blob: af249af4ccab1f487b4f600ee0a82b2ef2681e97 [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
725 * checking core_waiters and changing tsk->mm. The
726 * core-inducing thread will increment core_waiters for
727 * each thread whose ->mm == old_mm.
728 */
729 down_read(&old_mm->mmap_sem);
730 if (unlikely(old_mm->core_waiters)) {
731 up_read(&old_mm->mmap_sem);
732 return -EINTR;
733 }
734 }
735 task_lock(tsk);
736 active_mm = tsk->active_mm;
737 tsk->mm = mm;
738 tsk->active_mm = mm;
739 activate_mm(active_mm, mm);
740 task_unlock(tsk);
KOSAKI Motohiro4cd1a8f2008-05-12 14:02:31 -0700741 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 arch_pick_mmap_layout(mm);
743 if (old_mm) {
744 up_read(&old_mm->mmap_sem);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +0200745 BUG_ON(active_mm != old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 mmput(old_mm);
747 return 0;
748 }
749 mmdrop(active_mm);
750 return 0;
751}
752
753/*
754 * This function makes sure the current process has its own signal table,
755 * so that flush_signal_handlers can later reset the handlers without
756 * disturbing other processes. (Other processes might share the signal
757 * table via the CLONE_SIGHAND option to clone().)
758 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800759static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760{
761 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700762 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 spinlock_t *lock = &oldsighand->siglock;
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300764 struct task_struct *leader = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 int count;
766
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700767 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 goto no_thread_group;
769
770 /*
771 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800774 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 /*
776 * Another group action in progress, just
777 * return so that the signal is processed.
778 */
779 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 return -EAGAIN;
781 }
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800782 sig->group_exit_task = tsk;
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700783 zap_other_threads(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
Oleg Nesterovfea9d172008-02-08 04:19:19 -0800785 /* Account for the thread group leader hanging around: */
786 count = thread_group_leader(tsk) ? 1 : 2;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700787 sig->notify_count = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 while (atomic_read(&sig->count) > count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 __set_current_state(TASK_UNINTERRUPTIBLE);
790 spin_unlock_irq(lock);
791 schedule();
792 spin_lock_irq(lock);
793 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 spin_unlock_irq(lock);
795
796 /*
797 * At this point all other threads have exited, all we have to
798 * do is to wait for the thread group leader to become inactive,
799 * and to assume its PID:
800 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700801 if (!thread_group_leader(tsk)) {
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700802 leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700803
Oleg Nesterov2800d8d2008-04-30 00:53:12 -0700804 sig->notify_count = -1; /* for exit_notify() */
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700805 for (;;) {
806 write_lock_irq(&tasklist_lock);
807 if (likely(leader->exit_state))
808 break;
809 __set_current_state(TASK_UNINTERRUPTIBLE);
810 write_unlock_irq(&tasklist_lock);
811 schedule();
812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
Oleg Nesterov7a5e8732008-04-30 00:53:04 -0700814 if (unlikely(task_child_reaper(tsk) == leader))
815 task_active_pid_ns(tsk)->child_reaper = tsk;
Roland McGrathf5e90282006-04-10 22:54:16 -0700816 /*
817 * The only record we have of the real-time age of a
818 * process, regardless of execs it's done, is start_time.
819 * All the past CPU time is accumulated in signal_struct
820 * from sister threads now dead. But in this non-leader
821 * exec, nothing survives from the original leader thread,
822 * whose birth marks the true age of this process now.
823 * When we take on its identity by switching to its PID, we
824 * also take its birthdate (always earlier than our own).
825 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700826 tsk->start_time = leader->start_time;
Roland McGrathf5e90282006-04-10 22:54:16 -0700827
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700828 BUG_ON(!same_thread_group(leader, tsk));
829 BUG_ON(has_group_leader_pid(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 /*
831 * An exec() starts a new thread group with the
832 * TGID of the previous thread group. Rehash the
833 * two threads with a switched PID, and release
834 * the former thread group leader:
835 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800836
837 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -0700838 * The old leader becomes a thread of the this thread group.
839 * Note: The old leader also uses this pid until release_task
Eric W. Biedermand73d6522006-03-28 16:11:03 -0800840 * is called. Odd but simple and correct.
