blob: 99d33a1371e9aeaf7298c4548ed18a634b9f2427 [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>
45#include <linux/proc_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/mount.h>
47#include <linux/security.h>
48#include <linux/syscalls.h>
Jay Lan8f0ab512006-09-30 23:28:59 -070049#include <linux/tsacct_kern.h>
Matt Helsley9f460802005-11-07 00:59:16 -080050#include <linux/cn_proc.h>
Al Viro473ae302006-04-26 14:04:08 -040051#include <linux/audit.h>
Roland McGrath6341c392008-07-25 19:45:44 -070052#include <linux/tracehook.h>
Johannes Berg5f4123b2008-07-09 10:28:40 +020053#include <linux/kmod.h>
Eric Paris6110e3a2008-12-17 13:53:20 -050054#include <linux/fsnotify.h>
Al Viro5ad4e532009-03-29 19:50:06 -040055#include <linux/fs_struct.h>
Neil Horman61be2282009-09-23 15:56:58 -070056#include <linux/pipe_fs_i.h>
Ying Han3d5992d2010-10-26 14:21:23 -070057#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59#include <asm/uaccess.h>
60#include <asm/mmu_context.h>
Ollie Wildb6a2fea2007-07-19 01:48:16 -070061#include <asm/tlb.h>
David Howellsa6f76f22008-11-14 10:39:24 +110062#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Linus Torvalds1da177e2005-04-16 15:20:36 -070064int core_uses_pid;
Dan Aloni71ce92f2007-05-16 22:11:16 -070065char core_pattern[CORENAME_MAX_SIZE] = "core";
Neil Hormana2939802009-09-23 15:56:56 -070066unsigned int core_pipe_limit;
Alan Coxd6e71142005-06-23 00:09:43 -070067int suid_dumpable = 0;
68
Xiaotian Feng1b0d3002010-10-27 15:34:08 -070069struct core_name {
70 char *corename;
71 int used, size;
72};
73static atomic_t call_count = ATOMIC_INIT(1);
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075/* The maximal length of core_pattern is also specified in sysctl.c */
76
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070077static LIST_HEAD(formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078static DEFINE_RWLOCK(binfmt_lock);
79
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070080int __register_binfmt(struct linux_binfmt * fmt, int insert)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081{
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 if (!fmt)
83 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 write_lock(&binfmt_lock);
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070085 insert ? list_add(&fmt->lh, &formats) :
86 list_add_tail(&fmt->lh, &formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 write_unlock(&binfmt_lock);
88 return 0;
89}
90
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070091EXPORT_SYMBOL(__register_binfmt);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Alexey Dobriyanf6b450d2007-10-16 23:26:04 -070093void unregister_binfmt(struct linux_binfmt * fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094{
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 write_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070096 list_del(&fmt->lh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 write_unlock(&binfmt_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098}
99
100EXPORT_SYMBOL(unregister_binfmt);
101
102static inline void put_binfmt(struct linux_binfmt * fmt)
103{
104 module_put(fmt->module);
105}
106
107/*
108 * Note that a shared library must be both readable and executable due to
109 * security reasons.
110 *
111 * Also note that we take the address to load from from the file itself.
112 */
Heiko Carstens1e7bfb22009-01-14 14:14:29 +0100113SYSCALL_DEFINE1(uselib, const char __user *, library)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114{
Al Viro964bd182008-07-26 03:33:14 -0400115 struct file *file;
Al Viro964bd182008-07-26 03:33:14 -0400116 char *tmp = getname(library);
117 int error = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
Al Viro6e8341a2009-04-06 11:16:22 -0400119 if (IS_ERR(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 goto out;
121
Al Viro6e8341a2009-04-06 11:16:22 -0400122 file = do_filp_open(AT_FDCWD, tmp,
123 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
124 MAY_READ | MAY_EXEC | MAY_OPEN);
125 putname(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 error = PTR_ERR(file);
127 if (IS_ERR(file))
128 goto out;
129
Al Viro6e8341a2009-04-06 11:16:22 -0400130 error = -EINVAL;
131 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
132 goto exit;
133
134 error = -EACCES;
135 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
136 goto exit;
137
Eric Paris2a12a9d2009-12-17 21:24:21 -0500138 fsnotify_open(file);
Eric Paris6110e3a2008-12-17 13:53:20 -0500139
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 error = -ENOEXEC;
141 if(file->f_op) {
142 struct linux_binfmt * fmt;
143
144 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -0700145 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 if (!fmt->load_shlib)
147 continue;
148 if (!try_module_get(fmt->module))
149 continue;
150 read_unlock(&binfmt_lock);
151 error = fmt->load_shlib(file);
152 read_lock(&binfmt_lock);
153 put_binfmt(fmt);
154 if (error != -ENOEXEC)
155 break;
156 }
157 read_unlock(&binfmt_lock);
158 }
Al Viro6e8341a2009-04-06 11:16:22 -0400159exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 fput(file);
161out:
162 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163}
164
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700165#ifdef CONFIG_MMU
166
167static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
168 int write)
169{
170 struct page *page;
171 int ret;
172
173#ifdef CONFIG_STACK_GROWSUP
174 if (write) {
175 ret = expand_stack_downwards(bprm->vma, pos);
176 if (ret < 0)
177 return NULL;
178 }
179#endif
180 ret = get_user_pages(current, bprm->mm, pos,
181 1, write, 1, &page, NULL);
182 if (ret <= 0)
183 return NULL;
184
185 if (write) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700186 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800187 struct rlimit *rlim;
188
189 /*
190 * We've historically supported up to 32 pages (ARG_MAX)
191 * of argument strings even with small stacks
192 */
193 if (size <= ARG_MAX)
194 return page;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700195
196 /*
197 * Limit to 1/4-th the stack size for the argv+env strings.
198 * This ensures that:
199 * - the remaining binfmt code will not run out of stack space,
200 * - the program will have a reasonable amount of stack left
201 * to work from.
202 */
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800203 rlim = current->signal->rlim;
Jiri Slabyd554ed892010-03-05 13:42:42 -0800204 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700205 put_page(page);
206 return NULL;
207 }
208 }
209
210 return page;
211}
212
213static void put_arg_page(struct page *page)
214{
215 put_page(page);
216}
217
218static void free_arg_page(struct linux_binprm *bprm, int i)
219{
220}
221
222static void free_arg_pages(struct linux_binprm *bprm)
223{
224}
225
226static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
227 struct page *page)
228{
229 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
230}
231
232static int __bprm_mm_init(struct linux_binprm *bprm)
233{
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800234 int err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700235 struct vm_area_struct *vma = NULL;
236 struct mm_struct *mm = bprm->mm;
237
238 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
239 if (!vma)
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800240 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700241
242 down_write(&mm->mmap_sem);
243 vma->vm_mm = mm;
244
245 /*
246 * Place the stack at the largest stack address the architecture
247 * supports. Later, we'll move this to an appropriate place. We don't
248 * use STACK_TOP because that can depend on attributes which aren't
249 * configured yet.
250 */
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700251 BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700252 vma->vm_end = STACK_TOP_MAX;
253 vma->vm_start = vma->vm_end - PAGE_SIZE;
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700254 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
Coly Li3ed75eb2007-10-18 23:39:15 -0700255 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Rik van Riel5beb4932010-03-05 13:42:07 -0800256 INIT_LIST_HEAD(&vma->anon_vma_chain);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700257 err = insert_vm_struct(mm, vma);
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800258 if (err)
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700259 goto err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700260
261 mm->stack_vm = mm->total_vm = 1;
262 up_write(&mm->mmap_sem);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700263 bprm->p = vma->vm_end - sizeof(void *);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700264 return 0;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700265err:
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800266 up_write(&mm->mmap_sem);
267 bprm->vma = NULL;
268 kmem_cache_free(vm_area_cachep, vma);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700269 return err;
270}
271
272static bool valid_arg_len(struct linux_binprm *bprm, long len)
273{
274 return len <= MAX_ARG_STRLEN;
275}
276
277#else
278
279static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
280 int write)
281{
282 struct page *page;
283
284 page = bprm->page[pos / PAGE_SIZE];
285 if (!page && write) {
286 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
287 if (!page)
288 return NULL;
289 bprm->page[pos / PAGE_SIZE] = page;
290 }
291
292 return page;
293}
294
295static void put_arg_page(struct page *page)
296{
297}
298
299static void free_arg_page(struct linux_binprm *bprm, int i)
300{
301 if (bprm->page[i]) {
302 __free_page(bprm->page[i]);
303 bprm->page[i] = NULL;
304 }
305}
306
307static void free_arg_pages(struct linux_binprm *bprm)
308{
309 int i;
310
311 for (i = 0; i < MAX_ARG_PAGES; i++)
312 free_arg_page(bprm, i);
313}
314
315static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
316 struct page *page)
317{
318}
319
320static int __bprm_mm_init(struct linux_binprm *bprm)
321{
322 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
323 return 0;
324}
325
326static bool valid_arg_len(struct linux_binprm *bprm, long len)
327{
328 return len <= bprm->p;
329}
330
331#endif /* CONFIG_MMU */
332
333/*
334 * Create a new mm_struct and populate it with a temporary stack
335 * vm_area_struct. We don't have enough context at this point to set the stack
336 * flags, permissions, and offset, so we use temporary values. We'll update
337 * them later in setup_arg_pages().
