xref: /openbmc/linux/fs/binfmt_elf.c (revision 9d56dd3b083a3bec56e9da35ce07baca81030b03)
1 /*
2  * linux/fs/binfmt_elf.c
3  *
4  * These are the functions used to load ELF format executables as used
5  * on SVr4 machines.  Information on the format may be found in the book
6  * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
7  * Tools".
8  *
9  * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
10  */
11 
12 #include <linux/module.h>
13 #include <linux/kernel.h>
14 #include <linux/fs.h>
15 #include <linux/mm.h>
16 #include <linux/mman.h>
17 #include <linux/errno.h>
18 #include <linux/signal.h>
19 #include <linux/binfmts.h>
20 #include <linux/string.h>
21 #include <linux/file.h>
22 #include <linux/slab.h>
23 #include <linux/personality.h>
24 #include <linux/elfcore.h>
25 #include <linux/init.h>
26 #include <linux/highuid.h>
27 #include <linux/compiler.h>
28 #include <linux/highmem.h>
29 #include <linux/pagemap.h>
30 #include <linux/security.h>
31 #include <linux/random.h>
32 #include <linux/elf.h>
33 #include <linux/utsname.h>
34 #include <asm/uaccess.h>
35 #include <asm/param.h>
36 #include <asm/page.h>
37 
38 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
39 static int load_elf_library(struct file *);
40 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
41 				int, int, unsigned long);
42 
43 /*
44  * If we don't support core dumping, then supply a NULL so we
45  * don't even try.
46  */
47 #ifdef CONFIG_ELF_CORE
48 static int elf_core_dump(struct coredump_params *cprm);
49 #else
50 #define elf_core_dump	NULL
51 #endif
52 
53 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
54 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
55 #else
56 #define ELF_MIN_ALIGN	PAGE_SIZE
57 #endif
58 
59 #ifndef ELF_CORE_EFLAGS
60 #define ELF_CORE_EFLAGS	0
61 #endif
62 
63 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
64 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
65 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
66 
67 static struct linux_binfmt elf_format = {
68 		.module		= THIS_MODULE,
69 		.load_binary	= load_elf_binary,
70 		.load_shlib	= load_elf_library,
71 		.core_dump	= elf_core_dump,
72 		.min_coredump	= ELF_EXEC_PAGESIZE,
73 		.hasvdso	= 1
74 };
75 
76 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
77 
78 static int set_brk(unsigned long start, unsigned long end)
79 {
80 	start = ELF_PAGEALIGN(start);
81 	end = ELF_PAGEALIGN(end);
82 	if (end > start) {
83 		unsigned long addr;
84 		down_write(&current->mm->mmap_sem);
85 		addr = do_brk(start, end - start);
86 		up_write(&current->mm->mmap_sem);
87 		if (BAD_ADDR(addr))
88 			return addr;
89 	}
90 	current->mm->start_brk = current->mm->brk = end;
91 	return 0;
92 }
93 
94 /* We need to explicitly zero any fractional pages
95    after the data section (i.e. bss).  This would
96    contain the junk from the file that should not
97    be in memory
98  */
99 static int padzero(unsigned long elf_bss)
100 {
101 	unsigned long nbyte;
102 
103 	nbyte = ELF_PAGEOFFSET(elf_bss);
104 	if (nbyte) {
105 		nbyte = ELF_MIN_ALIGN - nbyte;
106 		if (clear_user((void __user *) elf_bss, nbyte))
107 			return -EFAULT;
108 	}
109 	return 0;
110 }
111 
112 /* Let's use some macros to make this stack manipulation a little clearer */
113 #ifdef CONFIG_STACK_GROWSUP
114 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
115 #define STACK_ROUND(sp, items) \
116 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
117 #define STACK_ALLOC(sp, len) ({ \
118 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
119 	old_sp; })
120 #else
121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
122 #define STACK_ROUND(sp, items) \
123 	(((unsigned long) (sp - items)) &~ 15UL)
124 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
125 #endif
126 
127 #ifndef ELF_BASE_PLATFORM
128 /*
129  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
130  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
131  * will be copied to the user stack in the same manner as AT_PLATFORM.
