Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | ||
1da177e4 LT |
25 | #include <linux/slab.h> |
26 | #include <linux/file.h> | |
9f3acc31 | 27 | #include <linux/fdtable.h> |
ba92a43d | 28 | #include <linux/mm.h> |
1da177e4 LT |
29 | #include <linux/stat.h> |
30 | #include <linux/fcntl.h> | |
31 | #include <linux/smp_lock.h> | |
ba92a43d | 32 | #include <linux/swap.h> |
74aadce9 | 33 | #include <linux/string.h> |
1da177e4 | 34 | #include <linux/init.h> |
ca5b172b | 35 | #include <linux/pagemap.h> |
cdd6c482 | 36 | #include <linux/perf_event.h> |
1da177e4 LT |
37 | #include <linux/highmem.h> |
38 | #include <linux/spinlock.h> | |
39 | #include <linux/key.h> | |
40 | #include <linux/personality.h> | |
41 | #include <linux/binfmts.h> | |
1da177e4 | 42 | #include <linux/utsname.h> |
84d73786 | 43 | #include <linux/pid_namespace.h> |
1da177e4 LT |
44 | #include <linux/module.h> |
45 | #include <linux/namei.h> | |
46 | #include <linux/proc_fs.h> | |
1da177e4 LT |
47 | #include <linux/mount.h> |
48 | #include <linux/security.h> | |
49 | #include <linux/syscalls.h> | |
8f0ab514 | 50 | #include <linux/tsacct_kern.h> |
9f46080c | 51 | #include <linux/cn_proc.h> |
473ae30b | 52 | #include <linux/audit.h> |
6341c393 | 53 | #include <linux/tracehook.h> |
5f4123be | 54 | #include <linux/kmod.h> |
6110e3ab | 55 | #include <linux/fsnotify.h> |
5ad4e53b | 56 | #include <linux/fs_struct.h> |
61be228a | 57 | #include <linux/pipe_fs_i.h> |
1da177e4 LT |
58 | |
59 | #include <asm/uaccess.h> | |
60 | #include <asm/mmu_context.h> | |
b6a2fea3 | 61 | #include <asm/tlb.h> |
a6f76f23 | 62 | #include "internal.h" |
1da177e4 | 63 | |
1da177e4 | 64 | int core_uses_pid; |
71ce92f3 | 65 | char core_pattern[CORENAME_MAX_SIZE] = "core"; |
a293980c | 66 | unsigned int core_pipe_limit; |
d6e71144 AC |
67 | int suid_dumpable = 0; |
68 | ||
1da177e4 LT |
69 | /* The maximal length of core_pattern is also specified in sysctl.c */ |
70 | ||
e4dc1b14 | 71 | static LIST_HEAD(formats); |
1da177e4 LT |
72 | static DEFINE_RWLOCK(binfmt_lock); |
73 | ||
74641f58 | 74 | int __register_binfmt(struct linux_binfmt * fmt, int insert) |
1da177e4 | 75 | { |
1da177e4 LT |
76 | if (!fmt) |
77 | return -EINVAL; | |
1da177e4 | 78 | write_lock(&binfmt_lock); |
74641f58 IK |
79 | insert ? list_add(&fmt->lh, &formats) : |
80 | list_add_tail(&fmt->lh, &formats); | |
1da177e4 LT |
81 | write_unlock(&binfmt_lock); |
82 | return 0; | |
83 | } | |
84 | ||
74641f58 | 85 | EXPORT_SYMBOL(__register_binfmt); |
1da177e4 | 86 | |
f6b450d4 | 87 | void unregister_binfmt(struct linux_binfmt * fmt) |
1da177e4 | 88 | { |
1da177e4 | 89 | write_lock(&binfmt_lock); |
e4dc1b14 | 90 | list_del(&fmt->lh); |
1da177e4 | 91 | write_unlock(&binfmt_lock); |
1da177e4 LT |
92 | } |
93 | ||
94 | EXPORT_SYMBOL(unregister_binfmt); | |
95 | ||
96 | static 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 | */ | |
1e7bfb21 | 107 | SYSCALL_DEFINE1(uselib, const char __user *, library) |
1da177e4 | 108 | { |
964bd183 | 109 | struct file *file; |
964bd183 AV |
110 | char *tmp = getname(library); |
111 | int error = PTR_ERR(tmp); | |
112 | ||
6e8341a1 AV |
113 | if (IS_ERR(tmp)) |
114 | goto out; | |
115 | ||
116 | file = do_filp_open(AT_FDCWD, tmp, | |
117 | O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0, | |
118 | MAY_READ | MAY_EXEC | MAY_OPEN); | |
119 | putname(tmp); | |
120 | error = PTR_ERR(file); | |
121 | if (IS_ERR(file)) | |
1da177e4 LT |
122 | goto out; |
123 | ||
124 | error = -EINVAL; | |
6e8341a1 | 125 | if (!S_ISREG(file->f_path.dentry->d_inode->i_mode)) |
1da177e4 LT |
126 | goto exit; |
127 | ||
30524472 | 128 | error = -EACCES; |
6e8341a1 | 129 | if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) |
1da177e4 LT |
130 | goto exit; |
131 | ||
6110e3ab EP |
132 | fsnotify_open(file->f_path.dentry); |
133 | ||
1da177e4 LT |
134 | error = -ENOEXEC; |
135 | if(file->f_op) { | |
136 | struct linux_binfmt * fmt; | |
137 | ||
138 | read_lock(&binfmt_lock); | |
e4dc1b14 | 139 | list_for_each_entry(fmt, &formats, lh) { |
1da177e4 LT |
140 | if (!fmt->load_shlib) |
141 | continue; | |
142 | if (!try_module_get(fmt->module)) | |
143 | continue; | |
144 | read_unlock(&binfmt_lock); | |
145 | error = fmt->load_shlib(file); | |
146 | read_lock(&binfmt_lock); | |
147 | put_binfmt(fmt); | |
148 | if (error != -ENOEXEC) | |
149 | break; | |
150 | } | |
151 | read_unlock(&binfmt_lock); | |
152 | } | |
6e8341a1 | 153 | exit: |
1da177e4 LT |
154 | fput(file); |
155 | out: | |
156 | return error; | |
1da177e4 LT |
157 | } |
158 | ||
b6a2fea3 OW |
159 | #ifdef CONFIG_MMU |
160 | ||
161 | static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos, | |
162 | int write) | |
163 | { | |
164 | struct page *page; | |
165 | int ret; | |
166 | ||
167 | #ifdef CONFIG_STACK_GROWSUP | |
168 | if (write) { | |
169 | ret = expand_stack_downwards(bprm->vma, pos); | |
170 | if (ret < 0) | |
171 | return NULL; | |
172 | } | |
173 | #endif | |
174 | ret = get_user_pages(current, bprm->mm, pos, | |
175 | 1, write, 1, &page, NULL); | |
176 | if (ret <= 0) | |
177 | return NULL; | |
178 | ||
179 | if (write) { | |
b6a2fea3 | 180 | unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start; |
a64e715f LT |
181 | struct rlimit *rlim; |
182 | ||
183 | /* | |
184 | * We've historically supported up to 32 pages (ARG_MAX) | |
185 | * of argument strings even with small stacks | |
186 | */ | |
187 | if (size <= ARG_MAX) | |
188 | return page; | |
b6a2fea3 OW |
189 | |
190 | /* | |
191 | * Limit to 1/4-th the stack size for the argv+env strings. | |
192 | * This ensures that: | |
193 | * - the remaining binfmt code will not run out of stack space, | |
194 | * - the program will have a reasonable amount of stack left | |
195 | * to work from. | |
196 | */ | |
a64e715f | 197 | rlim = current->signal->rlim; |
d554ed89 | 198 | if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) { |
b6a2fea3 OW |
199 | put_page(page); |
200 | return NULL; | |
201 | } | |
202 | } | |
203 | ||
204 | return page; | |
205 | } | |
206 | ||
207 | static void put_arg_page(struct page *page) | |
208 | { | |
209 | put_page(page); | |
210 | } | |
211 | ||
212 | static void free_arg_page(struct linux_binprm *bprm, int i) | |
213 | { | |
214 | } | |
215 | ||
216 | static void free_arg_pages(struct linux_binprm *bprm) | |
217 | { | |
218 | } | |
219 | ||
220 | static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos, | |
221 | struct page *page) | |
222 | { | |
223 | flush_cache_page(bprm->vma, pos, page_to_pfn(page)); | |
224 | } | |
225 | ||
226 | static int __bprm_mm_init(struct linux_binprm *bprm) | |
227 | { | |
eaccbfa5 | 228 | int err; |
b6a2fea3 OW |
229 | struct vm_area_struct *vma = NULL; |
230 | struct mm_struct *mm = bprm->mm; | |
231 | ||
232 | bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
233 | if (!vma) | |
eaccbfa5 | 234 | return -ENOMEM; |
b6a2fea3 OW |
235 | |
236 | down_write(&mm->mmap_sem); | |
237 | vma->vm_mm = mm; | |
238 | ||
239 | /* | |
240 | * Place the stack at the largest stack address the architecture | |
241 | * supports. Later, we'll move this to an appropriate place. We don't | |
242 | * use STACK_TOP because that can depend on attributes which aren't | |
243 | * configured yet. | |
244 | */ | |
245 | vma->vm_end = STACK_TOP_MAX; | |
246 | vma->vm_start = vma->vm_end - PAGE_SIZE; | |
b6a2fea3 | 247 | vma->vm_flags = VM_STACK_FLAGS; |
3ed75eb8 | 248 | vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); |
5beb4930 | 249 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
b6a2fea3 | 250 | err = insert_vm_struct(mm, vma); |
eaccbfa5 | 251 | if (err) |
b6a2fea3 | 252 | goto err; |
b6a2fea3 OW |
253 | |
254 | mm->stack_vm = mm->total_vm = 1; | |
255 | up_write(&mm->mmap_sem); | |
b6a2fea3 | 256 | bprm->p = vma->vm_end - sizeof(void *); |
b6a2fea3 | 257 | return 0; |
b6a2fea3 | 258 | err: |
eaccbfa5 LFC |
259 | up_write(&mm->mmap_sem); |
260 | bprm->vma = NULL; | |
261 | kmem_cache_free(vm_area_cachep, vma); | |
b6a2fea3 OW |
262 | return err; |
263 | } | |
264 | ||
265 | static bool valid_arg_len(struct linux_binprm *bprm, long len) | |
266 | { | |
267 | return len <= MAX_ARG_STRLEN; | |
268 | } | |
269 | ||
270 | #else | |
271 | ||
272 | static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos, | |
273 | int write) | |
274 | { | |
275 | struct page *page; | |
276 | ||
277 | page = bprm->page[pos / PAGE_SIZE]; | |
278 | if (!page && write) { | |
279 | page = alloc_page(GFP_HIGHUSER|__GFP_ZERO); | |
280 | if (!page) | |
281 | return NULL; | |
282 | bprm->page[pos / PAGE_SIZE] = page; | |
283 | } | |
284 | ||
285 | return page; | |
286 | } | |