338 */
339int bprm_mm_init(struct linux_binprm *bprm)
340{
341 int err;
342 struct mm_struct *mm = NULL;
343
344 bprm->mm = mm = mm_alloc();
345 err = -ENOMEM;
346 if (!mm)
347 goto err;
348
349 err = init_new_context(current, mm);
350 if (err)
351 goto err;
352
353 err = __bprm_mm_init(bprm);
354 if (err)
355 goto err;
356
357 return 0;
358
359err:
360 if (mm) {
361 bprm->mm = NULL;
362 mmdrop(mm);
363 }
364
365 return err;
366}
367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368/*
369 * count() counts the number of strings in array ARGV.
370 */
David Howellsd7627462010-08-17 23:52:56 +0100371static int count(const char __user * const __user * argv, int max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372{
373 int i = 0;
374
375 if (argv != NULL) {
376 for (;;) {
David Howellsd7627462010-08-17 23:52:56 +0100377 const char __user * p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
379 if (get_user(p, argv))
380 return -EFAULT;
381 if (!p)
382 break;
383 argv++;
Jason Baron362e6662008-10-15 22:01:52 -0700384 if (i++ >= max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 return -E2BIG;
Roland McGrath9aea5a62010-09-07 19:37:06 -0700386
387 if (fatal_signal_pending(current))
388 return -ERESTARTNOHAND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 cond_resched();
390 }
391 }
392 return i;
393}
394
395/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700396 * 'copy_strings()' copies argument/environment strings from the old
397 * processes's memory to the new process's stack. The call to get_user_pages()
398 * ensures the destination page is created and not swapped out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 */
David Howellsd7627462010-08-17 23:52:56 +0100400static int copy_strings(int argc, const char __user *const __user *argv,
Adrian Bunk75c96f82005-05-05 16:16:09 -0700401 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402{
403 struct page *kmapped_page = NULL;
404 char *kaddr = NULL;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700405 unsigned long kpos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 int ret;
407
408 while (argc-- > 0) {
David Howellsd7627462010-08-17 23:52:56 +0100409 const char __user *str;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 int len;
411 unsigned long pos;
412
413 if (get_user(str, argv+argc) ||
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700414 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 ret = -EFAULT;
416 goto out;
417 }
418
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700419 if (!valid_arg_len(bprm, len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 ret = -E2BIG;
421 goto out;
422 }
423
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700424 /* We're going to work our way backwords. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 pos = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700426 str += len;
427 bprm->p -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
429 while (len > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 int offset, bytes_to_copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431
Roland McGrath9aea5a62010-09-07 19:37:06 -0700432 if (fatal_signal_pending(current)) {
433 ret = -ERESTARTNOHAND;
434 goto out;
435 }
Roland McGrath7993bc12010-09-07 19:36:28 -0700436 cond_resched();
437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 offset = pos % PAGE_SIZE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700439 if (offset == 0)
440 offset = PAGE_SIZE;
441
442 bytes_to_copy = offset;
443 if (bytes_to_copy > len)
444 bytes_to_copy = len;
445
446 offset -= bytes_to_copy;
447 pos -= bytes_to_copy;
448 str -= bytes_to_copy;
449 len -= bytes_to_copy;
450
451 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
452 struct page *page;
453
454 page = get_arg_page(bprm, pos, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 if (!page) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700456 ret = -E2BIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 goto out;
458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700460 if (kmapped_page) {
461 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700463 put_arg_page(kmapped_page);
464 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 kmapped_page = page;
466 kaddr = kmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700467 kpos = pos & PAGE_MASK;
468 flush_arg_page(bprm, kpos, kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700470 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 ret = -EFAULT;
472 goto out;
473 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 }
475 }
476 ret = 0;
477out:
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700478 if (kmapped_page) {
479 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700481 put_arg_page(kmapped_page);
482 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 return ret;
484}
485
486/*
487 * Like copy_strings, but get argv and its values from kernel memory.
488 */
David Howellsd7627462010-08-17 23:52:56 +0100489int copy_strings_kernel(int argc, const char *const *argv,
490 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
492 int r;
493 mm_segment_t oldfs = get_fs();
494 set_fs(KERNEL_DS);
David Howellsd7627462010-08-17 23:52:56 +0100495 r = copy_strings(argc, (const char __user *const __user *)argv, bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 set_fs(oldfs);
497 return r;
498}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499EXPORT_SYMBOL(copy_strings_kernel);
500
501#ifdef CONFIG_MMU
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700502
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700504 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
505 * the binfmt code determines where the new stack should reside, we shift it to
506 * its final location. The process proceeds as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 *
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700508 * 1) Use shift to calculate the new vma endpoints.
509 * 2) Extend vma to cover both the old and new ranges. This ensures the
510 * arguments passed to subsequent functions are consistent.
511 * 3) Move vma's page tables to the new range.
512 * 4) Free up any cleared pgd range.
513 * 5) Shrink the vma to cover only the new range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700515static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516{
517 struct mm_struct *mm = vma->vm_mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700518 unsigned long old_start = vma->vm_start;
519 unsigned long old_end = vma->vm_end;
520 unsigned long length = old_end - old_start;
521 unsigned long new_start = old_start - shift;
522 unsigned long new_end = old_end - shift;
523 struct mmu_gather *tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700525 BUG_ON(new_start > new_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700527 /*
528 * ensure there are no vmas between where we want to go
529 * and where we are
530 */
531 if (vma != find_vma(mm, new_start))
532 return -EFAULT;
533
534 /*
535 * cover the whole range: [new_start, old_end)
536 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800537 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
538 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700539
540 /*
541 * move the page tables downwards, on failure we rely on
542 * process cleanup to remove whatever mess we made.
543 */
544 if (length != move_page_tables(vma, old_start,
545 vma, new_start, length))
546 return -ENOMEM;
547
548 lru_add_drain();
549 tlb = tlb_gather_mmu(mm, 0);
550 if (new_end > old_start) {
551 /*
552 * when the old and new regions overlap clear from new_end.
553 */
Jan Beulich42b77722008-07-23 21:27:10 -0700554 free_pgd_range(tlb, new_end, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700555 vma->vm_next ? vma->vm_next->vm_start : 0);
556 } else {
557 /*
558 * otherwise, clean from old_start; this is done to not touch
559 * the address space in [new_end, old_start) some architectures
560 * have constraints on va-space that make this illegal (IA64) -
561 * for the others its just a little faster.
562 */
Jan Beulich42b77722008-07-23 21:27:10 -0700563 free_pgd_range(tlb, old_start, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700564 vma->vm_next ? vma->vm_next->vm_start : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700566 tlb_finish_mmu(tlb, new_end, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700568 /*
Rik van Riel5beb4932010-03-05 13:42:07 -0800569 * Shrink the vma to just the new range. Always succeeds.
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700570 */
571 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
572
573 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574}
575
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700576/*
577 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
578 * the stack is optionally relocated, and some extra space is added.
579 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580int setup_arg_pages(struct linux_binprm *bprm,
581 unsigned long stack_top,
582 int executable_stack)
583{
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700584 unsigned long ret;
585 unsigned long stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 struct mm_struct *mm = current->mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700587 struct vm_area_struct *vma = bprm->vma;
588 struct vm_area_struct *prev = NULL;
589 unsigned long vm_flags;
590 unsigned long stack_base;
Michael Neuling803bf5e2010-02-10 13:56:42 -0800591 unsigned long stack_size;
592 unsigned long stack_expand;
593 unsigned long rlim_stack;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
595#ifdef CONFIG_STACK_GROWSUP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 /* Limit stack size to 1GB */
Jiri Slabyd554ed892010-03-05 13:42:42 -0800597 stack_base = rlimit_max(RLIMIT_STACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 if (stack_base > (1 << 30))
599 stack_base = 1 << 30;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700601 /* Make sure we didn't let the argument array grow too large. */
602 if (vma->vm_end - vma->vm_start > stack_base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 return -ENOMEM;
604
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700605 stack_base = PAGE_ALIGN(stack_top - stack_base);
606
607 stack_shift = vma->vm_start - stack_base;
608 mm->arg_start = bprm->p - stack_shift;
609 bprm->p = vma->vm_end - stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610#else
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700611 stack_top = arch_align_stack(stack_top);
612 stack_top = PAGE_ALIGN(stack_top);
Roland McGrath1b528182010-09-07 19:35:49 -0700613
614 if (unlikely(stack_top < mmap_min_addr) ||
615 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
616 return -ENOMEM;
617
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700618 stack_shift = vma->vm_end - stack_top;
619
620 bprm->p -= stack_shift;
621 mm->arg_start = bprm->p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622#endif
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700623
624 if (bprm->loader)
625 bprm->loader -= stack_shift;
626 bprm->exec -= stack_shift;
627
628 down_write(&mm->mmap_sem);
Hugh Dickins96a8e132008-07-10 21:19:20 +0100629 vm_flags = VM_STACK_FLAGS;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700630
631 /*
632 * Adjust stack execute permissions; explicitly enable for
633 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
634 * (arch default) otherwise.
