blob: 5e559013e303fbcb6f9666828cca450ab57de7f9 [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>
31#include <linux/smp_lock.h>
Hugh Dickinsba92a432008-07-25 01:45:43 -070032#include <linux/swap.h>
Neil Horman74aadce2007-10-16 23:26:35 -070033#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/highmem.h>
36#include <linux/spinlock.h>
37#include <linux/key.h>
38#include <linux/personality.h>
39#include <linux/binfmts.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <linux/utsname.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080041#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/module.h>
43#include <linux/namei.h>
44#include <linux/proc_fs.h>
45#include <linux/ptrace.h>
46#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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53#include <asm/uaccess.h>
54#include <asm/mmu_context.h>
Ollie Wildb6a2fea2007-07-19 01:48:16 -070055#include <asm/tlb.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
57#ifdef CONFIG_KMOD
58#include <linux/kmod.h>
59#endif
60
David Woodhouse702773b2008-06-16 12:11:54 +010061#ifdef __alpha__
62/* for /sbin/loader handling in search_binary_handler() */
63#include <linux/a.out.h>
64#endif
65
Linus Torvalds1da177e2005-04-16 15:20:36 -070066int core_uses_pid;
Dan Aloni71ce92f2007-05-16 22:11:16 -070067char core_pattern[CORENAME_MAX_SIZE] = "core";
Alan Coxd6e71142005-06-23 00:09:43 -070068int suid_dumpable = 0;
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070/* The maximal length of core_pattern is also specified in sysctl.c */
71
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070072static LIST_HEAD(formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070073static DEFINE_RWLOCK(binfmt_lock);
74
75int register_binfmt(struct linux_binfmt * fmt)
76{
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 if (!fmt)
78 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 write_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070080 list_add(&fmt->lh, &formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 write_unlock(&binfmt_lock);
82 return 0;
83}
84
85EXPORT_SYMBOL(register_binfmt);
86
Alexey Dobriyanf6b450d2007-10-16 23:26:04 -070087void unregister_binfmt(struct linux_binfmt * fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 write_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070090 list_del(&fmt->lh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 write_unlock(&binfmt_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092}
93
94EXPORT_SYMBOL(unregister_binfmt);
95
96static inline void put_binfmt(struct linux_binfmt * fmt)
97{
98 module_put(fmt->module);
99}
100
101/*
102 * Note that a shared library must be both readable and executable due to
103 * security reasons.
104 *
105 * Also note that we take the address to load from from the file itself.
106 */
107asmlinkage long sys_uselib(const char __user * library)
108{
109 struct file * file;
110 struct nameidata nd;
111 int error;
112
Oleg Drokinb5005312006-03-25 03:07:01 -0800113 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 if (error)
115 goto out;
116
117 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -0800118 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 goto exit;
120
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800121 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 if (error)
123 goto exit;
124
Andi Kleenabe8be32008-02-08 04:20:23 -0800125 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 error = PTR_ERR(file);
127 if (IS_ERR(file))
128 goto out;
129
130 error = -ENOEXEC;
131 if(file->f_op) {
132 struct linux_binfmt * fmt;
133
134 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -0700135 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 if (!fmt->load_shlib)
137 continue;
138 if (!try_module_get(fmt->module))
139 continue;
140 read_unlock(&binfmt_lock);
141 error = fmt->load_shlib(file);
142 read_lock(&binfmt_lock);
143 put_binfmt(fmt);
144 if (error != -ENOEXEC)
145 break;
146 }
147 read_unlock(&binfmt_lock);
148 }
149 fput(file);
150out:
151 return error;
152exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -0700153 release_open_intent(&nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800154 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 goto out;
156}
157
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700158#ifdef CONFIG_MMU
159
160static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
161 int write)
162{
163 struct page *page;
164 int ret;
165
166#ifdef CONFIG_STACK_GROWSUP
167 if (write) {
168 ret = expand_stack_downwards(bprm->vma, pos);
169 if (ret < 0)
170 return NULL;
171 }
172#endif
173 ret = get_user_pages(current, bprm->mm, pos,
174 1, write, 1, &page, NULL);
175 if (ret <= 0)
176 return NULL;
177
178 if (write) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700179 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800180 struct rlimit *rlim;
181
182 /*
183 * We've historically supported up to 32 pages (ARG_MAX)
184 * of argument strings even with small stacks
185 */
186 if (size <= ARG_MAX)
187 return page;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700188
189 /*
190 * Limit to 1/4-th the stack size for the argv+env strings.
191 * This ensures that:
192 * - the remaining binfmt code will not run out of stack space,
193 * - the program will have a reasonable amount of stack left
194 * to work from.
195 */
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800196 rlim = current->signal->rlim;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700197 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
198 put_page(page);
199 return NULL;
200 }
201 }
202
203 return page;
204}
205
206static void put_arg_page(struct page *page)
207{
208 put_page(page);
209}
210
211static void free_arg_page(struct linux_binprm *bprm, int i)
212{
213}
214
215static void free_arg_pages(struct linux_binprm *bprm)
216{
217}
218
219static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
220 struct page *page)
221{
222 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
223}
224
225static int __bprm_mm_init(struct linux_binprm *bprm)
226{
227 int err = -ENOMEM;
228 struct vm_area_struct *vma = NULL;
229 struct mm_struct *mm = bprm->mm;
230
231 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
232 if (!vma)
233 goto err;
234
235 down_write(&mm->mmap_sem);
236 vma->vm_mm = mm;
237
238 /*
239 * Place the stack at the largest stack address the architecture
240 * supports. Later, we'll move this to an appropriate place. We don't
241 * use STACK_TOP because that can depend on attributes which aren't
242 * configured yet.
243 */
244 vma->vm_end = STACK_TOP_MAX;
245 vma->vm_start = vma->vm_end - PAGE_SIZE;
246
247 vma->vm_flags = VM_STACK_FLAGS;
Coly Li3ed75eb2007-10-18 23:39:15 -0700248 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700249 err = insert_vm_struct(mm, vma);
250 if (err) {
251 up_write(&mm->mmap_sem);
252 goto err;
253 }
254
255 mm->stack_vm = mm->total_vm = 1;
256 up_write(&mm->mmap_sem);
257
258 bprm->p = vma->vm_end - sizeof(void *);
259
260 return 0;
261
262err:
263 if (vma) {
264 bprm->vma = NULL;
265 kmem_cache_free(vm_area_cachep, vma);
266 }
267
268 return err;
269}
270
271static bool valid_arg_len(struct linux_binprm *bprm, long len)
272{
273 return len <= MAX_ARG_STRLEN;
274}
275
276#else
277
278static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
279 int write)
280{
281 struct page *page;
282
283 page = bprm->page[pos / PAGE_SIZE];
284 if (!page && write) {
285 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
286 if (!page)
287 return NULL;
288 bprm->page[pos / PAGE_SIZE] = page;
289 }
290
291 return page;
292}
293
294static void put_arg_page(struct page *page)
295{
296}
297
298static void free_arg_page(struct linux_binprm *bprm, int i)
299{
300 if (bprm->page[i]) {
301 __free_page(bprm->page[i]);
302 bprm->page[i] = NULL;
303 }
304}
305
306static void free_arg_pages(struct linux_binprm *bprm)
307{
308 int i;
309
310 for (i = 0; i < MAX_ARG_PAGES; i++)
311 free_arg_page(bprm, i);
312}
313
314static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
315 struct page *page)
316{
317}
318
319static int __bprm_mm_init(struct linux_binprm *bprm)
320{
321 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
322 return 0;
323}
324
325static bool valid_arg_len(struct linux_binprm *bprm, long len)
326{
327 return len <= bprm->p;
328}
329
330#endif /* CONFIG_MMU */
331
332/*
333 * Create a new mm_struct and populate it with a temporary stack
334 * vm_area_struct. We don't have enough context at this point to set the stack
335 * flags, permissions, and offset, so we use temporary values. We'll update
336 * them later in setup_arg_pages().