287 | ||
288 | static void put_arg_page(struct page *page) | |
289 | { | |
290 | } | |
291 | ||
292 | static void free_arg_page(struct linux_binprm *bprm, int i) | |
293 | { | |
294 | if (bprm->page[i]) { | |
295 | __free_page(bprm->page[i]); | |
296 | bprm->page[i] = NULL; | |
297 | } | |
298 | } | |
299 | ||
300 | static void free_arg_pages(struct linux_binprm *bprm) | |
301 | { | |
302 | int i; | |
303 | ||
304 | for (i = 0; i < MAX_ARG_PAGES; i++) | |
305 | free_arg_page(bprm, i); | |
306 | } | |
307 | ||
308 | static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos, | |
309 | struct page *page) | |
310 | { | |
311 | } | |
312 | ||
313 | static int __bprm_mm_init(struct linux_binprm *bprm) | |
314 | { | |
315 | bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *); | |
316 | return 0; | |
317 | } | |
318 | ||
319 | static bool valid_arg_len(struct linux_binprm *bprm, long len) | |
320 | { | |
321 | return len <= bprm->p; | |
322 | } | |
323 | ||
324 | #endif /* CONFIG_MMU */ | |
325 | ||
326 | /* | |
327 | * Create a new mm_struct and populate it with a temporary stack | |
328 | * vm_area_struct. We don't have enough context at this point to set the stack | |
329 | * flags, permissions, and offset, so we use temporary values. We'll update | |
330 | * them later in setup_arg_pages(). | |
331 | */ | |
332 | int bprm_mm_init(struct linux_binprm *bprm) | |
333 | { | |
334 | int err; | |
335 | struct mm_struct *mm = NULL; | |
336 | ||
337 | bprm->mm = mm = mm_alloc(); | |
338 | err = -ENOMEM; | |
339 | if (!mm) | |
340 | goto err; | |
341 | ||
342 | err = init_new_context(current, mm); | |
343 | if (err) | |
344 | goto err; | |
345 | ||
346 | err = __bprm_mm_init(bprm); | |
347 | if (err) | |
348 | goto err; | |
349 | ||
350 | return 0; | |
351 | ||
352 | err: | |
353 | if (mm) { | |
354 | bprm->mm = NULL; | |
355 | mmdrop(mm); | |
356 | } | |
357 | ||
358 | return err; | |
359 | } | |
360 | ||
1da177e4 LT |
361 | /* |
362 | * count() counts the number of strings in array ARGV. | |
363 | */ | |
364 | static int count(char __user * __user * argv, int max) | |
365 | { | |
366 | int i = 0; | |
367 | ||
368 | if (argv != NULL) { | |
369 | for (;;) { | |
370 | char __user * p; | |
371 | ||
372 | if (get_user(p, argv)) | |
373 | return -EFAULT; | |
374 | if (!p) | |
375 | break; | |
376 | argv++; | |
362e6663 | 377 | if (i++ >= max) |
1da177e4 LT |
378 | return -E2BIG; |
379 | cond_resched(); | |
380 | } | |
381 | } | |
382 | return i; | |
383 | } | |
384 | ||
385 | /* | |
b6a2fea3 OW |
386 | * 'copy_strings()' copies argument/environment strings from the old |
387 | * processes's memory to the new process's stack. The call to get_user_pages() | |
388 | * ensures the destination page is created and not swapped out. | |
1da177e4 | 389 | */ |
75c96f85 AB |
390 | static int copy_strings(int argc, char __user * __user * argv, |
391 | struct linux_binprm *bprm) | |
1da177e4 LT |
392 | { |
393 | struct page *kmapped_page = NULL; | |
394 | char *kaddr = NULL; | |
b6a2fea3 | 395 | unsigned long kpos = 0; |
1da177e4 LT |
396 | int ret; |
397 | ||
398 | while (argc-- > 0) { | |
399 | char __user *str; | |
400 | int len; | |
401 | unsigned long pos; | |
402 | ||
403 | if (get_user(str, argv+argc) || | |
b6a2fea3 | 404 | !(len = strnlen_user(str, MAX_ARG_STRLEN))) { |
1da177e4 LT |
405 | ret = -EFAULT; |
406 | goto out; | |
407 | } | |
408 | ||
b6a2fea3 | 409 | if (!valid_arg_len(bprm, len)) { |
1da177e4 LT |
410 | ret = -E2BIG; |
411 | goto out; | |
412 | } | |
413 | ||
b6a2fea3 | 414 | /* We're going to work our way backwords. */ |
1da177e4 | 415 | pos = bprm->p; |
b6a2fea3 OW |
416 | str += len; |
417 | bprm->p -= len; | |
1da177e4 LT |
418 | |
419 | while (len > 0) { | |
1da177e4 | 420 | int offset, bytes_to_copy; |
1da177e4 LT |
421 | |
422 | offset = pos % PAGE_SIZE; | |
b6a2fea3 OW |
423 | if (offset == 0) |
424 | offset = PAGE_SIZE; | |
425 | ||
426 | bytes_to_copy = offset; | |
427 | if (bytes_to_copy > len) | |
428 | bytes_to_copy = len; | |
429 | ||
430 | offset -= bytes_to_copy; | |
431 | pos -= bytes_to_copy; | |
432 | str -= bytes_to_copy; | |
433 | len -= bytes_to_copy; | |
434 | ||
435 | if (!kmapped_page || kpos != (pos & PAGE_MASK)) { | |
436 | struct page *page; | |
437 | ||
438 | page = get_arg_page(bprm, pos, 1); | |
1da177e4 | 439 | if (!page) { |
b6a2fea3 | 440 | ret = -E2BIG; |
1da177e4 LT |
441 | goto out; |
442 | } | |
1da177e4 | 443 | |
b6a2fea3 OW |
444 | if (kmapped_page) { |
445 | flush_kernel_dcache_page(kmapped_page); | |
1da177e4 | 446 | kunmap(kmapped_page); |
b6a2fea3 OW |
447 | put_arg_page(kmapped_page); |
448 | } | |
1da177e4 LT |
449 | kmapped_page = page; |
450 | kaddr = kmap(kmapped_page); | |
b6a2fea3 OW |
451 | kpos = pos & PAGE_MASK; |
452 | flush_arg_page(bprm, kpos, kmapped_page); | |
1da177e4 | 453 | } |
b6a2fea3 | 454 | if (copy_from_user(kaddr+offset, str, bytes_to_copy)) { |
1da177e4 LT |
455 | ret = -EFAULT; |
456 | goto out; | |
457 | } | |
1da177e4 LT |
458 | } |
459 | } | |
460 | ret = 0; | |
461 | out: | |
b6a2fea3 OW |
462 | if (kmapped_page) { |
463 | flush_kernel_dcache_page(kmapped_page); | |
1da177e4 | 464 | kunmap(kmapped_page); |
b6a2fea3 OW |
465 | put_arg_page(kmapped_page); |
466 | } | |
1da177e4 LT |
467 | return ret; |
468 | } | |
469 | ||
470 | /* | |
471 | * Like copy_strings, but get argv and its values from kernel memory. | |
472 | */ | |
473 | int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm) | |
474 | { | |
475 | int r; | |
476 | mm_segment_t oldfs = get_fs(); | |
477 | set_fs(KERNEL_DS); | |
478 | r = copy_strings(argc, (char __user * __user *)argv, bprm); | |
479 | set_fs(oldfs); | |
480 | return r; | |
481 | } | |
1da177e4 LT |
482 | EXPORT_SYMBOL(copy_strings_kernel); |
483 | ||
484 | #ifdef CONFIG_MMU | |
b6a2fea3 | 485 | |
1da177e4 | 486 | /* |
b6a2fea3 OW |
487 | * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once |
488 | * the binfmt code determines where the new stack should reside, we shift it to | |
489 | * its final location. The process proceeds as follows: | |
1da177e4 | 490 | * |
b6a2fea3 OW |
491 | * 1) Use shift to calculate the new vma endpoints. |
492 | * 2) Extend vma to cover both the old and new ranges. This ensures the | |
493 | * arguments passed to subsequent functions are consistent. | |
494 | * 3) Move vma's page tables to the new range. | |
495 | * 4) Free up any cleared pgd range. | |
496 | * 5) Shrink the vma to cover only the new range. | |
1da177e4 | 497 | */ |
b6a2fea3 | 498 | static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift) |
1da177e4 LT |
499 | { |
500 | struct mm_struct *mm = vma->vm_mm; | |
b6a2fea3 OW |
501 | unsigned long old_start = vma->vm_start; |
502 | unsigned long old_end = vma->vm_end; | |
503 | unsigned long length = old_end - old_start; | |
504 | unsigned long new_start = old_start - shift; | |
505 | unsigned long new_end = old_end - shift; | |
506 | struct mmu_gather *tlb; | |
1da177e4 | 507 | |
b6a2fea3 | 508 | BUG_ON(new_start > new_end); |
1da177e4 | 509 | |
b6a2fea3 OW |
510 | /* |
511 | * ensure there are no vmas between where we want to go | |
512 | * and where we are | |
513 | */ | |
514 | if (vma != find_vma(mm, new_start)) | |
515 | return -EFAULT; | |
516 | ||
517 | /* | |
518 | * cover the whole range: [new_start, old_end) | |
519 | */ | |
5beb4930 RR |
520 | if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL)) |
521 | return -ENOMEM; | |
b6a2fea3 OW |
522 | |
523 | /* | |
524 | * move the page tables downwards, on failure we rely on | |
525 | * process cleanup to remove whatever mess we made. | |
526 | */ | |
527 | if (length != move_page_tables(vma, old_start, | |
528 | vma, new_start, length)) | |
529 | return -ENOMEM; | |
530 | ||
531 | lru_add_drain(); | |
532 | tlb = tlb_gather_mmu(mm, 0); | |
533 | if (new_end > old_start) { | |
534 | /* | |
535 | * when the old and new regions overlap clear from new_end. | |
536 | */ | |
42b77728 | 537 | free_pgd_range(tlb, new_end, old_end, new_end, |
b6a2fea3 OW |
538 | vma->vm_next ? vma->vm_next->vm_start : 0); |
539 | } else { | |
540 | /* | |
541 | * otherwise, clean from old_start; this is done to not touch | |
542 | * the address space in [new_end, old_start) some architectures | |
543 | * have constraints on va-space that make this illegal (IA64) - | |
544 | * for the others its just a little faster. | |
545 | */ | |
42b77728 | 546 | free_pgd_range(tlb, old_start, old_end, new_end, |
b6a2fea3 | 547 | vma->vm_next ? vma->vm_next->vm_start : 0); |
1da177e4 | 548 | } |
b6a2fea3 OW |
549 | tlb_finish_mmu(tlb, new_end, old_end); |
550 | ||
551 | /* | |