841 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700842 detach_pid(tsk, PIDTYPE_PID);
843 tsk->pid = leader->pid;
Sukadev Bhattiprolu3743ca02007-10-18 23:39:51 -0700844 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700845 transfer_pid(leader, tsk, PIDTYPE_PGID);
846 transfer_pid(leader, tsk, PIDTYPE_SID);
847 list_replace_rcu(&leader->tasks, &tsk->tasks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700849 tsk->group_leader = tsk;
850 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -0600851
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700852 tsk->exit_signal = SIGCHLD;
Oleg Nesterov962b5642005-11-23 13:37:43 -0800853
854 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
855 leader->exit_state = EXIT_DEAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
857 write_unlock_irq(&tasklist_lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -0800858 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859
Oleg Nesterov6db840f2007-10-16 23:27:23 -0700860 sig->group_exit_task = NULL;
861 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
863no_thread_group:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 exit_itimers(sig);
Oleg Nesterovcbaffba2008-05-26 20:55:42 +0400865 flush_itimer_signals();
Oleg Nesterov329f7db2005-11-07 21:12:43 +0300866 if (leader)
867 release_task(leader);
868
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700869 if (atomic_read(&oldsighand->count) != 1) {
870 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700872 * This ->sighand is shared with the CLONE_SIGHAND
873 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -0700875 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
876 if (!newsighand)
877 return -ENOMEM;
878
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 atomic_set(&newsighand->count, 1);
880 memcpy(newsighand->action, oldsighand->action,
881 sizeof(newsighand->action));
882
883 write_lock_irq(&tasklist_lock);
884 spin_lock(&oldsighand->siglock);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700885 rcu_assign_pointer(tsk->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 spin_unlock(&oldsighand->siglock);
887 write_unlock_irq(&tasklist_lock);
888
Davide Libenzifba2afa2007-05-10 22:23:13 -0700889 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 }
891
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -0700892 BUG_ON(!thread_group_leader(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 return 0;
894}
Oleg Nesterov0840a902007-10-16 23:27:22 -0700895
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896/*
897 * These functions flushes out all traces of the currently running executable
898 * so that a new one can be started
899 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800900static void flush_old_files(struct files_struct * files)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
902 long j = -1;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700903 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
905 spin_lock(&files->file_lock);
906 for (;;) {
907 unsigned long set, i;
908
909 j++;
910 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700911 fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800912 if (i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700914 set = fdt->close_on_exec->fds_bits[j];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 if (!set)
916 continue;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700917 fdt->close_on_exec->fds_bits[j] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 spin_unlock(&files->file_lock);
919 for ( ; set ; i++,set >>= 1) {
920 if (set & 1) {
921 sys_close(i);
922 }
923 }
924 spin_lock(&files->file_lock);
925
926 }
927 spin_unlock(&files->file_lock);
928}
929
Andrew Morton59714d62008-02-04 22:27:21 -0800930char *get_task_comm(char *buf, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931{
932 /* buf must be at least sizeof(tsk->comm) in size */
933 task_lock(tsk);
934 strncpy(buf, tsk->comm, sizeof(tsk->comm));
935 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -0800936 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937}
938
939void set_task_comm(struct task_struct *tsk, char *buf)
940{
941 task_lock(tsk);
942 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
943 task_unlock(tsk);
944}
945
946int flush_old_exec(struct linux_binprm * bprm)
947{
948 char * name;
949 int i, ch, retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 char tcomm[sizeof(current->comm)];
951
952 /*
953 * Make sure we have a private signal table and that
954 * we are unassociated from the previous thread group.
955 */
956 retval = de_thread(current);
957 if (retval)
958 goto out;
959
Matt Helsley925d1c42008-04-29 01:01:36 -0700960 set_mm_exe_file(bprm->mm, bprm->file);
961
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 * Release all of the old mmap stuff
964 */
965 retval = exec_mmap(bprm->mm);
966 if (retval)
Al Virofd8328b2008-04-22 05:11:59 -0400967 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
969 bprm->mm = NULL; /* We're using it now */
970
971 /* This is the point of no return */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 current->sas_ss_sp = current->sas_ss_size = 0;
973
974 if (current->euid == current->uid && current->egid == current->gid)
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -0700975 set_dumpable(current->mm, 1);
Alan Coxd6e71142005-06-23 00:09:43 -0700976 else
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -0700977 set_dumpable(current->mm, suid_dumpable);
Alan Coxd6e71142005-06-23 00:09:43 -0700978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 name = bprm->filename;
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -0700980
981 /* Copies the binary name from after last slash */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 for (i=0; (ch = *(name++)) != '\0';) {
983 if (ch == '/')
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -0700984 i = 0; /* overwrite what we wrote */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 else
986 if (i < (sizeof(tcomm) - 1))
987 tcomm[i++] = ch;
988 }
989 tcomm[i] = '\0';
990 set_task_comm(current, tcomm);
991
992 current->flags &= ~PF_RANDOMIZE;
993 flush_thread();
994
Benjamin Herrenschmidt0551fbd2006-02-28 16:59:19 -0800995 /* Set the new mm task size. We have to do that late because it may
996 * depend on TIF_32BIT which is only updated in flush_thread() on
997 * some architectures like powerpc
998 */
999 current->mm->task_size = TASK_SIZE;
1000
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001001 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1002 suid_keys(current);
1003 set_dumpable(current->mm, suid_dumpable);
1004 current->pdeath_signal = 0;
1005 } else if (file_permission(bprm->file, MAY_READ) ||
1006 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 suid_keys(current);
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001008 set_dumpable(current->mm, suid_dumpable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 }
1010
1011 /* An exec changes our domain. We are no longer part of the thread
1012 group */
1013
1014 current->self_exec_id++;
1015
1016 flush_signal_handlers(current, 0);
1017 flush_old_files(current->files);
1018
1019 return 0;
1020
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021out:
1022 return retval;
1023}
1024
1025EXPORT_SYMBOL(flush_old_exec);
1026
1027/*
1028 * Fill the binprm structure from the inode.