132  */
133 #define ELF_BASE_PLATFORM NULL
134 #endif
135 
136 static int
137 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
138 		unsigned long load_addr, unsigned long interp_load_addr)
139 {
140 	unsigned long p = bprm->p;
141 	int argc = bprm->argc;
142 	int envc = bprm->envc;
143 	elf_addr_t __user *argv;
144 	elf_addr_t __user *envp;
145 	elf_addr_t __user *sp;
146 	elf_addr_t __user *u_platform;
147 	elf_addr_t __user *u_base_platform;
148 	elf_addr_t __user *u_rand_bytes;
149 	const char *k_platform = ELF_PLATFORM;
150 	const char *k_base_platform = ELF_BASE_PLATFORM;
151 	unsigned char k_rand_bytes[16];
152 	int items;
153 	elf_addr_t *elf_info;
154 	int ei_index = 0;
155 	const struct cred *cred = current_cred();
156 	struct vm_area_struct *vma;
157 
158 	/*
159 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
160 	 * evictions by the processes running on the same package. One
161 	 * thing we can do is to shuffle the initial stack for them.
162 	 */
163 
164 	p = arch_align_stack(p);
165 
166 	/*
167 	 * If this architecture has a platform capability string, copy it
168 	 * to userspace.  In some cases (Sparc), this info is impossible
169 	 * for userspace to get any other way, in others (i386) it is
170 	 * merely difficult.
171 	 */
172 	u_platform = NULL;
173 	if (k_platform) {
174 		size_t len = strlen(k_platform) + 1;
175 
176 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
177 		if (__copy_to_user(u_platform, k_platform, len))
178 			return -EFAULT;
179 	}
180 
181 	/*
182 	 * If this architecture has a "base" platform capability
183 	 * string, copy it to userspace.
184 	 */
185 	u_base_platform = NULL;
186 	if (k_base_platform) {
187 		size_t len = strlen(k_base_platform) + 1;
188 
189 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
190 		if (__copy_to_user(u_base_platform, k_base_platform, len))
191 			return -EFAULT;
192 	}
193 
194 	/*
195 	 * Generate 16 random bytes for userspace PRNG seeding.
196 	 */
197 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
198 	u_rand_bytes = (elf_addr_t __user *)
199 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
200 	if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
201 		return -EFAULT;
202 
203 	/* Create the ELF interpreter info */
204 	elf_info = (elf_addr_t *)current->mm->saved_auxv;
205 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
206 #define NEW_AUX_ENT(id, val) \
207 	do { \
208 		elf_info[ei_index++] = id; \
209 		elf_info[ei_index++] = val; \
210 	} while (0)
211 
212 #ifdef ARCH_DLINFO
213 	/*
214 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
215 	 * AUXV.
216 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
217 	 * ARCH_DLINFO changes
218 	 */
219 	ARCH_DLINFO;
220 #endif
221 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
222 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
223 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
224 	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
225 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
226 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
227 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
228 	NEW_AUX_ENT(AT_FLAGS, 0);
229 	NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
230 	NEW_AUX_ENT(AT_UID, cred->uid);
231 	NEW_AUX_ENT(AT_EUID, cred->euid);
232 	NEW_AUX_ENT(AT_GID, cred->gid);
233 	NEW_AUX_ENT(AT_EGID, cred->egid);
234  	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
235 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
236 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
237 	if (k_platform) {
238 		NEW_AUX_ENT(AT_PLATFORM,
239 			    (elf_addr_t)(unsigned long)u_platform);
240 	}
241 	if (k_base_platform) {
242 		NEW_AUX_ENT(AT_BASE_PLATFORM,
243 			    (elf_addr_t)(unsigned long)u_base_platform);
244 	}
245 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
246 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
247 	}
248 #undef NEW_AUX_ENT
249 	/* AT_NULL is zero; clear the rest too */
250 	memset(&elf_info[ei_index], 0,
251 	       sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
252 
253 	/* And advance past the AT_NULL entry.  */
254 	ei_index += 2;
255 
256 	sp = STACK_ADD(p, ei_index);
257 
258 	items = (argc + 1) + (envc + 1) + 1;
259 	bprm->p = STACK_ROUND(sp, items);
260 
261 	/* Point sp at the lowest address on the stack */
262 #ifdef CONFIG_STACK_GROWSUP
263 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
264 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
265 #else
266 	sp = (elf_addr_t __user *)bprm->p;
267 #endif
268 
269 
270 	/*
271 	 * Grow the stack manually; some architectures have a limit on how
272 	 * far ahead a user-space access may be in order to grow the stack.