337 */
338int bprm_mm_init(struct linux_binprm *bprm)
339{
340 int err;
341 struct mm_struct *mm = NULL;
342
343 bprm->mm = mm = mm_alloc();
344 err = -ENOMEM;
345 if (!mm)
346 goto err;
347
348 err = init_new_context(current, mm);
349 if (err)
350 goto err;
351
352 err = __bprm_mm_init(bprm);
353 if (err)
354 goto err;
355
356 return 0;
357
358err:
359 if (mm) {
360 bprm->mm = NULL;
361 mmdrop(mm);
362 }
363
364 return err;
365}
366
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367/*
368 * count() counts the number of strings in array ARGV.
369 */
370static int count(char __user * __user * argv, int max)
371{
372 int i = 0;
373
374 if (argv != NULL) {
375 for (;;) {
376 char __user * p;
377
378 if (get_user(p, argv))
379 return -EFAULT;
380 if (!p)
381 break;
382 argv++;
383 if(++i > max)
384 return -E2BIG;
385 cond_resched();
386 }
387 }
388 return i;
389}
390
391/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700392 * 'copy_strings()' copies argument/environment strings from the old
393 * processes's memory to the new process's stack. The call to get_user_pages()
394 * ensures the destination page is created and not swapped out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 */
Adrian Bunk75c96f82005-05-05 16:16:09 -0700396static int copy_strings(int argc, char __user * __user * argv,
397 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
399 struct page *kmapped_page = NULL;
400 char *kaddr = NULL;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700401 unsigned long kpos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 int ret;
403
404 while (argc-- > 0) {
405 char __user *str;
406 int len;
407 unsigned long pos;
408
409 if (get_user(str, argv+argc) ||
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700410 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 ret = -EFAULT;
412 goto out;
413 }
414
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700415 if (!valid_arg_len(bprm, len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 ret = -E2BIG;
417 goto out;
418 }
419
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700420 /* We're going to work our way backwords. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 pos = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700422 str += len;
423 bprm->p -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
425 while (len > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 int offset, bytes_to_copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
428 offset = pos % PAGE_SIZE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700429 if (offset == 0)
430 offset = PAGE_SIZE;
431
432 bytes_to_copy = offset;
433 if (bytes_to_copy > len)
434 bytes_to_copy = len;
435
436 offset -= bytes_to_copy;
437 pos -= bytes_to_copy;
438 str -= bytes_to_copy;
439 len -= bytes_to_copy;
440
441 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
442 struct page *page;
443
444 page = get_arg_page(bprm, pos, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 if (!page) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700446 ret = -E2BIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 goto out;
448 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700450 if (kmapped_page) {
451 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700453 put_arg_page(kmapped_page);
454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 kmapped_page = page;
456 kaddr = kmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700457 kpos = pos & PAGE_MASK;
458 flush_arg_page(bprm, kpos, kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700460 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 ret = -EFAULT;
462 goto out;
463 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 }
465 }
466 ret = 0;
467out:
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700468 if (kmapped_page) {
469 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700471 put_arg_page(kmapped_page);
472 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 return ret;
474}
475
476/*
477 * Like copy_strings, but get argv and its values from kernel memory.
478 */
479int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
480{
481 int r;
482 mm_segment_t oldfs = get_fs();
483 set_fs(KERNEL_DS);
484 r = copy_strings(argc, (char __user * __user *)argv, bprm);
485 set_fs(oldfs);
486 return r;
487}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488EXPORT_SYMBOL(copy_strings_kernel);
489
490#ifdef CONFIG_MMU
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700491
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700493 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
494 * the binfmt code determines where the new stack should reside, we shift it to
495 * its final location. The process proceeds as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 *
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700497 * 1) Use shift to calculate the new vma endpoints.
498 * 2) Extend vma to cover both the old and new ranges. This ensures the
499 * arguments passed to subsequent functions are consistent.
500 * 3) Move vma's page tables to the new range.
501 * 4) Free up any cleared pgd range.
502 * 5) Shrink the vma to cover only the new range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700504static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505{
506 struct mm_struct *mm = vma->vm_mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700507 unsigned long old_start = vma->vm_start;
508 unsigned long old_end = vma->vm_end;
509 unsigned long length = old_end - old_start;
510 unsigned long new_start = old_start - shift;
511 unsigned long new_end = old_end - shift;
512 struct mmu_gather *tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700514 BUG_ON(new_start > new_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700516 /*
517 * ensure there are no vmas between where we want to go
518 * and where we are
519 */
520 if (vma != find_vma(mm, new_start))
521 return -EFAULT;
522
523 /*
524 * cover the whole range: [new_start, old_end)
525 */
526 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
527
528 /*
529 * move the page tables downwards, on failure we rely on
530 * process cleanup to remove whatever mess we made.
531 */
532 if (length != move_page_tables(vma, old_start,
533 vma, new_start, length))
534 return -ENOMEM;
535
536 lru_add_drain();
537 tlb = tlb_gather_mmu(mm, 0);
538 if (new_end > old_start) {
539 /*
540 * when the old and new regions overlap clear from new_end.
541 */
Jan Beulich42b77722008-07-23 21:27:10 -0700542 free_pgd_range(tlb, new_end, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700543 vma->vm_next ? vma->vm_next->vm_start : 0);
544 } else {
545 /*
546 * otherwise, clean from old_start; this is done to not touch
547 * the address space in [new_end, old_start) some architectures
548 * have constraints on va-space that make this illegal (IA64) -
549 * for the others its just a little faster.
550 */
Jan Beulich42b77722008-07-23 21:27:10 -0700551 free_pgd_range(tlb, old_start, old_end, new_end,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700552 vma->vm_next ? vma->vm_next->vm_start : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700554 tlb_finish_mmu(tlb, new_end, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700556 /*
557 * shrink the vma to just the new range.
558 */
559 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
560
561 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562}
563
564#define EXTRA_STACK_VM_PAGES 20 /* random */
565
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700566/*
567 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
568 * the stack is optionally relocated, and some extra space is added.