635 */
636 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
637 vm_flags |= VM_EXEC;
638 else if (executable_stack == EXSTACK_DISABLE_X)
639 vm_flags &= ~VM_EXEC;
640 vm_flags |= mm->def_flags;
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700641 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700642
643 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
644 vm_flags);
645 if (ret)
646 goto out_unlock;
647 BUG_ON(prev != vma);
648
649 /* Move stack pages down in memory. */
650 if (stack_shift) {
651 ret = shift_arg_pages(vma, stack_shift);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800652 if (ret)
653 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 }
655
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700656 /* mprotect_fixup is overkill to remove the temporary stack flags */
657 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
658
Michael Neuling5ef097d2010-03-05 13:42:57 -0800659 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
Michael Neuling803bf5e2010-02-10 13:56:42 -0800660 stack_size = vma->vm_end - vma->vm_start;
661 /*
662 * Align this down to a page boundary as expand_stack
663 * will align it up.
664 */
665 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700666#ifdef CONFIG_STACK_GROWSUP
Michael Neuling803bf5e2010-02-10 13:56:42 -0800667 if (stack_size + stack_expand > rlim_stack)
668 stack_base = vma->vm_start + rlim_stack;
669 else
670 stack_base = vma->vm_end + stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700671#else
Michael Neuling803bf5e2010-02-10 13:56:42 -0800672 if (stack_size + stack_expand > rlim_stack)
673 stack_base = vma->vm_end - rlim_stack;
674 else
675 stack_base = vma->vm_start - stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700676#endif
Eric B Munson3af9e852010-05-18 15:30:49 +0100677 current->mm->start_stack = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700678 ret = expand_stack(vma, stack_base);
679 if (ret)
680 ret = -EFAULT;
681
682out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 up_write(&mm->mmap_sem);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800684 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686EXPORT_SYMBOL(setup_arg_pages);
687
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688#endif /* CONFIG_MMU */
689
690struct file *open_exec(const char *name)
691{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 struct file *file;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200693 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Al Viro6e8341a2009-04-06 11:16:22 -0400695 file = do_filp_open(AT_FDCWD, name,
696 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
697 MAY_EXEC | MAY_OPEN);
698 if (IS_ERR(file))
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200699 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200701 err = -EACCES;
Al Viro6e8341a2009-04-06 11:16:22 -0400702 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
703 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200704
Al Viro6e8341a2009-04-06 11:16:22 -0400705 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
706 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200707
Eric Paris2a12a9d2009-12-17 21:24:21 -0500708 fsnotify_open(file);
Eric Paris6110e3a2008-12-17 13:53:20 -0500709
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200710 err = deny_write_access(file);
Al Viro6e8341a2009-04-06 11:16:22 -0400711 if (err)
712 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
Al Viro6e8341a2009-04-06 11:16:22 -0400714out:
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200715 return file;
716
Al Viro6e8341a2009-04-06 11:16:22 -0400717exit:
718 fput(file);
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200719 return ERR_PTR(err);
720}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721EXPORT_SYMBOL(open_exec);
722
Mimi Zohar6777d772009-08-21 14:32:48 -0400723int kernel_read(struct file *file, loff_t offset,
724 char *addr, unsigned long count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
726 mm_segment_t old_fs;
727 loff_t pos = offset;
728 int result;
729
730 old_fs = get_fs();
731 set_fs(get_ds());
732 /* The cast to a user pointer is valid due to the set_fs() */
733 result = vfs_read(file, (void __user *)addr, count, &pos);
734 set_fs(old_fs);
735 return result;
736}
737
738EXPORT_SYMBOL(kernel_read);
739
740static int exec_mmap(struct mm_struct *mm)
741{
742 struct task_struct *tsk;
743 struct mm_struct * old_mm, *active_mm;
744
745 /* Notify parent that we're no longer interested in the old VM */
746 tsk = current;
747 old_mm = current->mm;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800748 sync_mm_rss(tsk, old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 mm_release(tsk, old_mm);
750
751 if (old_mm) {
752 /*
753 * Make sure that if there is a core dump in progress
754 * for the old mm, we get out and die instead of going
755 * through with the exec. We must hold mmap_sem around
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700756 * checking core_state and changing tsk->mm.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 */
758 down_read(&old_mm->mmap_sem);
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700759 if (unlikely(old_mm->core_state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 up_read(&old_mm->mmap_sem);
761 return -EINTR;
762 }
763 }
764 task_lock(tsk);
765 active_mm = tsk->active_mm;
766 tsk->mm = mm;
767 tsk->active_mm = mm;
768 activate_mm(active_mm, mm);
Ying Han3d5992d2010-10-26 14:21:23 -0700769 if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
770 atomic_dec(&old_mm->oom_disable_count);
771 atomic_inc(&tsk->mm->oom_disable_count);
772 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 task_unlock(tsk);
774 arch_pick_mmap_layout(mm);
775 if (old_mm) {
776 up_read(&old_mm->mmap_sem);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +0200777 BUG_ON(active_mm != old_mm);
Balbir Singh31a78f22008-09-28 23:09:31 +0100778 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 mmput(old_mm);
780 return 0;
781 }
782 mmdrop(active_mm);
783 return 0;
784}
785
786/*
787 * This function makes sure the current process has its own signal table,
788 * so that flush_signal_handlers can later reset the handlers without
789 * disturbing other processes. (Other processes might share the signal
790 * table via the CLONE_SIGHAND option to clone().)
791 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800792static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
794 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700795 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 spinlock_t *lock = &oldsighand->siglock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700798 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 goto no_thread_group;
800
801 /*
802 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800805 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 /*
807 * Another group action in progress, just
808 * return so that the signal is processed.
809 */
810 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 return -EAGAIN;
812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
Oleg Nesterovd3441932010-05-26 14:43:11 -0700814 sig->group_exit_task = tsk;
815 sig->notify_count = zap_other_threads(tsk);
816 if (!thread_group_leader(tsk))
817 sig->notify_count--;
818
819 while (sig->notify_count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 __set_current_state(TASK_UNINTERRUPTIBLE);
821 spin_unlock_irq(lock);
822 schedule();
823 spin_lock_irq(lock);
824 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 spin_unlock_irq(lock);
826
827 /*
828 * At this point all other threads have exited, all we have to
829 * do is to wait for the thread group leader to become inactive,
830 * and to assume its PID:
831 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700832 if (!thread_group_leader(tsk)) {
Oleg Nesterov81879262008-12-01 14:18:16 -0800833 struct task_struct *leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700834
Oleg Nesterov2800d8d2008-04-30 00:53:12 -0700835 sig->notify_count = -1; /* for exit_notify() */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700836 for (;;) {
837 write_lock_irq(&tasklist_lock);
838 if (likely(leader->exit_state))
839 break;
840 __set_current_state(TASK_UNINTERRUPTIBLE);
841 write_unlock_irq(&tasklist_lock);
842 schedule();
843 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
Roland McGrathf5e90282006-04-10 22:54:16 -0700845 /*
846 * The only record we have of the real-time age of a
847 * process, regardless of execs it's done, is start_time.
848 * All the past CPU time is accumulated in signal_struct
849 * from sister threads now dead. But in this non-leader
850 * exec, nothing survives from the original leader thread,
851 * whose birth marks the true age of this process now.
852 * When we take on its identity by switching to its PID, we
853 * also take its birthdate (always earlier than our own).
854 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700855 tsk->start_time = leader->start_time;
Roland McGrathf5e90282006-04-10 22:54:16 -0700856
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700857 BUG_ON(!same_thread_group(leader, tsk));
858 BUG_ON(has_group_leader_pid(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 /*
860 * An exec() starts a new thread group with the
861 * TGID of the previous thread group. Rehash the
862 * two threads with a switched PID, and release
863 * the former thread group leader:
864 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800865
866 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700867 * The old leader becomes a thread of the this thread group.
868 * Note: The old leader also uses this pid until release_task
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800869 * is called. Odd but simple and correct.