1029 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1030 */
1031int prepare_binprm(struct linux_binprm *bprm)
1032{
1033 int mode;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001034 struct inode * inode = bprm->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 int retval;
1036
1037 mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 if (bprm->file->f_op == NULL)
1039 return -EACCES;
1040
1041 bprm->e_uid = current->euid;
1042 bprm->e_gid = current->egid;
1043
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001044 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 /* Set-uid? */
1046 if (mode & S_ISUID) {
1047 current->personality &= ~PER_CLEAR_ON_SETID;
1048 bprm->e_uid = inode->i_uid;
1049 }
1050
1051 /* Set-gid? */
1052 /*
1053 * If setgid is set but no group execute bit then this
1054 * is a candidate for mandatory locking, not a setgid
1055 * executable.
1056 */
1057 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1058 current->personality &= ~PER_CLEAR_ON_SETID;
1059 bprm->e_gid = inode->i_gid;
1060 }
1061 }
1062
1063 /* fill in binprm security blob */
1064 retval = security_bprm_set(bprm);
1065 if (retval)
1066 return retval;
1067
1068 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1069 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1070}
1071
1072EXPORT_SYMBOL(prepare_binprm);
1073
Arjan van de Ven858119e2006-01-14 13:20:43 -08001074static int unsafe_exec(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075{
1076 int unsafe = 0;
1077 if (p->ptrace & PT_PTRACED) {
1078 if (p->ptrace & PT_PTRACE_CAP)
1079 unsafe |= LSM_UNSAFE_PTRACE_CAP;
1080 else
1081 unsafe |= LSM_UNSAFE_PTRACE;
1082 }
1083 if (atomic_read(&p->fs->count) > 1 ||
1084 atomic_read(&p->files->count) > 1 ||
1085 atomic_read(&p->sighand->count) > 1)
1086 unsafe |= LSM_UNSAFE_SHARE;
1087
1088 return unsafe;
1089}
1090
1091void compute_creds(struct linux_binprm *bprm)
1092{
1093 int unsafe;
1094
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001095 if (bprm->e_uid != current->uid) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 suid_keys(current);
Marcel Holtmannd2d56c52007-08-17 21:47:58 +02001097 current->pdeath_signal = 0;
1098 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 exec_keys(current);
1100
1101 task_lock(current);
1102 unsafe = unsafe_exec(current);
1103 security_bprm_apply_creds(bprm, unsafe);
1104 task_unlock(current);
1105 security_bprm_post_apply_creds(bprm);
1106}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107EXPORT_SYMBOL(compute_creds);
1108
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001109/*
1110 * Arguments are '\0' separated strings found at the location bprm->p
1111 * points to; chop off the first by relocating brpm->p to right after
1112 * the first '\0' encountered.