569 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570int setup_arg_pages(struct linux_binprm *bprm,
571 unsigned long stack_top,
572 int executable_stack)
573{
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700574 unsigned long ret;
575 unsigned long stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 struct mm_struct *mm = current->mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700577 struct vm_area_struct *vma = bprm->vma;
578 struct vm_area_struct *prev = NULL;
579 unsigned long vm_flags;
580 unsigned long stack_base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
582#ifdef CONFIG_STACK_GROWSUP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 /* Limit stack size to 1GB */
584 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
585 if (stack_base > (1 << 30))
586 stack_base = 1 << 30;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700588 /* Make sure we didn't let the argument array grow too large. */
589 if (vma->vm_end - vma->vm_start > stack_base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 return -ENOMEM;
591
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700592 stack_base = PAGE_ALIGN(stack_top - stack_base);
593
594 stack_shift = vma->vm_start - stack_base;
595 mm->arg_start = bprm->p - stack_shift;
596 bprm->p = vma->vm_end - stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597#else
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700598 stack_top = arch_align_stack(stack_top);
599 stack_top = PAGE_ALIGN(stack_top);
600 stack_shift = vma->vm_end - stack_top;
601
602 bprm->p -= stack_shift;
603 mm->arg_start = bprm->p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604#endif
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700605
606 if (bprm->loader)
607 bprm->loader -= stack_shift;
608 bprm->exec -= stack_shift;
609
610 down_write(&mm->mmap_sem);
Hugh Dickins96a8e132008-07-10 21:19:20 +0100611 vm_flags = VM_STACK_FLAGS;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700612
613 /*
614 * Adjust stack execute permissions; explicitly enable for
615 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
616 * (arch default) otherwise.
617 */
618 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
619 vm_flags |= VM_EXEC;
620 else if (executable_stack == EXSTACK_DISABLE_X)
621 vm_flags &= ~VM_EXEC;
622 vm_flags |= mm->def_flags;
623
624 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
625 vm_flags);
626 if (ret)
627 goto out_unlock;
628 BUG_ON(prev != vma);
629
630 /* Move stack pages down in memory. */
631 if (stack_shift) {
632 ret = shift_arg_pages(vma, stack_shift);
633 if (ret) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 up_write(&mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 return ret;
636 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 }
638
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700639#ifdef CONFIG_STACK_GROWSUP
640 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
641#else
642 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
643#endif
644 ret = expand_stack(vma, stack_base);
645 if (ret)
646 ret = -EFAULT;
647
648out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 up_write(&mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 return 0;
651}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652EXPORT_SYMBOL(setup_arg_pages);
653
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654#endif /* CONFIG_MMU */
655
656struct file *open_exec(const char *name)
657{
658 struct nameidata nd;
659 int err;
660 struct file *file;
661
Oleg Drokinb5005312006-03-25 03:07:01 -0800662 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 file = ERR_PTR(err);
664
665 if (!err) {
Jan Blunck4ac91372008-02-14 19:34:32 -0800666 struct inode *inode = nd.path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 file = ERR_PTR(-EACCES);
Miklos Szeredi1a159dd2007-10-16 23:27:08 -0700668 if (S_ISREG(inode->i_mode)) {
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800669 int err = vfs_permission(&nd, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 file = ERR_PTR(err);
671 if (!err) {
Andi Kleenabe8be32008-02-08 04:20:23 -0800672 file = nameidata_to_filp(&nd,
673 O_RDONLY|O_LARGEFILE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 if (!IS_ERR(file)) {
675 err = deny_write_access(file);
676 if (err) {
677 fput(file);
678 file = ERR_PTR(err);
679 }
680 }
681out:
682 return file;
683 }
684 }
Trond Myklebust834f2a42005-10-18 14:20:16 -0700685 release_open_intent(&nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800686 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 }
688 goto out;
689}
690
691EXPORT_SYMBOL(open_exec);
692
693int kernel_read(struct file *file, unsigned long offset,
694 char *addr, unsigned long count)
695{
696 mm_segment_t old_fs;
697 loff_t pos = offset;
698 int result;
699
700 old_fs = get_fs();
701 set_fs(get_ds());
702 /* The cast to a user pointer is valid due to the set_fs() */
703 result = vfs_read(file, (void __user *)addr, count, &pos);
704 set_fs(old_fs);
705 return result;
706}
707
708EXPORT_SYMBOL(kernel_read);
709
710static int exec_mmap(struct mm_struct *mm)
711{
712 struct task_struct *tsk;
713 struct mm_struct * old_mm, *active_mm;
714
715 /* Notify parent that we're no longer interested in the old VM */
716 tsk = current;
717 old_mm = current->mm;
718 mm_release(tsk, old_mm);
719
720 if (old_mm) {
721 /*
722 * Make sure that if there is a core dump in progress
723 * for the old mm, we get out and die instead of going
724 * through with the exec. We must hold mmap_sem around
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700725 * checking core_state and changing tsk->mm.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 */
727 down_read(&old_mm->mmap_sem);
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700728 if (unlikely(old_mm->core_state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 up_read(&old_mm->mmap_sem);
730 return -EINTR;
731 }
732 }
733 task_lock(tsk);
734 active_mm = tsk->active_mm;
735 tsk->mm = mm;
736 tsk->active_mm = mm;
737 activate_mm(active_mm, mm);
738 task_unlock(tsk);
KOSAKI Motohiro4cd1a8f2008-05-12 14:02:31 -0700739 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 arch_pick_mmap_layout(mm);
741 if (old_mm) {
742 up_read(&old_mm->mmap_sem);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +0200743 BUG_ON(active_mm != old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 mmput(old_mm);
745 return 0;
746 }
747 mmdrop(active_mm);
748 return 0;
749}
750
751/*
752 * This function makes sure the current process has its own signal table,
753 * so that flush_signal_handlers can later reset the handlers without
754 * disturbing other processes. (Other processes might share the signal
755 * table via the CLONE_SIGHAND option to clone().)
756 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800757static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758{
759 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700760 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 spinlock_t *lock = &oldsighand->siglock;
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300762 struct task_struct *leader = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 int count;
764
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700765 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 goto no_thread_group;
767
768 /*
769 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800772 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 /*
774 * Another group action in progress, just
775 * return so that the signal is processed.
776 */
777 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 return -EAGAIN;
779 }
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800780 sig->group_exit_task = tsk;
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700781 zap_other_threads(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
Oleg Nesterovfea9d172008-02-08 04:19:19 -0800783 /* Account for the thread group leader hanging around: */
784 count = thread_group_leader(tsk) ? 1 : 2;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700785 sig->notify_count = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 while (atomic_read(&sig->count) > count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 __set_current_state(TASK_UNINTERRUPTIBLE);
788 spin_unlock_irq(lock);
789 schedule();
790 spin_lock_irq(lock);
791 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 spin_unlock_irq(lock);
793
794 /*
795 * At this point all other threads have exited, all we have to
796 * do is to wait for the thread group leader to become inactive,
797 * and to assume its PID:
798 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700799 if (!thread_group_leader(tsk)) {
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700800 leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700801
Oleg Nesterov2800d8d2008-04-30 00:53:12 -0700802 sig->notify_count = -1; /* for exit_notify() */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700803 for (;;) {
804 write_lock_irq(&tasklist_lock);
805 if (likely(leader->exit_state))
806 break;
807 __set_current_state(TASK_UNINTERRUPTIBLE);
808 write_unlock_irq(&tasklist_lock);
809 schedule();
810 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
Oleg Nesterov7a5e8732008-04-30 00:53:04 -0700812 if (unlikely(task_child_reaper(tsk) == leader))
813 task_active_pid_ns(tsk)->child_reaper = tsk;
Roland McGrathf5e90282006-04-10 22:54:16 -0700814 /*
815 * The only record we have of the real-time age of a
816 * process, regardless of execs it's done, is start_time.