273 	 */
274 	vma = find_extend_vma(current->mm, bprm->p);
275 	if (!vma)
276 		return -EFAULT;
277 
278 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
279 	if (__put_user(argc, sp++))
280 		return -EFAULT;
281 	argv = sp;
282 	envp = argv + argc + 1;
283 
284 	/* Populate argv and envp */
285 	p = current->mm->arg_end = current->mm->arg_start;
286 	while (argc-- > 0) {
287 		size_t len;
288 		if (__put_user((elf_addr_t)p, argv++))
289 			return -EFAULT;
290 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
291 		if (!len || len > MAX_ARG_STRLEN)
292 			return -EINVAL;
293 		p += len;
294 	}
295 	if (__put_user(0, argv))
296 		return -EFAULT;
297 	current->mm->arg_end = current->mm->env_start = p;
298 	while (envc-- > 0) {
299 		size_t len;
300 		if (__put_user((elf_addr_t)p, envp++))
301 			return -EFAULT;
302 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
303 		if (!len || len > MAX_ARG_STRLEN)
304 			return -EINVAL;
305 		p += len;
306 	}
307 	if (__put_user(0, envp))
308 		return -EFAULT;
309 	current->mm->env_end = p;
310 
311 	/* Put the elf_info on the stack in the right place.  */
312 	sp = (elf_addr_t __user *)envp + 1;
313 	if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
314 		return -EFAULT;
315 	return 0;
316 }
317 
318 #ifndef elf_map
319 
320 static unsigned long elf_map(struct file *filep, unsigned long addr,
321 		struct elf_phdr *eppnt, int prot, int type,
322 		unsigned long total_size)
323 {
324 	unsigned long map_addr;
325 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
326 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
327 	addr = ELF_PAGESTART(addr);
328 	size = ELF_PAGEALIGN(size);
329 
330 	/* mmap() will return -EINVAL if given a zero size, but a
331 	 * segment with zero filesize is perfectly valid */
332 	if (!size)
333 		return addr;
334 
335 	down_write(&current->mm->mmap_sem);
336 	/*
337 	* total_size is the size of the ELF (interpreter) image.
338 	* The _first_ mmap needs to know the full size, otherwise
339 	* randomization might put this image into an overlapping
340 	* position with the ELF binary image. (since size < total_size)
341 	* So we first map the 'big' image - and unmap the remainder at
342 	* the end. (which unmap is needed for ELF images with holes.)
343 	*/
344 	if (total_size) {
345 		total_size = ELF_PAGEALIGN(total_size);
346 		map_addr = do_mmap(filep, addr, total_size, prot, type, off);
347 		if (!BAD_ADDR(map_addr))
348 			do_munmap(current->mm, map_addr+size, total_size-size);
349 	} else
350 		map_addr = do_mmap(filep, addr, size, prot, type, off);
351 
352 	up_write(&current->mm->mmap_sem);
353 	return(map_addr);
354 }
355 
356 #endif /* !elf_map */
357 
358 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
359 {
360 	int i, first_idx = -1, last_idx = -1;
361 
362 	for (i = 0; i < nr; i++) {
363 		if (cmds[i].p_type == PT_LOAD) {
364 			last_idx = i;
365 			if (first_idx == -1)
366 				first_idx = i;
367 		}
368 	}
369 	if (first_idx == -1)
370 		return 0;
371 
372 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
373 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
374 }
375 
376 
377 /* This is much more generalized than the library routine read function,
378    so we keep this separate.  Technically the library read function
379    is only provided so that we can read a.out libraries that have
380    an ELF header */
381 
382 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
383 		struct file *interpreter, unsigned long *interp_map_addr,
384 		unsigned long no_base)
385 {
386 	struct elf_phdr *elf_phdata;
387 	struct elf_phdr *eppnt;
388 	unsigned long load_addr = 0;
389 	int load_addr_set = 0;
390 	unsigned long last_bss = 0, elf_bss = 0;
391 	unsigned long error = ~0UL;
392 	unsigned long total_size;
393 	int retval, i, size;
394 
395 	/* First of all, some simple consistency checks */
396 	if (interp_elf_ex->e_type != ET_EXEC &&
397 	    interp_elf_ex->e_type != ET_DYN)
398 		goto out;
399 	if (!elf_check_arch(interp_elf_ex))
400 		goto out;
401 	if (!interpreter->f_op || !interpreter->f_op->mmap)
402 		goto out;
403 
404 	/*
405 	 * If the size of this structure has changed, then punt, since
406 	 * we will be doing the wrong thing.