1113 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001114int remove_arg_zero(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115{
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001116 int ret = 0;
1117 unsigned long offset;
1118 char *kaddr;
1119 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001121 if (!bprm->argc)
1122 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001124 do {
1125 offset = bprm->p & ~PAGE_MASK;
1126 page = get_arg_page(bprm, bprm->p, 0);
1127 if (!page) {
1128 ret = -EFAULT;
1129 goto out;
1130 }
1131 kaddr = kmap_atomic(page, KM_USER0);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001132
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001133 for (; offset < PAGE_SIZE && kaddr[offset];
1134 offset++, bprm->p++)
1135 ;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001136
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001137 kunmap_atomic(kaddr, KM_USER0);
1138 put_arg_page(page);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001139
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001140 if (offset == PAGE_SIZE)
1141 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1142 } while (offset == PAGE_SIZE);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001143
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001144 bprm->p++;
1145 bprm->argc--;
1146 ret = 0;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001147
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001148out:
1149 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151EXPORT_SYMBOL(remove_arg_zero);
1152
1153/*
1154 * cycle the list of binary formats handler, until one recognizes the image
1155 */
1156int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1157{
1158 int try,retval;
1159 struct linux_binfmt *fmt;
David Woodhouse702773b2008-06-16 12:11:54 +01001160#ifdef __alpha__
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 /* handle /sbin/loader.. */
1162 {
1163 struct exec * eh = (struct exec *) bprm->buf;
1164
1165 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1166 (eh->fh.f_flags & 0x3000) == 0x3000)
1167 {
1168 struct file * file;
1169 unsigned long loader;
1170
1171 allow_write_access(bprm->file);
1172 fput(bprm->file);
1173 bprm->file = NULL;
1174
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001175 loader = bprm->vma->vm_end - sizeof(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
1177 file = open_exec("/sbin/loader");
1178 retval = PTR_ERR(file);
1179 if (IS_ERR(file))
1180 return retval;
1181
1182 /* Remember if the application is TASO. */
1183 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1184
1185 bprm->file = file;
1186 bprm->loader = loader;
1187 retval = prepare_binprm(bprm);
1188 if (retval<0)
1189 return retval;
1190 /* should call search_binary_handler recursively here,
1191 but it does not matter */
1192 }
1193 }
1194#endif
1195 retval = security_bprm_check(bprm);
1196 if (retval)
1197 return retval;
1198
1199 /* kernel module loader fixup */
1200 /* so we don't try to load run modprobe in kernel space. */
1201 set_fs(USER_DS);
Al Viro473ae302006-04-26 14:04:08 -04001202
1203 retval = audit_bprm(bprm);
1204 if (retval)
1205 return retval;
1206
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 retval = -ENOENT;
1208 for (try=0; try<2; try++) {
1209 read_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -07001210 list_for_each_entry(fmt, &formats, lh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1212 if (!fn)
1213 continue;
1214 if (!try_module_get(fmt->module))
1215 continue;
1216 read_unlock(&binfmt_lock);
1217 retval = fn(bprm, regs);
1218 if (retval >= 0) {
1219 put_binfmt(fmt);
1220 allow_write_access(bprm->file);
1221 if (bprm->file)
1222 fput(bprm->file);
1223 bprm->file = NULL;
1224 current->did_exec = 1;
Matt Helsley9f460802005-11-07 00:59:16 -08001225 proc_exec_connector(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 return retval;
1227 }
1228 read_lock(&binfmt_lock);
1229 put_binfmt(fmt);
1230 if (retval != -ENOEXEC || bprm->mm == NULL)
1231 break;
1232 if (!bprm->file) {
1233 read_unlock(&binfmt_lock);
1234 return retval;
1235 }
1236 }
1237 read_unlock(&binfmt_lock);
1238 if (retval != -ENOEXEC || bprm->mm == NULL) {
1239 break;
1240#ifdef CONFIG_KMOD
1241 }else{
1242#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1243 if (printable(bprm->buf[0]) &&
1244 printable(bprm->buf[1]) &&
1245 printable(bprm->buf[2]) &&
1246 printable(bprm->buf[3]))
1247 break; /* -ENOEXEC */
1248 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1249#endif
1250 }
1251 }
1252 return retval;
1253}
1254
1255EXPORT_SYMBOL(search_binary_handler);
1256
Al Viro08a6fac2008-05-10 16:38:25 -04001257void free_bprm(struct linux_binprm *bprm)
1258{
1259 free_arg_pages(bprm);
1260 kfree(bprm);
1261}
1262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263/*
1264 * sys_execve() executes a new program.