817 * All the past CPU time is accumulated in signal_struct
818 * from sister threads now dead. But in this non-leader
819 * exec, nothing survives from the original leader thread,
820 * whose birth marks the true age of this process now.
821 * When we take on its identity by switching to its PID, we
822 * also take its birthdate (always earlier than our own).
823 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700824 tsk->start_time = leader->start_time;
Roland McGrathf5e90282006-04-10 22:54:16 -0700825
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700826 BUG_ON(!same_thread_group(leader, tsk));
827 BUG_ON(has_group_leader_pid(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 /*
829 * An exec() starts a new thread group with the
830 * TGID of the previous thread group. Rehash the
831 * two threads with a switched PID, and release
832 * the former thread group leader:
833 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800834
835 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700836 * The old leader becomes a thread of the this thread group.
837 * Note: The old leader also uses this pid until release_task
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800838 * is called. Odd but simple and correct.
839 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700840 detach_pid(tsk, PIDTYPE_PID);
841 tsk->pid = leader->pid;
Sukadev Bhattiprolu3743ca02007-10-18 23:39:51 -0700842 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700843 transfer_pid(leader, tsk, PIDTYPE_PGID);
844 transfer_pid(leader, tsk, PIDTYPE_SID);
845 list_replace_rcu(&leader->tasks, &tsk->tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700847 tsk->group_leader = tsk;
848 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -0600849
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700850 tsk->exit_signal = SIGCHLD;
Oleg Nesterov962b5642005-11-23 13:37:43 -0800851
852 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
853 leader->exit_state = EXIT_DEAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
855 write_unlock_irq(&tasklist_lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700858 sig->group_exit_task = NULL;
859 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
861no_thread_group:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 exit_itimers(sig);
Oleg Nesterovcbaffba2008-05-26 20:55:42 +0400863 flush_itimer_signals();
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300864 if (leader)
865 release_task(leader);
866
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700867 if (atomic_read(&oldsighand->count) != 1) {
868 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700870 * This ->sighand is shared with the CLONE_SIGHAND
871 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700873 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
874 if (!newsighand)
875 return -ENOMEM;
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 atomic_set(&newsighand->count, 1);
878 memcpy(newsighand->action, oldsighand->action,
879 sizeof(newsighand->action));
880
881 write_lock_irq(&tasklist_lock);
882 spin_lock(&oldsighand->siglock);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700883 rcu_assign_pointer(tsk->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 spin_unlock(&oldsighand->siglock);
885 write_unlock_irq(&tasklist_lock);
886
Davide Libenzifba2afa2007-05-10 22:23:13 -0700887 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 }
889
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700890 BUG_ON(!thread_group_leader(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 return 0;
892}
Oleg Nesterov0840a902007-10-16 23:27:22 -0700893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894/*
895 * These functions flushes out all traces of the currently running executable
896 * so that a new one can be started
897 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800898static void flush_old_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899{
900 long j = -1;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700901 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
903 spin_lock(&files->file_lock);
904 for (;;) {
905 unsigned long set, i;
906
907 j++;
908 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700909 fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800910 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700912 set = fdt->close_on_exec->fds_bits[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 if (!set)
914 continue;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700915 fdt->close_on_exec->fds_bits[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 spin_unlock(&files->file_lock);
917 for ( ; set ; i++,set >>= 1) {
918 if (set & 1) {
919 sys_close(i);
920 }
921 }
922 spin_lock(&files->file_lock);
923
924 }
925 spin_unlock(&files->file_lock);
926}
927
Andrew Morton59714d62008-02-04 22:27:21 -0800928char *get_task_comm(char *buf, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929{
930 /* buf must be at least sizeof(tsk->comm) in size */
931 task_lock(tsk);
932 strncpy(buf, tsk->comm, sizeof(tsk->comm));
933 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -0800934 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935}
936
937void set_task_comm(struct task_struct *tsk, char *buf)
938{
939 task_lock(tsk);
940 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
941 task_unlock(tsk);
942}
943
944int flush_old_exec(struct linux_binprm * bprm)
945{
946 char * name;
947 int i, ch, retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 char tcomm[sizeof(current->comm)];
949
950 /*
951 * Make sure we have a private signal table and that
952 * we are unassociated from the previous thread group.
953 */
954 retval = de_thread(current);
955 if (retval)
956 goto out;
957
Matt Helsley925d1c42008-04-29 01:01:36 -0700958 set_mm_exe_file(bprm->mm, bprm->file);
959
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 * Release all of the old mmap stuff
962 */
963 retval = exec_mmap(bprm->mm);
964 if (retval)
Al Virofd8328b2008-04-22 05:11:59 -0400965 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
967 bprm->mm = NULL; /* We're using it now */
968
969 /* This is the point of no return */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 current->sas_ss_sp = current->sas_ss_size = 0;
971
972 if (current->euid == current->uid && current->egid == current->gid)
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -0700973 set_dumpable(current->mm, 1);
Alan Coxd6e71142005-06-23 00:09:43 -0700974 else
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -0700975 set_dumpable(current->mm, suid_dumpable);
Alan Coxd6e71142005-06-23 00:09:43 -0700976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 name = bprm->filename;
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -0700978
979 /* Copies the binary name from after last slash */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 for (i=0; (ch = *(name++)) != '\0';) {
981 if (ch == '/')
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -0700982 i = 0; /* overwrite what we wrote */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 else
984 if (i < (sizeof(tcomm) - 1))
985 tcomm[i++] = ch;
986 }
987 tcomm[i] = '\0';
988 set_task_comm(current, tcomm);
989
990 current->flags &= ~PF_RANDOMIZE;
991 flush_thread();
992
Benjamin Herrenschmidt0551fbd2006-02-28 16:59:19 -0800993 /* Set the new mm task size. We have to do that late because it may
994 * depend on TIF_32BIT which is only updated in flush_thread() on
995 * some architectures like powerpc
996 */
997 current->mm->task_size = TASK_SIZE;
998
Marcel Holtmannd2d56c52007-08-17 21:47:58 +0200999 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1000 suid_keys(current);
1001 set_dumpable(current->mm, suid_dumpable);
1002 current->pdeath_signal = 0;
1003 } else if (file_permission(bprm->file, MAY_READ) ||
1004 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 suid_keys(current);
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001006 set_dumpable(current->mm, suid_dumpable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 }
1008
1009 /* An exec changes our domain. We are no longer part of the thread
1010 group */
1011
1012 current->self_exec_id++;
1013
1014 flush_signal_handlers(current, 0);
1015 flush_old_files(current->files);
1016
1017 return 0;
1018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019out:
1020 return retval;
1021}
1022
1023EXPORT_SYMBOL(flush_old_exec);
1024
1025/*
1026 * Fill the binprm structure from the inode.