870 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700871 detach_pid(tsk, PIDTYPE_PID);
872 tsk->pid = leader->pid;
Sukadev Bhattiprolu3743ca02007-10-18 23:39:51 -0700873 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700874 transfer_pid(leader, tsk, PIDTYPE_PGID);
875 transfer_pid(leader, tsk, PIDTYPE_SID);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800876
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700877 list_replace_rcu(&leader->tasks, &tsk->tasks);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800878 list_replace_init(&leader->sibling, &tsk->sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700880 tsk->group_leader = tsk;
881 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -0600882
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700883 tsk->exit_signal = SIGCHLD;
Oleg Nesterov962b5642005-11-23 13:37:43 -0800884
885 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
886 leader->exit_state = EXIT_DEAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 write_unlock_irq(&tasklist_lock);
Oleg Nesterov81879262008-12-01 14:18:16 -0800888
889 release_task(leader);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800890 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700892 sig->group_exit_task = NULL;
893 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
895no_thread_group:
Jiri Pirko1f102062009-09-22 16:44:10 -0700896 if (current->mm)
897 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
898
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 exit_itimers(sig);
Oleg Nesterovcbaffba2008-05-26 20:55:42 +0400900 flush_itimer_signals();
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300901
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700902 if (atomic_read(&oldsighand->count) != 1) {
903 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700905 * This ->sighand is shared with the CLONE_SIGHAND
906 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700908 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
909 if (!newsighand)
910 return -ENOMEM;
911
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 atomic_set(&newsighand->count, 1);
913 memcpy(newsighand->action, oldsighand->action,
914 sizeof(newsighand->action));
915
916 write_lock_irq(&tasklist_lock);
917 spin_lock(&oldsighand->siglock);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700918 rcu_assign_pointer(tsk->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 spin_unlock(&oldsighand->siglock);
920 write_unlock_irq(&tasklist_lock);
921
Davide Libenzifba2afa2007-05-10 22:23:13 -0700922 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 }
924
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700925 BUG_ON(!thread_group_leader(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 return 0;
927}
Oleg Nesterov0840a902007-10-16 23:27:22 -0700928
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929/*
930 * These functions flushes out all traces of the currently running executable
931 * so that a new one can be started
932 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800933static void flush_old_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934{
935 long j = -1;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700936 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937
938 spin_lock(&files->file_lock);
939 for (;;) {
940 unsigned long set, i;
941
942 j++;
943 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700944 fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800945 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700947 set = fdt->close_on_exec->fds_bits[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 if (!set)
949 continue;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700950 fdt->close_on_exec->fds_bits[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 spin_unlock(&files->file_lock);
952 for ( ; set ; i++,set >>= 1) {
953 if (set & 1) {
954 sys_close(i);
955 }
956 }
957 spin_lock(&files->file_lock);
958
959 }
960 spin_unlock(&files->file_lock);
961}
962
Andrew Morton59714d62008-02-04 22:27:21 -0800963char *get_task_comm(char *buf, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964{
965 /* buf must be at least sizeof(tsk->comm) in size */
966 task_lock(tsk);
967 strncpy(buf, tsk->comm, sizeof(tsk->comm));
968 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -0800969 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970}
971
972void set_task_comm(struct task_struct *tsk, char *buf)
973{
974 task_lock(tsk);
john stultz4614a696b2009-12-14 18:00:05 -0800975
976 /*
977 * Threads may access current->comm without holding
978 * the task lock, so write the string carefully.
979 * Readers without a lock may see incomplete new
980 * names but are safe from non-terminating string reads.
981 */
982 memset(tsk->comm, 0, TASK_COMM_LEN);
983 wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
985 task_unlock(tsk);
Ingo Molnarcdd6c482009-09-21 12:02:48 +0200986 perf_event_comm(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987}
988
989int flush_old_exec(struct linux_binprm * bprm)
990{
Linus Torvalds221af7f2010-01-28 22:14:42 -0800991 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
993 /*
994 * Make sure we have a private signal table and that
995 * we are unassociated from the previous thread group.
996 */
997 retval = de_thread(current);
998 if (retval)
999 goto out;
1000
Matt Helsley925d1c42008-04-29 01:01:36 -07001001 set_mm_exe_file(bprm->mm, bprm->file);
1002
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 * Release all of the old mmap stuff
1005 */
1006 retval = exec_mmap(bprm->mm);
1007 if (retval)
Al Virofd8328b2008-04-22 05:11:59 -04001008 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
1010 bprm->mm = NULL; /* We're using it now */
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001011
KOSAKI Motohiro98391cf2010-10-27 15:34:16 -07001012 current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001013 flush_thread();
1014 current->personality &= ~bprm->per_clear;
1015
Linus Torvalds221af7f2010-01-28 22:14:42 -08001016 return 0;
1017
1018out:
1019 return retval;
1020}
1021EXPORT_SYMBOL(flush_old_exec);
1022
1023void setup_new_exec(struct linux_binprm * bprm)
1024{
1025 int i, ch;
David Howellsd7627462010-08-17 23:52:56 +01001026 const char *name;
Linus Torvalds221af7f2010-01-28 22:14:42 -08001027 char tcomm[sizeof(current->comm)];
1028
1029 arch_pick_mmap_layout(current->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
1031 /* This is the point of no return */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 current->sas_ss_sp = current->sas_ss_size = 0;
1033
David Howellsda9592e2008-11-14 10:39:05 +11001034 if (current_euid() == current_uid() && current_egid() == current_gid())
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001035 set_dumpable(current->mm, 1);
Alan Coxd6e71142005-06-23 00:09:43 -07001036 else
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001037 set_dumpable(current->mm, suid_dumpable);
Alan Coxd6e71142005-06-23 00:09:43 -07001038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 name = bprm->filename;
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -07001040
1041 /* Copies the binary name from after last slash */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 for (i=0; (ch = *(name++)) != '\0';) {
1043 if (ch == '/')
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -07001044 i = 0; /* overwrite what we wrote */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 else
1046 if (i < (sizeof(tcomm) - 1))
1047 tcomm[i++] = ch;
1048 }
1049 tcomm[i] = '\0';
1050 set_task_comm(current, tcomm);
1051
Benjamin Herrenschmidt0551fbd2006-02-28 16:59:19 -08001052 /* Set the new mm task size. We have to do that late because it may
1053 * depend on TIF_32BIT which is only updated in flush_thread() on
1054 * some architectures like powerpc
1055 */
1056 current->mm->task_size = TASK_SIZE;
1057
David Howellsa6f76f22008-11-14 10:39:24 +11001058 /* install the new credentials */
1059 if (bprm->cred->uid != current_euid() ||
1060 bprm->cred->gid != current_egid()) {
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001061 current->pdeath_signal = 0;
1062 } else if (file_permission(bprm->file, MAY_READ) ||
David Howellsa6f76f22008-11-14 10:39:24 +11001063 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001064 set_dumpable(current->mm, suid_dumpable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 }
1066
Ingo Molnarf65cb452008-12-16 13:40:44 +01001067 /*
1068 * Flush performance counters when crossing a
1069 * security domain:
1070 */
1071 if (!get_dumpable(current->mm))
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001072 perf_event_exit_task(current);
Ingo Molnarf65cb452008-12-16 13:40:44 +01001073
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 /* An exec changes our domain. We are no longer part of the thread
1075 group */
1076
1077 current->self_exec_id++;
1078
1079 flush_signal_handlers(current, 0);
1080 flush_old_files(current->files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081}
Linus Torvalds221af7f2010-01-28 22:14:42 -08001082EXPORT_SYMBOL(setup_new_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083
David Howellsa6f76f22008-11-14 10:39:24 +11001084/*
Oleg Nesterova2a84742009-09-05 11:17:13 -07001085 * Prepare credentials and lock ->cred_guard_mutex.
1086 * install_exec_creds() commits the new creds and drops the lock.
1087 * Or, if exec fails before, free_bprm() should release ->cred and
1088 * and unlock.
1089 */
1090int prepare_bprm_creds(struct linux_binprm *bprm)
1091{
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001092 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
Oleg Nesterova2a84742009-09-05 11:17:13 -07001093 return -ERESTARTNOINTR;
1094
1095 bprm->cred = prepare_exec_creds();
1096 if (likely(bprm->cred))
1097 return 0;
1098
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001099 mutex_unlock(&current->signal->cred_guard_mutex);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001100 return -ENOMEM;
1101}
1102
1103void free_bprm(struct linux_binprm *bprm)
1104{
1105 free_arg_pages(bprm);
1106 if (bprm->cred) {
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001107 mutex_unlock(&current->signal->cred_guard_mutex);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001108 abort_creds(bprm->cred);
1109 }
1110 kfree(bprm);
1111}
1112
1113/*
David Howellsa6f76f22008-11-14 10:39:24 +11001114 * install the new credentials for this executable
1115 */
1116void install_exec_creds(struct linux_binprm *bprm)
1117{
1118 security_bprm_committing_creds(bprm);
1119
1120 commit_creds(bprm->cred);
1121 bprm->cred = NULL;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001122 /*
1123 * cred_guard_mutex must be held at least to this point to prevent
David Howellsa6f76f22008-11-14 10:39:24 +11001124 * ptrace_attach() from altering our determination of the task's
Oleg Nesterova2a84742009-09-05 11:17:13 -07001125 * credentials; any time after this it may be unlocked.