1265 */
1266int do_execve(char * filename,
1267 char __user *__user *argv,
1268 char __user *__user *envp,
1269 struct pt_regs * regs)
1270{
1271 struct linux_binprm *bprm;
1272 struct file *file;
Al Viro3b125382008-04-22 05:31:30 -04001273 struct files_struct *displaced;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Al Viro3b125382008-04-22 05:31:30 -04001276 retval = unshare_files(&displaced);
Al Virofd8328b2008-04-22 05:11:59 -04001277 if (retval)
1278 goto out_ret;
1279
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 retval = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001281 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 if (!bprm)
Al Virofd8328b2008-04-22 05:11:59 -04001283 goto out_files;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
1285 file = open_exec(filename);
1286 retval = PTR_ERR(file);
1287 if (IS_ERR(file))
1288 goto out_kfree;
1289
1290 sched_exec();
1291
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 bprm->file = file;
1293 bprm->filename = filename;
1294 bprm->interp = filename;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001295
1296 retval = bprm_mm_init(bprm);
1297 if (retval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 goto out_file;
1299
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001300 bprm->argc = count(argv, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 if ((retval = bprm->argc) < 0)
1302 goto out_mm;
1303
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001304 bprm->envc = count(envp, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 if ((retval = bprm->envc) < 0)
1306 goto out_mm;
1307
1308 retval = security_bprm_alloc(bprm);
1309 if (retval)
1310 goto out;
1311
1312 retval = prepare_binprm(bprm);
1313 if (retval < 0)
1314 goto out;
1315
1316 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1317 if (retval < 0)
1318 goto out;
1319
1320 bprm->exec = bprm->p;
1321 retval = copy_strings(bprm->envc, envp, bprm);
1322 if (retval < 0)
1323 goto out;
1324
1325 retval = copy_strings(bprm->argc, argv, bprm);
1326 if (retval < 0)
1327 goto out;
1328
1329 retval = search_binary_handler(bprm,regs);
1330 if (retval >= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 /* execve success */
1332 security_bprm_free(bprm);
1333 acct_update_integrals(current);
Al Viro08a6fac2008-05-10 16:38:25 -04001334 free_bprm(bprm);
Al Viro3b125382008-04-22 05:31:30 -04001335 if (displaced)
1336 put_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 return retval;
1338 }
1339
1340out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 if (bprm->security)
1342 security_bprm_free(bprm);
1343
1344out_mm:
1345 if (bprm->mm)
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001346 mmput (bprm->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
1348out_file:
1349 if (bprm->file) {
1350 allow_write_access(bprm->file);
1351 fput(bprm->file);
1352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353out_kfree:
Al Viro08a6fac2008-05-10 16:38:25 -04001354 free_bprm(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
Al Virofd8328b2008-04-22 05:11:59 -04001356out_files:
Al Viro3b125382008-04-22 05:31:30 -04001357 if (displaced)
1358 reset_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359out_ret:
1360 return retval;
1361}
1362
1363int set_binfmt(struct linux_binfmt *new)
1364{
1365 struct linux_binfmt *old = current->binfmt;
1366
1367 if (new) {
1368 if (!try_module_get(new->module))
1369 return -1;
1370 }
1371 current->binfmt = new;
1372 if (old)
1373 module_put(old->module);
1374 return 0;
1375}
1376
1377EXPORT_SYMBOL(set_binfmt);
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379/* format_corename will inspect the pattern parameter, and output a
1380 * name into corename, which must have space for at least
1381 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1382 */
Alan Coxc4bbafd2007-04-16 22:53:13 -07001383static int format_corename(char *corename, const char *pattern, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384{
1385 const char *pat_ptr = pattern;
1386 char *out_ptr = corename;
1387 char *const out_end = corename + CORENAME_MAX_SIZE;
1388 int rc;
1389 int pid_in_pattern = 0;
Alan Coxc4bbafd2007-04-16 22:53:13 -07001390 int ispipe = 0;
1391
1392 if (*pattern == '|')
1393 ispipe = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
1395 /* Repeat as long as we have more pattern to process and more output
1396 space */
1397 while (*pat_ptr) {
1398 if (*pat_ptr != '%') {
1399 if (out_ptr == out_end)
1400 goto out;
1401 *out_ptr++ = *pat_ptr++;
1402 } else {
1403 switch (*++pat_ptr) {
1404 case 0:
1405 goto out;
1406 /* Double percent, output one percent */
1407 case '%':
1408 if (out_ptr == out_end)
1409 goto out;
1410 *out_ptr++ = '%';
1411 break;
1412 /* pid */
1413 case 'p':
1414 pid_in_pattern = 1;
1415 rc = snprintf(out_ptr, out_end - out_ptr,
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001416 "%d", task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 if (rc > out_end - out_ptr)
1418 goto out;
1419 out_ptr += rc;
1420 break;
1421 /* uid */
1422 case 'u':
1423 rc = snprintf(out_ptr, out_end - out_ptr,
1424 "%d", current->uid);
1425 if (rc > out_end - out_ptr)
1426 goto out;
1427 out_ptr += rc;
1428 break;
1429 /* gid */
1430 case 'g':
1431 rc = snprintf(out_ptr, out_end - out_ptr,
1432 "%d", current->gid);
1433 if (rc > out_end - out_ptr)
1434 goto out;
1435 out_ptr += rc;
1436 break;
1437 /* signal that caused the coredump */
1438 case 's':