5beb4930 | 552 | * Shrink the vma to just the new range. Always succeeds. |
b6a2fea3 OW |
553 | */ |
554 | vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL); | |
555 | ||
556 | return 0; | |
1da177e4 LT |
557 | } |
558 | ||
559 | #define EXTRA_STACK_VM_PAGES 20 /* random */ | |
560 | ||
b6a2fea3 OW |
561 | /* |
562 | * Finalizes the stack vm_area_struct. The flags and permissions are updated, | |
563 | * the stack is optionally relocated, and some extra space is added. | |
564 | */ | |
1da177e4 LT |
565 | int setup_arg_pages(struct linux_binprm *bprm, |
566 | unsigned long stack_top, | |
567 | int executable_stack) | |
568 | { | |
b6a2fea3 OW |
569 | unsigned long ret; |
570 | unsigned long stack_shift; | |
1da177e4 | 571 | struct mm_struct *mm = current->mm; |
b6a2fea3 OW |
572 | struct vm_area_struct *vma = bprm->vma; |
573 | struct vm_area_struct *prev = NULL; | |
574 | unsigned long vm_flags; | |
575 | unsigned long stack_base; | |
803bf5ec MN |
576 | unsigned long stack_size; |
577 | unsigned long stack_expand; | |
578 | unsigned long rlim_stack; | |
1da177e4 LT |
579 | |
580 | #ifdef CONFIG_STACK_GROWSUP | |
1da177e4 | 581 | /* Limit stack size to 1GB */ |
d554ed89 | 582 | stack_base = rlimit_max(RLIMIT_STACK); |
1da177e4 LT |
583 | if (stack_base > (1 << 30)) |
584 | stack_base = 1 << 30; | |
1da177e4 | 585 | |
b6a2fea3 OW |
586 | /* Make sure we didn't let the argument array grow too large. */ |
587 | if (vma->vm_end - vma->vm_start > stack_base) | |
588 | return -ENOMEM; | |
1da177e4 | 589 | |
b6a2fea3 | 590 | stack_base = PAGE_ALIGN(stack_top - stack_base); |
1da177e4 | 591 | |
b6a2fea3 OW |
592 | stack_shift = vma->vm_start - stack_base; |
593 | mm->arg_start = bprm->p - stack_shift; | |
594 | bprm->p = vma->vm_end - stack_shift; | |
1da177e4 | 595 | #else |
b6a2fea3 OW |
596 | stack_top = arch_align_stack(stack_top); |
597 | stack_top = PAGE_ALIGN(stack_top); | |
598 | stack_shift = vma->vm_end - stack_top; | |
599 | ||
600 | bprm->p -= stack_shift; | |
1da177e4 | 601 | mm->arg_start = bprm->p; |
1da177e4 LT |
602 | #endif |
603 | ||
1da177e4 | 604 | if (bprm->loader) |
b6a2fea3 OW |
605 | bprm->loader -= stack_shift; |
606 | bprm->exec -= stack_shift; | |
1da177e4 | 607 | |
1da177e4 | 608 | down_write(&mm->mmap_sem); |
96a8e13e | 609 | vm_flags = VM_STACK_FLAGS; |
b6a2fea3 OW |
610 | |
611 | /* | |
612 | * Adjust stack execute permissions; explicitly enable for | |
613 | * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone | |
614 | * (arch default) otherwise. | |
615 | */ | |
616 | if (unlikely(executable_stack == EXSTACK_ENABLE_X)) | |
617 | vm_flags |= VM_EXEC; | |
618 | else if (executable_stack == EXSTACK_DISABLE_X) | |
619 | vm_flags &= ~VM_EXEC; | |
620 | vm_flags |= mm->def_flags; | |
621 | ||
622 | ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end, | |
623 | vm_flags); | |
624 | if (ret) | |
625 | goto out_unlock; | |
626 | BUG_ON(prev != vma); | |
627 | ||
628 | /* Move stack pages down in memory. */ | |
629 | if (stack_shift) { | |
630 | ret = shift_arg_pages(vma, stack_shift); | |
fc63cf23 AB |
631 | if (ret) |
632 | goto out_unlock; | |
1da177e4 LT |
633 | } |
634 | ||
803bf5ec MN |
635 | stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE; |
636 | stack_size = vma->vm_end - vma->vm_start; | |
637 | /* | |
638 | * Align this down to a page boundary as expand_stack | |
639 | * will align it up. | |
640 | */ | |
641 | rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK; | |
b6a2fea3 | 642 | #ifdef CONFIG_STACK_GROWSUP |
803bf5ec MN |
643 | if (stack_size + stack_expand > rlim_stack) |
644 | stack_base = vma->vm_start + rlim_stack; | |
645 | else | |
646 | stack_base = vma->vm_end + stack_expand; | |
b6a2fea3 | 647 | #else |
803bf5ec MN |
648 | if (stack_size + stack_expand > rlim_stack) |
649 | stack_base = vma->vm_end - rlim_stack; | |
650 | else | |
651 | stack_base = vma->vm_start - stack_expand; | |
b6a2fea3 OW |
652 | #endif |
653 | ret = expand_stack(vma, stack_base); | |
654 | if (ret) | |
655 | ret = -EFAULT; | |
656 | ||
657 | out_unlock: | |
1da177e4 | 658 | up_write(&mm->mmap_sem); |
fc63cf23 | 659 | return ret; |
1da177e4 | 660 | } |
1da177e4 LT |
661 | EXPORT_SYMBOL(setup_arg_pages); |
662 | ||
1da177e4 LT |
663 | #endif /* CONFIG_MMU */ |
664 | ||
665 | struct file *open_exec(const char *name) | |
666 | { | |
1da177e4 | 667 | struct file *file; |
e56b6a5d | 668 | int err; |
1da177e4 | 669 | |
6e8341a1 AV |
670 | file = do_filp_open(AT_FDCWD, name, |
671 | O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0, | |
672 | MAY_EXEC | MAY_OPEN); | |
673 | if (IS_ERR(file)) | |
e56b6a5d CH |
674 | goto out; |
675 | ||
676 | err = -EACCES; | |
6e8341a1 AV |
677 | if (!S_ISREG(file->f_path.dentry->d_inode->i_mode)) |
678 | goto exit; | |
e56b6a5d | 679 | |
6e8341a1 AV |
680 | if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) |
681 | goto exit; | |
e56b6a5d | 682 | |
6110e3ab EP |
683 | fsnotify_open(file->f_path.dentry); |
684 | ||
e56b6a5d | 685 | err = deny_write_access(file); |
6e8341a1 AV |
686 | if (err) |
687 | goto exit; | |
1da177e4 | 688 | |
6e8341a1 | 689 | out: |
e56b6a5d CH |
690 | return file; |
691 | ||
6e8341a1 AV |
692 | exit: |
693 | fput(file); | |
e56b6a5d CH |
694 | return ERR_PTR(err); |
695 | } | |
1da177e4 LT |
696 | EXPORT_SYMBOL(open_exec); |
697 | ||
6777d773 MZ |
698 | int kernel_read(struct file *file, loff_t offset, |
699 | char *addr, unsigned long count) | |
1da177e4 LT |
700 | { |
701 | mm_segment_t old_fs; | |
702 | loff_t pos = offset; | |
703 | int result; | |
704 | ||
705 | old_fs = get_fs(); | |
706 | set_fs(get_ds()); | |
707 | /* The cast to a user pointer is valid due to the set_fs() */ | |
708 | result = vfs_read(file, (void __user *)addr, count, &pos); | |
709 | set_fs(old_fs); | |
710 | return result; | |
711 | } | |
712 | ||
713 | EXPORT_SYMBOL(kernel_read); | |
714 | ||
715 | static int exec_mmap(struct mm_struct *mm) | |
716 | { | |
717 | struct task_struct *tsk; | |
718 | struct mm_struct * old_mm, *active_mm; | |
719 | ||
720 | /* Notify parent that we're no longer interested in the old VM */ | |
721 | tsk = current; | |
722 | old_mm = current->mm; | |
34e55232 | 723 | sync_mm_rss(tsk, old_mm); |
1da177e4 LT |
724 | mm_release(tsk, old_mm); |
725 | ||
726 | if (old_mm) { | |
727 | /* | |
728 | * Make sure that if there is a core dump in progress | |
729 | * for the old mm, we get out and die instead of going | |
730 | * through with the exec. We must hold mmap_sem around | |
999d9fc1 | 731 | * checking core_state and changing tsk->mm. |
1da177e4 LT |
732 | */ |
733 | down_read(&old_mm->mmap_sem); | |
999d9fc1 | 734 | if (unlikely(old_mm->core_state)) { |
1da177e4 LT |
735 | up_read(&old_mm->mmap_sem); |
736 | return -EINTR; | |
737 | } | |
738 | } | |
739 | task_lock(tsk); | |
740 | active_mm = tsk->active_mm; | |
741 | tsk->mm = mm; | |
742 | tsk->active_mm = mm; | |
743 | activate_mm(active_mm, mm); | |
744 | task_unlock(tsk); | |
745 | arch_pick_mmap_layout(mm); | |
746 | if (old_mm) { | |
747 | up_read(&old_mm->mmap_sem); | |
7dddb12c | 748 | BUG_ON(active_mm != old_mm); |
31a78f23 | 749 | mm_update_next_owner(old_mm); |
1da177e4 LT |
750 | mmput(old_mm); |
751 | return 0; | |
752 | } | |
753 | mmdrop(active_mm); | |
754 | return 0; | |
755 | } | |
756 | ||
757 | /* | |
758 | * This function makes sure the current process has its own signal table, | |
759 | * so that flush_signal_handlers can later reset the handlers without | |
760 | * disturbing other processes. (Other processes might share the signal | |
761 | * table via the CLONE_SIGHAND option to clone().) | |
762 | */ | |
858119e1 | 763 | static int de_thread(struct task_struct *tsk) |
1da177e4 LT |
764 | { |
765 | struct signal_struct *sig = tsk->signal; | |
b2c903b8 | 766 | struct sighand_struct *oldsighand = tsk->sighand; |
1da177e4 LT |
767 | spinlock_t *lock = &oldsighand->siglock; |
768 | int count; | |
769 | ||
aafe6c2a | 770 | if (thread_group_empty(tsk)) |
1da177e4 LT |
771 | goto no_thread_group; |
772 | ||
773 | /* | |
774 | * Kill all other threads in the thread group. | |
1da177e4 | 775 | */ |
1da177e4 | 776 | spin_lock_irq(lock); |
ed5d2cac | 777 | if (signal_group_exit(sig)) { |
1da177e4 LT |
778 | /* |
779 | * Another group action in progress, just | |
780 | * return so that the signal is processed. | |
781 | */ | |
782 | spin_unlock_irq(lock); | |
1da177e4 LT |
783 | return -EAGAIN; |
784 | } | |
ed5d2cac | 785 | sig->group_exit_task = tsk; |
aafe6c2a | 786 | zap_other_threads(tsk); |
1da177e4 | 787 | |
fea9d175 ON |
788 | /* Account for the thread group leader hanging around: */ |