407 	 */
408 	if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
409 		goto out;
410 	if (interp_elf_ex->e_phnum < 1 ||
411 		interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
412 		goto out;
413 
414 	/* Now read in all of the header information */
415 	size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
416 	if (size > ELF_MIN_ALIGN)
417 		goto out;
418 	elf_phdata = kmalloc(size, GFP_KERNEL);
419 	if (!elf_phdata)
420 		goto out;
421 
422 	retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
423 			     (char *)elf_phdata,size);
424 	error = -EIO;
425 	if (retval != size) {
426 		if (retval < 0)
427 			error = retval;
428 		goto out_close;
429 	}
430 
431 	total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
432 	if (!total_size) {
433 		error = -EINVAL;
434 		goto out_close;
435 	}
436 
437 	eppnt = elf_phdata;
438 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
439 		if (eppnt->p_type == PT_LOAD) {
440 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
441 			int elf_prot = 0;
442 			unsigned long vaddr = 0;
443 			unsigned long k, map_addr;
444 
445 			if (eppnt->p_flags & PF_R)
446 		    		elf_prot = PROT_READ;
447 			if (eppnt->p_flags & PF_W)
448 				elf_prot |= PROT_WRITE;
449 			if (eppnt->p_flags & PF_X)
450 				elf_prot |= PROT_EXEC;
451 			vaddr = eppnt->p_vaddr;
452 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
453 				elf_type |= MAP_FIXED;
454 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
455 				load_addr = -vaddr;
456 
457 			map_addr = elf_map(interpreter, load_addr + vaddr,
458 					eppnt, elf_prot, elf_type, total_size);
459 			total_size = 0;
460 			if (!*interp_map_addr)
461 				*interp_map_addr = map_addr;
462 			error = map_addr;
463 			if (BAD_ADDR(map_addr))
464 				goto out_close;
465 
466 			if (!load_addr_set &&
467 			    interp_elf_ex->e_type == ET_DYN) {
468 				load_addr = map_addr - ELF_PAGESTART(vaddr);
469 				load_addr_set = 1;
470 			}
471 
472 			/*
473 			 * Check to see if the section's size will overflow the
474 			 * allowed task size. Note that p_filesz must always be
475 			 * <= p_memsize so it's only necessary to check p_memsz.
476 			 */
477 			k = load_addr + eppnt->p_vaddr;
478 			if (BAD_ADDR(k) ||
479 			    eppnt->p_filesz > eppnt->p_memsz ||
480 			    eppnt->p_memsz > TASK_SIZE ||
481 			    TASK_SIZE - eppnt->p_memsz < k) {
482 				error = -ENOMEM;
483 				goto out_close;
484 			}
485 
486 			/*
487 			 * Find the end of the file mapping for this phdr, and
488 			 * keep track of the largest address we see for this.
489 			 */
490 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
491 			if (k > elf_bss)
492 				elf_bss = k;
493 
494 			/*
495 			 * Do the same thing for the memory mapping - between
496 			 * elf_bss and last_bss is the bss section.
497 			 */
498 			k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
499 			if (k > last_bss)
500 				last_bss = k;
501 		}
502 	}
503 
504 	if (last_bss > elf_bss) {
505 		/*
506 		 * Now fill out the bss section.  First pad the last page up
507 		 * to the page boundary, and then perform a mmap to make sure
508 		 * that there are zero-mapped pages up to and including the
509 		 * last bss page.
