172 * right pid if a thread in a multi-threaded
2173 * core_pattern process dies.
2174 */
2175 printk(KERN_WARNING
Neil Horman898b3742010-05-26 14:42:59 -07002176 "Process %d(%s) has RLIMIT_CORE set to 1\n",
Neil Horman725eae32009-09-23 15:56:54 -07002177 task_tgid_vnr(current), current->comm);
2178 printk(KERN_WARNING "Aborting core\n");
2179 goto fail_unlock;
2180 }
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002181 cprm.limit = RLIM_INFINITY;
Neil Horman725eae32009-09-23 15:56:54 -07002182
Neil Hormana2939802009-09-23 15:56:56 -07002183 dump_count = atomic_inc_return(&core_dump_count);
2184 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2185 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2186 task_tgid_vnr(current), current->comm);
2187 printk(KERN_WARNING "Skipping core dump\n");
2188 goto fail_dropcount;
2189 }
2190
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002191 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
Tetsuo Handa350eaf72009-01-06 14:41:11 -08002192 if (!helper_argv) {
2193 printk(KERN_WARNING "%s failed to allocate memory\n",
2194 __func__);
Neil Hormana2939802009-09-23 15:56:56 -07002195 goto fail_dropcount;
Tetsuo Handa350eaf72009-01-06 14:41:11 -08002196 }
Neil Horman32321132007-10-16 23:26:36 -07002197
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002198 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2199 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2200 NULL, &cprm);
2201 argv_free(helper_argv);
2202 if (retval) {
Andi Kleend025c9d2006-09-30 23:29:28 -07002203 printk(KERN_INFO "Core dump to %s pipe failed\n",
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002204 cn.corename);
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002205 goto close_fail;
Andi Kleend025c9d2006-09-30 23:29:28 -07002206 }
Oleg Nesterovc7135412010-05-26 14:43:05 -07002207 } else {
2208 struct inode *inode;
2209
2210 if (cprm.limit < binfmt->min_coredump)
2211 goto fail_unlock;
2212
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002213 cprm.file = filp_open(cn.corename,
Alexey Dobriyan6d4df672006-12-06 20:40:39 -08002214 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2215 0600);
Oleg Nesterovc7135412010-05-26 14:43:05 -07002216 if (IS_ERR(cprm.file))
2217 goto fail_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218
Oleg Nesterovc7135412010-05-26 14:43:05 -07002219 inode = cprm.file->f_path.dentry->d_inode;
2220 if (inode->i_nlink > 1)
2221 goto close_fail;
2222 if (d_unhashed(cprm.file->f_path.dentry))
2223 goto close_fail;
2224 /*
2225 * AK: actually i see no reason to not allow this for named
2226 * pipes etc, but keep the previous behaviour for now.
2227 */
2228 if (!S_ISREG(inode->i_mode))
2229 goto close_fail;
2230 /*
2231 * Dont allow local users get cute and trick others to coredump
2232 * into their pre-created files.
2233 */
2234 if (inode->i_uid != current_fsuid())
2235 goto close_fail;
2236 if (!cprm.file->f_op || !cprm.file->f_op->write)
2237 goto close_fail;
2238 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2239 goto close_fail;
2240 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002242 retval = binfmt->core_dump(&cprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 if (retval)
2244 current->signal->group_exit_code |= 0x80;
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002245
Neil Horman61be2282009-09-23 15:56:58 -07002246 if (ispipe && core_pipe_limit)
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002247 wait_for_dump_helpers(cprm.file);
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002248close_fail:
2249 if (cprm.file)
2250 filp_close(cprm.file, NULL);
Neil Hormana2939802009-09-23 15:56:56 -07002251fail_dropcount:
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002252 if (ispipe)
Neil Hormana2939802009-09-23 15:56:56 -07002253 atomic_dec(&core_dump_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254fail_unlock:
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002255 kfree(cn.corename);
2256fail_corename:
Oleg Nesterova94e2d42008-07-25 01:47:46 -07002257 coredump_finish(mm);
Oleg Nesterov5e43aef52010-05-26 14:43:07 -07002258 revert_creds(old_cred);
2259fail_creds:
2260 put_cred(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261fail:
WANG Cong8cd3ac32009-01-06 14:42:48 -08002262 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263}
Linus Torvalds3aa0ce82010-10-14 14:32:06 -07002264
2265/*
2266 * Core dumping helper functions. These are the only things you should
2267 * do on a core-file: use only these functions to write out all the
2268 * necessary info.
2269 */
2270int dump_write(struct file *file, const void *addr, int nr)
2271{
2272 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2273}
Linus Torvalds8fd01d62010-10-14 19:15:28 -07002274EXPORT_SYMBOL(dump_write);
Linus Torvalds3aa0ce82010-10-14 14:32:06 -07002275
2276int dump_seek(struct file *file, loff_t off)
2277{
2278 int ret = 1;
2279
2280 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2281 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2282 return 0;
2283 } else {
2284 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2285
2286 if (!buf)
2287 return 0;
2288 while (off > 0) {
2289 unsigned long n = off;
2290
2291 if (n > PAGE_SIZE)
2292 n = PAGE_SIZE;
2293 if (!dump_write(file, buf, n)) {
2294 ret = 0;
2295 break;
2296 }
2297 off -= n;
2298 }
2299 free_page((unsigned long)buf);
2300 }
2301 return ret;
2302}
Linus Torvalds8fd01d62010-10-14 19:15:28 -07002303EXPORT_SYMBOL(dump_seek);