1126 */
David Howellsa6f76f22008-11-14 10:39:24 +11001127 security_bprm_committed_creds(bprm);
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001128 mutex_unlock(&current->signal->cred_guard_mutex);
David Howellsa6f76f22008-11-14 10:39:24 +11001129}
1130EXPORT_SYMBOL(install_exec_creds);
1131
1132/*
1133 * determine how safe it is to execute the proposed program
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001134 * - the caller must hold ->cred_guard_mutex to protect against
David Howellsa6f76f22008-11-14 10:39:24 +11001135 * PTRACE_ATTACH
1136 */
Al Viro498052b2009-03-30 07:20:30 -04001137int check_unsafe_exec(struct linux_binprm *bprm)
David Howellsa6f76f22008-11-14 10:39:24 +11001138{
David Howells0bf2f3a2009-02-06 11:45:46 +00001139 struct task_struct *p = current, *t;
Al Virof1191b52009-03-30 07:35:18 -04001140 unsigned n_fs;
Al Viro498052b2009-03-30 07:20:30 -04001141 int res = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001142
1143 bprm->unsafe = tracehook_unsafe_exec(p);
1144
David Howells0bf2f3a2009-02-06 11:45:46 +00001145 n_fs = 1;
Nick Piggin2a4419b2010-08-18 04:37:33 +10001146 spin_lock(&p->fs->lock);
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001147 rcu_read_lock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001148 for (t = next_thread(p); t != p; t = next_thread(t)) {
1149 if (t->fs == p->fs)
1150 n_fs++;
David Howells0bf2f3a2009-02-06 11:45:46 +00001151 }
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001152 rcu_read_unlock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001153
Al Virof1191b52009-03-30 07:35:18 -04001154 if (p->fs->users > n_fs) {
David Howellsa6f76f22008-11-14 10:39:24 +11001155 bprm->unsafe |= LSM_UNSAFE_SHARE;
Al Viro498052b2009-03-30 07:20:30 -04001156 } else {
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001157 res = -EAGAIN;
1158 if (!p->fs->in_exec) {
1159 p->fs->in_exec = 1;
1160 res = 1;
1161 }
Al Viro498052b2009-03-30 07:20:30 -04001162 }
Nick Piggin2a4419b2010-08-18 04:37:33 +10001163 spin_unlock(&p->fs->lock);
Al Viro498052b2009-03-30 07:20:30 -04001164
1165 return res;
David Howellsa6f76f22008-11-14 10:39:24 +11001166}
1167
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168/*
1169 * Fill the binprm structure from the inode.
1170 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
David Howellsa6f76f22008-11-14 10:39:24 +11001171 *
1172 * This may be called multiple times for binary chains (scripts for example).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 */
1174int prepare_binprm(struct linux_binprm *bprm)
1175{
David Howellsa6f76f22008-11-14 10:39:24 +11001176 umode_t mode;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001177 struct inode * inode = bprm->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 int retval;
1179
1180 mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 if (bprm->file->f_op == NULL)
1182 return -EACCES;
1183
David Howellsa6f76f22008-11-14 10:39:24 +11001184 /* clear any previous set[ug]id data from a previous binary */
1185 bprm->cred->euid = current_euid();
1186 bprm->cred->egid = current_egid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
David Howellsa6f76f22008-11-14 10:39:24 +11001188 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 /* Set-uid? */
1190 if (mode & S_ISUID) {
David Howellsa6f76f22008-11-14 10:39:24 +11001191 bprm->per_clear |= PER_CLEAR_ON_SETID;
1192 bprm->cred->euid = inode->i_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 }
1194
1195 /* Set-gid? */
1196 /*
1197 * If setgid is set but no group execute bit then this
1198 * is a candidate for mandatory locking, not a setgid
1199 * executable.
1200 */
1201 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
David Howellsa6f76f22008-11-14 10:39:24 +11001202 bprm->per_clear |= PER_CLEAR_ON_SETID;
1203 bprm->cred->egid = inode->i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 }
1205 }
1206
1207 /* fill in binprm security blob */
David Howellsa6f76f22008-11-14 10:39:24 +11001208 retval = security_bprm_set_creds(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 if (retval)
1210 return retval;
David Howellsa6f76f22008-11-14 10:39:24 +11001211 bprm->cred_prepared = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
David Howellsa6f76f22008-11-14 10:39:24 +11001213 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1214 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215}
1216
1217EXPORT_SYMBOL(prepare_binprm);
1218
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001219/*
1220 * Arguments are '\0' separated strings found at the location bprm->p
1221 * points to; chop off the first by relocating brpm->p to right after
1222 * the first '\0' encountered.
1223 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001224int remove_arg_zero(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225{
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001226 int ret = 0;
1227 unsigned long offset;
1228 char *kaddr;
1229 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001231 if (!bprm->argc)
1232 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001234 do {
1235 offset = bprm->p & ~PAGE_MASK;
1236 page = get_arg_page(bprm, bprm->p, 0);
1237 if (!page) {
1238 ret = -EFAULT;
1239 goto out;
1240 }
1241 kaddr = kmap_atomic(page, KM_USER0);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001242
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001243 for (; offset < PAGE_SIZE && kaddr[offset];
1244 offset++, bprm->p++)
1245 ;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001246
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001247 kunmap_atomic(kaddr, KM_USER0);
1248 put_arg_page(page);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001249
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001250 if (offset == PAGE_SIZE)
1251 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1252 } while (offset == PAGE_SIZE);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001253
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001254 bprm->p++;
1255 bprm->argc--;
1256 ret = 0;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001257
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001258out:
1259 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261EXPORT_SYMBOL(remove_arg_zero);
1262
1263/*
1264 * cycle the list of binary formats handler, until one recognizes the image
1265 */
1266int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1267{
Roland McGrath85f33462008-12-09 19:36:38 -08001268 unsigned int depth = bprm->recursion_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 int try,retval;
1270 struct linux_binfmt *fmt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 retval = security_bprm_check(bprm);
1273 if (retval)
1274 return retval;
1275
1276 /* kernel module loader fixup */
1277 /* so we don't try to load run modprobe in kernel space. */
1278 set_fs(USER_DS);
Al Viro473ae302006-04-26 14:04:08 -04001279
1280 retval = audit_bprm(bprm);
1281 if (retval)
1282 return retval;
1283
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 retval = -ENOENT;
1285 for (try=0; try<2; try++) {
1286 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -07001287 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1289 if (!fn)
1290 continue;
1291 if (!try_module_get(fmt->module))
1292 continue;
1293 read_unlock(&binfmt_lock);
1294 retval = fn(bprm, regs);
Roland McGrath85f33462008-12-09 19:36:38 -08001295 /*
1296 * Restore the depth counter to its starting value
1297 * in this call, so we don't have to rely on every
1298 * load_binary function to restore it on return.
1299 */
1300 bprm->recursion_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 if (retval >= 0) {
Roland McGrath85f33462008-12-09 19:36:38 -08001302 if (depth == 0)
1303 tracehook_report_exec(fmt, bprm, regs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 put_binfmt(fmt);
1305 allow_write_access(bprm->file);
1306 if (bprm->file)
1307 fput(bprm->file);
1308 bprm->file = NULL;
1309 current->did_exec = 1;
Matt Helsley9f460802005-11-07 00:59:16 -08001310 proc_exec_connector(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 return retval;
1312 }
1313 read_lock(&binfmt_lock);
1314 put_binfmt(fmt);
1315 if (retval != -ENOEXEC || bprm->mm == NULL)
1316 break;
1317 if (!bprm->file) {
1318 read_unlock(&binfmt_lock);
1319 return retval;
1320 }
1321 }
1322 read_unlock(&binfmt_lock);
1323 if (retval != -ENOEXEC || bprm->mm == NULL) {
1324 break;
Johannes Berg5f4123b2008-07-09 10:28:40 +02001325#ifdef CONFIG_MODULES
1326 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1328 if (printable(bprm->buf[0]) &&
1329 printable(bprm->buf[1]) &&
1330 printable(bprm->buf[2]) &&
1331 printable(bprm->buf[3]))
1332 break; /* -ENOEXEC */
1333 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1334#endif
1335 }
1336 }
1337 return retval;
1338}
1339
1340EXPORT_SYMBOL(search_binary_handler);
1341
1342/*
1343 * sys_execve() executes a new program.