1439 rc = snprintf(out_ptr, out_end - out_ptr,
1440 "%ld", signr);
1441 if (rc > out_end - out_ptr)
1442 goto out;
1443 out_ptr += rc;
1444 break;
1445 /* UNIX time of coredump */
1446 case 't': {
1447 struct timeval tv;
1448 do_gettimeofday(&tv);
1449 rc = snprintf(out_ptr, out_end - out_ptr,
1450 "%lu", tv.tv_sec);
1451 if (rc > out_end - out_ptr)
1452 goto out;
1453 out_ptr += rc;
1454 break;
1455 }
1456 /* hostname */
1457 case 'h':
1458 down_read(&uts_sem);
1459 rc = snprintf(out_ptr, out_end - out_ptr,
Serge E. Hallyne9ff3992006-10-02 02:18:11 -07001460 "%s", utsname()->nodename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 up_read(&uts_sem);
1462 if (rc > out_end - out_ptr)
1463 goto out;
1464 out_ptr += rc;
1465 break;
1466 /* executable */
1467 case 'e':
1468 rc = snprintf(out_ptr, out_end - out_ptr,
1469 "%s", current->comm);
1470 if (rc > out_end - out_ptr)
1471 goto out;
1472 out_ptr += rc;
1473 break;
Neil Horman74aadce2007-10-16 23:26:35 -07001474 /* core limit size */
1475 case 'c':
1476 rc = snprintf(out_ptr, out_end - out_ptr,
1477 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1478 if (rc > out_end - out_ptr)
1479 goto out;
1480 out_ptr += rc;
1481 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 default:
1483 break;
1484 }
1485 ++pat_ptr;
1486 }
1487 }
1488 /* Backward compatibility with core_uses_pid:
1489 *
1490 * If core_pattern does not include a %p (as is the default)
1491 * and core_uses_pid is set, then .%pid will be appended to
Alan Coxc4bbafd2007-04-16 22:53:13 -07001492 * the filename. Do not do this for piped commands. */
1493 if (!ispipe && !pid_in_pattern
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1495 rc = snprintf(out_ptr, out_end - out_ptr,
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001496 ".%d", task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 if (rc > out_end - out_ptr)
1498 goto out;
1499 out_ptr += rc;
1500 }
Alan Coxc4bbafd2007-04-16 22:53:13 -07001501out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 *out_ptr = 0;
Alan Coxc4bbafd2007-04-16 22:53:13 -07001503 return ispipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504}
1505
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001506static void zap_process(struct task_struct *start)
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001507{
1508 struct task_struct *t;
Oleg Nesterov281de332006-06-26 00:26:06 -07001509
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001510 start->signal->flags = SIGNAL_GROUP_EXIT;
1511 start->signal->group_stop_count = 0;
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001512
1513 t = start;
1514 do {
1515 if (t != current && t->mm) {
1516 t->mm->core_waiters++;
Oleg Nesterov281de332006-06-26 00:26:06 -07001517 sigaddset(&t->pending.signal, SIGKILL);
1518 signal_wake_up(t, 1);
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001519 }
Oleg Nesterove4901f92008-07-25 01:47:31 -07001520 } while_each_thread(start, t);
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001521}
1522
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001523static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1524 int exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525{
1526 struct task_struct *g, *p;
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001527 unsigned long flags;
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001528 int err = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001530 spin_lock_irq(&tsk->sighand->siglock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001531 if (!signal_group_exit(tsk->signal)) {
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001532 tsk->signal->group_exit_code = exit_code;
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001533 zap_process(tsk);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001534 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 }
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001536 spin_unlock_irq(&tsk->sighand->siglock);
1537 if (err)
1538 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001540 if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1541 goto done;
Oleg Nesterove4901f92008-07-25 01:47:31 -07001542 /*
1543 * We should find and kill all tasks which use this mm, and we should
1544 * count them correctly into mm->core_waiters. We don't take tasklist
1545 * lock, but this is safe wrt:
1546 *
1547 * fork:
1548 * None of sub-threads can fork after zap_process(leader). All
1549 * processes which were created before this point should be
1550 * visible to zap_threads() because copy_process() adds the new
1551 * process to the tail of init_task.tasks list, and lock/unlock
1552 * of ->siglock provides a memory barrier.
1553 *
1554 * do_exit:
1555 * The caller holds mm->mmap_sem. This means that the task which
1556 * uses this mm can't pass exit_mm(), so it can't exit or clear
1557 * its ->mm.
1558 *
1559 * de_thread:
1560 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1561 * we must see either old or new leader, this does not matter.
1562 * However, it can change p->sighand, so lock_task_sighand(p)
1563 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1564 * it can't fail.
1565 *
1566 * Note also that "g" can be the old leader with ->mm == NULL
1567 * and already unhashed and thus removed from ->thread_group.
1568 * This is OK, __unhash_process()->list_del_rcu() does not
1569 * clear the ->next pointer, we will find the new leader via
1570 * next_thread().