510 		 */
511 		if (padzero(elf_bss)) {
512 			error = -EFAULT;
513 			goto out_close;
514 		}
515 
516 		/* What we have mapped so far */
517 		elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
518 
519 		/* Map the last of the bss segment */
520 		down_write(&current->mm->mmap_sem);
521 		error = do_brk(elf_bss, last_bss - elf_bss);
522 		up_write(&current->mm->mmap_sem);
523 		if (BAD_ADDR(error))
524 			goto out_close;
525 	}
526 
527 	error = load_addr;
528 
529 out_close:
530 	kfree(elf_phdata);
531 out:
532 	return error;
533 }
534 
535 /*
536  * These are the functions used to load ELF style executables and shared
537  * libraries.  There is no binary dependent code anywhere else.
538  */
539 
540 #define INTERPRETER_NONE 0
541 #define INTERPRETER_ELF 2
542 
543 #ifndef STACK_RND_MASK
544 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))	/* 8MB of VA */
545 #endif
546 
547 static unsigned long randomize_stack_top(unsigned long stack_top)
548 {
549 	unsigned int random_variable = 0;
550 
551 	if ((current->flags & PF_RANDOMIZE) &&
552 		!(current->personality & ADDR_NO_RANDOMIZE)) {
553 		random_variable = get_random_int() & STACK_RND_MASK;
554 		random_variable <<= PAGE_SHIFT;
555 	}
556 #ifdef CONFIG_STACK_GROWSUP
557 	return PAGE_ALIGN(stack_top) + random_variable;
558 #else
559 	return PAGE_ALIGN(stack_top) - random_variable;
560 #endif
561 }
562 
563 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
564 {
565 	struct file *interpreter = NULL; /* to shut gcc up */
566  	unsigned long load_addr = 0, load_bias = 0;
567 	int load_addr_set = 0;
568 	char * elf_interpreter = NULL;
569 	unsigned long error;
570 	struct elf_phdr *elf_ppnt, *elf_phdata;
571 	unsigned long elf_bss, elf_brk;
572 	int retval, i;
573 	unsigned int size;
574 	unsigned long elf_entry;
575 	unsigned long interp_load_addr = 0;
576 	unsigned long start_code, end_code, start_data, end_data;
577 	unsigned long reloc_func_desc = 0;
578 	int executable_stack = EXSTACK_DEFAULT;
579 	unsigned long def_flags = 0;
580 	struct {
581 		struct elfhdr elf_ex;
582 		struct elfhdr interp_elf_ex;
583 	} *loc;
584 
585 	loc = kmalloc(sizeof(*loc), GFP_KERNEL);
586 	if (!loc) {
587 		retval = -ENOMEM;
588 		goto out_ret;
589 	}
590 
591 	/* Get the exec-header */
592 	loc->elf_ex = *((struct elfhdr *)bprm->buf);
593 
594 	retval = -ENOEXEC;
595 	/* First of all, some simple consistency checks */
596 	if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
597 		goto out;
598 
599 	if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
600 		goto out;
601 	if (!elf_check_arch(&loc->elf_ex))
602 		goto out;
603 	if (!bprm->file->f_op||!bprm->file->f_op->mmap)
604 		goto out;
605 
606 	/* Now read in all of the header information */
607 	if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
608 		goto out;
609 	if (loc->elf_ex.e_phnum < 1 ||
610 	 	loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
611 		goto out;
612 	size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
613 	retval = -ENOMEM;
614 	elf_phdata = kmalloc(size, GFP_KERNEL);
615 	if (!elf_phdata)
616 		goto out;
617 
618 	retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
619 			     (char *)elf_phdata, size);
620 	if (retval != size) {
621 		if (retval >= 0)
622 			retval = -EIO;
623 		goto out_free_ph;
624 	}
625 
626 	elf_ppnt = elf_phdata;
627 	elf_bss = 0;
628 	elf_brk = 0;
629 
630 	start_code = ~0UL;
631 	end_code = 0;
632 	start_data = 0;
633 	end_data = 0;
634 
635 	for (i = 0; i < loc->elf_ex.e_phnum; i++) {
636 		if (elf_ppnt->p_type == PT_INTERP) {
637 			/* This is the program interpreter used for
638 			 * shared libraries - for now assume that this
639 			 * is an a.out format binary
640 			 */
641 			retval = -ENOEXEC;
642 			if (elf_ppnt->p_filesz > PATH_MAX ||
643 			    elf_ppnt->p_filesz < 2)
644 				goto out_free_ph;
645 
646 			retval = -ENOMEM;
647 			elf_interpreter = kmalloc(elf_ppnt->p_filesz,
648 						  GFP_KERNEL);
649 			if (!elf_interpreter)
650 				goto out_free_ph;
651 
652 			retval = kernel_read(bprm->file, elf_ppnt->p_offset,
653 					     elf_interpreter,
654 					     elf_ppnt->p_filesz);
655 			if (retval != elf_ppnt->p_filesz) {
656 				if (retval >= 0)
657 					retval = -EIO;
658 				goto out_free_interp;
659 			}
660 			/* make sure path is NULL terminated */
661 			retval = -ENOEXEC;
662 			if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
663 				goto out_free_interp;
664 
665 			/*
666 			 * The early SET_PERSONALITY here is so that the lookup
667 			 * for the interpreter happens in the namespace of the
668 			 * to-be-execed image.  SET_PERSONALITY can select an
669 			 * alternate root.