1027 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1028 */
1029int prepare_binprm(struct linux_binprm *bprm)
1030{
1031 int mode;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001032 struct inode * inode = bprm->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 int retval;
1034
1035 mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 if (bprm->file->f_op == NULL)
1037 return -EACCES;
1038
1039 bprm->e_uid = current->euid;
1040 bprm->e_gid = current->egid;
1041
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001042 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 /* Set-uid? */
1044 if (mode & S_ISUID) {
1045 current->personality &= ~PER_CLEAR_ON_SETID;
1046 bprm->e_uid = inode->i_uid;
1047 }
1048
1049 /* Set-gid? */
1050 /*
1051 * If setgid is set but no group execute bit then this
1052 * is a candidate for mandatory locking, not a setgid
1053 * executable.
1054 */
1055 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1056 current->personality &= ~PER_CLEAR_ON_SETID;
1057 bprm->e_gid = inode->i_gid;
1058 }
1059 }
1060
1061 /* fill in binprm security blob */
1062 retval = security_bprm_set(bprm);
1063 if (retval)
1064 return retval;
1065
1066 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1067 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1068}
1069
1070EXPORT_SYMBOL(prepare_binprm);
1071
Arjan van de Ven858119e2006-01-14 13:20:43 -08001072static int unsafe_exec(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
1074 int unsafe = 0;
1075 if (p->ptrace & PT_PTRACED) {
1076 if (p->ptrace & PT_PTRACE_CAP)
1077 unsafe |= LSM_UNSAFE_PTRACE_CAP;
1078 else
1079 unsafe |= LSM_UNSAFE_PTRACE;
1080 }
1081 if (atomic_read(&p->fs->count) > 1 ||
1082 atomic_read(&p->files->count) > 1 ||
1083 atomic_read(&p->sighand->count) > 1)
1084 unsafe |= LSM_UNSAFE_SHARE;
1085
1086 return unsafe;
1087}
1088
1089void compute_creds(struct linux_binprm *bprm)
1090{
1091 int unsafe;
1092
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001093 if (bprm->e_uid != current->uid) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 suid_keys(current);
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001095 current->pdeath_signal = 0;
1096 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 exec_keys(current);
1098
1099 task_lock(current);
1100 unsafe = unsafe_exec(current);
1101 security_bprm_apply_creds(bprm, unsafe);
1102 task_unlock(current);
1103 security_bprm_post_apply_creds(bprm);
1104}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105EXPORT_SYMBOL(compute_creds);
1106
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001107/*
1108 * Arguments are '\0' separated strings found at the location bprm->p
1109 * points to; chop off the first by relocating brpm->p to right after
1110 * the first '\0' encountered.
1111 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001112int remove_arg_zero(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113{
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001114 int ret = 0;
1115 unsigned long offset;
1116 char *kaddr;
1117 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001119 if (!bprm->argc)
1120 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001122 do {
1123 offset = bprm->p & ~PAGE_MASK;
1124 page = get_arg_page(bprm, bprm->p, 0);
1125 if (!page) {
1126 ret = -EFAULT;
1127 goto out;
1128 }
1129 kaddr = kmap_atomic(page, KM_USER0);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001130
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001131 for (; offset < PAGE_SIZE && kaddr[offset];
1132 offset++, bprm->p++)
1133 ;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001134
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001135 kunmap_atomic(kaddr, KM_USER0);
1136 put_arg_page(page);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001137
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001138 if (offset == PAGE_SIZE)
1139 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1140 } while (offset == PAGE_SIZE);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001141
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001142 bprm->p++;
1143 bprm->argc--;
1144 ret = 0;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001145
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001146out:
1147 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149EXPORT_SYMBOL(remove_arg_zero);
1150
1151/*
1152 * cycle the list of binary formats handler, until one recognizes the image
1153 */
1154int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1155{
1156 int try,retval;
1157 struct linux_binfmt *fmt;
David Woodhouse702773b2008-06-16 12:11:54 +01001158#ifdef __alpha__
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 /* handle /sbin/loader.. */
1160 {
1161 struct exec * eh = (struct exec *) bprm->buf;
1162
1163 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1164 (eh->fh.f_flags & 0x3000) == 0x3000)
1165 {
1166 struct file * file;
1167 unsigned long loader;
1168
1169 allow_write_access(bprm->file);
1170 fput(bprm->file);
1171 bprm->file = NULL;
1172
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001173 loader = bprm->vma->vm_end - sizeof(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
1175 file = open_exec("/sbin/loader");
1176 retval = PTR_ERR(file);
1177 if (IS_ERR(file))
1178 return retval;
1179
1180 /* Remember if the application is TASO. */
1181 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1182
1183 bprm->file = file;
1184 bprm->loader = loader;
1185 retval = prepare_binprm(bprm);
1186 if (retval<0)
1187 return retval;
1188 /* should call search_binary_handler recursively here,
1189 but it does not matter */
1190 }
1191 }
1192#endif
1193 retval = security_bprm_check(bprm);
1194 if (retval)
1195 return retval;
1196
1197 /* kernel module loader fixup */
1198 /* so we don't try to load run modprobe in kernel space. */
1199 set_fs(USER_DS);
Al Viro473ae302006-04-26 14:04:08 -04001200
1201 retval = audit_bprm(bprm);
1202 if (retval)
1203 return retval;
1204
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 retval = -ENOENT;
1206 for (try=0; try<2; try++) {
1207 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -07001208 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1210 if (!fn)
1211 continue;
1212 if (!try_module_get(fmt->module))
1213 continue;
1214 read_unlock(&binfmt_lock);
1215 retval = fn(bprm, regs);
1216 if (retval >= 0) {
1217 put_binfmt(fmt);
1218 allow_write_access(bprm->file);
1219 if (bprm->file)
1220 fput(bprm->file);
1221 bprm->file = NULL;
1222 current->did_exec = 1;
Matt Helsley9f460802005-11-07 00:59:16 -08001223 proc_exec_connector(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 return retval;
1225 }
1226 read_lock(&binfmt_lock);
1227 put_binfmt(fmt);
1228 if (retval != -ENOEXEC || bprm->mm == NULL)
1229 break;
1230 if (!bprm->file) {
1231 read_unlock(&binfmt_lock);
1232 return retval;
1233 }
1234 }
1235 read_unlock(&binfmt_lock);
1236 if (retval != -ENOEXEC || bprm->mm == NULL) {
1237 break;
1238#ifdef CONFIG_KMOD
1239 }else{
1240#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1241 if (printable(bprm->buf[0]) &&
1242 printable(bprm->buf[1]) &&
1243 printable(bprm->buf[2]) &&
1244 printable(bprm->buf[3]))
1245 break; /* -ENOEXEC */
1246 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1247#endif
1248 }
1249 }
1250 return retval;
1251}
1252
1253EXPORT_SYMBOL(search_binary_handler);
1254
Al Viro08a6fac2008-05-10 16:38:25 -04001255void free_bprm(struct linux_binprm *bprm)
1256{
1257 free_arg_pages(bprm);
1258 kfree(bprm);
1259}
1260
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261/*
1262 * sys_execve() executes a new program.