1571 */
Oleg Nesterov7b1c6152006-06-26 00:26:08 -07001572 rcu_read_lock();
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001573 for_each_process(g) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001574 if (g == tsk->group_leader)
1575 continue;
1576
Oleg Nesterovaceecc02006-06-26 00:26:05 -07001577 p = g;
1578 do {
1579 if (p->mm) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001580 if (p->mm == mm) {
Oleg Nesterov5debfa62006-06-26 00:26:09 -07001581 lock_task_sighand(p, &flags);
Oleg Nesterovd5f70c02006-06-26 00:26:07 -07001582 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 Nesterovdcf560c2006-06-26 00:26:08 -07001591 return mm->core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592}
1593
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001594static int coredump_wait(int exit_code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595{
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001596 struct task_struct *tsk = current;
1597 struct mm_struct *mm = tsk->mm;
1598 struct completion startup_done;
1599 struct completion *vfork_done;
Oleg Nesterov2384f552005-10-30 15:02:47 -08001600 int core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001602 init_completion(&mm->core_done);
1603 init_completion(&startup_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 mm->core_startup_done = &startup_done;
1605
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001606 core_waiters = zap_threads(tsk, mm, exit_code);
Oleg Nesterov2384f552005-10-30 15:02:47 -08001607 up_write(&mm->mmap_sem);
1608
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001609 if (unlikely(core_waiters < 0))
1610 goto fail;
1611
1612 /*
1613 * Make sure nobody is waiting for us to release the VM,
1614 * otherwise we can deadlock when we wait on each other
1615 */
1616 vfork_done = tsk->vfork_done;
1617 if (vfork_done) {
1618 tsk->vfork_done = NULL;
1619 complete(vfork_done);
1620 }
1621
Oleg Nesterov2384f552005-10-30 15:02:47 -08001622 if (core_waiters)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 wait_for_completion(&startup_done);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001624fail:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 BUG_ON(mm->core_waiters);
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001626 return core_waiters;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627}
1628
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001629/*
1630 * set_dumpable converts traditional three-value dumpable to two flags and
1631 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1632 * these bits are not changed atomically. So get_dumpable can observe the
1633 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1634 * return either old dumpable or new one by paying attention to the order of
1635 * modifying the bits.
1636 *
1637 * dumpable | mm->flags (binary)
1638 * old new | initial interim final
1639 * ---------+-----------------------
1640 * 0 1 | 00 01 01
1641 * 0 2 | 00 10(*) 11
1642 * 1 0 | 01 00 00
1643 * 1 2 | 01 11 11
1644 * 2 0 | 11 10(*) 00
1645 * 2 1 | 11 11 01
1646 *
1647 * (*) get_dumpable regards interim value of 10 as 11.
1648 */
1649void set_dumpable(struct mm_struct *mm, int value)
1650{
1651 switch (value) {
1652 case 0:
1653 clear_bit(MMF_DUMPABLE, &mm->flags);
1654 smp_wmb();
1655 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1656 break;
1657 case 1:
1658 set_bit(MMF_DUMPABLE, &mm->flags);
1659 smp_wmb();
1660 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1661 break;
1662 case 2:
1663 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1664 smp_wmb();
1665 set_bit(MMF_DUMPABLE, &mm->flags);
1666 break;
1667 }
1668}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001669
1670int get_dumpable(struct mm_struct *mm)
1671{
1672 int ret;
1673
1674 ret = mm->flags & 0x3;
1675 return (ret >= 2) ? 2 : ret;
1676}
1677
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1679{
1680 char corename[CORENAME_MAX_SIZE + 1];
1681 struct mm_struct *mm = current->mm;
1682 struct linux_binfmt * binfmt;
1683 struct inode * inode;
1684 struct file * file;
1685 int retval = 0;
Alan Coxd6e71142005-06-23 00:09:43 -07001686 int fsuid = current->fsuid;
1687 int flag = 0;
Andi Kleend025c9d2006-09-30 23:29:28 -07001688 int ispipe = 0;
Neil Horman7dc0b222007-10-16 23:26:34 -07001689 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
Neil Horman74aadce2007-10-16 23:26:35 -07001690 char **helper_argv = NULL;
1691 int helper_argc = 0;
1692 char *delimit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693
Steve Grubb0a4ff8c2007-04-19 10:28:21 -04001694 audit_core_dumps(signr);
1695
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 binfmt = current->binfmt;
1697 if (!binfmt || !binfmt->core_dump)
1698 goto fail;
1699 down_write(&mm->mmap_sem);
Roland McGrath00ec99d2007-11-11 19:13:43 -08001700 /*
1701 * If another thread got here first, or we are not dumpable, bail out.
1702 */
1703 if (mm->core_waiters || !get_dumpable(mm)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 up_write(&mm->mmap_sem);
1705 goto fail;
1706 }
Alan Coxd6e71142005-06-23 00:09:43 -07001707
1708 /*
1709 * We cannot trust fsuid as being the "true" uid of the
1710 * process nor do we know its entire history. We only know it
1711 * was tainted so we dump it as root in mode 2.