670 			 *
671 			 * However, SET_PERSONALITY is NOT allowed to switch
672 			 * this task into the new images's memory mapping
673 			 * policy - that is, TASK_SIZE must still evaluate to
674 			 * that which is appropriate to the execing application.
675 			 * This is because exit_mmap() needs to have TASK_SIZE
676 			 * evaluate to the size of the old image.
677 			 *
678 			 * So if (say) a 64-bit application is execing a 32-bit
679 			 * application it is the architecture's responsibility
680 			 * to defer changing the value of TASK_SIZE until the
681 			 * switch really is going to happen - do this in
682 			 * flush_thread().	- akpm
683 			 */
684 			SET_PERSONALITY(loc->elf_ex);
685 
686 			interpreter = open_exec(elf_interpreter);
687 			retval = PTR_ERR(interpreter);
688 			if (IS_ERR(interpreter))
689 				goto out_free_interp;
690 
691 			/*
692 			 * If the binary is not readable then enforce
693 			 * mm->dumpable = 0 regardless of the interpreter's
694 			 * permissions.
695 			 */
696 			if (file_permission(interpreter, MAY_READ) < 0)
697 				bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
698 
699 			retval = kernel_read(interpreter, 0, bprm->buf,
700 					     BINPRM_BUF_SIZE);
701 			if (retval != BINPRM_BUF_SIZE) {
702 				if (retval >= 0)
703 					retval = -EIO;
704 				goto out_free_dentry;
705 			}
706 
707 			/* Get the exec headers */
708 			loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
709 			break;
710 		}
711 		elf_ppnt++;
712 	}
713 
714 	elf_ppnt = elf_phdata;
715 	for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
716 		if (elf_ppnt->p_type == PT_GNU_STACK) {
717 			if (elf_ppnt->p_flags & PF_X)
718 				executable_stack = EXSTACK_ENABLE_X;
719 			else
720 				executable_stack = EXSTACK_DISABLE_X;
721 			break;
722 		}
723 
724 	/* Some simple consistency checks for the interpreter */
725 	if (elf_interpreter) {
726 		retval = -ELIBBAD;
727 		/* Not an ELF interpreter */
728 		if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
729 			goto out_free_dentry;
730 		/* Verify the interpreter has a valid arch */
731 		if (!elf_check_arch(&loc->interp_elf_ex))
732 			goto out_free_dentry;
733 	} else {
734 		/* Executables without an interpreter also need a personality  */
735 		SET_PERSONALITY(loc->elf_ex);
736 	}
737 
738 	/* Flush all traces of the currently running executable */
739 	retval = flush_old_exec(bprm);
740 	if (retval)
741 		goto out_free_dentry;
742 
743 	/* OK, This is the point of no return */
744 	current->flags &= ~PF_FORKNOEXEC;
745 	current->mm->def_flags = def_flags;
746 
747 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
748 	   may depend on the personality.  */
749 	SET_PERSONALITY(loc->elf_ex);
750 	if (elf_read_implies_exec(loc->elf_ex, executable_stack))
751 		current->personality |= READ_IMPLIES_EXEC;
752 
753 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
754 		current->flags |= PF_RANDOMIZE;
755 	arch_pick_mmap_layout(current->mm);
756 
757 	/* Do this so that we can load the interpreter, if need be.  We will
758 	   change some of these later */
759 	current->mm->free_area_cache = current->mm->mmap_base;
760 	current->mm->cached_hole_size = 0;