1344 */
David Howellsd7627462010-08-17 23:52:56 +01001345int do_execve(const char * filename,
1346 const char __user *const __user *argv,
1347 const char __user *const __user *envp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 struct pt_regs * regs)
1349{
1350 struct linux_binprm *bprm;
1351 struct file *file;
Al Viro3b125382008-04-22 05:31:30 -04001352 struct files_struct *displaced;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001353 bool clear_in_exec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
Al Viro3b125382008-04-22 05:31:30 -04001356 retval = unshare_files(&displaced);
Al Virofd8328b2008-04-22 05:11:59 -04001357 if (retval)
1358 goto out_ret;
1359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 retval = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001361 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 if (!bprm)
Al Virofd8328b2008-04-22 05:11:59 -04001363 goto out_files;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Oleg Nesterova2a84742009-09-05 11:17:13 -07001365 retval = prepare_bprm_creds(bprm);
1366 if (retval)
David Howellsa6f76f22008-11-14 10:39:24 +11001367 goto out_free;
Al Viro498052b2009-03-30 07:20:30 -04001368
1369 retval = check_unsafe_exec(bprm);
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001370 if (retval < 0)
Oleg Nesterova2a84742009-09-05 11:17:13 -07001371 goto out_free;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001372 clear_in_exec = retval;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001373 current->in_execve = 1;
David Howellsa6f76f22008-11-14 10:39:24 +11001374
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 file = open_exec(filename);
1376 retval = PTR_ERR(file);
1377 if (IS_ERR(file))
Al Viro498052b2009-03-30 07:20:30 -04001378 goto out_unmark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379
1380 sched_exec();
1381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 bprm->file = file;
1383 bprm->filename = filename;
1384 bprm->interp = filename;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001385
1386 retval = bprm_mm_init(bprm);
1387 if (retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 goto out_file;
1389
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001390 bprm->argc = count(argv, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 if ((retval = bprm->argc) < 0)
David Howellsa6f76f22008-11-14 10:39:24 +11001392 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001394 bprm->envc = count(envp, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 if ((retval = bprm->envc) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 goto out;
1397
1398 retval = prepare_binprm(bprm);
1399 if (retval < 0)
1400 goto out;
1401
1402 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1403 if (retval < 0)
1404 goto out;
1405
1406 bprm->exec = bprm->p;
1407 retval = copy_strings(bprm->envc, envp, bprm);
1408 if (retval < 0)
1409 goto out;
1410
1411 retval = copy_strings(bprm->argc, argv, bprm);
1412 if (retval < 0)
1413 goto out;
1414
1415 retval = search_binary_handler(bprm,regs);
David Howellsa6f76f22008-11-14 10:39:24 +11001416 if (retval < 0)
1417 goto out;
1418
1419 /* execve succeeded */
Al Viro498052b2009-03-30 07:20:30 -04001420 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001421 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001422 acct_update_integrals(current);
1423 free_bprm(bprm);
1424 if (displaced)
1425 put_files_struct(displaced);
1426 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
1428out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 if (bprm->mm)
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001430 mmput (bprm->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
1432out_file:
1433 if (bprm->file) {
1434 allow_write_access(bprm->file);
1435 fput(bprm->file);
1436 }
David Howellsa6f76f22008-11-14 10:39:24 +11001437
Al Viro498052b2009-03-30 07:20:30 -04001438out_unmark:
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001439 if (clear_in_exec)
1440 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001441 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001442
1443out_free:
Al Viro08a6fac2008-05-10 16:38:25 -04001444 free_bprm(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
Al Virofd8328b2008-04-22 05:11:59 -04001446out_files:
Al Viro3b125382008-04-22 05:31:30 -04001447 if (displaced)
1448 reset_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449out_ret:
1450 return retval;
1451}
1452
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001453void set_binfmt(struct linux_binfmt *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454{
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001455 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001457 if (mm->binfmt)
1458 module_put(mm->binfmt->module);
1459
1460 mm->binfmt = new;
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001461 if (new)
1462 __module_get(new->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463}
1464
1465EXPORT_SYMBOL(set_binfmt);
1466
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001467static int expand_corename(struct core_name *cn)
1468{
1469 char *old_corename = cn->corename;
1470
1471 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1472 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1473
1474 if (!cn->corename) {
1475 kfree(old_corename);
1476 return -ENOMEM;
1477 }
1478
1479 return 0;
1480}
1481
1482static int cn_printf(struct core_name *cn, const char *fmt, ...)
1483{
1484 char *cur;
1485 int need;
1486 int ret;
1487 va_list arg;
1488
1489 va_start(arg, fmt);
1490 need = vsnprintf(NULL, 0, fmt, arg);
1491 va_end(arg);
1492
1493 if (likely(need < cn->size - cn->used - 1))
1494 goto out_printf;
1495
1496 ret = expand_corename(cn);
1497 if (ret)
1498 goto expand_fail;
1499
1500out_printf:
1501 cur = cn->corename + cn->used;
1502 va_start(arg, fmt);
1503 vsnprintf(cur, need + 1, fmt, arg);
1504 va_end(arg);
1505 cn->used += need;
1506 return 0;
1507
1508expand_fail:
1509 return ret;
1510}
1511
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512/* format_corename will inspect the pattern parameter, and output a
1513 * name into corename, which must have space for at least
1514 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1515 */
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001516static int format_corename(struct core_name *cn, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517{
David Howells86a264a2008-11-14 10:39:18 +11001518 const struct cred *cred = current_cred();
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001519 const char *pat_ptr = core_pattern;
1520 int ispipe = (*pat_ptr == '|');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 int pid_in_pattern = 0;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001522 int err = 0;
1523
1524 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1525 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1526 cn->used = 0;
1527
1528 if (!cn->corename)
1529 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
1531 /* Repeat as long as we have more pattern to process and more output
1532 space */
1533 while (*pat_ptr) {
1534 if (*pat_ptr != '%') {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001535 if (*pat_ptr == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 goto out;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001537 err = cn_printf(cn, "%c", *pat_ptr++);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 } else {
1539 switch (*++pat_ptr) {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001540 /* single % at the end, drop that */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 case 0:
1542 goto out;
1543 /* Double percent, output one percent */
1544 case '%':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001545 err = cn_printf(cn, "%c", '%');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 break;
1547 /* pid */
1548 case 'p':
1549 pid_in_pattern = 1;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001550 err = cn_printf(cn, "%d",
1551 task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 break;
1553 /* uid */
1554 case 'u':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001555 err = cn_printf(cn, "%d", cred->uid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 break;
1557 /* gid */
1558 case 'g':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001559 err = cn_printf(cn, "%d", cred->gid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 break;
1561 /* signal that caused the coredump */
1562 case 's':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001563 err = cn_printf(cn, "%ld", signr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 break;
1565 /* UNIX time of coredump */
1566 case 't': {
1567 struct timeval tv;
1568 do_gettimeofday(&tv);
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001569 err = cn_printf(cn, "%lu", tv.tv_sec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 break;
1571 }
1572 /* hostname */
1573 case 'h':
1574 down_read(&uts_sem);
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001575 err = cn_printf(cn, "%s",
1576 utsname()->nodename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 up_read(&uts_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 break;
1579 /* executable */
1580 case 'e':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001581 err = cn_printf(cn, "%s", current->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 break;
Neil Horman74aadce2007-10-16 23:26:35 -07001583 /* core limit size */
1584 case 'c':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001585 err = cn_printf(cn, "%lu",
1586 rlimit(RLIMIT_CORE));
Neil Horman74aadce2007-10-16 23:26:35 -07001587 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 default:
1589 break;
1590 }
1591 ++pat_ptr;
1592 }
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001593
1594 if (err)
1595 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 }
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001597
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 /* Backward compatibility with core_uses_pid:
1599 *
1600 * If core_pattern does not include a %p (as is the default)
1601 * and core_uses_pid is set, then .%pid will be appended to
Alan Coxc4bbafd2007-04-16 22:53:13 -07001602 * the filename. Do not do this for piped commands. */
Oleg Nesterov64093242008-10-18 20:28:22 -07001603 if (!ispipe && !pid_in_pattern && core_uses_pid) {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001604 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1605 if (err)
1606 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 }
Alan Coxc4bbafd2007-04-16 22:53:13 -07001608out:
Alan Coxc4bbafd2007-04-16 22:53:13 -07001609 return ispipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610}
1611
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001612static int zap_process(struct task_struct *start, int exit_code)
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001613{
1614 struct task_struct *t;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001615 int nr = 0;
Oleg Nesterov281de332006-06-26 00:26:06 -07001616
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001617 start->signal->flags = SIGNAL_GROUP_EXIT;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001618 start->signal->group_exit_code = exit_code;
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001619 start->signal->group_stop_count = 0;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001620
1621 t = start;
1622 do {
1623 if (t != current && t->mm) {
Oleg Nesterov281de332006-06-26 00:26:06 -07001624 sigaddset(&t->pending.signal, SIGKILL);
1625 signal_wake_up(t, 1);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001626 nr++;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001627 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001628 } while_each_thread(start, t);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001629
1630 return nr;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001631}
1632
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001633static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001634 struct core_state *core_state, int exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635{
1636 struct task_struct *g, *p;
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001637 unsigned long flags;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001638 int nr = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001640 spin_lock_irq(&tsk->sighand->siglock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001641 if (!signal_group_exit(tsk->signal)) {
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001642 mm->core_state = core_state;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001643 nr = zap_process(tsk, exit_code);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 }
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001645 spin_unlock_irq(&tsk->sighand->siglock);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001646 if (unlikely(nr < 0))
1647 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001649 if (atomic_read(&mm->mm_users) == nr + 1)
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001650 goto done;
Oleg Nesterove4901f92008-07-25 01:47:31 -07001651 /*
1652 * We should find and kill all tasks which use this mm, and we should
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001653 * count them correctly into ->nr_threads. We don't take tasklist
Oleg Nesterove4901f92008-07-25 01:47:31 -07001654 * lock, but this is safe wrt:
1655 *
1656 * fork:
1657 * None of sub-threads can fork after zap_process(leader). All
1658 * processes which were created before this point should be
1659 * visible to zap_threads() because copy_process() adds the new
1660 * process to the tail of init_task.tasks list, and lock/unlock
1661 * of ->siglock provides a memory barrier.