789 | count = thread_group_leader(tsk) ? 1 : 2; | |
6db840fa | 790 | sig->notify_count = count; |
1da177e4 | 791 | while (atomic_read(&sig->count) > count) { |
1da177e4 LT |
792 | __set_current_state(TASK_UNINTERRUPTIBLE); |
793 | spin_unlock_irq(lock); | |
794 | schedule(); | |
795 | spin_lock_irq(lock); | |
796 | } | |
1da177e4 LT |
797 | spin_unlock_irq(lock); |
798 | ||
799 | /* | |
800 | * At this point all other threads have exited, all we have to | |
801 | * do is to wait for the thread group leader to become inactive, | |
802 | * and to assume its PID: | |
803 | */ | |
aafe6c2a | 804 | if (!thread_group_leader(tsk)) { |
8187926b | 805 | struct task_struct *leader = tsk->group_leader; |
6db840fa | 806 | |
2800d8d1 | 807 | sig->notify_count = -1; /* for exit_notify() */ |
6db840fa ON |
808 | for (;;) { |
809 | write_lock_irq(&tasklist_lock); | |
810 | if (likely(leader->exit_state)) | |
811 | break; | |
812 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
813 | write_unlock_irq(&tasklist_lock); | |
814 | schedule(); | |
815 | } | |
1da177e4 | 816 | |
f5e90281 RM |
817 | /* |
818 | * The only record we have of the real-time age of a | |
819 | * process, regardless of execs it's done, is start_time. | |
820 | * All the past CPU time is accumulated in signal_struct | |
821 | * from sister threads now dead. But in this non-leader | |
822 | * exec, nothing survives from the original leader thread, | |
823 | * whose birth marks the true age of this process now. | |
824 | * When we take on its identity by switching to its PID, we | |
825 | * also take its birthdate (always earlier than our own). | |
826 | */ | |
aafe6c2a | 827 | tsk->start_time = leader->start_time; |
f5e90281 | 828 | |
bac0abd6 PE |
829 | BUG_ON(!same_thread_group(leader, tsk)); |
830 | BUG_ON(has_group_leader_pid(tsk)); | |
1da177e4 LT |
831 | /* |
832 | * An exec() starts a new thread group with the | |
833 | * TGID of the previous thread group. Rehash the | |
834 | * two threads with a switched PID, and release | |
835 | * the former thread group leader: | |
836 | */ | |
d73d6529 EB |
837 | |
838 | /* Become a process group leader with the old leader's pid. | |
c18258c6 EB |
839 | * The old leader becomes a thread of the this thread group. |
840 | * Note: The old leader also uses this pid until release_task | |
d73d6529 EB |
841 | * is called. Odd but simple and correct. |
842 | */ | |
aafe6c2a EB |
843 | detach_pid(tsk, PIDTYPE_PID); |
844 | tsk->pid = leader->pid; | |
3743ca05 | 845 | attach_pid(tsk, PIDTYPE_PID, task_pid(leader)); |
aafe6c2a EB |
846 | transfer_pid(leader, tsk, PIDTYPE_PGID); |
847 | transfer_pid(leader, tsk, PIDTYPE_SID); | |
9cd80bbb | 848 | |
aafe6c2a | 849 | list_replace_rcu(&leader->tasks, &tsk->tasks); |
9cd80bbb | 850 | list_replace_init(&leader->sibling, &tsk->sibling); |
1da177e4 | 851 | |
aafe6c2a EB |
852 | tsk->group_leader = tsk; |
853 | leader->group_leader = tsk; | |
de12a787 | 854 | |
aafe6c2a | 855 | tsk->exit_signal = SIGCHLD; |
962b564c ON |
856 | |
857 | BUG_ON(leader->exit_state != EXIT_ZOMBIE); | |
858 | leader->exit_state = EXIT_DEAD; | |
1da177e4 | 859 | write_unlock_irq(&tasklist_lock); |
8187926b ON |
860 | |
861 | release_task(leader); | |
ed5d2cac | 862 | } |
1da177e4 | 863 | |
6db840fa ON |
864 | sig->group_exit_task = NULL; |
865 | sig->notify_count = 0; | |
1da177e4 LT |
866 | |
867 | no_thread_group: | |
1f10206c JP |
868 | if (current->mm) |
869 | setmax_mm_hiwater_rss(&sig->maxrss, current->mm); | |
870 | ||
1da177e4 | 871 | exit_itimers(sig); |
cbaffba1 | 872 | flush_itimer_signals(); |
329f7dba | 873 | |
b2c903b8 ON |
874 | if (atomic_read(&oldsighand->count) != 1) { |
875 | struct sighand_struct *newsighand; | |
1da177e4 | 876 | /* |
b2c903b8 ON |
877 | * This ->sighand is shared with the CLONE_SIGHAND |
878 | * but not CLONE_THREAD task, switch to the new one. | |
1da177e4 | 879 | */ |
b2c903b8 ON |
880 | newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); |
881 | if (!newsighand) | |
882 | return -ENOMEM; | |
883 | ||
1da177e4 LT |
884 | atomic_set(&newsighand->count, 1); |
885 | memcpy(newsighand->action, oldsighand->action, | |
886 | sizeof(newsighand->action)); | |
887 | ||
888 | write_lock_irq(&tasklist_lock); | |
889 | spin_lock(&oldsighand->siglock); | |
aafe6c2a | 890 | rcu_assign_pointer(tsk->sighand, newsighand); |
1da177e4 LT |
891 | spin_unlock(&oldsighand->siglock); |
892 | write_unlock_irq(&tasklist_lock); | |
893 | ||
fba2afaa | 894 | __cleanup_sighand(oldsighand); |
1da177e4 LT |
895 | } |
896 | ||
aafe6c2a | 897 | BUG_ON(!thread_group_leader(tsk)); |
1da177e4 LT |
898 | return 0; |
899 | } | |
0840a90d | 900 | |
1da177e4 LT |
901 | /* |
902 | * These functions flushes out all traces of the currently running executable | |
903 | * so that a new one can be started | |
904 | */ | |
858119e1 | 905 | static void flush_old_files(struct files_struct * files) |
1da177e4 LT |
906 | { |
907 | long j = -1; | |
badf1662 | 908 | struct fdtable *fdt; |
1da177e4 LT |
909 | |
910 | spin_lock(&files->file_lock); | |
911 | for (;;) { | |
912 | unsigned long set, i; | |
913 | ||
914 | j++; | |
915 | i = j * __NFDBITS; | |
badf1662 | 916 | fdt = files_fdtable(files); |
bbea9f69 | 917 | if (i >= fdt->max_fds) |
1da177e4 | 918 | break; |
badf1662 | 919 | set = fdt->close_on_exec->fds_bits[j]; |
1da177e4 LT |
920 | if (!set) |
921 | continue; | |
badf1662 | 922 | fdt->close_on_exec->fds_bits[j] = 0; |
1da177e4 LT |
923 | spin_unlock(&files->file_lock); |
924 | for ( ; set ; i++,set >>= 1) { | |
925 | if (set & 1) { | |
926 | sys_close(i); | |
927 | } | |
928 | } | |
929 | spin_lock(&files->file_lock); | |
930 | ||
931 | } | |
932 | spin_unlock(&files->file_lock); | |
933 | } | |
934 | ||
59714d65 | 935 | char *get_task_comm(char *buf, struct task_struct *tsk) |
1da177e4 LT |
936 | { |
937 | /* buf must be at least sizeof(tsk->comm) in size */ | |
938 | task_lock(tsk); | |
939 | strncpy(buf, tsk->comm, sizeof(tsk->comm)); | |
940 | task_unlock(tsk); | |
59714d65 | 941 | return buf; |
1da177e4 LT |
942 | } |
943 | ||
944 | void set_task_comm(struct task_struct *tsk, char *buf) | |
945 | { | |
946 | task_lock(tsk); | |
4614a696 | 947 | |
948 | /* | |
949 | * Threads may access current->comm without holding | |
950 | * the task lock, so write the string carefully. | |
951 | * Readers without a lock may see incomplete new | |
952 | * names but are safe from non-terminating string reads. | |
953 | */ | |
954 | memset(tsk->comm, 0, TASK_COMM_LEN); | |
955 | wmb(); | |
1da177e4 LT |
956 | strlcpy(tsk->comm, buf, sizeof(tsk->comm)); |
957 | task_unlock(tsk); | |
cdd6c482 | 958 | perf_event_comm(tsk); |
1da177e4 LT |
959 | } |
960 | ||
961 | int flush_old_exec(struct linux_binprm * bprm) | |
962 | { | |
221af7f8 | 963 | int retval; |
1da177e4 LT |
964 | |
965 | /* | |
966 | * Make sure we have a private signal table and that | |
967 | * we are unassociated from the previous thread group. | |
968 | */ | |
969 | retval = de_thread(current); | |
970 | if (retval) | |
971 | goto out; | |
972 | ||
925d1c40 MH |
973 | set_mm_exe_file(bprm->mm, bprm->file); |
974 | ||
1da177e4 LT |
975 | /* |
976 | * Release all of the old mmap stuff | |
977 | */ | |
978 | retval = exec_mmap(bprm->mm); | |
979 | if (retval) | |
fd8328be | 980 | goto out; |
1da177e4 LT |
981 | |
982 | bprm->mm = NULL; /* We're using it now */ | |
7ab02af4 LT |
983 | |
984 | current->flags &= ~PF_RANDOMIZE; | |
985 | flush_thread(); | |
986 | current->personality &= ~bprm->per_clear; | |
987 | ||
221af7f8 LT |
988 | return 0; |
989 | ||
990 | out: | |
991 | return retval; | |
992 | } | |
993 | EXPORT_SYMBOL(flush_old_exec); | |
994 | ||
995 | void setup_new_exec(struct linux_binprm * bprm) | |
996 | { | |
997 | int i, ch; | |
998 | char * name; | |
999 | char tcomm[sizeof(current->comm)]; | |
1000 | ||
1001 | arch_pick_mmap_layout(current->mm); | |
1da177e4 LT |
1002 | |
1003 | /* This is the point of no return */ | |
1da177e4 LT |
1004 | current->sas_ss_sp = current->sas_ss_size = 0; |
1005 | ||
da9592ed | 1006 | if (current_euid() == current_uid() && current_egid() == current_gid()) |
6c5d5238 | 1007 | set_dumpable(current->mm, 1); |
d6e71144 | 1008 | else |
6c5d5238 | 1009 | set_dumpable(current->mm, suid_dumpable); |
d6e71144 | 1010 | |
1da177e4 | 1011 | name = bprm->filename; |
36772092 PBG |
1012 | |
1013 | /* Copies the binary name from after last slash */ | |
1da177e4 LT |
1014 | for (i=0; (ch = *(name++)) != '\0';) { |
1015 | if (ch == '/') | |
36772092 | 1016 | i = 0; /* overwrite what we wrote */ |
1da177e4 LT |
1017 | else |
1018 | if (i < (sizeof(tcomm) - 1)) | |
