blob: c62efcb959c73fc458367d811c576e634fb49634 [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
Oleg Nesterov114279b2010-11-30 20:56:02 +0100167void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100168{
169 struct mm_struct *mm = current->mm;
170 long diff = (long)(pages - bprm->vma_pages);
171
172 if (!mm || !diff)
173 return;
174
175 bprm->vma_pages = pages;
176
177#ifdef SPLIT_RSS_COUNTING
178 add_mm_counter(mm, MM_ANONPAGES, diff);
179#else
180 spin_lock(&mm->page_table_lock);
181 add_mm_counter(mm, MM_ANONPAGES, diff);
182 spin_unlock(&mm->page_table_lock);
183#endif
184}
185
Oleg Nesterov114279b2010-11-30 20:56:02 +0100186struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700187 int write)
188{
189 struct page *page;
190 int ret;
191
192#ifdef CONFIG_STACK_GROWSUP
193 if (write) {
194 ret = expand_stack_downwards(bprm->vma, pos);
195 if (ret < 0)
196 return NULL;
197 }
198#endif
199 ret = get_user_pages(current, bprm->mm, pos,
200 1, write, 1, &page, NULL);
201 if (ret <= 0)
202 return NULL;
203
204 if (write) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700205 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800206 struct rlimit *rlim;
207
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100208 acct_arg_size(bprm, size / PAGE_SIZE);
209
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800210 /*
211 * We've historically supported up to 32 pages (ARG_MAX)
212 * of argument strings even with small stacks
213 */
214 if (size <= ARG_MAX)
215 return page;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700216
217 /*
218 * Limit to 1/4-th the stack size for the argv+env strings.
219 * This ensures that:
220 * - the remaining binfmt code will not run out of stack space,
221 * - the program will have a reasonable amount of stack left
222 * to work from.
223 */
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800224 rlim = current->signal->rlim;
Jiri Slabyd554ed892010-03-05 13:42:42 -0800225 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700226 put_page(page);
227 return NULL;
228 }
229 }
230
231 return page;
232}
233
234static void put_arg_page(struct page *page)
235{
236 put_page(page);
237}
238
239static void free_arg_page(struct linux_binprm *bprm, int i)
240{
241}
242
243static void free_arg_pages(struct linux_binprm *bprm)
244{
245}
246
247static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
248 struct page *page)
249{
250 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
251}
252
253static int __bprm_mm_init(struct linux_binprm *bprm)
254{
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800255 int err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700256 struct vm_area_struct *vma = NULL;
257 struct mm_struct *mm = bprm->mm;
258
259 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
260 if (!vma)
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800261 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700262
263 down_write(&mm->mmap_sem);
264 vma->vm_mm = mm;
265
266 /*
267 * Place the stack at the largest stack address the architecture
268 * supports. Later, we'll move this to an appropriate place. We don't
269 * use STACK_TOP because that can depend on attributes which aren't
270 * configured yet.
271 */
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700272 BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700273 vma->vm_end = STACK_TOP_MAX;
274 vma->vm_start = vma->vm_end - PAGE_SIZE;
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700275 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
Coly Li3ed75eb2007-10-18 23:39:15 -0700276 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Rik van Riel5beb4932010-03-05 13:42:07 -0800277 INIT_LIST_HEAD(&vma->anon_vma_chain);
Tavis Ormandy462e635e2010-12-09 15:29:42 +0100278
279 err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
280 if (err)
281 goto err;
282
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700283 err = insert_vm_struct(mm, vma);
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800284 if (err)
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700285 goto err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700286
287 mm->stack_vm = mm->total_vm = 1;
288 up_write(&mm->mmap_sem);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700289 bprm->p = vma->vm_end - sizeof(void *);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700290 return 0;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700291err:
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800292 up_write(&mm->mmap_sem);
293 bprm->vma = NULL;
294 kmem_cache_free(vm_area_cachep, vma);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700295 return err;
296}
297
298static bool valid_arg_len(struct linux_binprm *bprm, long len)
299{
300 return len <= MAX_ARG_STRLEN;
301}
302
303#else
304
Oleg Nesterov114279b2010-11-30 20:56:02 +0100305void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100306{
307}
308
Oleg Nesterov114279b2010-11-30 20:56:02 +0100309struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700310 int write)
311{
312 struct page *page;
313
314 page = bprm->page[pos / PAGE_SIZE];
315 if (!page && write) {
316 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
317 if (!page)
318 return NULL;
319 bprm->page[pos / PAGE_SIZE] = page;
320 }
321
322 return page;
323}
324
325static void put_arg_page(struct page *page)
326{
327}
328
329static void free_arg_page(struct linux_binprm *bprm, int i)
330{
331 if (bprm->page[i]) {
332 __free_page(bprm->page[i]);
333 bprm->page[i] = NULL;
334 }
335}
336
337static void free_arg_pages(struct linux_binprm *bprm)
338{
339 int i;
340
341 for (i = 0; i < MAX_ARG_PAGES; i++)
342 free_arg_page(bprm, i);
343}
344
345static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
346 struct page *page)
347{
348}
349
350static int __bprm_mm_init(struct linux_binprm *bprm)
351{
352 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
353 return 0;
354}
355
356static bool valid_arg_len(struct linux_binprm *bprm, long len)
357{
358 return len <= bprm->p;
359}
360
361#endif /* CONFIG_MMU */
362
363/*
364 * Create a new mm_struct and populate it with a temporary stack
365 * vm_area_struct. We don't have enough context at this point to set the stack
366 * flags, permissions, and offset, so we use temporary values. We'll update
367 * them later in setup_arg_pages().
368 */
369int bprm_mm_init(struct linux_binprm *bprm)
370{
371 int err;
372 struct mm_struct *mm = NULL;
373
374 bprm->mm = mm = mm_alloc();
375 err = -ENOMEM;
376 if (!mm)
377 goto err;
378
379 err = init_new_context(current, mm);
380 if (err)
381 goto err;
382
383 err = __bprm_mm_init(bprm);
384 if (err)
385 goto err;
386
387 return 0;
388
389err:
390 if (mm) {
391 bprm->mm = NULL;
392 mmdrop(mm);
393 }
394
395 return err;
396}
397
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398/*
399 * count() counts the number of strings in array ARGV.
400 */
David Howellsd7627462010-08-17 23:52:56 +0100401static int count(const char __user * const __user * argv, int max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402{
403 int i = 0;
404
405 if (argv != NULL) {
406 for (;;) {
David Howellsd7627462010-08-17 23:52:56 +0100407 const char __user * p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
409 if (get_user(p, argv))
410 return -EFAULT;
411 if (!p)
412 break;
413 argv++;
Jason Baron362e6662008-10-15 22:01:52 -0700414 if (i++ >= max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 return -E2BIG;
Roland McGrath9aea5a62010-09-07 19:37:06 -0700416
417 if (fatal_signal_pending(current))
418 return -ERESTARTNOHAND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 cond_resched();
420 }
421 }
422 return i;
423}
424
425/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700426 * 'copy_strings()' copies argument/environment strings from the old
427 * processes's memory to the new process's stack. The call to get_user_pages()
428 * ensures the destination page is created and not swapped out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 */
David Howellsd7627462010-08-17 23:52:56 +0100430static int copy_strings(int argc, const char __user *const __user *argv,
Adrian Bunk75c96f82005-05-05 16:16:09 -0700431 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
433 struct page *kmapped_page = NULL;
434 char *kaddr = NULL;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700435 unsigned long kpos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 int ret;
437
438 while (argc-- > 0) {
David Howellsd7627462010-08-17 23:52:56 +0100439 const char __user *str;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 int len;
441 unsigned long pos;
442
443 if (get_user(str, argv+argc) ||
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700444 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 ret = -EFAULT;
446 goto out;
447 }
448
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700449 if (!valid_arg_len(bprm, len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 ret = -E2BIG;
451 goto out;
452 }
453
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700454 /* We're going to work our way backwords. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 pos = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700456 str += len;
457 bprm->p -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
459 while (len > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 int offset, bytes_to_copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Roland McGrath9aea5a62010-09-07 19:37:06 -0700462 if (fatal_signal_pending(current)) {
463 ret = -ERESTARTNOHAND;
464 goto out;
465 }
Roland McGrath7993bc12010-09-07 19:36:28 -0700466 cond_resched();
467
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 offset = pos % PAGE_SIZE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700469 if (offset == 0)
470 offset = PAGE_SIZE;
471
472 bytes_to_copy = offset;
473 if (bytes_to_copy > len)
474 bytes_to_copy = len;
475
476 offset -= bytes_to_copy;
477 pos -= bytes_to_copy;
478 str -= bytes_to_copy;
479 len -= bytes_to_copy;
480
481 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
482 struct page *page;
483
484 page = get_arg_page(bprm, pos, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 if (!page) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700486 ret = -E2BIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 goto out;
488 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700490 if (kmapped_page) {
491 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700493 put_arg_page(kmapped_page);
494 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 kmapped_page = page;
496 kaddr = kmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700497 kpos = pos & PAGE_MASK;
498 flush_arg_page(bprm, kpos, kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700500 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 ret = -EFAULT;
502 goto out;
503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 }
505 }
506 ret = 0;
507out:
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700508 if (kmapped_page) {
509 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700511 put_arg_page(kmapped_page);
512 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 return ret;
514}
515
516/*
517 * Like copy_strings, but get argv and its values from kernel memory.