761 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
762 				 executable_stack);
763 	if (retval < 0) {
764 		send_sig(SIGKILL, current, 0);
765 		goto out_free_dentry;
766 	}
767 
768 	current->mm->start_stack = bprm->p;
769 
770 	/* Now we do a little grungy work by mmapping the ELF image into
771 	   the correct location in memory. */
772 	for(i = 0, elf_ppnt = elf_phdata;
773 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
774 		int elf_prot = 0, elf_flags;
775 		unsigned long k, vaddr;
776 
777 		if (elf_ppnt->p_type != PT_LOAD)
778 			continue;
779 
780 		if (unlikely (elf_brk > elf_bss)) {
781 			unsigned long nbyte;
782 
783 			/* There was a PT_LOAD segment with p_memsz > p_filesz
784 			   before this one. Map anonymous pages, if needed,
785 			   and clear the area.  */
786 			retval = set_brk (elf_bss + load_bias,
787 					  elf_brk + load_bias);
788 			if (retval) {
789 				send_sig(SIGKILL, current, 0);
790 				goto out_free_dentry;
791 			}
792 			nbyte = ELF_PAGEOFFSET(elf_bss);
793 			if (nbyte) {
794 				nbyte = ELF_MIN_ALIGN - nbyte;
795 				if (nbyte > elf_brk - elf_bss)
796 					nbyte = elf_brk - elf_bss;
797 				if (clear_user((void __user *)elf_bss +
798 							load_bias, nbyte)) {
799 					/*
800 					 * This bss-zeroing can fail if the ELF
801 					 * file specifies odd protections. So
802 					 * we don't check the return value
803 					 */
804 				}
805 			}
806 		}
807 
808 		if (elf_ppnt->p_flags & PF_R)
809 			elf_prot |= PROT_READ;
810 		if (elf_ppnt->p_flags & PF_W)
811 			elf_prot |= PROT_WRITE;
812 		if (elf_ppnt->p_flags & PF_X)
813 			elf_prot |= PROT_EXEC;
814 
815 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
816 
817 		vaddr = elf_ppnt->p_vaddr;
818 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
819 			elf_flags |= MAP_FIXED;
820 		} else if (loc->elf_ex.e_type == ET_DYN) {
821 			/* Try and get dynamic programs out of the way of the
822 			 * default mmap base, as well as whatever program they
823 			 * might try to exec.  This is because the brk will
824 			 * follow the loader, and is not movable.  */
825 #ifdef CONFIG_X86
826 			load_bias = 0;
827 #else
828 			load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
829 #endif
830 		}
831 
832 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
833 				elf_prot, elf_flags, 0);
834 		if (BAD_ADDR(error)) {
835 			send_sig(SIGKILL, current, 0);
836 			retval = IS_ERR((void *)error) ?
837 				PTR_ERR((void*)error) : -EINVAL;
838 			goto out_free_dentry;
839 		}
840 
841 		if (!load_addr_set) {
842 			load_addr_set = 1;
843 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
844 			if (loc->elf_ex.e_type == ET_DYN) {
845 				load_bias += error -
846 				             ELF_PAGESTART(load_bias + vaddr);
847 				load_addr += load_bias;
848 				reloc_func_desc = load_bias;
849 			}
850 		}
851 		k = elf_ppnt->p_vaddr;
852 		if (k < start_code)
853 			start_code = k;
854 		if (start_data < k)
855 			start_data = k;
856 
857 		/*
858 		 * Check to see if the section's size will overflow the
859 		 * allowed task size. Note that p_filesz must always be
860 		 * <= p_memsz so it is only necessary to check p_memsz.