1019 | tcomm[i++] = ch; | |
1020 | } | |
1021 | tcomm[i] = '\0'; | |
1022 | set_task_comm(current, tcomm); | |
1023 | ||
0551fbd2 BH |
1024 | /* Set the new mm task size. We have to do that late because it may |
1025 | * depend on TIF_32BIT which is only updated in flush_thread() on | |
1026 | * some architectures like powerpc | |
1027 | */ | |
1028 | current->mm->task_size = TASK_SIZE; | |
1029 | ||
a6f76f23 DH |
1030 | /* install the new credentials */ |
1031 | if (bprm->cred->uid != current_euid() || | |
1032 | bprm->cred->gid != current_egid()) { | |
d2d56c5f MH |
1033 | current->pdeath_signal = 0; |
1034 | } else if (file_permission(bprm->file, MAY_READ) || | |
a6f76f23 | 1035 | bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) { |
6c5d5238 | 1036 | set_dumpable(current->mm, suid_dumpable); |
1da177e4 LT |
1037 | } |
1038 | ||
f65cb45c IM |
1039 | /* |
1040 | * Flush performance counters when crossing a | |
1041 | * security domain: | |
1042 | */ | |
1043 | if (!get_dumpable(current->mm)) | |
cdd6c482 | 1044 | perf_event_exit_task(current); |
f65cb45c | 1045 | |
1da177e4 LT |
1046 | /* An exec changes our domain. We are no longer part of the thread |
1047 | group */ | |
1048 | ||
1049 | current->self_exec_id++; | |
1050 | ||
1051 | flush_signal_handlers(current, 0); | |
1052 | flush_old_files(current->files); | |
1da177e4 | 1053 | } |
221af7f8 | 1054 | EXPORT_SYMBOL(setup_new_exec); |
1da177e4 | 1055 | |
a2a8474c ON |
1056 | /* |
1057 | * Prepare credentials and lock ->cred_guard_mutex. | |
1058 | * install_exec_creds() commits the new creds and drops the lock. | |
1059 | * Or, if exec fails before, free_bprm() should release ->cred and | |
1060 | * and unlock. | |
1061 | */ | |
1062 | int prepare_bprm_creds(struct linux_binprm *bprm) | |
1063 | { | |
1064 | if (mutex_lock_interruptible(¤t->cred_guard_mutex)) | |
1065 | return -ERESTARTNOINTR; | |
1066 | ||
1067 | bprm->cred = prepare_exec_creds(); | |
1068 | if (likely(bprm->cred)) | |
1069 | return 0; | |
1070 | ||
1071 | mutex_unlock(¤t->cred_guard_mutex); | |
1072 | return -ENOMEM; | |
1073 | } | |
1074 | ||
1075 | void free_bprm(struct linux_binprm *bprm) | |
1076 | { | |
1077 | free_arg_pages(bprm); | |
1078 | if (bprm->cred) { | |
1079 | mutex_unlock(¤t->cred_guard_mutex); | |
1080 | abort_creds(bprm->cred); | |
1081 | } | |
1082 | kfree(bprm); | |
1083 | } | |
1084 | ||
a6f76f23 DH |
1085 | /* |
1086 | * install the new credentials for this executable | |
1087 | */ | |
1088 | void install_exec_creds(struct linux_binprm *bprm) | |
1089 | { | |
1090 | security_bprm_committing_creds(bprm); | |
1091 | ||
1092 | commit_creds(bprm->cred); | |
1093 | bprm->cred = NULL; | |
a2a8474c ON |
1094 | /* |
1095 | * cred_guard_mutex must be held at least to this point to prevent | |
a6f76f23 | 1096 | * ptrace_attach() from altering our determination of the task's |
a2a8474c ON |
1097 | * credentials; any time after this it may be unlocked. |
1098 | */ | |
a6f76f23 | 1099 | security_bprm_committed_creds(bprm); |
a2a8474c | 1100 | mutex_unlock(¤t->cred_guard_mutex); |
a6f76f23 DH |
1101 | } |
1102 | EXPORT_SYMBOL(install_exec_creds); | |
1103 | ||
1104 | /* | |
1105 | * determine how safe it is to execute the proposed program | |
5e751e99 | 1106 | * - the caller must hold current->cred_guard_mutex to protect against |
a6f76f23 DH |
1107 | * PTRACE_ATTACH |
1108 | */ | |
498052bb | 1109 | int check_unsafe_exec(struct linux_binprm *bprm) |
a6f76f23 | 1110 | { |
0bf2f3ae | 1111 | struct task_struct *p = current, *t; |
f1191b50 | 1112 | unsigned n_fs; |
498052bb | 1113 | int res = 0; |
a6f76f23 DH |
1114 | |
1115 | bprm->unsafe = tracehook_unsafe_exec(p); | |
1116 | ||
0bf2f3ae | 1117 | n_fs = 1; |
498052bb | 1118 | write_lock(&p->fs->lock); |
437f7fdb | 1119 | rcu_read_lock(); |
0bf2f3ae DH |
1120 | for (t = next_thread(p); t != p; t = next_thread(t)) { |
1121 | if (t->fs == p->fs) | |
1122 | n_fs++; | |
0bf2f3ae | 1123 | } |
437f7fdb | 1124 | rcu_read_unlock(); |
0bf2f3ae | 1125 | |
f1191b50 | 1126 | if (p->fs->users > n_fs) { |
a6f76f23 | 1127 | bprm->unsafe |= LSM_UNSAFE_SHARE; |
498052bb | 1128 | } else { |
8c652f96 ON |
1129 | res = -EAGAIN; |
1130 | if (!p->fs->in_exec) { | |
1131 | p->fs->in_exec = 1; | |
1132 | res = 1; | |
1133 | } | |
498052bb | 1134 | } |
498052bb AV |
1135 | write_unlock(&p->fs->lock); |
1136 | ||
1137 | return res; | |
a6f76f23 DH |
1138 | } |
1139 | ||
1da177e4 LT |
1140 | /* |
1141 | * Fill the binprm structure from the inode. | |
1142 | * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes | |
a6f76f23 DH |
1143 | * |
1144 | * This may be called multiple times for binary chains (scripts for example). | |
1da177e4 LT |
1145 | */ |
1146 | int prepare_binprm(struct linux_binprm *bprm) | |
1147 | { | |
a6f76f23 | 1148 | umode_t mode; |
0f7fc9e4 | 1149 | struct inode * inode = bprm->file->f_path.dentry->d_inode; |
1da177e4 LT |
1150 | int retval; |
1151 | ||
1152 | mode = inode->i_mode; | |
1da177e4 LT |
1153 | if (bprm->file->f_op == NULL) |
1154 | return -EACCES; | |
1155 | ||
a6f76f23 DH |
1156 | /* clear any previous set[ug]id data from a previous binary */ |
1157 | bprm->cred->euid = current_euid(); | |
1158 | bprm->cred->egid = current_egid(); | |
1da177e4 | 1159 | |
a6f76f23 | 1160 | if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) { |
1da177e4 LT |
1161 | /* Set-uid? */ |
1162 | if (mode & S_ISUID) { | |
a6f76f23 DH |
1163 | bprm->per_clear |= PER_CLEAR_ON_SETID; |
1164 | bprm->cred->euid = inode->i_uid; | |
1da177e4 LT |
1165 | } |
1166 | ||
1167 | /* Set-gid? */ | |
1168 | /* | |
1169 | * If setgid is set but no group execute bit then this | |
1170 | * is a candidate for mandatory locking, not a setgid | |
1171 | * executable. | |
1172 | */ | |
1173 | if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) { | |
a6f76f23 DH |
1174 | bprm->per_clear |= PER_CLEAR_ON_SETID; |
1175 | bprm->cred->egid = inode->i_gid; | |
1da177e4 LT |
1176 | } |
1177 | } | |
1178 | ||
1179 | /* fill in binprm security blob */ | |
a6f76f23 | 1180 | retval = security_bprm_set_creds(bprm); |
1da177e4 LT |
1181 | if (retval) |
1182 | return retval; | |
a6f76f23 | 1183 | bprm->cred_prepared = 1; |
1da177e4 | 1184 | |
a6f76f23 DH |
1185 | memset(bprm->buf, 0, BINPRM_BUF_SIZE); |
1186 | return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE); | |
1da177e4 LT |
1187 | } |
1188 | ||
1189 | EXPORT_SYMBOL(prepare_binprm); | |
1190 | ||
4fc75ff4 NP |
1191 | /* |
1192 | * Arguments are '\0' separated strings found at the location bprm->p | |
1193 | * points to; chop off the first by relocating brpm->p to right after | |
1194 | * the first '\0' encountered. | |
1195 | */ | |
b6a2fea3 | 1196 | int remove_arg_zero(struct linux_binprm *bprm) |
1da177e4 | 1197 | { |
b6a2fea3 OW |
1198 | int ret = 0; |
1199 | unsigned long offset; | |
1200 | char *kaddr; | |
1201 | struct page *page; | |
4fc75ff4 | 1202 | |
b6a2fea3 OW |
1203 | if (!bprm->argc) |
1204 | return 0; | |
1da177e4 | 1205 | |
b6a2fea3 OW |
1206 | do { |
1207 | offset = bprm->p & ~PAGE_MASK; | |
1208 | page = get_arg_page(bprm, bprm->p, 0); | |
1209 | if (!page) { | |
1210 | ret = -EFAULT; | |
1211 | goto out; | |
1212 | } | |
1213 | kaddr = kmap_atomic(page, KM_USER0); | |
4fc75ff4 | 1214 | |
b6a2fea3 OW |
1215 | for (; offset < PAGE_SIZE && kaddr[offset]; |
1216 | offset++, bprm->p++) | |
1217 | ; | |
4fc75ff4 | 1218 | |
b6a2fea3 OW |
1219 | kunmap_atomic(kaddr, KM_USER0); |
1220 | put_arg_page(page); | |
4fc75ff4 | 1221 | |
b6a2fea3 OW |
1222 | if (offset == PAGE_SIZE) |
1223 | free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1); | |
1224 | } while (offset == PAGE_SIZE); | |
4fc75ff4 | 1225 | |
b6a2fea3 OW |
1226 | bprm->p++; |
1227 | bprm->argc--; | |
1228 | ret = 0; | |
4fc75ff4 | 1229 | |
b6a2fea3 OW |
1230 | out: |
1231 | return ret; | |
1da177e4 | 1232 | } |
1da177e4 LT |
1233 | EXPORT_SYMBOL(remove_arg_zero); |
1234 | ||
1235 | /* | |
1236 | * cycle the list of binary formats handler, until one recognizes the image | |
1237 | */ | |
1238 | int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs) | |
1239 | { | |
85f33466 | 1240 | unsigned int depth = bprm->recursion_depth; |
1da177e4 LT |
1241 | int try,retval; |
1242 | struct linux_binfmt *fmt; | |
1da177e4 | 1243 | |
1da177e4 LT |
1244 | retval = security_bprm_check(bprm); |
1245 | if (retval) | |
1246 | return retval; | |
1247 | ||
1248 | /* kernel module loader fixup */ | |
1249 | /* so we don't try to load run modprobe in kernel space. */ | |
1250 | set_fs(USER_DS); | |
473ae30b AV |
1251 | |
1252 | retval = audit_bprm(bprm); | |
1253 | if (retval) | |
1254 | return retval; | |