1662 *
1663 * do_exit:
1664 * The caller holds mm->mmap_sem. This means that the task which
1665 * uses this mm can't pass exit_mm(), so it can't exit or clear
1666 * its ->mm.
1667 *
1668 * de_thread:
1669 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1670 * we must see either old or new leader, this does not matter.
1671 * However, it can change p->sighand, so lock_task_sighand(p)
1672 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1673 * it can't fail.
1674 *
1675 * Note also that "g" can be the old leader with ->mm == NULL
1676 * and already unhashed and thus removed from ->thread_group.
1677 * This is OK, __unhash_process()->list_del_rcu() does not
1678 * clear the ->next pointer, we will find the new leader via
1679 * next_thread().
1680 */
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001681 rcu_read_lock();
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001682 for_each_process(g) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001683 if (g == tsk->group_leader)
1684 continue;
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001685 if (g->flags & PF_KTHREAD)
1686 continue;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001687 p = g;
1688 do {
1689 if (p->mm) {
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001690 if (unlikely(p->mm == mm)) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001691 lock_task_sighand(p, &flags);
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001692 nr += zap_process(p, exit_code);
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001693 unlock_task_sighand(p, &flags);
1694 }
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001695 break;
1696 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001697 } while_each_thread(g, p);
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001698 }
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001699 rcu_read_unlock();
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001700done:
Oleg Nesterovc5f1cc82008-07-25 01:47:42 -07001701 atomic_set(&core_state->nr_threads, nr);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001702 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703}
1704
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001705static int coredump_wait(int exit_code, struct core_state *core_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706{
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001707 struct task_struct *tsk = current;
1708 struct mm_struct *mm = tsk->mm;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001709 struct completion *vfork_done;
Oleg Nesterov269b0052010-05-26 14:43:08 -07001710 int core_waiters = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001712 init_completion(&core_state->startup);
Oleg Nesterovb564daf2008-07-25 01:47:44 -07001713 core_state->dumper.task = tsk;
1714 core_state->dumper.next = NULL;
Oleg Nesterov269b0052010-05-26 14:43:08 -07001715
1716 down_write(&mm->mmap_sem);
1717 if (!mm->core_state)
1718 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001719 up_write(&mm->mmap_sem);
1720
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001721 if (unlikely(core_waiters < 0))
1722 goto fail;
1723
1724 /*
1725 * Make sure nobody is waiting for us to release the VM,
1726 * otherwise we can deadlock when we wait on each other
1727 */
1728 vfork_done = tsk->vfork_done;
1729 if (vfork_done) {
1730 tsk->vfork_done = NULL;
1731 complete(vfork_done);
1732 }
1733
Oleg Nesterov2384f552005-10-30 15:02:47 -08001734 if (core_waiters)
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001735 wait_for_completion(&core_state->startup);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001736fail:
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001737 return core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738}
1739
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001740static void coredump_finish(struct mm_struct *mm)
1741{
1742 struct core_thread *curr, *next;
1743 struct task_struct *task;
1744
1745 next = mm->core_state->dumper.next;
1746 while ((curr = next) != NULL) {
1747 next = curr->next;
1748 task = curr->task;
1749 /*
1750 * see exit_mm(), curr->task must not see
1751 * ->task == NULL before we read ->next.
1752 */
1753 smp_mb();
1754 curr->task = NULL;
1755 wake_up_process(task);
1756 }
1757
1758 mm->core_state = NULL;
1759}
1760
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001761/*
1762 * set_dumpable converts traditional three-value dumpable to two flags and
1763 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1764 * these bits are not changed atomically. So get_dumpable can observe the
1765 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1766 * return either old dumpable or new one by paying attention to the order of
1767 * modifying the bits.
1768 *
1769 * dumpable | mm->flags (binary)
1770 * old new | initial interim final
1771 * ---------+-----------------------
1772 * 0 1 | 00 01 01
1773 * 0 2 | 00 10(*) 11
1774 * 1 0 | 01 00 00
1775 * 1 2 | 01 11 11
1776 * 2 0 | 11 10(*) 00
1777 * 2 1 | 11 11 01
1778 *
1779 * (*) get_dumpable regards interim value of 10 as 11.
1780 */
1781void set_dumpable(struct mm_struct *mm, int value)
1782{
1783 switch (value) {
1784 case 0:
1785 clear_bit(MMF_DUMPABLE, &mm->flags);
1786 smp_wmb();
1787 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1788 break;
1789 case 1:
1790 set_bit(MMF_DUMPABLE, &mm->flags);
1791 smp_wmb();
1792 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1793 break;
1794 case 2:
1795 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1796 smp_wmb();
1797 set_bit(MMF_DUMPABLE, &mm->flags);
1798 break;
1799 }
1800}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001801
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001802static int __get_dumpable(unsigned long mm_flags)
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001803{
1804 int ret;
1805
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001806 ret = mm_flags & MMF_DUMPABLE_MASK;
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001807 return (ret >= 2) ? 2 : ret;
1808}
1809
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001810int get_dumpable(struct mm_struct *mm)
1811{
1812 return __get_dumpable(mm->flags);
1813}
1814
Neil Horman61be2282009-09-23 15:56:58 -07001815static void wait_for_dump_helpers(struct file *file)
1816{
1817 struct pipe_inode_info *pipe;
1818
1819 pipe = file->f_path.dentry->d_inode->i_pipe;
1820
1821 pipe_lock(pipe);
1822 pipe->readers++;
1823 pipe->writers--;
1824
1825 while ((pipe->readers > 1) && (!signal_pending(current))) {
1826 wake_up_interruptible_sync(&pipe->wait);
1827 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1828 pipe_wait(pipe);
1829 }
1830
1831 pipe->readers--;
1832 pipe->writers++;
1833 pipe_unlock(pipe);
1834
1835}
1836
1837
Neil Horman898b3742010-05-26 14:42:59 -07001838/*
1839 * uhm_pipe_setup
1840 * helper function to customize the process used
1841 * to collect the core in userspace. Specifically
1842 * it sets up a pipe and installs it as fd 0 (stdin)
1843 * for the process. Returns 0 on success, or
1844 * PTR_ERR on failure.
1845 * Note that it also sets the core limit to 1. This
1846 * is a special value that we use to trap recursive
1847 * core dumps
1848 */
1849static int umh_pipe_setup(struct subprocess_info *info)
1850{
1851 struct file *rp, *wp;
1852 struct fdtable *fdt;
1853 struct coredump_params *cp = (struct coredump_params *)info->data;
1854 struct files_struct *cf = current->files;
1855
1856 wp = create_write_pipe(0);
1857 if (IS_ERR(wp))
1858 return PTR_ERR(wp);
1859
1860 rp = create_read_pipe(wp, 0);
1861 if (IS_ERR(rp)) {
1862 free_write_pipe(wp);
1863 return PTR_ERR(rp);
1864 }
1865
1866 cp->file = wp;
1867
1868 sys_close(0);
1869 fd_install(0, rp);
1870 spin_lock(&cf->file_lock);
1871 fdt = files_fdtable(cf);
1872 FD_SET(0, fdt->open_fds);
1873 FD_CLR(0, fdt->close_on_exec);
1874 spin_unlock(&cf->file_lock);
1875
1876 /* and disallow core files too */
1877 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1878
1879 return 0;
1880}
1881
WANG Cong8cd3ac32009-01-06 14:42:48 -08001882void do_coredump(long signr, int exit_code, struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883{
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001884 struct core_state core_state;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001885 struct core_name cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 struct mm_struct *mm = current->mm;
1887 struct linux_binfmt * binfmt;
David Howellsd84f4f92008-11-14 10:39:23 +11001888 const struct cred *old_cred;
1889 struct cred *cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 int retval = 0;
Alan Coxd6e71142005-06-23 00:09:43 -07001891 int flag = 0;
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07001892 int ispipe;
Neil Hormana2939802009-09-23 15:56:56 -07001893 static atomic_t core_dump_count = ATOMIC_INIT(0);
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001894 struct coredump_params cprm = {
1895 .signr = signr,
1896 .regs = regs,
Jiri Slabyd554ed892010-03-05 13:42:42 -08001897 .limit = rlimit(RLIMIT_CORE),
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001898 /*
1899 * We must use the same mm->flags while dumping core to avoid
1900 * inconsistency of bit flags, since this flag is not protected
1901 * by any locks.