1712 */
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001713 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
Alan Coxd6e71142005-06-23 00:09:43 -07001714 flag = O_EXCL; /* Stop rewrite attacks */
1715 current->fsuid = 0; /* Dump root private */
1716 }
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001717
Oleg Nesterovdcf560c2006-06-26 00:26:08 -07001718 retval = coredump_wait(exit_code);
1719 if (retval < 0)
Oleg Nesterov1291cf42005-10-30 15:02:54 -08001720 goto fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
1722 /*
1723 * Clear any false indication of pending signals that might
1724 * be seen by the filesystem code called to write the core file.
1725 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 clear_thread_flag(TIF_SIGPENDING);
1727
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 /*
1729 * lock_kernel() because format_corename() is controlled by sysctl, which
1730 * uses lock_kernel()
1731 */
1732 lock_kernel();
Alan Coxc4bbafd2007-04-16 22:53:13 -07001733 ispipe = format_corename(corename, core_pattern, signr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 unlock_kernel();
Neil Horman7dc0b222007-10-16 23:26:34 -07001735 /*
1736 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1737 * to a pipe. Since we're not writing directly to the filesystem
1738 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1739 * created unless the pipe reader choses to write out the core file
1740 * at which point file size limits and permissions will be imposed
1741 * as it does with any other process
1742 */
Neil Horman74aadce2007-10-16 23:26:35 -07001743 if ((!ispipe) && (core_limit < binfmt->min_coredump))
Neil Horman7dc0b222007-10-16 23:26:34 -07001744 goto fail_unlock;
1745
Alan Coxc4bbafd2007-04-16 22:53:13 -07001746 if (ispipe) {
Neil Horman74aadce2007-10-16 23:26:35 -07001747 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1748 /* Terminate the string before the first option */
1749 delimit = strchr(corename, ' ');
1750 if (delimit)
1751 *delimit = '\0';
Neil Horman32321132007-10-16 23:26:36 -07001752 delimit = strrchr(helper_argv[0], '/');
1753 if (delimit)
1754 delimit++;
1755 else
1756 delimit = helper_argv[0];
1757 if (!strcmp(delimit, current->comm)) {
1758 printk(KERN_NOTICE "Recursive core dump detected, "
1759 "aborting\n");
1760 goto fail_unlock;
1761 }
1762
1763 core_limit = RLIM_INFINITY;
1764
Andi Kleend025c9d2006-09-30 23:29:28 -07001765 /* SIGPIPE can happen, but it's just never processed */
Neil Horman32321132007-10-16 23:26:36 -07001766 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1767 &file)) {
Andi Kleend025c9d2006-09-30 23:29:28 -07001768 printk(KERN_INFO "Core dump to %s pipe failed\n",
1769 corename);
1770 goto fail_unlock;
1771 }
Andi Kleend025c9d2006-09-30 23:29:28 -07001772 } else
1773 file = filp_open(corename,
Alexey Dobriyan6d4df672006-12-06 20:40:39 -08001774 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1775 0600);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 if (IS_ERR(file))
1777 goto fail_unlock;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001778 inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 if (inode->i_nlink > 1)
1780 goto close_fail; /* multiple links - don't dump */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001781 if (!ispipe && d_unhashed(file->f_path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 goto close_fail;
1783
Andi Kleend025c9d2006-09-30 23:29:28 -07001784 /* AK: actually i see no reason to not allow this for named pipes etc.,
1785 but keep the previous behaviour for now. */
1786 if (!ispipe && !S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 goto close_fail;
Ingo Molnarc46f7392007-11-28 13:59:18 +01001788 /*
1789 * Dont allow local users get cute and trick others to coredump
1790 * into their pre-created files:
1791 */
1792 if (inode->i_uid != current->fsuid)
1793 goto close_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 if (!file->f_op)
1795 goto close_fail;
1796 if (!file->f_op->write)
1797 goto close_fail;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001798 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 goto close_fail;
1800
Neil Horman7dc0b222007-10-16 23:26:34 -07001801 retval = binfmt->core_dump(signr, regs, file, core_limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
1803 if (retval)
1804 current->signal->group_exit_code |= 0x80;
1805close_fail:
1806 filp_close(file, NULL);
1807fail_unlock:
Neil Horman74aadce2007-10-16 23:26:35 -07001808 if (helper_argv)
1809 argv_free(helper_argv);
1810
Alan Coxd6e71142005-06-23 00:09:43 -07001811 current->fsuid = fsuid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 complete_all(&mm->core_done);
1813fail:
1814 return retval;
1815}