518 */
David Howellsd7627462010-08-17 23:52:56 +0100519int copy_strings_kernel(int argc, const char *const *argv,
520 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521{
522 int r;
523 mm_segment_t oldfs = get_fs();
524 set_fs(KERNEL_DS);
David Howellsd7627462010-08-17 23:52:56 +0100525 r = copy_strings(argc, (const char __user *const __user *)argv, bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 set_fs(oldfs);
527 return r;
528}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529EXPORT_SYMBOL(copy_strings_kernel);
530
531#ifdef CONFIG_MMU
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700532
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700534 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
535 * the binfmt code determines where the new stack should reside, we shift it to
536 * its final location. The process proceeds as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 *
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700538 * 1) Use shift to calculate the new vma endpoints.
539 * 2) Extend vma to cover both the old and new ranges. This ensures the
540 * arguments passed to subsequent functions are consistent.
541 * 3) Move vma's page tables to the new range.
542 * 4) Free up any cleared pgd range.
543 * 5) Shrink the vma to cover only the new range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700545static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
547 struct mm_struct *mm = vma->vm_mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700548 unsigned long old_start = vma->vm_start;
549 unsigned long old_end = vma->vm_end;
550 unsigned long length = old_end - old_start;
551 unsigned long new_start = old_start - shift;
552 unsigned long new_end = old_end - shift;
553 struct mmu_gather *tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700555 BUG_ON(new_start > new_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700557 /*
558 * ensure there are no vmas between where we want to go
559 * and where we are
560 */
561 if (vma != find_vma(mm, new_start))
562 return -EFAULT;
563
564 /*
565 * cover the whole range: [new_start, old_end)
566 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800567 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
568 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700569
570 /*
571 * move the page tables downwards, on failure we rely on
572 * process cleanup to remove whatever mess we made.
573 */
574 if (length != move_page_tables(vma, old_start,
575 vma, new_start, length))
576 return -ENOMEM;
577
578 lru_add_drain();
579 tlb = tlb_gather_mmu(mm, 0);
580 if (new_end > old_start) {
581 /*
582 * when the old and new regions overlap clear from new_end.
583 */
Jan Beulich42b77722008-07-23 21:27:10 -0700584 free_pgd_range(tlb, new_end, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700585 vma->vm_next ? vma->vm_next->vm_start : 0);
586 } else {
587 /*
588 * otherwise, clean from old_start; this is done to not touch
589 * the address space in [new_end, old_start) some architectures
590 * have constraints on va-space that make this illegal (IA64) -
591 * for the others its just a little faster.
592 */
Jan Beulich42b77722008-07-23 21:27:10 -0700593 free_pgd_range(tlb, old_start, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700594 vma->vm_next ? vma->vm_next->vm_start : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700596 tlb_finish_mmu(tlb, new_end, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700598 /*
Rik van Riel5beb4932010-03-05 13:42:07 -0800599 * Shrink the vma to just the new range. Always succeeds.
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700600 */
601 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
602
603 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
605
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700606/*
607 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
608 * the stack is optionally relocated, and some extra space is added.
609 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610int setup_arg_pages(struct linux_binprm *bprm,
611 unsigned long stack_top,
612 int executable_stack)
613{
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700614 unsigned long ret;
615 unsigned long stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 struct mm_struct *mm = current->mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700617 struct vm_area_struct *vma = bprm->vma;
618 struct vm_area_struct *prev = NULL;
619 unsigned long vm_flags;
620 unsigned long stack_base;
Michael Neuling803bf5e2010-02-10 13:56:42 -0800621 unsigned long stack_size;
622 unsigned long stack_expand;
623 unsigned long rlim_stack;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
625#ifdef CONFIG_STACK_GROWSUP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 /* Limit stack size to 1GB */
Jiri Slabyd554ed892010-03-05 13:42:42 -0800627 stack_base = rlimit_max(RLIMIT_STACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 if (stack_base > (1 << 30))
629 stack_base = 1 << 30;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700631 /* Make sure we didn't let the argument array grow too large. */
632 if (vma->vm_end - vma->vm_start > stack_base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 return -ENOMEM;
634
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700635 stack_base = PAGE_ALIGN(stack_top - stack_base);
636
637 stack_shift = vma->vm_start - stack_base;
638 mm->arg_start = bprm->p - stack_shift;
639 bprm->p = vma->vm_end - stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640#else
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700641 stack_top = arch_align_stack(stack_top);
642 stack_top = PAGE_ALIGN(stack_top);
Roland McGrath1b528182010-09-07 19:35:49 -0700643
644 if (unlikely(stack_top < mmap_min_addr) ||
645 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
646 return -ENOMEM;
647
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700648 stack_shift = vma->vm_end - stack_top;
649
650 bprm->p -= stack_shift;
651 mm->arg_start = bprm->p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652#endif
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700653
654 if (bprm->loader)
655 bprm->loader -= stack_shift;
656 bprm->exec -= stack_shift;
657
658 down_write(&mm->mmap_sem);
Hugh Dickins96a8e132008-07-10 21:19:20 +0100659 vm_flags = VM_STACK_FLAGS;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700660
661 /*
662 * Adjust stack execute permissions; explicitly enable for
663 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
664 * (arch default) otherwise.
665 */
666 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
667 vm_flags |= VM_EXEC;
668 else if (executable_stack == EXSTACK_DISABLE_X)
669 vm_flags &= ~VM_EXEC;
670 vm_flags |= mm->def_flags;
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700671 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700672
673 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
674 vm_flags);
675 if (ret)
676 goto out_unlock;
677 BUG_ON(prev != vma);
678
679 /* Move stack pages down in memory. */
680 if (stack_shift) {
681 ret = shift_arg_pages(vma, stack_shift);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800682 if (ret)
683 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 }
685
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700686 /* mprotect_fixup is overkill to remove the temporary stack flags */
687 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
688
Michael Neuling5ef097d2010-03-05 13:42:57 -0800689 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
Michael Neuling803bf5e2010-02-10 13:56:42 -0800690 stack_size = vma->vm_end - vma->vm_start;
691 /*
692 * Align this down to a page boundary as expand_stack
693 * will align it up.