1902 */
1903 .mm_flags = mm->flags,
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001904 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
Steve Grubb0a4ff8c2007-04-19 10:28:21 -04001906 audit_core_dumps(signr);
1907
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001908 binfmt = mm->binfmt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 if (!binfmt || !binfmt->core_dump)
1910 goto fail;
Oleg Nesterov269b0052010-05-26 14:43:08 -07001911 if (!__get_dumpable(cprm.mm_flags))
1912 goto fail;
David Howellsd84f4f92008-11-14 10:39:23 +11001913
1914 cred = prepare_creds();
Oleg Nesterov5e43aef2010-05-26 14:43:07 -07001915 if (!cred)
David Howellsd84f4f92008-11-14 10:39:23 +11001916 goto fail;
Alan Coxd6e71142005-06-23 00:09:43 -07001917 /*
1918 * We cannot trust fsuid as being the "true" uid of the
1919 * process nor do we know its entire history. We only know it
1920 * was tainted so we dump it as root in mode 2.
1921 */
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001922 if (__get_dumpable(cprm.mm_flags) == 2) {
1923 /* Setuid core dump mode */
Alan Coxd6e71142005-06-23 00:09:43 -07001924 flag = O_EXCL; /* Stop rewrite attacks */
David Howellsd84f4f92008-11-14 10:39:23 +11001925 cred->fsuid = 0; /* Dump root private */
Alan Coxd6e71142005-06-23 00:09:43 -07001926 }
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001927
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001928 retval = coredump_wait(exit_code, &core_state);
Oleg Nesterov5e43aef2010-05-26 14:43:07 -07001929 if (retval < 0)
1930 goto fail_creds;
David Howellsd84f4f92008-11-14 10:39:23 +11001931
1932 old_cred = override_creds(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933
1934 /*
1935 * Clear any false indication of pending signals that might
1936 * be seen by the filesystem code called to write the core file.
1937 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 clear_thread_flag(TIF_SIGPENDING);
1939
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001940 ispipe = format_corename(&cn, signr);
1941
1942 if (ispipe == -ENOMEM) {
1943 printk(KERN_WARNING "format_corename failed\n");
1944 printk(KERN_WARNING "Aborting core\n");
1945 goto fail_corename;
1946 }
Neil Horman725eae32009-09-23 15:56:54 -07001947
Alan Coxc4bbafd2007-04-16 22:53:13 -07001948 if (ispipe) {
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07001949 int dump_count;
1950 char **helper_argv;
1951
Neil Horman898b3742010-05-26 14:42:59 -07001952 if (cprm.limit == 1) {
Neil Horman725eae32009-09-23 15:56:54 -07001953 /*
1954 * Normally core limits are irrelevant to pipes, since
1955 * we're not writing to the file system, but we use
Neil Horman898b3742010-05-26 14:42:59 -07001956 * cprm.limit of 1 here as a speacial value. Any
1957 * non-1 limit gets set to RLIM_INFINITY below, but
Neil Horman725eae32009-09-23 15:56:54 -07001958 * a limit of 0 skips the dump. This is a consistent
1959 * way to catch recursive crashes. We can still crash
Neil Horman898b3742010-05-26 14:42:59 -07001960 * if the core_pattern binary sets RLIM_CORE = !1
Neil Horman725eae32009-09-23 15:56:54 -07001961 * but it runs as root, and can do lots of stupid things
1962 * Note that we use task_tgid_vnr here to grab the pid
1963 * of the process group leader. That way we get the
1964 * right pid if a thread in a multi-threaded
1965 * core_pattern process dies.
1966 */
1967 printk(KERN_WARNING
Neil Horman898b3742010-05-26 14:42:59 -07001968 "Process %d(%s) has RLIMIT_CORE set to 1\n",
Neil Horman725eae32009-09-23 15:56:54 -07001969 task_tgid_vnr(current), current->comm);
1970 printk(KERN_WARNING "Aborting core\n");
1971 goto fail_unlock;
1972 }
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07001973 cprm.limit = RLIM_INFINITY;
Neil Horman725eae32009-09-23 15:56:54 -07001974
Neil Hormana2939802009-09-23 15:56:56 -07001975 dump_count = atomic_inc_return(&core_dump_count);
1976 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1977 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1978 task_tgid_vnr(current), current->comm);
1979 printk(KERN_WARNING "Skipping core dump\n");
1980 goto fail_dropcount;
1981 }
1982
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001983 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
Tetsuo Handa350eaf72009-01-06 14:41:11 -08001984 if (!helper_argv) {
1985 printk(KERN_WARNING "%s failed to allocate memory\n",
1986 __func__);
Neil Hormana2939802009-09-23 15:56:56 -07001987 goto fail_dropcount;
Tetsuo Handa350eaf72009-01-06 14:41:11 -08001988 }
Neil Horman32321132007-10-16 23:26:36 -07001989
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07001990 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
1991 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
1992 NULL, &cprm);
1993 argv_free(helper_argv);
1994 if (retval) {
Andi Kleend025c9d2006-09-30 23:29:28 -07001995 printk(KERN_INFO "Core dump to %s pipe failed\n",
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001996 cn.corename);
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07001997 goto close_fail;
Andi Kleend025c9d2006-09-30 23:29:28 -07001998 }
Oleg Nesterovc7135412010-05-26 14:43:05 -07001999 } else {
2000 struct inode *inode;
2001
2002 if (cprm.limit < binfmt->min_coredump)
2003 goto fail_unlock;
2004
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002005 cprm.file = filp_open(cn.corename,
Alexey Dobriyan6d4df672006-12-06 20:40:39 -08002006 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2007 0600);
Oleg Nesterovc7135412010-05-26 14:43:05 -07002008 if (IS_ERR(cprm.file))
2009 goto fail_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
Oleg Nesterovc7135412010-05-26 14:43:05 -07002011 inode = cprm.file->f_path.dentry->d_inode;
2012 if (inode->i_nlink > 1)
2013 goto close_fail;
2014 if (d_unhashed(cprm.file->f_path.dentry))
2015 goto close_fail;
2016 /*
2017 * AK: actually i see no reason to not allow this for named
2018 * pipes etc, but keep the previous behaviour for now.
2019 */
2020 if (!S_ISREG(inode->i_mode))
2021 goto close_fail;
2022 /*
2023 * Dont allow local users get cute and trick others to coredump
2024 * into their pre-created files.
2025 */
2026 if (inode->i_uid != current_fsuid())
2027 goto close_fail;
2028 if (!cprm.file->f_op || !cprm.file->f_op->write)
2029 goto close_fail;
2030 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2031 goto close_fail;
2032 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002034 retval = binfmt->core_dump(&cprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 if (retval)
2036 current->signal->group_exit_code |= 0x80;
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002037
Neil Horman61be2282009-09-23 15:56:58 -07002038 if (ispipe && core_pipe_limit)
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002039 wait_for_dump_helpers(cprm.file);
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002040close_fail:
2041 if (cprm.file)
2042 filp_close(cprm.file, NULL);
Neil Hormana2939802009-09-23 15:56:56 -07002043fail_dropcount:
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002044 if (ispipe)
Neil Hormana2939802009-09-23 15:56:56 -07002045 atomic_dec(&core_dump_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046fail_unlock:
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002047 kfree(cn.corename);
2048fail_corename:
Oleg Nesterova94e2d42008-07-25 01:47:46 -07002049 coredump_finish(mm);
Oleg Nesterov5e43aef2010-05-26 14:43:07 -07002050 revert_creds(old_cred);
2051fail_creds:
2052 put_cred(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053fail:
WANG Cong8cd3ac32009-01-06 14:42:48 -08002054 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055}
Linus Torvalds3aa0ce82010-10-14 14:32:06 -07002056
2057/*
2058 * Core dumping helper functions. These are the only things you should
2059 * do on a core-file: use only these functions to write out all the
2060 * necessary info.
2061 */
2062int dump_write(struct file *file, const void *addr, int nr)
2063{
2064 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2065}
Linus Torvalds8fd01d62010-10-14 19:15:28 -07002066EXPORT_SYMBOL(dump_write);
Linus Torvalds3aa0ce82010-10-14 14:32:06 -07002067
2068int dump_seek(struct file *file, loff_t off)
2069{
2070 int ret = 1;
2071
2072 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2073 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2074 return 0;
2075 } else {
2076 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2077
2078 if (!buf)
2079 return 0;
2080 while (off > 0) {
2081 unsigned long n = off;
2082
2083 if (n > PAGE_SIZE)
2084 n = PAGE_SIZE;
2085 if (!dump_write(file, buf, n)) {
2086 ret = 0;
2087 break;
2088 }
2089 off -= n;
2090 }
2091 free_page((unsigned long)buf);
2092 }
2093 return ret;
2094}
Linus Torvalds8fd01d62010-10-14 19:15:28 -07002095EXPORT_SYMBOL(dump_seek);