1255 | ||
1da177e4 LT |
1256 | retval = -ENOENT; |
1257 | for (try=0; try<2; try++) { | |
1258 | read_lock(&binfmt_lock); | |
e4dc1b14 | 1259 | list_for_each_entry(fmt, &formats, lh) { |
1da177e4 LT |
1260 | int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary; |
1261 | if (!fn) | |
1262 | continue; | |
1263 | if (!try_module_get(fmt->module)) | |
1264 | continue; | |
1265 | read_unlock(&binfmt_lock); | |
1266 | retval = fn(bprm, regs); | |
85f33466 RM |
1267 | /* |
1268 | * Restore the depth counter to its starting value | |
1269 | * in this call, so we don't have to rely on every | |
1270 | * load_binary function to restore it on return. | |
1271 | */ | |
1272 | bprm->recursion_depth = depth; | |
1da177e4 | 1273 | if (retval >= 0) { |
85f33466 RM |
1274 | if (depth == 0) |
1275 | tracehook_report_exec(fmt, bprm, regs); | |
1da177e4 LT |
1276 | put_binfmt(fmt); |
1277 | allow_write_access(bprm->file); | |
1278 | if (bprm->file) | |
1279 | fput(bprm->file); | |
1280 | bprm->file = NULL; | |
1281 | current->did_exec = 1; | |
9f46080c | 1282 | proc_exec_connector(current); |
1da177e4 LT |
1283 | return retval; |
1284 | } | |
1285 | read_lock(&binfmt_lock); | |
1286 | put_binfmt(fmt); | |
1287 | if (retval != -ENOEXEC || bprm->mm == NULL) | |
1288 | break; | |
1289 | if (!bprm->file) { | |
1290 | read_unlock(&binfmt_lock); | |
1291 | return retval; | |
1292 | } | |
1293 | } | |
1294 | read_unlock(&binfmt_lock); | |
1295 | if (retval != -ENOEXEC || bprm->mm == NULL) { | |
1296 | break; | |
5f4123be JB |
1297 | #ifdef CONFIG_MODULES |
1298 | } else { | |
1da177e4 LT |
1299 | #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e)) |
1300 | if (printable(bprm->buf[0]) && | |
1301 | printable(bprm->buf[1]) && | |
1302 | printable(bprm->buf[2]) && | |
1303 | printable(bprm->buf[3])) | |
1304 | break; /* -ENOEXEC */ | |
1305 | request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2])); | |
1306 | #endif | |
1307 | } | |
1308 | } | |
1309 | return retval; | |
1310 | } | |
1311 | ||
1312 | EXPORT_SYMBOL(search_binary_handler); | |
1313 | ||
1314 | /* | |
1315 | * sys_execve() executes a new program. | |
1316 | */ | |
1317 | int do_execve(char * filename, | |
1318 | char __user *__user *argv, | |
1319 | char __user *__user *envp, | |
1320 | struct pt_regs * regs) | |
1321 | { | |
1322 | struct linux_binprm *bprm; | |
1323 | struct file *file; | |
3b125388 | 1324 | struct files_struct *displaced; |
8c652f96 | 1325 | bool clear_in_exec; |
1da177e4 | 1326 | int retval; |
1da177e4 | 1327 | |
3b125388 | 1328 | retval = unshare_files(&displaced); |
fd8328be AV |
1329 | if (retval) |
1330 | goto out_ret; | |
1331 | ||
1da177e4 | 1332 | retval = -ENOMEM; |
11b0b5ab | 1333 | bprm = kzalloc(sizeof(*bprm), GFP_KERNEL); |
1da177e4 | 1334 | if (!bprm) |
fd8328be | 1335 | goto out_files; |
1da177e4 | 1336 | |
a2a8474c ON |
1337 | retval = prepare_bprm_creds(bprm); |
1338 | if (retval) | |
a6f76f23 | 1339 | goto out_free; |
498052bb AV |
1340 | |
1341 | retval = check_unsafe_exec(bprm); | |
8c652f96 | 1342 | if (retval < 0) |
a2a8474c | 1343 | goto out_free; |
8c652f96 | 1344 | clear_in_exec = retval; |
a2a8474c | 1345 | current->in_execve = 1; |
a6f76f23 | 1346 | |
1da177e4 LT |
1347 | file = open_exec(filename); |
1348 | retval = PTR_ERR(file); | |
1349 | if (IS_ERR(file)) | |
498052bb | 1350 | goto out_unmark; |
1da177e4 LT |
1351 | |
1352 | sched_exec(); | |
1353 | ||
1da177e4 LT |
1354 | bprm->file = file; |
1355 | bprm->filename = filename; | |
1356 | bprm->interp = filename; | |
1da177e4 | 1357 | |
b6a2fea3 OW |
1358 | retval = bprm_mm_init(bprm); |
1359 | if (retval) | |
1360 | goto out_file; | |
1da177e4 | 1361 | |
b6a2fea3 | 1362 | bprm->argc = count(argv, MAX_ARG_STRINGS); |
1da177e4 | 1363 | if ((retval = bprm->argc) < 0) |
a6f76f23 | 1364 | goto out; |
1da177e4 | 1365 | |
b6a2fea3 | 1366 | bprm->envc = count(envp, MAX_ARG_STRINGS); |
1da177e4 | 1367 | if ((retval = bprm->envc) < 0) |
1da177e4 LT |
1368 | goto out; |
1369 | ||
1370 | retval = prepare_binprm(bprm); | |
1371 | if (retval < 0) | |
1372 | goto out; | |
1373 | ||
1374 | retval = copy_strings_kernel(1, &bprm->filename, bprm); | |
1375 | if (retval < 0) | |
1376 | goto out; | |
1377 | ||
1378 | bprm->exec = bprm->p; | |
1379 | retval = copy_strings(bprm->envc, envp, bprm); | |
1380 | if (retval < 0) | |
1381 | goto out; | |
1382 | ||
1383 | retval = copy_strings(bprm->argc, argv, bprm); | |
1384 | if (retval < 0) | |
1385 | goto out; | |
1386 | ||
7b34e428 | 1387 | current->flags &= ~PF_KTHREAD; |
1da177e4 | 1388 | retval = search_binary_handler(bprm,regs); |
a6f76f23 DH |
1389 | if (retval < 0) |
1390 | goto out; | |
1da177e4 | 1391 | |
d899bf7b SS |
1392 | current->stack_start = current->mm->start_stack; |
1393 | ||
a6f76f23 | 1394 | /* execve succeeded */ |
498052bb | 1395 | current->fs->in_exec = 0; |
f9ce1f1c | 1396 | current->in_execve = 0; |
a6f76f23 DH |
1397 | acct_update_integrals(current); |
1398 | free_bprm(bprm); | |
1399 | if (displaced) | |
1400 | put_files_struct(displaced); | |
1401 | return retval; | |
1da177e4 | 1402 | |
a6f76f23 | 1403 | out: |
1da177e4 | 1404 | if (bprm->mm) |
b6a2fea3 | 1405 | mmput (bprm->mm); |
1da177e4 LT |
1406 | |
1407 | out_file: | |
1408 | if (bprm->file) { | |
1409 | allow_write_access(bprm->file); | |
1410 | fput(bprm->file); | |
1411 | } | |
a6f76f23 | 1412 | |
498052bb | 1413 | out_unmark: |
8c652f96 ON |
1414 | if (clear_in_exec) |
1415 | current->fs->in_exec = 0; | |
f9ce1f1c | 1416 | current->in_execve = 0; |
a6f76f23 DH |
1417 | |
1418 | out_free: | |
08a6fac1 | 1419 | free_bprm(bprm); |
1da177e4 | 1420 | |
fd8328be | 1421 | out_files: |
3b125388 AV |
1422 | if (displaced) |
1423 | reset_files_struct(displaced); | |
1da177e4 LT |
1424 | out_ret: |
1425 | return retval; | |
1426 | } | |
1427 | ||
964ee7df | 1428 | void set_binfmt(struct linux_binfmt *new) |
1da177e4 | 1429 | { |
801460d0 HS |
1430 | struct mm_struct *mm = current->mm; |
1431 | ||
1432 | if (mm->binfmt) | |
1433 | module_put(mm->binfmt->module); | |
1da177e4 | 1434 | |
801460d0 | 1435 | mm->binfmt = new; |
964ee7df ON |
1436 | if (new) |
1437 | __module_get(new->module); | |
1da177e4 LT |
1438 | } |
1439 | ||
1440 | EXPORT_SYMBOL(set_binfmt); | |
1441 | ||
1da177e4 LT |
1442 | /* format_corename will inspect the pattern parameter, and output a |
1443 | * name into corename, which must have space for at least | |
1444 | * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator. | |
1445 | */ | |
6409324b | 1446 | static int format_corename(char *corename, long signr) |
1da177e4 | 1447 | { |
86a264ab | 1448 | const struct cred *cred = current_cred(); |
565b9b14 ON |
1449 | const char *pat_ptr = core_pattern; |
1450 | int ispipe = (*pat_ptr == '|'); | |
1da177e4 LT |
1451 | char *out_ptr = corename; |
1452 | char *const out_end = corename + CORENAME_MAX_SIZE; | |
1453 | int rc; | |
1454 | int pid_in_pattern = 0; | |
1455 | ||
1456 | /* Repeat as long as we have more pattern to process and more output | |
1457 | space */ | |
1458 | while (*pat_ptr) { | |
1459 | if (*pat_ptr != '%') { | |
1460 | if (out_ptr == out_end) | |
1461 | goto out; | |
1462 | *out_ptr++ = *pat_ptr++; | |
1463 | } else { | |
1464 | switch (*++pat_ptr) { | |
1465 | case 0: | |
1466 | goto out; | |
1467 | /* Double percent, output one percent */ | |
1468 | case '%': | |
1469 | if (out_ptr == out_end) | |
1470 | goto out; | |
1471 | *out_ptr++ = '%'; | |
1472 | break; | |
1473 | /* pid */ | |
1474 | case 'p': | |
1475 | pid_in_pattern = 1; | |
1476 | rc = snprintf(out_ptr, out_end - out_ptr, | |
b488893a | 1477 | "%d", task_tgid_vnr(current)); |
1da177e4 LT |
1478 | if (rc > out_end - out_ptr) |
1479 | goto out; | |
1480 | out_ptr += rc; | |
1481 | break; | |
1482 | /* uid */ | |
1483 | case 'u': | |
1484 | rc = snprintf(out_ptr, out_end - out_ptr, | |
86a264ab | 1485 | "%d", cred->uid); |
1da177e4 LT |
1486 | if (rc > out_end - out_ptr) |
1487 | goto out; | |
1488 | out_ptr += rc; | |
1489 | break; | |
1490 | /* gid */ | |
1491 | case 'g': | |
1492 | rc = snprintf(out_ptr, out_end - out_ptr, | |
86a264ab | 1493 | "%d", cred->gid); |
1da177e4 LT |
1494 | if (rc > out_end - out_ptr) |
1495 | goto out; | |
1496 | out_ptr += rc; | |
1497 | break; | |
1498 | /* signal that caused the coredump */ | |
1499 | case 's': | |
1500 | rc = snprintf(out_ptr, out_end - out_ptr, | |
1501 | "%ld", signr); | |
1502 | if (rc > out_end - out_ptr) | |
1503 | goto out; | |
1504 | out_ptr += rc; | |
1505 | break; | |
1506 | /* UNIX time of coredump */ | |
1507 | case 't': { | |
1508 | struct timeval tv; | |
1509 | do_gettimeofday(&tv); | |