694 */
695 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700696#ifdef CONFIG_STACK_GROWSUP
Michael Neuling803bf5e2010-02-10 13:56:42 -0800697 if (stack_size + stack_expand > rlim_stack)
698 stack_base = vma->vm_start + rlim_stack;
699 else
700 stack_base = vma->vm_end + stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700701#else
Michael Neuling803bf5e2010-02-10 13:56:42 -0800702 if (stack_size + stack_expand > rlim_stack)
703 stack_base = vma->vm_end - rlim_stack;
704 else
705 stack_base = vma->vm_start - stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700706#endif
Eric B Munson3af9e852010-05-18 15:30:49 +0100707 current->mm->start_stack = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700708 ret = expand_stack(vma, stack_base);
709 if (ret)
710 ret = -EFAULT;
711
712out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 up_write(&mm->mmap_sem);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800714 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716EXPORT_SYMBOL(setup_arg_pages);
717
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718#endif /* CONFIG_MMU */
719
720struct file *open_exec(const char *name)
721{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 struct file *file;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200723 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
Al Viro6e8341a2009-04-06 11:16:22 -0400725 file = do_filp_open(AT_FDCWD, name,
726 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
727 MAY_EXEC | MAY_OPEN);
728 if (IS_ERR(file))
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200729 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200731 err = -EACCES;
Al Viro6e8341a2009-04-06 11:16:22 -0400732 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
733 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200734
Al Viro6e8341a2009-04-06 11:16:22 -0400735 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
736 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200737
Eric Paris2a12a9d2009-12-17 21:24:21 -0500738 fsnotify_open(file);
Eric Paris6110e3a2008-12-17 13:53:20 -0500739
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200740 err = deny_write_access(file);
Al Viro6e8341a2009-04-06 11:16:22 -0400741 if (err)
742 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
Al Viro6e8341a2009-04-06 11:16:22 -0400744out:
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200745 return file;
746
Al Viro6e8341a2009-04-06 11:16:22 -0400747exit:
748 fput(file);
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200749 return ERR_PTR(err);
750}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751EXPORT_SYMBOL(open_exec);
752
Mimi Zohar6777d772009-08-21 14:32:48 -0400753int kernel_read(struct file *file, loff_t offset,
754 char *addr, unsigned long count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755{
756 mm_segment_t old_fs;
757 loff_t pos = offset;
758 int result;
759
760 old_fs = get_fs();
761 set_fs(get_ds());
762 /* The cast to a user pointer is valid due to the set_fs() */
763 result = vfs_read(file, (void __user *)addr, count, &pos);
764 set_fs(old_fs);
765 return result;
766}
767
768EXPORT_SYMBOL(kernel_read);
769
770static int exec_mmap(struct mm_struct *mm)
771{
772 struct task_struct *tsk;
773 struct mm_struct * old_mm, *active_mm;
774
775 /* Notify parent that we're no longer interested in the old VM */
776 tsk = current;
777 old_mm = current->mm;
KAMEZAWA Hiroyuki34e55232010-03-05 13:41:40 -0800778 sync_mm_rss(tsk, old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 mm_release(tsk, old_mm);
780
781 if (old_mm) {
782 /*
783 * Make sure that if there is a core dump in progress
784 * for the old mm, we get out and die instead of going
785 * through with the exec. We must hold mmap_sem around
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700786 * checking core_state and changing tsk->mm.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 */
788 down_read(&old_mm->mmap_sem);
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700789 if (unlikely(old_mm->core_state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 up_read(&old_mm->mmap_sem);
791 return -EINTR;
792 }
793 }
794 task_lock(tsk);
795 active_mm = tsk->active_mm;
796 tsk->mm = mm;
797 tsk->active_mm = mm;
798 activate_mm(active_mm, mm);
Ying Han3d5992d2010-10-26 14:21:23 -0700799 if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
800 atomic_dec(&old_mm->oom_disable_count);
801 atomic_inc(&tsk->mm->oom_disable_count);
802 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 task_unlock(tsk);
804 arch_pick_mmap_layout(mm);
805 if (old_mm) {
806 up_read(&old_mm->mmap_sem);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +0200807 BUG_ON(active_mm != old_mm);
Balbir Singh31a78f22008-09-28 23:09:31 +0100808 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 mmput(old_mm);
810 return 0;
811 }
812 mmdrop(active_mm);
813 return 0;
814}
815
816/*
817 * This function makes sure the current process has its own signal table,
818 * so that flush_signal_handlers can later reset the handlers without
819 * disturbing other processes. (Other processes might share the signal
820 * table via the CLONE_SIGHAND option to clone().)
821 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800822static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823{
824 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700825 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 spinlock_t *lock = &oldsighand->siglock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700828 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 goto no_thread_group;
830
831 /*
832 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800835 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 /*
837 * Another group action in progress, just
838 * return so that the signal is processed.
839 */
840 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 return -EAGAIN;
842 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
Oleg Nesterovd3441932010-05-26 14:43:11 -0700844 sig->group_exit_task = tsk;
845 sig->notify_count = zap_other_threads(tsk);
846 if (!thread_group_leader(tsk))
847 sig->notify_count--;
848
849 while (sig->notify_count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 __set_current_state(TASK_UNINTERRUPTIBLE);
851 spin_unlock_irq(lock);
852 schedule();
853 spin_lock_irq(lock);
854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 spin_unlock_irq(lock);
856
857 /*
858 * At this point all other threads have exited, all we have to
859 * do is to wait for the thread group leader to become inactive,
860 * and to assume its PID:
861 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700862 if (!thread_group_leader(tsk)) {
Oleg Nesterov81879262008-12-01 14:18:16 -0800863 struct task_struct *leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700864
Oleg Nesterov2800d8d2008-04-30 00:53:12 -0700865 sig->notify_count = -1; /* for exit_notify() */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700866 for (;;) {
867 write_lock_irq(&tasklist_lock);
868 if (likely(leader->exit_state))
869 break;
870 __set_current_state(TASK_UNINTERRUPTIBLE);
871 write_unlock_irq(&tasklist_lock);
872 schedule();
873 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
Roland McGrathf5e90282006-04-10 22:54:16 -0700875 /*
876 * The only record we have of the real-time age of a
877 * process, regardless of execs it's done, is start_time.
878 * All the past CPU time is accumulated in signal_struct
879 * from sister threads now dead. But in this non-leader
880 * exec, nothing survives from the original leader thread,
881 * whose birth marks the true age of this process now.
882 * When we take on its identity by switching to its PID, we
883 * also take its birthdate (always earlier than our own).
884 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700885 tsk->start_time = leader->start_time;
Roland McGrathf5e90282006-04-10 22:54:16 -0700886
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700887 BUG_ON(!same_thread_group(leader, tsk));
888 BUG_ON(has_group_leader_pid(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 /*
890 * An exec() starts a new thread group with the
891 * TGID of the previous thread group. Rehash the
892 * two threads with a switched PID, and release
893 * the former thread group leader:
894 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800895
896 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700897 * The old leader becomes a thread of the this thread group.
898 * Note: The old leader also uses this pid until release_task
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800899 * is called. Odd but simple and correct.
900 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700901 detach_pid(tsk, PIDTYPE_PID);
902 tsk->pid = leader->pid;
Sukadev Bhattiprolu3743ca02007-10-18 23:39:51 -0700903 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700904 transfer_pid(leader, tsk, PIDTYPE_PGID);
905 transfer_pid(leader, tsk, PIDTYPE_SID);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800906
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700907 list_replace_rcu(&leader->tasks, &tsk->tasks);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -0800908 list_replace_init(&leader->sibling, &tsk->sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700910 tsk->group_leader = tsk;
911 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -0600912
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700913 tsk->exit_signal = SIGCHLD;
Oleg Nesterov962b5642005-11-23 13:37:43 -0800914
915 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
916 leader->exit_state = EXIT_DEAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 write_unlock_irq(&tasklist_lock);
Oleg Nesterov81879262008-12-01 14:18:16 -0800918
919 release_task(leader);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800920 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700922 sig->group_exit_task = NULL;
923 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
925no_thread_group:
Jiri Pirko1f102062009-09-22 16:44:10 -0700926 if (current->mm)
927 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
928
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 exit_itimers(sig);
Oleg Nesterovcbaffba2008-05-26 20:55:42 +0400930 flush_itimer_signals();
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300931
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700932 if (atomic_read(&oldsighand->count) != 1) {
933 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700935 * This ->sighand is shared with the CLONE_SIGHAND
936 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700938 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
939 if (!newsighand)
940 return -ENOMEM;
941
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 atomic_set(&newsighand->count, 1);
943 memcpy(newsighand->action, oldsighand->action,
944 sizeof(newsighand->action));
945
946 write_lock_irq(&tasklist_lock);
947 spin_lock(&oldsighand->siglock);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700948 rcu_assign_pointer(tsk->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 spin_unlock(&oldsighand->siglock);
950 write_unlock_irq(&tasklist_lock);
951
Davide Libenzifba2afa2007-05-10 22:23:13 -0700952 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 }
954
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700955 BUG_ON(!thread_group_leader(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 return 0;
957}
Oleg Nesterov0840a902007-10-16 23:27:22 -0700958
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959/*
960 * These functions flushes out all traces of the currently running executable
961 * so that a new one can be started
962 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800963static void flush_old_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964{
965 long j = -1;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700966 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967
968 spin_lock(&files->file_lock);
969 for (;;) {
970 unsigned long set, i;
971
972 j++;
973 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700974 fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800975 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700977 set = fdt->close_on_exec->fds_bits[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 if (!set)
979 continue;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700980 fdt->close_on_exec->fds_bits[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 spin_unlock(&files->file_lock);
982 for ( ; set ; i++,set >>= 1) {
983 if (set & 1) {
984 sys_close(i);
985 }
986 }
987 spin_lock(&files->file_lock);
988
989 }
990 spin_unlock(&files->file_lock);
991}
992
Andrew Morton59714d62008-02-04 22:27:21 -0800993char *get_task_comm(char *buf, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
995 /* buf must be at least sizeof(tsk->comm) in size */
996 task_lock(tsk);
997 strncpy(buf, tsk->comm, sizeof(tsk->comm));
998 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -0800999 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000}
1001
1002void set_task_comm(struct task_struct *tsk, char *buf)
1003{
1004 task_lock(tsk);
john stultz4614a696b2009-12-14 18:00:05 -08001005
1006 /*
1007 * Threads may access current->comm without holding
1008 * the task lock, so write the string carefully.