1263 */
1264int do_execve(char * filename,
1265 char __user *__user *argv,
1266 char __user *__user *envp,
1267 struct pt_regs * regs)
1268{
1269 struct linux_binprm *bprm;
1270 struct file *file;
Al Viro3b125382008-04-22 05:31:30 -04001271 struct files_struct *displaced;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
Al Viro3b125382008-04-22 05:31:30 -04001274 retval = unshare_files(&displaced);
Al Virofd8328b2008-04-22 05:11:59 -04001275 if (retval)
1276 goto out_ret;
1277
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 retval = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001279 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 if (!bprm)
Al Virofd8328b2008-04-22 05:11:59 -04001281 goto out_files;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
1283 file = open_exec(filename);
1284 retval = PTR_ERR(file);
1285 if (IS_ERR(file))
1286 goto out_kfree;
1287
1288 sched_exec();
1289
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 bprm->file = file;
1291 bprm->filename = filename;
1292 bprm->interp = filename;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001293
1294 retval = bprm_mm_init(bprm);
1295 if (retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 goto out_file;
1297
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001298 bprm->argc = count(argv, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 if ((retval = bprm->argc) < 0)
1300 goto out_mm;
1301
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001302 bprm->envc = count(envp, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 if ((retval = bprm->envc) < 0)
1304 goto out_mm;
1305
1306 retval = security_bprm_alloc(bprm);
1307 if (retval)
1308 goto out;
1309
1310 retval = prepare_binprm(bprm);
1311 if (retval < 0)
1312 goto out;
1313
1314 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1315 if (retval < 0)
1316 goto out;
1317
1318 bprm->exec = bprm->p;
1319 retval = copy_strings(bprm->envc, envp, bprm);
1320 if (retval < 0)
1321 goto out;
1322
1323 retval = copy_strings(bprm->argc, argv, bprm);
1324 if (retval < 0)
1325 goto out;
1326
Oleg Nesterov7b34e422008-07-25 01:47:37 -07001327 current->flags &= ~PF_KTHREAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 retval = search_binary_handler(bprm,regs);
1329 if (retval >= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 /* execve success */
1331 security_bprm_free(bprm);
1332 acct_update_integrals(current);
Al Viro08a6fac2008-05-10 16:38:25 -04001333 free_bprm(bprm);
Al Viro3b125382008-04-22 05:31:30 -04001334 if (displaced)
1335 put_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 return retval;
1337 }
1338
1339out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 if (bprm->security)
1341 security_bprm_free(bprm);
1342
1343out_mm:
1344 if (bprm->mm)
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001345 mmput (bprm->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
1347out_file:
1348 if (bprm->file) {
1349 allow_write_access(bprm->file);
1350 fput(bprm->file);
1351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352out_kfree:
Al Viro08a6fac2008-05-10 16:38:25 -04001353 free_bprm(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
Al Virofd8328b2008-04-22 05:11:59 -04001355out_files:
Al Viro3b125382008-04-22 05:31:30 -04001356 if (displaced)
1357 reset_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358out_ret:
1359 return retval;
1360}
1361
1362int set_binfmt(struct linux_binfmt *new)
1363{
1364 struct linux_binfmt *old = current->binfmt;
1365
1366 if (new) {
1367 if (!try_module_get(new->module))
1368 return -1;
1369 }
1370 current->binfmt = new;
1371 if (old)
1372 module_put(old->module);
1373 return 0;
1374}
1375
1376EXPORT_SYMBOL(set_binfmt);
1377
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378/* format_corename will inspect the pattern parameter, and output a
1379 * name into corename, which must have space for at least
1380 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1381 */
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001382static int format_corename(char *corename, int nr_threads, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383{
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001384 const char *pat_ptr = core_pattern;
1385 int ispipe = (*pat_ptr == '|');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 char *out_ptr = corename;
1387 char *const out_end = corename + CORENAME_MAX_SIZE;
1388 int rc;
1389 int pid_in_pattern = 0;
1390
1391 /* Repeat as long as we have more pattern to process and more output
1392 space */
1393 while (*pat_ptr) {
1394 if (*pat_ptr != '%') {
1395 if (out_ptr == out_end)
1396 goto out;
1397 *out_ptr++ = *pat_ptr++;
1398 } else {
1399 switch (*++pat_ptr) {
1400 case 0:
1401 goto out;
1402 /* Double percent, output one percent */
1403 case '%':
1404 if (out_ptr == out_end)
1405 goto out;
1406 *out_ptr++ = '%';
1407 break;
1408 /* pid */
1409 case 'p':
1410 pid_in_pattern = 1;
1411 rc = snprintf(out_ptr, out_end - out_ptr,
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001412 "%d", task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 if (rc > out_end - out_ptr)
1414 goto out;
1415 out_ptr += rc;
1416 break;
1417 /* uid */
1418 case 'u':
1419 rc = snprintf(out_ptr, out_end - out_ptr,
1420 "%d", current->uid);
1421 if (rc > out_end - out_ptr)
1422 goto out;
1423 out_ptr += rc;
1424 break;
1425 /* gid */
1426 case 'g':
1427 rc = snprintf(out_ptr, out_end - out_ptr,
1428 "%d", current->gid);
1429 if (rc > out_end - out_ptr)
1430 goto out;
1431 out_ptr += rc;
1432 break;
1433 /* signal that caused the coredump */
1434 case 's':
1435 rc = snprintf(out_ptr, out_end - out_ptr,
1436 "%ld", signr);
1437 if (rc > out_end - out_ptr)
1438 goto out;
1439 out_ptr += rc;
1440 break;
1441 /* UNIX time of coredump */
1442 case 't': {
1443 struct timeval tv;
1444 do_gettimeofday(&tv);
1445 rc = snprintf(out_ptr, out_end - out_ptr,
1446 "%lu", tv.tv_sec);
1447 if (rc > out_end - out_ptr)
1448 goto out;
1449 out_ptr += rc;
1450 break;
1451 }
1452 /* hostname */
1453 case 'h':
1454 down_read(&uts_sem);
1455 rc = snprintf(out_ptr, out_end - out_ptr,
Serge E. Hallyne9ff3992006-10-02 02:18:11 -07001456 "%s", utsname()->nodename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 up_read(&uts_sem);
1458 if (rc > out_end - out_ptr)
1459 goto out;
1460 out_ptr += rc;
1461 break;
1462 /* executable */
1463 case 'e':
1464 rc = snprintf(out_ptr, out_end - out_ptr,
1465 "%s", current->comm);
1466 if (rc > out_end - out_ptr)
1467 goto out;
1468 out_ptr += rc;
1469 break;
Neil Horman74aadce2007-10-16 23:26:35 -07001470 /* core limit size */
1471 case 'c':
1472 rc = snprintf(out_ptr, out_end - out_ptr,
1473 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1474 if (rc > out_end - out_ptr)
1475 goto out;
1476 out_ptr += rc;
1477 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 default:
1479 break;
1480 }
1481 ++pat_ptr;
1482 }
1483 }
1484 /* Backward compatibility with core_uses_pid:
1485 *
1486 * If core_pattern does not include a %p (as is the default)
1487 * and core_uses_pid is set, then .%pid will be appended to
Alan Coxc4bbafd2007-04-16 22:53:13 -07001488 * the filename. Do not do this for piped commands. */
1489 if (!ispipe && !pid_in_pattern
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001490 && (core_uses_pid || nr_threads)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 rc = snprintf(out_ptr, out_end - out_ptr,
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001492 ".%d", task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 if (rc > out_end - out_ptr)
1494 goto out;
1495 out_ptr += rc;
1496 }
Alan Coxc4bbafd2007-04-16 22:53:13 -07001497out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 *out_ptr = 0;
Alan Coxc4bbafd2007-04-16 22:53:13 -07001499 return ispipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500}
1501
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001502static int zap_process(struct task_struct *start)