1510 | rc = snprintf(out_ptr, out_end - out_ptr, | |
1511 | "%lu", tv.tv_sec); | |
1512 | if (rc > out_end - out_ptr) | |
1513 | goto out; | |
1514 | out_ptr += rc; | |
1515 | break; | |
1516 | } | |
1517 | /* hostname */ | |
1518 | case 'h': | |
1519 | down_read(&uts_sem); | |
1520 | rc = snprintf(out_ptr, out_end - out_ptr, | |
e9ff3990 | 1521 | "%s", utsname()->nodename); |
1da177e4 LT |
1522 | up_read(&uts_sem); |
1523 | if (rc > out_end - out_ptr) | |
1524 | goto out; | |
1525 | out_ptr += rc; | |
1526 | break; | |
1527 | /* executable */ | |
1528 | case 'e': | |
1529 | rc = snprintf(out_ptr, out_end - out_ptr, | |
1530 | "%s", current->comm); | |
1531 | if (rc > out_end - out_ptr) | |
1532 | goto out; | |
1533 | out_ptr += rc; | |
1534 | break; | |
74aadce9 NH |
1535 | /* core limit size */ |
1536 | case 'c': | |
1537 | rc = snprintf(out_ptr, out_end - out_ptr, | |
d554ed89 | 1538 | "%lu", rlimit(RLIMIT_CORE)); |
74aadce9 NH |
1539 | if (rc > out_end - out_ptr) |
1540 | goto out; | |
1541 | out_ptr += rc; | |
1542 | break; | |
1da177e4 LT |
1543 | default: |
1544 | break; | |
1545 | } | |
1546 | ++pat_ptr; | |
1547 | } | |
1548 | } | |
1549 | /* Backward compatibility with core_uses_pid: | |
1550 | * | |
1551 | * If core_pattern does not include a %p (as is the default) | |
1552 | * and core_uses_pid is set, then .%pid will be appended to | |
c4bbafda | 1553 | * the filename. Do not do this for piped commands. */ |
6409324b | 1554 | if (!ispipe && !pid_in_pattern && core_uses_pid) { |
1da177e4 | 1555 | rc = snprintf(out_ptr, out_end - out_ptr, |
b488893a | 1556 | ".%d", task_tgid_vnr(current)); |
1da177e4 LT |
1557 | if (rc > out_end - out_ptr) |
1558 | goto out; | |
1559 | out_ptr += rc; | |
1560 | } | |
c4bbafda | 1561 | out: |
1da177e4 | 1562 | *out_ptr = 0; |
c4bbafda | 1563 | return ispipe; |
1da177e4 LT |
1564 | } |
1565 | ||
8cd9c249 | 1566 | static int zap_process(struct task_struct *start) |
aceecc04 ON |
1567 | { |
1568 | struct task_struct *t; | |
8cd9c249 | 1569 | int nr = 0; |
281de339 | 1570 | |
d5f70c00 ON |
1571 | start->signal->flags = SIGNAL_GROUP_EXIT; |
1572 | start->signal->group_stop_count = 0; | |
aceecc04 ON |
1573 | |
1574 | t = start; | |
1575 | do { | |
1576 | if (t != current && t->mm) { | |
281de339 ON |
1577 | sigaddset(&t->pending.signal, SIGKILL); |
1578 | signal_wake_up(t, 1); | |
8cd9c249 | 1579 | nr++; |
aceecc04 | 1580 | } |
e4901f92 | 1581 | } while_each_thread(start, t); |
8cd9c249 ON |
1582 | |
1583 | return nr; | |
aceecc04 ON |
1584 | } |
1585 | ||
dcf560c5 | 1586 | static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm, |
8cd9c249 | 1587 | struct core_state *core_state, int exit_code) |
1da177e4 LT |
1588 | { |
1589 | struct task_struct *g, *p; | |
5debfa6d | 1590 | unsigned long flags; |
8cd9c249 | 1591 | int nr = -EAGAIN; |
dcf560c5 ON |
1592 | |
1593 | spin_lock_irq(&tsk->sighand->siglock); | |
ed5d2cac | 1594 | if (!signal_group_exit(tsk->signal)) { |
8cd9c249 | 1595 | mm->core_state = core_state; |
dcf560c5 | 1596 | tsk->signal->group_exit_code = exit_code; |
8cd9c249 | 1597 | nr = zap_process(tsk); |
1da177e4 | 1598 | } |
dcf560c5 | 1599 | spin_unlock_irq(&tsk->sighand->siglock); |
8cd9c249 ON |
1600 | if (unlikely(nr < 0)) |
1601 | return nr; | |
1da177e4 | 1602 | |
8cd9c249 | 1603 | if (atomic_read(&mm->mm_users) == nr + 1) |
5debfa6d | 1604 | goto done; |
e4901f92 ON |
1605 | /* |
1606 | * We should find and kill all tasks which use this mm, and we should | |
999d9fc1 | 1607 | * count them correctly into ->nr_threads. We don't take tasklist |
e4901f92 ON |
1608 | * lock, but this is safe wrt: |
1609 | * | |
1610 | * fork: | |
1611 | * None of sub-threads can fork after zap_process(leader). All | |
1612 | * processes which were created before this point should be | |
1613 | * visible to zap_threads() because copy_process() adds the new | |
1614 | * process to the tail of init_task.tasks list, and lock/unlock | |
1615 | * of ->siglock provides a memory barrier. | |
1616 | * | |
1617 | * do_exit: | |
1618 | * The caller holds mm->mmap_sem. This means that the task which | |
1619 | * uses this mm can't pass exit_mm(), so it can't exit or clear | |
1620 | * its ->mm. | |
1621 | * | |
1622 | * de_thread: | |
1623 | * It does list_replace_rcu(&leader->tasks, ¤t->tasks), | |
1624 | * we must see either old or new leader, this does not matter. | |
1625 | * However, it can change p->sighand, so lock_task_sighand(p) | |
1626 | * must be used. Since p->mm != NULL and we hold ->mmap_sem | |
1627 | * it can't fail. | |
1628 | * | |
1629 | * Note also that "g" can be the old leader with ->mm == NULL | |
1630 | * and already unhashed and thus removed from ->thread_group. | |
1631 | * This is OK, __unhash_process()->list_del_rcu() does not | |
1632 | * clear the ->next pointer, we will find the new leader via | |
1633 | * next_thread(). | |
1634 | */ | |
7b1c6154 | 1635 | rcu_read_lock(); |
aceecc04 | 1636 | for_each_process(g) { |
5debfa6d ON |
1637 | if (g == tsk->group_leader) |
1638 | continue; | |
15b9f360 ON |
1639 | if (g->flags & PF_KTHREAD) |
1640 | continue; | |
aceecc04 ON |
1641 | p = g; |
1642 | do { | |
1643 | if (p->mm) { | |
15b9f360 | 1644 | if (unlikely(p->mm == mm)) { |
5debfa6d | 1645 | lock_task_sighand(p, &flags); |
8cd9c249 | 1646 | nr += zap_process(p); |
5debfa6d ON |
1647 | unlock_task_sighand(p, &flags); |
1648 | } | |
aceecc04 ON |
1649 | break; |
1650 | } | |
e4901f92 | 1651 | } while_each_thread(g, p); |
aceecc04 | 1652 | } |
7b1c6154 | 1653 | rcu_read_unlock(); |
5debfa6d | 1654 | done: |
c5f1cc8c | 1655 | atomic_set(&core_state->nr_threads, nr); |
8cd9c249 | 1656 | return nr; |
1da177e4 LT |
1657 | } |
1658 | ||
9d5b327b | 1659 | static int coredump_wait(int exit_code, struct core_state *core_state) |
1da177e4 | 1660 | { |
dcf560c5 ON |
1661 | struct task_struct *tsk = current; |
1662 | struct mm_struct *mm = tsk->mm; | |
dcf560c5 | 1663 | struct completion *vfork_done; |
2384f55f | 1664 | int core_waiters; |
1da177e4 | 1665 | |
9d5b327b | 1666 | init_completion(&core_state->startup); |
b564daf8 ON |
1667 | core_state->dumper.task = tsk; |
1668 | core_state->dumper.next = NULL; | |
9d5b327b | 1669 | core_waiters = zap_threads(tsk, mm, core_state, exit_code); |
2384f55f ON |
1670 | up_write(&mm->mmap_sem); |
1671 | ||
dcf560c5 ON |
1672 | if (unlikely(core_waiters < 0)) |
1673 | goto fail; | |
1674 | ||
1675 | /* | |
1676 | * Make sure nobody is waiting for us to release the VM, | |
1677 | * otherwise we can deadlock when we wait on each other | |
1678 | */ | |
1679 | vfork_done = tsk->vfork_done; | |
1680 | if (vfork_done) { | |
1681 | tsk->vfork_done = NULL; | |
1682 | complete(vfork_done); | |
1683 | } | |
1684 | ||
2384f55f | 1685 | if (core_waiters) |
9d5b327b | 1686 | wait_for_completion(&core_state->startup); |
dcf560c5 | 1687 | fail: |
dcf560c5 | 1688 | return core_waiters; |
1da177e4 LT |
1689 | } |
1690 | ||
a94e2d40 ON |
1691 | static void coredump_finish(struct mm_struct *mm) |
1692 | { | |
1693 | struct core_thread *curr, *next; | |
1694 | struct task_struct *task; | |
1695 | ||
1696 | next = mm->core_state->dumper.next; | |
1697 | while ((curr = next) != NULL) { | |
1698 | next = curr->next; | |
1699 | task = curr->task; | |
1700 | /* | |
1701 | * see exit_mm(), curr->task must not see | |
1702 | * ->task == NULL before we read ->next. | |
1703 | */ | |
1704 | smp_mb(); | |
1705 | curr->task = NULL; | |
1706 | wake_up_process(task); | |
1707 | } | |
1708 | ||
1709 | mm->core_state = NULL; | |
1710 | } | |
1711 | ||
6c5d5238 KH |
1712 | /* |
1713 | * set_dumpable converts traditional three-value dumpable to two flags and | |
1714 | * stores them into mm->flags. It modifies lower two bits of mm->flags, but | |
1715 | * these bits are not changed atomically. So get_dumpable can observe the | |
1716 | * intermediate state. To avoid doing unexpected behavior, get get_dumpable | |
1717 | * return either old dumpable or new one by paying attention to the order of | |
1718 | * modifying the bits. | |
1719 | * | |
1720 | * dumpable | mm->flags (binary) | |
1721 | * old new | initial interim final | |
1722 | * ---------+----------------------- | |
1723 | * 0 1 | 00 01 01 | |
1724 | * 0 2 | 00 10(*) 11 | |
1725 | * 1 0 | 01 00 00 | |
1726 | * 1 2 | 01 11 11 | |
1727 | * 2 0 | 11 10(*) 00 | |
1728 | * 2 1 | 11 11 01 | |
1729 | * | |
1730 | * (*) get_dumpable regards interim value of 10 as 11. | |
1731 | */ | |
1732 | void set_dumpable(struct mm_struct *mm, int value) | |
1733 | { | |
1734 | switch (value) { | |
1735 | case 0: | |
1736 | clear_bit(MMF_DUMPABLE, &mm->flags); | |
1737 | smp_wmb(); | |