1009 * Readers without a lock may see incomplete new
1010 * names but are safe from non-terminating string reads.
1011 */
1012 memset(tsk->comm, 0, TASK_COMM_LEN);
1013 wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1015 task_unlock(tsk);
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001016 perf_event_comm(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017}
1018
1019int flush_old_exec(struct linux_binprm * bprm)
1020{
Linus Torvalds221af7f2010-01-28 22:14:42 -08001021 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
1023 /*
1024 * Make sure we have a private signal table and that
1025 * we are unassociated from the previous thread group.
1026 */
1027 retval = de_thread(current);
1028 if (retval)
1029 goto out;
1030
Matt Helsley925d1c42008-04-29 01:01:36 -07001031 set_mm_exe_file(bprm->mm, bprm->file);
1032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 * Release all of the old mmap stuff
1035 */
Oleg Nesterov3c77f842010-11-30 20:55:34 +01001036 acct_arg_size(bprm, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 retval = exec_mmap(bprm->mm);
1038 if (retval)
Al Virofd8328b2008-04-22 05:11:59 -04001039 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 bprm->mm = NULL; /* We're using it now */
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001042
KOSAKI Motohiro98391cf2010-10-27 15:34:16 -07001043 current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001044 flush_thread();
1045 current->personality &= ~bprm->per_clear;
1046
Linus Torvalds221af7f2010-01-28 22:14:42 -08001047 return 0;
1048
1049out:
1050 return retval;
1051}
1052EXPORT_SYMBOL(flush_old_exec);
1053
1054void setup_new_exec(struct linux_binprm * bprm)
1055{
1056 int i, ch;
David Howellsd7627462010-08-17 23:52:56 +01001057 const char *name;
Linus Torvalds221af7f2010-01-28 22:14:42 -08001058 char tcomm[sizeof(current->comm)];
1059
1060 arch_pick_mmap_layout(current->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
1062 /* This is the point of no return */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 current->sas_ss_sp = current->sas_ss_size = 0;
1064
David Howellsda9592e2008-11-14 10:39:05 +11001065 if (current_euid() == current_uid() && current_egid() == current_gid())
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001066 set_dumpable(current->mm, 1);
Alan Coxd6e71142005-06-23 00:09:43 -07001067 else
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001068 set_dumpable(current->mm, suid_dumpable);
Alan Coxd6e71142005-06-23 00:09:43 -07001069
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 name = bprm->filename;
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -07001071
1072 /* Copies the binary name from after last slash */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 for (i=0; (ch = *(name++)) != '\0';) {
1074 if (ch == '/')
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -07001075 i = 0; /* overwrite what we wrote */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 else
1077 if (i < (sizeof(tcomm) - 1))
1078 tcomm[i++] = ch;
1079 }
1080 tcomm[i] = '\0';
1081 set_task_comm(current, tcomm);
1082
Benjamin Herrenschmidt0551fbd2006-02-28 16:59:19 -08001083 /* Set the new mm task size. We have to do that late because it may
1084 * depend on TIF_32BIT which is only updated in flush_thread() on
1085 * some architectures like powerpc
1086 */
1087 current->mm->task_size = TASK_SIZE;
1088
David Howellsa6f76f22008-11-14 10:39:24 +11001089 /* install the new credentials */
1090 if (bprm->cred->uid != current_euid() ||
1091 bprm->cred->gid != current_egid()) {
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001092 current->pdeath_signal = 0;
1093 } else if (file_permission(bprm->file, MAY_READ) ||
David Howellsa6f76f22008-11-14 10:39:24 +11001094 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001095 set_dumpable(current->mm, suid_dumpable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 }
1097
Ingo Molnarf65cb452008-12-16 13:40:44 +01001098 /*
1099 * Flush performance counters when crossing a
1100 * security domain:
1101 */
1102 if (!get_dumpable(current->mm))
Ingo Molnarcdd6c482009-09-21 12:02:48 +02001103 perf_event_exit_task(current);
Ingo Molnarf65cb452008-12-16 13:40:44 +01001104
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 /* An exec changes our domain. We are no longer part of the thread
1106 group */
1107
1108 current->self_exec_id++;
1109
1110 flush_signal_handlers(current, 0);
1111 flush_old_files(current->files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112}
Linus Torvalds221af7f2010-01-28 22:14:42 -08001113EXPORT_SYMBOL(setup_new_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
David Howellsa6f76f22008-11-14 10:39:24 +11001115/*
Oleg Nesterova2a84742009-09-05 11:17:13 -07001116 * Prepare credentials and lock ->cred_guard_mutex.
1117 * install_exec_creds() commits the new creds and drops the lock.
1118 * Or, if exec fails before, free_bprm() should release ->cred and
1119 * and unlock.
1120 */
1121int prepare_bprm_creds(struct linux_binprm *bprm)
1122{
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001123 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
Oleg Nesterova2a84742009-09-05 11:17:13 -07001124 return -ERESTARTNOINTR;
1125
1126 bprm->cred = prepare_exec_creds();
1127 if (likely(bprm->cred))
1128 return 0;
1129
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001130 mutex_unlock(&current->signal->cred_guard_mutex);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001131 return -ENOMEM;
1132}
1133
1134void free_bprm(struct linux_binprm *bprm)
1135{
1136 free_arg_pages(bprm);
1137 if (bprm->cred) {
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001138 mutex_unlock(&current->signal->cred_guard_mutex);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001139 abort_creds(bprm->cred);
1140 }
1141 kfree(bprm);
1142}
1143
1144/*
David Howellsa6f76f22008-11-14 10:39:24 +11001145 * install the new credentials for this executable
1146 */
1147void install_exec_creds(struct linux_binprm *bprm)
1148{
1149 security_bprm_committing_creds(bprm);
1150
1151 commit_creds(bprm->cred);
1152 bprm->cred = NULL;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001153 /*
1154 * cred_guard_mutex must be held at least to this point to prevent
David Howellsa6f76f22008-11-14 10:39:24 +11001155 * ptrace_attach() from altering our determination of the task's
Oleg Nesterova2a84742009-09-05 11:17:13 -07001156 * credentials; any time after this it may be unlocked.
1157 */
David Howellsa6f76f22008-11-14 10:39:24 +11001158 security_bprm_committed_creds(bprm);
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001159 mutex_unlock(&current->signal->cred_guard_mutex);
David Howellsa6f76f22008-11-14 10:39:24 +11001160}
1161EXPORT_SYMBOL(install_exec_creds);
1162
1163/*
1164 * determine how safe it is to execute the proposed program
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001165 * - the caller must hold ->cred_guard_mutex to protect against
David Howellsa6f76f22008-11-14 10:39:24 +11001166 * PTRACE_ATTACH
1167 */
Al Viro498052b2009-03-30 07:20:30 -04001168int check_unsafe_exec(struct linux_binprm *bprm)
David Howellsa6f76f22008-11-14 10:39:24 +11001169{
David Howells0bf2f3a2009-02-06 11:45:46 +00001170 struct task_struct *p = current, *t;
Al Virof1191b52009-03-30 07:35:18 -04001171 unsigned n_fs;
Al Viro498052b2009-03-30 07:20:30 -04001172 int res = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001173
1174 bprm->unsafe = tracehook_unsafe_exec(p);
1175
David Howells0bf2f3a2009-02-06 11:45:46 +00001176 n_fs = 1;
Nick Piggin2a4419b2010-08-18 04:37:33 +10001177 spin_lock(&p->fs->lock);
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001178 rcu_read_lock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001179 for (t = next_thread(p); t != p; t = next_thread(t)) {
1180 if (t->fs == p->fs)
1181 n_fs++;
David Howells0bf2f3a2009-02-06 11:45:46 +00001182 }
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001183 rcu_read_unlock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001184
Al Virof1191b52009-03-30 07:35:18 -04001185 if (p->fs->users > n_fs) {
David Howellsa6f76f22008-11-14 10:39:24 +11001186 bprm->unsafe |= LSM_UNSAFE_SHARE;
Al Viro498052b2009-03-30 07:20:30 -04001187 } else {
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001188 res = -EAGAIN;
1189 if (!p->fs->in_exec) {
1190 p->fs->in_exec = 1;
1191 res = 1;
1192 }
Al Viro498052b2009-03-30 07:20:30 -04001193 }
Nick Piggin2a4419b2010-08-18 04:37:33 +10001194 spin_unlock(&p->fs->lock);
Al Viro498052b2009-03-30 07:20:30 -04001195
1196 return res;
David Howellsa6f76f22008-11-14 10:39:24 +11001197}
1198
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199/*
1200 * Fill the binprm structure from the inode.