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001503{
1504 struct task_struct *t;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001505 int nr = 0;
Oleg Nesterov281de332006-06-26 00:26:06 -07001506
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001507 start->signal->flags = SIGNAL_GROUP_EXIT;
1508 start->signal->group_stop_count = 0;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001509
1510 t = start;
1511 do {
1512 if (t != current && t->mm) {
Oleg Nesterov281de332006-06-26 00:26:06 -07001513 sigaddset(&t->pending.signal, SIGKILL);
1514 signal_wake_up(t, 1);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001515 nr++;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001516 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001517 } while_each_thread(start, t);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001518
1519 return nr;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001520}
1521
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001522static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001523 struct core_state *core_state, int exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
1525 struct task_struct *g, *p;
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001526 unsigned long flags;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001527 int nr = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001529 spin_lock_irq(&tsk->sighand->siglock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001530 if (!signal_group_exit(tsk->signal)) {
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001531 mm->core_state = core_state;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001532 tsk->signal->group_exit_code = exit_code;
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001533 nr = zap_process(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 }
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001535 spin_unlock_irq(&tsk->sighand->siglock);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001536 if (unlikely(nr < 0))
1537 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001539 if (atomic_read(&mm->mm_users) == nr + 1)
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001540 goto done;
Oleg Nesterove4901f92008-07-25 01:47:31 -07001541 /*
1542 * We should find and kill all tasks which use this mm, and we should
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001543 * count them correctly into ->nr_threads. We don't take tasklist
Oleg Nesterove4901f92008-07-25 01:47:31 -07001544 * lock, but this is safe wrt:
1545 *
1546 * fork:
1547 * None of sub-threads can fork after zap_process(leader). All
1548 * processes which were created before this point should be
1549 * visible to zap_threads() because copy_process() adds the new
1550 * process to the tail of init_task.tasks list, and lock/unlock
1551 * of ->siglock provides a memory barrier.
1552 *
1553 * do_exit:
1554 * The caller holds mm->mmap_sem. This means that the task which
1555 * uses this mm can't pass exit_mm(), so it can't exit or clear
1556 * its ->mm.
1557 *
1558 * de_thread:
1559 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1560 * we must see either old or new leader, this does not matter.
1561 * However, it can change p->sighand, so lock_task_sighand(p)
1562 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1563 * it can't fail.
1564 *
1565 * Note also that "g" can be the old leader with ->mm == NULL
1566 * and already unhashed and thus removed from ->thread_group.
1567 * This is OK, __unhash_process()->list_del_rcu() does not
1568 * clear the ->next pointer, we will find the new leader via
1569 * next_thread().
1570 */
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001571 rcu_read_lock();
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001572 for_each_process(g) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001573 if (g == tsk->group_leader)
1574 continue;
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001575 if (g->flags & PF_KTHREAD)
1576 continue;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001577 p = g;
1578 do {
1579 if (p->mm) {
Oleg Nesterov15b9f362008-07-25 01:47:39 -07001580 if (unlikely(p->mm == mm)) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001581 lock_task_sighand(p, &flags);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001582 nr += zap_process(p);
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001583 unlock_task_sighand(p, &flags);
1584 }
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001585 break;
1586 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001587 } while_each_thread(g, p);
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001588 }
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001589 rcu_read_unlock();
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001590done:
Oleg Nesterovc5f1cc82008-07-25 01:47:42 -07001591 atomic_set(&core_state->nr_threads, nr);
Oleg Nesterov8cd9c242008-07-25 01:47:42 -07001592 return nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593}
1594
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001595static int coredump_wait(int exit_code, struct core_state *core_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596{
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001597 struct task_struct *tsk = current;
1598 struct mm_struct *mm = tsk->mm;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001599 struct completion *vfork_done;
Oleg Nesterov2384f552005-10-30 15:02:47 -08001600 int core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001602 init_completion(&core_state->startup);
Oleg Nesterovb564daf2008-07-25 01:47:44 -07001603 core_state->dumper.task = tsk;
1604 core_state->dumper.next = NULL;
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001605 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001606 up_write(&mm->mmap_sem);
1607
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001608 if (unlikely(core_waiters < 0))
1609 goto fail;
1610
1611 /*
1612 * Make sure nobody is waiting for us to release the VM,
1613 * otherwise we can deadlock when we wait on each other
1614 */
1615 vfork_done = tsk->vfork_done;
1616 if (vfork_done) {
1617 tsk->vfork_done = NULL;
1618 complete(vfork_done);
1619 }
1620
Oleg Nesterov2384f552005-10-30 15:02:47 -08001621 if (core_waiters)
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001622 wait_for_completion(&core_state->startup);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001623fail:
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001624 return core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625}
1626
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001627static void coredump_finish(struct mm_struct *mm)
1628{
1629 struct core_thread *curr, *next;
1630 struct task_struct *task;
1631
1632 next = mm->core_state->dumper.next;
1633 while ((curr = next) != NULL) {
1634 next = curr->next;
1635 task = curr->task;
1636 /*
1637 * see exit_mm(), curr->task must not see
1638 * ->task == NULL before we read ->next.
1639 */
1640 smp_mb();
1641 curr->task = NULL;
1642 wake_up_process(task);
1643 }
1644
1645 mm->core_state = NULL;
1646}
1647
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001648/*
1649 * set_dumpable converts traditional three-value dumpable to two flags and
1650 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1651 * these bits are not changed atomically. So get_dumpable can observe the
1652 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1653 * return either old dumpable or new one by paying attention to the order of
1654 * modifying the bits.
1655 *
1656 * dumpable | mm->flags (binary)
1657 * old new | initial interim final
1658 * ---------+-----------------------
1659 * 0 1 | 00 01 01
1660 * 0 2 | 00 10(*) 11
1661 * 1 0 | 01 00 00
1662 * 1 2 | 01 11 11
1663 * 2 0 | 11 10(*) 00
1664 * 2 1 | 11 11 01
1665 *
1666 * (*) get_dumpable regards interim value of 10 as 11.