1738 | clear_bit(MMF_DUMP_SECURELY, &mm->flags); | |
1739 | break; | |
1740 | case 1: | |
1741 | set_bit(MMF_DUMPABLE, &mm->flags); | |
1742 | smp_wmb(); | |
1743 | clear_bit(MMF_DUMP_SECURELY, &mm->flags); | |
1744 | break; | |
1745 | case 2: | |
1746 | set_bit(MMF_DUMP_SECURELY, &mm->flags); | |
1747 | smp_wmb(); | |
1748 | set_bit(MMF_DUMPABLE, &mm->flags); | |
1749 | break; | |
1750 | } | |
1751 | } | |
6c5d5238 KH |
1752 | |
1753 | int get_dumpable(struct mm_struct *mm) | |
1754 | { | |
1755 | int ret; | |
1756 | ||
1757 | ret = mm->flags & 0x3; | |
1758 | return (ret >= 2) ? 2 : ret; | |
1759 | } | |
1760 | ||
61be228a NH |
1761 | static void wait_for_dump_helpers(struct file *file) |
1762 | { | |
1763 | struct pipe_inode_info *pipe; | |
1764 | ||
1765 | pipe = file->f_path.dentry->d_inode->i_pipe; | |
1766 | ||
1767 | pipe_lock(pipe); | |
1768 | pipe->readers++; | |
1769 | pipe->writers--; | |
1770 | ||
1771 | while ((pipe->readers > 1) && (!signal_pending(current))) { | |
1772 | wake_up_interruptible_sync(&pipe->wait); | |
1773 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
1774 | pipe_wait(pipe); | |
1775 | } | |
1776 | ||
1777 | pipe->readers--; | |
1778 | pipe->writers++; | |
1779 | pipe_unlock(pipe); | |
1780 | ||
1781 | } | |
1782 | ||
1783 | ||
8cd3ac3a | 1784 | void do_coredump(long signr, int exit_code, struct pt_regs *regs) |
1da177e4 | 1785 | { |
9d5b327b | 1786 | struct core_state core_state; |
1da177e4 LT |
1787 | char corename[CORENAME_MAX_SIZE + 1]; |
1788 | struct mm_struct *mm = current->mm; | |
1789 | struct linux_binfmt * binfmt; | |
1790 | struct inode * inode; | |
d84f4f99 DH |
1791 | const struct cred *old_cred; |
1792 | struct cred *cred; | |
1da177e4 | 1793 | int retval = 0; |
d6e71144 | 1794 | int flag = 0; |
d025c9db | 1795 | int ispipe = 0; |
74aadce9 NH |
1796 | char **helper_argv = NULL; |
1797 | int helper_argc = 0; | |
a293980c NH |
1798 | int dump_count = 0; |
1799 | static atomic_t core_dump_count = ATOMIC_INIT(0); | |
f6151dfe MH |
1800 | struct coredump_params cprm = { |
1801 | .signr = signr, | |
1802 | .regs = regs, | |
d554ed89 | 1803 | .limit = rlimit(RLIMIT_CORE), |
f6151dfe | 1804 | }; |
1da177e4 | 1805 | |
0a4ff8c2 SG |
1806 | audit_core_dumps(signr); |
1807 | ||
801460d0 | 1808 | binfmt = mm->binfmt; |
1da177e4 LT |
1809 | if (!binfmt || !binfmt->core_dump) |
1810 | goto fail; | |
d84f4f99 DH |
1811 | |
1812 | cred = prepare_creds(); | |
1813 | if (!cred) { | |
1814 | retval = -ENOMEM; | |
1815 | goto fail; | |
1816 | } | |
1817 | ||
1da177e4 | 1818 | down_write(&mm->mmap_sem); |
00ec99da RM |
1819 | /* |
1820 | * If another thread got here first, or we are not dumpable, bail out. | |
1821 | */ | |
999d9fc1 | 1822 | if (mm->core_state || !get_dumpable(mm)) { |
1da177e4 | 1823 | up_write(&mm->mmap_sem); |
d84f4f99 | 1824 | put_cred(cred); |
1da177e4 LT |
1825 | goto fail; |
1826 | } | |
d6e71144 AC |
1827 | |
1828 | /* | |
1829 | * We cannot trust fsuid as being the "true" uid of the | |
1830 | * process nor do we know its entire history. We only know it | |
1831 | * was tainted so we dump it as root in mode 2. | |
1832 | */ | |
6c5d5238 | 1833 | if (get_dumpable(mm) == 2) { /* Setuid core dump mode */ |
d6e71144 | 1834 | flag = O_EXCL; /* Stop rewrite attacks */ |
d84f4f99 | 1835 | cred->fsuid = 0; /* Dump root private */ |
d6e71144 | 1836 | } |
1291cf41 | 1837 | |
9d5b327b | 1838 | retval = coredump_wait(exit_code, &core_state); |
d84f4f99 DH |
1839 | if (retval < 0) { |
1840 | put_cred(cred); | |
1291cf41 | 1841 | goto fail; |
d84f4f99 DH |
1842 | } |
1843 | ||
1844 | old_cred = override_creds(cred); | |
1da177e4 LT |
1845 | |
1846 | /* | |
1847 | * Clear any false indication of pending signals that might | |
1848 | * be seen by the filesystem code called to write the core file. | |
1849 | */ | |
1da177e4 LT |
1850 | clear_thread_flag(TIF_SIGPENDING); |
1851 | ||
1da177e4 LT |
1852 | /* |
1853 | * lock_kernel() because format_corename() is controlled by sysctl, which | |
1854 | * uses lock_kernel() | |
1855 | */ | |
1856 | lock_kernel(); | |
6409324b | 1857 | ispipe = format_corename(corename, signr); |
1da177e4 | 1858 | unlock_kernel(); |
725eae32 | 1859 | |
f6151dfe | 1860 | if ((!ispipe) && (cprm.limit < binfmt->min_coredump)) |
7dc0b22e NH |
1861 | goto fail_unlock; |
1862 | ||
c4bbafda | 1863 | if (ispipe) { |
f6151dfe | 1864 | if (cprm.limit == 0) { |
725eae32 NH |
1865 | /* |
1866 | * Normally core limits are irrelevant to pipes, since | |
1867 | * we're not writing to the file system, but we use | |
f6151dfe | 1868 | * cprm.limit of 0 here as a speacial value. Any |
725eae32 NH |
1869 | * non-zero limit gets set to RLIM_INFINITY below, but |
1870 | * a limit of 0 skips the dump. This is a consistent | |
1871 | * way to catch recursive crashes. We can still crash | |
1872 | * if the core_pattern binary sets RLIM_CORE = !0 | |
1873 | * but it runs as root, and can do lots of stupid things | |
1874 | * Note that we use task_tgid_vnr here to grab the pid | |
1875 | * of the process group leader. That way we get the | |
1876 | * right pid if a thread in a multi-threaded | |
1877 | * core_pattern process dies. | |
1878 | */ | |
1879 | printk(KERN_WARNING | |
1880 | "Process %d(%s) has RLIMIT_CORE set to 0\n", | |
1881 | task_tgid_vnr(current), current->comm); | |
1882 | printk(KERN_WARNING "Aborting core\n"); | |
1883 | goto fail_unlock; | |
1884 | } | |
1885 | ||
a293980c NH |
1886 | dump_count = atomic_inc_return(&core_dump_count); |
1887 | if (core_pipe_limit && (core_pipe_limit < dump_count)) { | |
1888 | printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n", | |
1889 | task_tgid_vnr(current), current->comm); | |
1890 | printk(KERN_WARNING "Skipping core dump\n"); | |
1891 | goto fail_dropcount; | |
1892 | } | |
1893 | ||
74aadce9 | 1894 | helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc); |
350eaf79 TH |
1895 | if (!helper_argv) { |
1896 | printk(KERN_WARNING "%s failed to allocate memory\n", | |
1897 | __func__); | |
a293980c | 1898 | goto fail_dropcount; |
350eaf79 | 1899 | } |
32321137 | 1900 | |
f6151dfe | 1901 | cprm.limit = RLIM_INFINITY; |
32321137 | 1902 | |
d025c9db | 1903 | /* SIGPIPE can happen, but it's just never processed */ |
a293980c | 1904 | if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL, |
f6151dfe | 1905 | &cprm.file)) { |
d025c9db AK |
1906 | printk(KERN_INFO "Core dump to %s pipe failed\n", |
1907 | corename); | |
a293980c | 1908 | goto fail_dropcount; |
d025c9db | 1909 | } |
d025c9db | 1910 | } else |
f6151dfe | 1911 | cprm.file = filp_open(corename, |
6d4df677 AD |
1912 | O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, |
1913 | 0600); | |
f6151dfe | 1914 | if (IS_ERR(cprm.file)) |
a293980c | 1915 | goto fail_dropcount; |
f6151dfe | 1916 | inode = cprm.file->f_path.dentry->d_inode; |
1da177e4 LT |
1917 | if (inode->i_nlink > 1) |
1918 | goto close_fail; /* multiple links - don't dump */ | |
f6151dfe | 1919 | if (!ispipe && d_unhashed(cprm.file->f_path.dentry)) |
1da177e4 LT |
1920 | goto close_fail; |
1921 | ||
d025c9db AK |
1922 | /* AK: actually i see no reason to not allow this for named pipes etc., |
1923 | but keep the previous behaviour for now. */ | |
1924 | if (!ispipe && !S_ISREG(inode->i_mode)) | |
1da177e4 | 1925 | goto close_fail; |
c46f739d IM |
1926 | /* |
1927 | * Dont allow local users get cute and trick others to coredump | |
1928 | * into their pre-created files: | |
1929 | */ | |
da9592ed | 1930 | if (inode->i_uid != current_fsuid()) |
c46f739d | 1931 | goto close_fail; |
f6151dfe | 1932 | if (!cprm.file->f_op) |
1da177e4 | 1933 | goto close_fail; |
f6151dfe | 1934 | if (!cprm.file->f_op->write) |
1da177e4 | 1935 | goto close_fail; |
f6151dfe MH |
1936 | if (!ispipe && |
1937 | do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file) != 0) | |
1da177e4 LT |
1938 | goto close_fail; |
1939 | ||
f6151dfe | 1940 | retval = binfmt->core_dump(&cprm); |
1da177e4 LT |
1941 | |
1942 | if (retval) | |
1943 | current->signal->group_exit_code |= 0x80; | |
1944 | close_fail: | |
61be228a | 1945 | if (ispipe && core_pipe_limit) |
f6151dfe MH |
1946 | wait_for_dump_helpers(cprm.file); |
1947 | filp_close(cprm.file, NULL); | |
a293980c NH |
1948 | fail_dropcount: |
1949 | if (dump_count) | |
1950 | atomic_dec(&core_dump_count); | |
1da177e4 | 1951 | fail_unlock: |
74aadce9 NH |
1952 | if (helper_argv) |
1953 | argv_free(helper_argv); | |
1954 | ||
d84f4f99 DH |
1955 | revert_creds(old_cred); |
1956 | put_cred(cred); | |
a94e2d40 | 1957 | coredump_finish(mm); |
1da177e4 | 1958 | fail: |
8cd3ac3a | 1959 | return; |
1da177e4 | 1960 | } |