1201 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
David Howellsa6f76f22008-11-14 10:39:24 +11001202 *
1203 * This may be called multiple times for binary chains (scripts for example).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 */
1205int prepare_binprm(struct linux_binprm *bprm)
1206{
David Howellsa6f76f22008-11-14 10:39:24 +11001207 umode_t mode;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001208 struct inode * inode = bprm->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 int retval;
1210
1211 mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 if (bprm->file->f_op == NULL)
1213 return -EACCES;
1214
David Howellsa6f76f22008-11-14 10:39:24 +11001215 /* clear any previous set[ug]id data from a previous binary */
1216 bprm->cred->euid = current_euid();
1217 bprm->cred->egid = current_egid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218
David Howellsa6f76f22008-11-14 10:39:24 +11001219 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 /* Set-uid? */
1221 if (mode & S_ISUID) {
David Howellsa6f76f22008-11-14 10:39:24 +11001222 bprm->per_clear |= PER_CLEAR_ON_SETID;
1223 bprm->cred->euid = inode->i_uid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 }
1225
1226 /* Set-gid? */
1227 /*
1228 * If setgid is set but no group execute bit then this
1229 * is a candidate for mandatory locking, not a setgid
1230 * executable.
1231 */
1232 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
David Howellsa6f76f22008-11-14 10:39:24 +11001233 bprm->per_clear |= PER_CLEAR_ON_SETID;
1234 bprm->cred->egid = inode->i_gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 }
1236 }
1237
1238 /* fill in binprm security blob */
David Howellsa6f76f22008-11-14 10:39:24 +11001239 retval = security_bprm_set_creds(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 if (retval)
1241 return retval;
David Howellsa6f76f22008-11-14 10:39:24 +11001242 bprm->cred_prepared = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
David Howellsa6f76f22008-11-14 10:39:24 +11001244 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1245 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246}
1247
1248EXPORT_SYMBOL(prepare_binprm);
1249
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001250/*
1251 * Arguments are '\0' separated strings found at the location bprm->p
1252 * points to; chop off the first by relocating brpm->p to right after
1253 * the first '\0' encountered.
1254 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001255int remove_arg_zero(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256{
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001257 int ret = 0;
1258 unsigned long offset;
1259 char *kaddr;
1260 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001262 if (!bprm->argc)
1263 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001265 do {
1266 offset = bprm->p & ~PAGE_MASK;
1267 page = get_arg_page(bprm, bprm->p, 0);
1268 if (!page) {
1269 ret = -EFAULT;
1270 goto out;
1271 }
1272 kaddr = kmap_atomic(page, KM_USER0);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001273
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001274 for (; offset < PAGE_SIZE && kaddr[offset];
1275 offset++, bprm->p++)
1276 ;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001277
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001278 kunmap_atomic(kaddr, KM_USER0);
1279 put_arg_page(page);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001280
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001281 if (offset == PAGE_SIZE)
1282 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1283 } while (offset == PAGE_SIZE);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001284
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001285 bprm->p++;
1286 bprm->argc--;
1287 ret = 0;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001288
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001289out:
1290 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292EXPORT_SYMBOL(remove_arg_zero);
1293
1294/*
1295 * cycle the list of binary formats handler, until one recognizes the image
1296 */
1297int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1298{
Roland McGrath85f33462008-12-09 19:36:38 -08001299 unsigned int depth = bprm->recursion_depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 int try,retval;
1301 struct linux_binfmt *fmt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 retval = security_bprm_check(bprm);
1304 if (retval)
1305 return retval;
1306
1307 /* kernel module loader fixup */
1308 /* so we don't try to load run modprobe in kernel space. */
1309 set_fs(USER_DS);
Al Viro473ae302006-04-26 14:04:08 -04001310
1311 retval = audit_bprm(bprm);
1312 if (retval)
1313 return retval;
1314
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 retval = -ENOENT;
1316 for (try=0; try<2; try++) {
1317 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -07001318 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1320 if (!fn)
1321 continue;
1322 if (!try_module_get(fmt->module))
1323 continue;
1324 read_unlock(&binfmt_lock);
1325 retval = fn(bprm, regs);
Roland McGrath85f33462008-12-09 19:36:38 -08001326 /*
1327 * Restore the depth counter to its starting value
1328 * in this call, so we don't have to rely on every
1329 * load_binary function to restore it on return.
1330 */
1331 bprm->recursion_depth = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 if (retval >= 0) {
Roland McGrath85f33462008-12-09 19:36:38 -08001333 if (depth == 0)
1334 tracehook_report_exec(fmt, bprm, regs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 put_binfmt(fmt);
1336 allow_write_access(bprm->file);
1337 if (bprm->file)
1338 fput(bprm->file);
1339 bprm->file = NULL;
1340 current->did_exec = 1;
Matt Helsley9f460802005-11-07 00:59:16 -08001341 proc_exec_connector(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 return retval;
1343 }
1344 read_lock(&binfmt_lock);
1345 put_binfmt(fmt);
1346 if (retval != -ENOEXEC || bprm->mm == NULL)
1347 break;
1348 if (!bprm->file) {
1349 read_unlock(&binfmt_lock);
1350 return retval;
1351 }
1352 }
1353 read_unlock(&binfmt_lock);
1354 if (retval != -ENOEXEC || bprm->mm == NULL) {
1355 break;
Johannes Berg5f4123b2008-07-09 10:28:40 +02001356#ifdef CONFIG_MODULES
1357 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1359 if (printable(bprm->buf[0]) &&
1360 printable(bprm->buf[1]) &&
1361 printable(bprm->buf[2]) &&
1362 printable(bprm->buf[3]))
1363 break; /* -ENOEXEC */
1364 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1365#endif
1366 }
1367 }
1368 return retval;
1369}
1370
1371EXPORT_SYMBOL(search_binary_handler);
1372
1373/*
1374 * sys_execve() executes a new program.
1375 */
David Howellsd7627462010-08-17 23:52:56 +01001376int do_execve(const char * filename,
1377 const char __user *const __user *argv,
1378 const char __user *const __user *envp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 struct pt_regs * regs)
1380{
1381 struct linux_binprm *bprm;
1382 struct file *file;
Al Viro3b125382008-04-22 05:31:30 -04001383 struct files_struct *displaced;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001384 bool clear_in_exec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
Al Viro3b125382008-04-22 05:31:30 -04001387 retval = unshare_files(&displaced);
Al Virofd8328b2008-04-22 05:11:59 -04001388 if (retval)
1389 goto out_ret;
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 retval = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001392 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 if (!bprm)
Al Virofd8328b2008-04-22 05:11:59 -04001394 goto out_files;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395
Oleg Nesterova2a84742009-09-05 11:17:13 -07001396 retval = prepare_bprm_creds(bprm);
1397 if (retval)
David Howellsa6f76f22008-11-14 10:39:24 +11001398 goto out_free;
Al Viro498052b2009-03-30 07:20:30 -04001399
1400 retval = check_unsafe_exec(bprm);
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001401 if (retval < 0)
Oleg Nesterova2a84742009-09-05 11:17:13 -07001402 goto out_free;
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001403 clear_in_exec = retval;
Oleg Nesterova2a84742009-09-05 11:17:13 -07001404 current->in_execve = 1;
David Howellsa6f76f22008-11-14 10:39:24 +11001405
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 file = open_exec(filename);
1407 retval = PTR_ERR(file);
1408 if (IS_ERR(file))
Al Viro498052b2009-03-30 07:20:30 -04001409 goto out_unmark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
1411 sched_exec();
1412
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 bprm->file = file;
1414 bprm->filename = filename;
1415 bprm->interp = filename;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001416
1417 retval = bprm_mm_init(bprm);
1418 if (retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 goto out_file;
1420
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001421 bprm->argc = count(argv, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 if ((retval = bprm->argc) < 0)
David Howellsa6f76f22008-11-14 10:39:24 +11001423 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001425 bprm->envc = count(envp, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 if ((retval = bprm->envc) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 goto out;
1428
1429 retval = prepare_binprm(bprm);
1430 if (retval < 0)
1431 goto out;
1432
1433 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1434 if (retval < 0)
1435 goto out;
1436
1437 bprm->exec = bprm->p;
1438 retval = copy_strings(bprm->envc, envp, bprm);
1439 if (retval < 0)
1440 goto out;
1441
1442 retval = copy_strings(bprm->argc, argv, bprm);
1443 if (retval < 0)
1444 goto out;
1445
1446 retval = search_binary_handler(bprm,regs);
David Howellsa6f76f22008-11-14 10:39:24 +11001447 if (retval < 0)
1448 goto out;
1449
1450 /* execve succeeded */
Al Viro498052b2009-03-30 07:20:30 -04001451 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001452 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001453 acct_update_integrals(current);
1454 free_bprm(bprm);
1455 if (displaced)
1456 put_files_struct(displaced);
1457 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
1459out:
Oleg Nesterov3c77f842010-11-30 20:55:34 +01001460 if (bprm->mm) {
1461 acct_arg_size(bprm, 0);
1462 mmput(bprm->mm);
1463 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
1465out_file:
1466 if (bprm->file) {
1467 allow_write_access(bprm->file);
1468 fput(bprm->file);
1469 }
David Howellsa6f76f22008-11-14 10:39:24 +11001470
Al Viro498052b2009-03-30 07:20:30 -04001471out_unmark:
Oleg Nesterov8c652f92009-04-24 01:01:56 +02001472 if (clear_in_exec)
1473 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001474 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001475
1476out_free:
Al Viro08a6fac2008-05-10 16:38:25 -04001477 free_bprm(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
Al Virofd8328b2008-04-22 05:11:59 -04001479out_files:
Al Viro3b125382008-04-22 05:31:30 -04001480 if (displaced)
1481 reset_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482out_ret:
1483 return retval;
1484}
1485
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001486void set_binfmt(struct linux_binfmt *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487{
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001488 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001490 if (mm->binfmt)
1491 module_put(mm->binfmt->module);
1492
1493 mm->binfmt = new;
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001494 if (new)
1495 __module_get(new->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496}
1497
1498EXPORT_SYMBOL(set_binfmt);
1499
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001500static int expand_corename(struct core_name *cn)
1501{
1502 char *old_corename = cn->corename;
1503
1504 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1505 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1506
1507 if (!cn->corename) {
1508 kfree(old_corename);
1509 return -ENOMEM;
1510 }
1511
1512 return 0;
1513}
1514
1515static int cn_printf(struct core_name *cn, const char *fmt, ...)