1667 */
1668void set_dumpable(struct mm_struct *mm, int value)
1669{
1670 switch (value) {
1671 case 0:
1672 clear_bit(MMF_DUMPABLE, &mm->flags);
1673 smp_wmb();
1674 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1675 break;
1676 case 1:
1677 set_bit(MMF_DUMPABLE, &mm->flags);
1678 smp_wmb();
1679 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1680 break;
1681 case 2:
1682 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1683 smp_wmb();
1684 set_bit(MMF_DUMPABLE, &mm->flags);
1685 break;
1686 }
1687}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001688
1689int get_dumpable(struct mm_struct *mm)
1690{
1691 int ret;
1692
1693 ret = mm->flags & 0x3;
1694 return (ret >= 2) ? 2 : ret;
1695}
1696
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1698{
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001699 struct core_state core_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 char corename[CORENAME_MAX_SIZE + 1];
1701 struct mm_struct *mm = current->mm;
1702 struct linux_binfmt * binfmt;
1703 struct inode * inode;
1704 struct file * file;
1705 int retval = 0;
Alan Coxd6e71142005-06-23 00:09:43 -07001706 int fsuid = current->fsuid;
1707 int flag = 0;
Andi Kleend025c9d2006-09-30 23:29:28 -07001708 int ispipe = 0;
Neil Horman7dc0b222007-10-16 23:26:34 -07001709 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
Neil Horman74aadce2007-10-16 23:26:35 -07001710 char **helper_argv = NULL;
1711 int helper_argc = 0;
1712 char *delimit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713
Steve Grubb0a4ff8c2007-04-19 10:28:21 -04001714 audit_core_dumps(signr);
1715
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 binfmt = current->binfmt;
1717 if (!binfmt || !binfmt->core_dump)
1718 goto fail;
1719 down_write(&mm->mmap_sem);
Roland McGrath00ec99d2007-11-11 19:13:43 -08001720 /*
1721 * If another thread got here first, or we are not dumpable, bail out.
1722 */
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001723 if (mm->core_state || !get_dumpable(mm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 up_write(&mm->mmap_sem);
1725 goto fail;
1726 }
Alan Coxd6e71142005-06-23 00:09:43 -07001727
1728 /*
1729 * We cannot trust fsuid as being the "true" uid of the
1730 * process nor do we know its entire history. We only know it
1731 * was tainted so we dump it as root in mode 2.
1732 */
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001733 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
Alan Coxd6e71142005-06-23 00:09:43 -07001734 flag = O_EXCL; /* Stop rewrite attacks */
1735 current->fsuid = 0; /* Dump root private */
1736 }
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001737
Oleg Nesterov9d5b3272008-07-25 01:47:43 -07001738 retval = coredump_wait(exit_code, &core_state);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001739 if (retval < 0)
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001740 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741
1742 /*
1743 * Clear any false indication of pending signals that might
1744 * be seen by the filesystem code called to write the core file.
1745 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 clear_thread_flag(TIF_SIGPENDING);
1747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 /*
1749 * lock_kernel() because format_corename() is controlled by sysctl, which
1750 * uses lock_kernel()
1751 */
1752 lock_kernel();
Oleg Nesterov565b9b12008-07-25 01:47:47 -07001753 ispipe = format_corename(corename, retval, signr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 unlock_kernel();
Neil Horman7dc0b222007-10-16 23:26:34 -07001755 /*
1756 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1757 * to a pipe. Since we're not writing directly to the filesystem
1758 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1759 * created unless the pipe reader choses to write out the core file
1760 * at which point file size limits and permissions will be imposed
1761 * as it does with any other process
1762 */
Neil Horman74aadce2007-10-16 23:26:35 -07001763 if ((!ispipe) && (core_limit < binfmt->min_coredump))
Neil Horman7dc0b222007-10-16 23:26:34 -07001764 goto fail_unlock;
1765
Alan Coxc4bbafd2007-04-16 22:53:13 -07001766 if (ispipe) {
Neil Horman74aadce2007-10-16 23:26:35 -07001767 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1768 /* Terminate the string before the first option */
1769 delimit = strchr(corename, ' ');
1770 if (delimit)
1771 *delimit = '\0';
Neil Horman32321132007-10-16 23:26:36 -07001772 delimit = strrchr(helper_argv[0], '/');
1773 if (delimit)
1774 delimit++;
1775 else
1776 delimit = helper_argv[0];
1777 if (!strcmp(delimit, current->comm)) {
1778 printk(KERN_NOTICE "Recursive core dump detected, "
1779 "aborting\n");
1780 goto fail_unlock;
1781 }
1782
1783 core_limit = RLIM_INFINITY;
1784
Andi Kleend025c9d2006-09-30 23:29:28 -07001785 /* SIGPIPE can happen, but it's just never processed */
Neil Horman32321132007-10-16 23:26:36 -07001786 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1787 &file)) {
Andi Kleend025c9d2006-09-30 23:29:28 -07001788 printk(KERN_INFO "Core dump to %s pipe failed\n",
1789 corename);
1790 goto fail_unlock;
1791 }
Andi Kleend025c9d2006-09-30 23:29:28 -07001792 } else
1793 file = filp_open(corename,
Alexey Dobriyan6d4df672006-12-06 20:40:39 -08001794 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1795 0600);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 if (IS_ERR(file))
1797 goto fail_unlock;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001798 inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 if (inode->i_nlink > 1)
1800 goto close_fail; /* multiple links - don't dump */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001801 if (!ispipe && d_unhashed(file->f_path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 goto close_fail;
1803
Andi Kleend025c9d2006-09-30 23:29:28 -07001804 /* AK: actually i see no reason to not allow this for named pipes etc.,
1805 but keep the previous behaviour for now. */
1806 if (!ispipe && !S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 goto close_fail;
Ingo Molnarc46f7392007-11-28 13:59:18 +01001808 /*
1809 * Dont allow local users get cute and trick others to coredump
1810 * into their pre-created files:
1811 */
1812 if (inode->i_uid != current->fsuid)
1813 goto close_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 if (!file->f_op)
1815 goto close_fail;
1816 if (!file->f_op->write)
1817 goto close_fail;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001818 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 goto close_fail;
1820
Neil Horman7dc0b222007-10-16 23:26:34 -07001821 retval = binfmt->core_dump(signr, regs, file, core_limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
1823 if (retval)
1824 current->signal->group_exit_code |= 0x80;
1825close_fail:
1826 filp_close(file, NULL);
1827fail_unlock:
Neil Horman74aadce2007-10-16 23:26:35 -07001828 if (helper_argv)
1829 argv_free(helper_argv);
1830
Alan Coxd6e71142005-06-23 00:09:43 -07001831 current->fsuid = fsuid;
Oleg Nesterova94e2d42008-07-25 01:47:46 -07001832 coredump_finish(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833fail:
1834 return retval;
1835}