1516{
1517 char *cur;
1518 int need;
1519 int ret;
1520 va_list arg;
1521
1522 va_start(arg, fmt);
1523 need = vsnprintf(NULL, 0, fmt, arg);
1524 va_end(arg);
1525
1526 if (likely(need < cn->size - cn->used - 1))
1527 goto out_printf;
1528
1529 ret = expand_corename(cn);
1530 if (ret)
1531 goto expand_fail;
1532
1533out_printf:
1534 cur = cn->corename + cn->used;
1535 va_start(arg, fmt);
1536 vsnprintf(cur, need + 1, fmt, arg);
1537 va_end(arg);
1538 cn->used += need;
1539 return 0;
1540
1541expand_fail:
1542 return ret;
1543}
1544
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545/* format_corename will inspect the pattern parameter, and output a
1546 * name into corename, which must have space for at least
1547 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1548 */
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001549static int format_corename(struct core_name *cn, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550{
David Howells86a264a2008-11-14 10:39:18 +11001551 const struct cred *cred = current_cred();
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001552 const char *pat_ptr = core_pattern;
1553 int ispipe = (*pat_ptr == '|');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 int pid_in_pattern = 0;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001555 int err = 0;
1556
1557 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1558 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1559 cn->used = 0;
1560
1561 if (!cn->corename)
1562 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563
1564 /* Repeat as long as we have more pattern to process and more output
1565 space */
1566 while (*pat_ptr) {
1567 if (*pat_ptr != '%') {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001568 if (*pat_ptr == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 goto out;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001570 err = cn_printf(cn, "%c", *pat_ptr++);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 } else {
1572 switch (*++pat_ptr) {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001573 /* single % at the end, drop that */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 case 0:
1575 goto out;
1576 /* Double percent, output one percent */
1577 case '%':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001578 err = cn_printf(cn, "%c", '%');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 break;
1580 /* pid */
1581 case 'p':
1582 pid_in_pattern = 1;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001583 err = cn_printf(cn, "%d",
1584 task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 break;
1586 /* uid */
1587 case 'u':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001588 err = cn_printf(cn, "%d", cred->uid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 break;
1590 /* gid */
1591 case 'g':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001592 err = cn_printf(cn, "%d", cred->gid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 break;
1594 /* signal that caused the coredump */
1595 case 's':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001596 err = cn_printf(cn, "%ld", signr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 break;
1598 /* UNIX time of coredump */
1599 case 't': {
1600 struct timeval tv;
1601 do_gettimeofday(&tv);
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001602 err = cn_printf(cn, "%lu", tv.tv_sec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 break;
1604 }
1605 /* hostname */
1606 case 'h':
1607 down_read(&uts_sem);
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001608 err = cn_printf(cn, "%s",
1609 utsname()->nodename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 up_read(&uts_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 break;
1612 /* executable */
1613 case 'e':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001614 err = cn_printf(cn, "%s", current->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 break;
Neil Horman74aadce2007-10-16 23:26:35 -07001616 /* core limit size */
1617 case 'c':
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001618 err = cn_printf(cn, "%lu",
1619 rlimit(RLIMIT_CORE));
Neil Horman74aadce2007-10-16 23:26:35 -07001620 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 default:
1622 break;
1623 }
1624 ++pat_ptr;
1625 }
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001626
1627 if (err)
1628 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 }
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001630
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 /* Backward compatibility with core_uses_pid:
1632 *
1633 * If core_pattern does not include a %p (as is the default)
1634 * and core_uses_pid is set, then .%pid will be appended to
Alan Coxc4bbafd2007-04-16 22:53:13 -07001635 * the filename. Do not do this for piped commands. */
Oleg Nesterov64093242008-10-18 20:28:22 -07001636 if (!ispipe && !pid_in_pattern && core_uses_pid) {
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001637 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1638 if (err)
1639 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 }
Alan Coxc4bbafd2007-04-16 22:53:13 -07001641out:
Alan Coxc4bbafd2007-04-16 22:53:13 -07001642 return ispipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643}
1644
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001645static int zap_process(struct task_struct *start, int exit_code)
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001646{
1647 struct task_struct *t;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001648 int nr = 0;
Oleg Nesterov281de332006-06-26 00:26:06 -07001649
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001650 start->signal->flags = SIGNAL_GROUP_EXIT;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001651 start->signal->group_exit_code = exit_code;
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001652 start->signal->group_stop_count = 0;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001653
1654 t = start;
1655 do {
1656 if (t != current && t->mm) {
Oleg Nesterov281de332006-06-26 00:26:06 -07001657 sigaddset(&t->pending.signal, SIGKILL);
1658 signal_wake_up(t, 1);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001659 nr++;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001660 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001661 } while_each_thread(start, t);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001662
1663 return nr;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001664}
1665
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001666static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001667 struct core_state *core_state, int exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668{
1669 struct task_struct *g, *p;
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001670 unsigned long flags;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001671 int nr = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001673 spin_lock_irq(&tsk->sighand->siglock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001674 if (!signal_group_exit(tsk->signal)) {
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001675 mm->core_state = core_state;
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001676 nr = zap_process(tsk, exit_code);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 }
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001678 spin_unlock_irq(&tsk->sighand->siglock);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001679 if (unlikely(nr < 0))
1680 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001682 if (atomic_read(&mm->mm_users) == nr + 1)
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001683 goto done;
Oleg Nesterove4901f92008-07-25 01:47:31 -07001684 /*
1685 * We should find and kill all tasks which use this mm, and we should
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001686 * count them correctly into ->nr_threads. We don't take tasklist
Oleg Nesterove4901f92008-07-25 01:47:31 -07001687 * lock, but this is safe wrt:
1688 *
1689 * fork:
1690 * None of sub-threads can fork after zap_process(leader). All
1691 * processes which were created before this point should be
1692 * visible to zap_threads() because copy_process() adds the new
1693 * process to the tail of init_task.tasks list, and lock/unlock
1694 * of ->siglock provides a memory barrier.
1695 *
1696 * do_exit:
1697 * The caller holds mm->mmap_sem. This means that the task which
1698 * uses this mm can't pass exit_mm(), so it can't exit or clear
1699 * its ->mm.
1700 *
1701 * de_thread:
1702 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1703 * we must see either old or new leader, this does not matter.
1704 * However, it can change p->sighand, so lock_task_sighand(p)
1705 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1706 * it can't fail.
1707 *
1708 * Note also that "g" can be the old leader with ->mm == NULL
1709 * and already unhashed and thus removed from ->thread_group.
1710 * This is OK, __unhash_process()->list_del_rcu() does not
1711 * clear the ->next pointer, we will find the new leader via
1712 * next_thread().
1713 */
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001714 rcu_read_lock();
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001715 for_each_process(g) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001716 if (g == tsk->group_leader)
1717 continue;
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001718 if (g->flags & PF_KTHREAD)
1719 continue;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001720 p = g;
1721 do {
1722 if (p->mm) {
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001723 if (unlikely(p->mm == mm)) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001724 lock_task_sighand(p, &flags);
Oleg Nesterov5c99cbf2010-03-05 13:44:14 -08001725 nr += zap_process(p, exit_code);
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001726 unlock_task_sighand(p, &flags);
1727 }
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001728 break;
1729 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001730 } while_each_thread(g, p);
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001731 }
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001732 rcu_read_unlock();
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001733done:
Oleg Nesterovc5f1cc82008-07-25 01:47:42 -07001734 atomic_set(&core_state->nr_threads, nr);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001735 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736}
1737
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001738static int coredump_wait(int exit_code, struct core_state *core_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739{
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001740 struct task_struct *tsk = current;
1741 struct mm_struct *mm = tsk->mm;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001742 struct completion *vfork_done;
Oleg Nesterov269b0052010-05-26 14:43:08 -07001743 int core_waiters = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001745 init_completion(&core_state->startup);
Oleg Nesterovb564daf2008-07-25 01:47:44 -07001746 core_state->dumper.task = tsk;
1747 core_state->dumper.next = NULL;
Oleg Nesterov269b0052010-05-26 14:43:08 -07001748
1749 down_write(&mm->mmap_sem);
1750 if (!mm->core_state)
1751 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001752 up_write(&mm->mmap_sem);
1753
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001754 if (unlikely(core_waiters < 0))
1755 goto fail;
1756
1757 /*
1758 * Make sure nobody is waiting for us to release the VM,
1759 * otherwise we can deadlock when we wait on each other
1760 */
1761 vfork_done = tsk->vfork_done;
1762 if (vfork_done) {
1763 tsk->vfork_done = NULL;
1764 complete(vfork_done);
1765 }
1766
Oleg Nesterov2384f552005-10-30 15:02:47 -08001767 if (core_waiters)
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001768 wait_for_completion(&core_state->startup);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001769fail:
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001770 return core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771}
1772
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001773static void coredump_finish(struct mm_struct *mm)
1774{
1775 struct core_thread *curr, *next;
1776 struct task_struct *task;
1777
1778 next = mm->core_state->dumper.next;
1779 while ((curr = next) != NULL) {
1780 next = curr->next;
1781 task = curr->task;
1782 /*
1783 * see exit_mm(), curr->task must not see
1784 * ->task == NULL before we read ->next.
1785 */
1786 smp_mb();
1787 curr->task = NULL;
1788 wake_up_process(task);
1789 }
1790
1791 mm->core_state = NULL;
1792}
1793
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001794/*
1795 * set_dumpable converts traditional three-value dumpable to two flags and
1796 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1797 * these bits are not changed atomically. So get_dumpable can observe the
1798 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1799 * return either old dumpable or new one by paying attention to the order of
1800 * modifying the bits.
1801 *
1802 * dumpable | mm->flags (binary)
1803 * old new | initial interim final
1804 * ---------+-----------------------
1805 * 0 1 | 00 01 01
1806 * 0 2 | 00 10(*) 11
1807 * 1 0 | 01 00 00
1808 * 1 2 | 01 11 11
1809 * 2 0 | 11 10(*) 00
1810 * 2 1 | 11 11 01
1811 *
1812 * (*) get_dumpable regards interim value of 10 as 11.
1813 */
1814void set_dumpable(struct mm_struct *mm, int value)
1815{
1816 switch (value) {
1817 case 0:
1818 clear_bit(MMF_DUMPABLE, &mm->flags);
1819 smp_wmb();
1820 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1821 break;
1822 case 1:
1823 set_bit(MMF_DUMPABLE, &mm->flags);
1824 smp_wmb();
1825 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1826 break;
1827 case 2:
1828 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1829 smp_wmb();
1830 set_bit(MMF_DUMPABLE, &mm->flags);
1831 break;
1832 }
1833}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001834
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001835static int __get_dumpable(unsigned long mm_flags)
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001836{
1837 int ret;
1838
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001839 ret = mm_flags & MMF_DUMPABLE_MASK;
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001840 return (ret >= 2) ? 2 : ret;
1841}
1842
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001843int get_dumpable(struct mm_struct *mm)
1844{
1845 return __get_dumpable(mm->flags);
1846}
1847
Neil Horman61be2282009-09-23 15:56:58 -07001848static void wait_for_dump_helpers(struct file *file)
1849{
1850 struct pipe_inode_info *pipe;
1851
1852 pipe = file->f_path.dentry->d_inode->i_pipe;
1853
1854 pipe_lock(pipe);
1855 pipe->readers++;
1856 pipe->writers--;
1857
1858 while ((pipe->readers > 1) && (!signal_pending(current))) {
1859 wake_up_interruptible_sync(&pipe->wait);
1860 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1861 pipe_wait(pipe);
1862 }
1863
1864 pipe->readers--;
1865 pipe->writers++;
1866 pipe_unlock(pipe);
1867
1868}
1869
1870
Neil Horman898b3742010-05-26 14:42:59 -07001871/*
1872 * uhm_pipe_setup
1873 * helper function to customize the process used
1874 * to collect the core in userspace. Specifically
1875 * it sets up a pipe and installs it as fd 0 (stdin)
1876 * for the process. Returns 0 on success, or
1877 * PTR_ERR on failure.
1878 * Note that it also sets the core limit to 1. This
1879 * is a special value that we use to trap recursive
1880 * core dumps
1881 */
1882static int umh_pipe_setup(struct subprocess_info *info)
1883{
1884 struct file *rp, *wp;
1885 struct fdtable *fdt;
1886 struct coredump_params *cp = (struct coredump_params *)info->data;
1887 struct files_struct *cf = current->files;
1888
1889 wp = create_write_pipe(0);
1890 if (IS_ERR(wp))
1891 return PTR_ERR(wp);
1892
1893 rp = create_read_pipe(wp, 0);
1894 if (IS_ERR(rp)) {
1895 free_write_pipe(wp);
1896 return PTR_ERR(rp);
1897 }
1898
1899 cp->file = wp;
1900
1901 sys_close(0);
1902 fd_install(0, rp);
1903 spin_lock(&cf->file_lock);
1904 fdt = files_fdtable(cf);
1905 FD_SET(0, fdt->open_fds);
1906 FD_CLR(0, fdt->close_on_exec);
1907 spin_unlock(&cf->file_lock);
1908
1909 /* and disallow core files too */
1910 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1911
1912 return 0;
1913}
1914
WANG Cong8cd3ac32009-01-06 14:42:48 -08001915void do_coredump(long signr, int exit_code, struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916{
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001917 struct core_state core_state;
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001918 struct core_name cn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 struct mm_struct *mm = current->mm;
1920 struct linux_binfmt * binfmt;
David Howellsd84f4f92008-11-14 10:39:23 +11001921 const struct cred *old_cred;
1922 struct cred *cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 int retval = 0;
Alan Coxd6e71142005-06-23 00:09:43 -07001924 int flag = 0;
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07001925 int ispipe;
Neil Hormana2939802009-09-23 15:56:56 -07001926 static atomic_t core_dump_count = ATOMIC_INIT(0);
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001927 struct coredump_params cprm = {
1928 .signr = signr,
1929 .regs = regs,
Jiri Slabyd554ed892010-03-05 13:42:42 -08001930 .limit = rlimit(RLIMIT_CORE),
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001931 /*
1932 * We must use the same mm->flags while dumping core to avoid
1933 * inconsistency of bit flags, since this flag is not protected
1934 * by any locks.
1935 */
1936 .mm_flags = mm->flags,
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001937 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938
Steve Grubb0a4ff8c2007-04-19 10:28:21 -04001939 audit_core_dumps(signr);
1940
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001941 binfmt = mm->binfmt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 if (!binfmt || !binfmt->core_dump)
1943 goto fail;
Oleg Nesterov269b0052010-05-26 14:43:08 -07001944 if (!__get_dumpable(cprm.mm_flags))
1945 goto fail;
David Howellsd84f4f92008-11-14 10:39:23 +11001946
1947 cred = prepare_creds();
Oleg Nesterov5e43aef2010-05-26 14:43:07 -07001948 if (!cred)
David Howellsd84f4f92008-11-14 10:39:23 +11001949 goto fail;
Alan Coxd6e71142005-06-23 00:09:43 -07001950 /*
1951 * We cannot trust fsuid as being the "true" uid of the
1952 * process nor do we know its entire history. We only know it
1953 * was tainted so we dump it as root in mode 2.
1954 */
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001955 if (__get_dumpable(cprm.mm_flags) == 2) {
1956 /* Setuid core dump mode */
Alan Coxd6e71142005-06-23 00:09:43 -07001957 flag = O_EXCL; /* Stop rewrite attacks */
David Howellsd84f4f92008-11-14 10:39:23 +11001958 cred->fsuid = 0; /* Dump root private */
Alan Coxd6e71142005-06-23 00:09:43 -07001959 }
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001960
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001961 retval = coredump_wait(exit_code, &core_state);
Oleg Nesterov5e43aef2010-05-26 14:43:07 -07001962 if (retval < 0)
1963 goto fail_creds;
David Howellsd84f4f92008-11-14 10:39:23 +11001964
1965 old_cred = override_creds(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966
1967 /*
1968 * Clear any false indication of pending signals that might
1969 * be seen by the filesystem code called to write the core file.
1970 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 clear_thread_flag(TIF_SIGPENDING);
1972
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07001973 ispipe = format_corename(&cn, signr);
1974
1975 if (ispipe == -ENOMEM) {
1976 printk(KERN_WARNING "format_corename failed\n");
1977 printk(KERN_WARNING "Aborting core\n");
1978 goto fail_corename;
1979 }
Neil Horman725eae32009-09-23 15:56:54 -07001980
Alan Coxc4bbafd2007-04-16 22:53:13 -07001981 if (ispipe) {
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07001982 int dump_count;
1983 char **helper_argv;
1984
Neil Horman898b3742010-05-26 14:42:59 -07001985 if (cprm.limit == 1) {
Neil Horman725eae32009-09-23 15:56:54 -07001986 /*
1987 * Normally core limits are irrelevant to pipes, since
1988 * we're not writing to the file system, but we use
Neil Horman898b3742010-05-26 14:42:59 -07001989 * cprm.limit of 1 here as a speacial value. Any
1990 * non-1 limit gets set to RLIM_INFINITY below, but
Neil Horman725eae32009-09-23 15:56:54 -07001991 * a limit of 0 skips the dump. This is a consistent
1992 * way to catch recursive crashes. We can still crash
Neil Horman898b3742010-05-26 14:42:59 -07001993 * if the core_pattern binary sets RLIM_CORE = !1
Neil Horman725eae32009-09-23 15:56:54 -07001994 * but it runs as root, and can do lots of stupid things
1995 * Note that we use task_tgid_vnr here to grab the pid
1996 * of the process group leader. That way we get the
1997 * right pid if a thread in a multi-threaded
1998 * core_pattern process dies.
1999 */
2000 printk(KERN_WARNING
Neil Horman898b3742010-05-26 14:42:59 -07002001 "Process %d(%s) has RLIMIT_CORE set to 1\n",
Neil Horman725eae32009-09-23 15:56:54 -07002002 task_tgid_vnr(current), current->comm);
2003 printk(KERN_WARNING "Aborting core\n");
2004 goto fail_unlock;
2005 }
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002006 cprm.limit = RLIM_INFINITY;
Neil Horman725eae32009-09-23 15:56:54 -07002007
Neil Hormana2939802009-09-23 15:56:56 -07002008 dump_count = atomic_inc_return(&core_dump_count);
2009 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2010 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2011 task_tgid_vnr(current), current->comm);
2012 printk(KERN_WARNING "Skipping core dump\n");
2013 goto fail_dropcount;
2014 }
2015
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002016 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
Tetsuo Handa350eaf72009-01-06 14:41:11 -08002017 if (!helper_argv) {
2018 printk(KERN_WARNING "%s failed to allocate memory\n",
2019 __func__);
Neil Hormana2939802009-09-23 15:56:56 -07002020 goto fail_dropcount;
Tetsuo Handa350eaf72009-01-06 14:41:11 -08002021 }
Neil Horman32321132007-10-16 23:26:36 -07002022
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002023 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2024 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2025 NULL, &cprm);
2026 argv_free(helper_argv);
2027 if (retval) {
Andi Kleend025c9d2006-09-30 23:29:28 -07002028 printk(KERN_INFO "Core dump to %s pipe failed\n",
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002029 cn.corename);
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002030 goto close_fail;
Andi Kleend025c9d2006-09-30 23:29:28 -07002031 }
Oleg Nesterovc7135412010-05-26 14:43:05 -07002032 } else {
2033 struct inode *inode;
2034
2035 if (cprm.limit < binfmt->min_coredump)
2036 goto fail_unlock;
2037
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002038 cprm.file = filp_open(cn.corename,
Alexey Dobriyan6d4df672006-12-06 20:40:39 -08002039 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2040 0600);
Oleg Nesterovc7135412010-05-26 14:43:05 -07002041 if (IS_ERR(cprm.file))
2042 goto fail_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
Oleg Nesterovc7135412010-05-26 14:43:05 -07002044 inode = cprm.file->f_path.dentry->d_inode;
2045 if (inode->i_nlink > 1)
2046 goto close_fail;
2047 if (d_unhashed(cprm.file->f_path.dentry))
2048 goto close_fail;
2049 /*
2050 * AK: actually i see no reason to not allow this for named
2051 * pipes etc, but keep the previous behaviour for now.
2052 */
2053 if (!S_ISREG(inode->i_mode))
2054 goto close_fail;
2055 /*
2056 * Dont allow local users get cute and trick others to coredump
2057 * into their pre-created files.
2058 */
2059 if (inode->i_uid != current_fsuid())
2060 goto close_fail;
2061 if (!cprm.file->f_op || !cprm.file->f_op->write)
2062 goto close_fail;
2063 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2064 goto close_fail;
2065 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002067 retval = binfmt->core_dump(&cprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 if (retval)
2069 current->signal->group_exit_code |= 0x80;
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002070
Neil Horman61be2282009-09-23 15:56:58 -07002071 if (ispipe && core_pipe_limit)
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002072 wait_for_dump_helpers(cprm.file);
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002073close_fail:
2074 if (cprm.file)
2075 filp_close(cprm.file, NULL);
Neil Hormana2939802009-09-23 15:56:56 -07002076fail_dropcount:
Oleg Nesterovd5bf4c42010-05-26 14:43:06 -07002077 if (ispipe)
Neil Hormana2939802009-09-23 15:56:56 -07002078 atomic_dec(&core_dump_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079fail_unlock:
Xiaotian Feng1b0d3002010-10-27 15:34:08 -07002080 kfree(cn.corename);
2081fail_corename:
Oleg Nesterova94e2d42008-07-25 01:47:46 -07002082 coredump_finish(mm);
Oleg Nesterov5e43aef2010-05-26 14:43:07 -07002083 revert_creds(old_cred);
2084fail_creds:
2085 put_cred(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086fail:
WANG Cong8cd3ac32009-01-06 14:42:48 -08002087 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088}
Linus Torvalds3aa0ce82010-10-14 14:32:06 -07002089
2090/*
2091 * Core dumping helper functions. These are the only things you should
2092 * do on a core-file: use only these functions to write out all the
2093 * necessary info.
2094 */
2095int dump_write(struct file *file, const void *addr, int nr)
2096{
2097 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2098}
Linus Torvalds8fd01d62010-10-14 19:15:28 -07002099EXPORT_SYMBOL(dump_write);
Linus Torvalds3aa0ce82010-10-14 14:32:06 -07002100
2101int dump_seek(struct file *file, loff_t off)
2102{
2103 int ret = 1;
2104
2105 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2106 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2107 return 0;
2108 } else {
2109 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2110
2111 if (!buf)
2112 return 0;
2113 while (off > 0) {
2114 unsigned long n = off;
2115
2116 if (n > PAGE_SIZE)
2117 n = PAGE_SIZE;
2118 if (!dump_write(file, buf, n)) {
2119 ret = 0;
2120 break;
2121 }
2122 off -= n;
2123 }
2124 free_page((unsigned long)buf);
2125 }
2126 return ret;
2127}
Linus Torvalds8fd01d62010-10-14 19:15:28 -07002128EXPORT_SYMBOL(dump_seek);