861 		 */
862 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
863 		    elf_ppnt->p_memsz > TASK_SIZE ||
864 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
865 			/* set_brk can never work. Avoid overflows. */
866 			send_sig(SIGKILL, current, 0);
867 			retval = -EINVAL;
868 			goto out_free_dentry;
869 		}
870 
871 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
872 
873 		if (k > elf_bss)
874 			elf_bss = k;
875 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
876 			end_code = k;
877 		if (end_data < k)
878 			end_data = k;
879 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
880 		if (k > elf_brk)
881 			elf_brk = k;
882 	}
883 
884 	loc->elf_ex.e_entry += load_bias;
885 	elf_bss += load_bias;
886 	elf_brk += load_bias;
887 	start_code += load_bias;
888 	end_code += load_bias;
889 	start_data += load_bias;
890 	end_data += load_bias;
891 
892 	/* Calling set_brk effectively mmaps the pages that we need
893 	 * for the bss and break sections.  We must do this before
894 	 * mapping in the interpreter, to make sure it doesn't wind
895 	 * up getting placed where the bss needs to go.
896 	 */
897 	retval = set_brk(elf_bss, elf_brk);
898 	if (retval) {
899 		send_sig(SIGKILL, current, 0);
900 		goto out_free_dentry;
901 	}
902 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
903 		send_sig(SIGSEGV, current, 0);
904 		retval = -EFAULT; /* Nobody gets to see this, but.. */
905 		goto out_free_dentry;
906 	}
907 
908 	if (elf_interpreter) {
909 		unsigned long uninitialized_var(interp_map_addr);
910 
911 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
912 					    interpreter,
913 					    &interp_map_addr,
914 					    load_bias);
915 		if (!IS_ERR((void *)elf_entry)) {
916 			/*
917 			 * load_elf_interp() returns relocation
918 			 * adjustment
919 			 */
920 			interp_load_addr = elf_entry;
921 			elf_entry += loc->interp_elf_ex.e_entry;
922 		}
923 		if (BAD_ADDR(elf_entry)) {
924 			force_sig(SIGSEGV, current);
925 			retval = IS_ERR((void *)elf_entry) ?
926 					(int)elf_entry : -EINVAL;
927 			goto out_free_dentry;
928 		}
929 		reloc_func_desc = interp_load_addr;
930 
931 		allow_write_access(interpreter);
932 		fput(interpreter);
933 		kfree(elf_interpreter);
934 	} else {
935 		elf_entry = loc->elf_ex.e_entry;
936 		if (BAD_ADDR(elf_entry)) {
937 			force_sig(SIGSEGV, current);
938 			retval = -EINVAL;
939 			goto out_free_dentry;
940 		}
941 	}
942 
943 	kfree(elf_phdata);
944 
945 	set_binfmt(&elf_format);
946 
947 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
948 	retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
949 	if (retval < 0) {
950 		send_sig(SIGKILL, current, 0);
951 		goto out;
952 	}
953 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
954 
955 	install_exec_creds(bprm);
956 	current->flags &= ~PF_FORKNOEXEC;
957 	retval = create_elf_tables(bprm, &loc->elf_ex,
958 			  load_addr, interp_load_addr);
959 	if (retval < 0) {
960 		send_sig(SIGKILL, current, 0);
961 		goto out;
962 	}
963 	/* N.B. passed_fileno might not be initialized? */
964 	current->mm->end_code = end_code;
965 	current->mm->start_code = start_code;
966 	current->mm->start_data = start_data;
967 	current->mm->end_data = end_data;
968 	current->mm->start_stack = bprm->p;
969 
970 #ifdef arch_randomize_brk
971 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1))
972 		current->mm->brk = current->mm->start_brk =
973 			arch_randomize_brk(current->mm);
974 #endif
975 
976 	if (current->personality & MMAP_PAGE_ZERO) {
977 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
978 		   and some applications "depend" upon this behavior.
979 		   Since we do not have the power to recompile these, we
980 		   emulate the SVr4 behavior. Sigh. */
981 		down_write(&current->mm->mmap_sem);
982 		error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
983 				MAP_FIXED | MAP_PRIVATE, 0);
984 		up_write(&current->mm->mmap_sem);
985 	}
986 
987 #ifdef ELF_PLAT_INIT
988 	/*
989 	 * The ABI may specify that certain registers be set up in special
990 	 * ways (on i386 %edx is the address of a DT_FINI function, for
991 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
992 	 * that the e_entry field is the address of the function descriptor
993 	 * for the startup routine, rather than the address of the startup
994 	 * routine itself.  This macro performs whatever initialization to
995 	 * the regs structure is required as well as any relocations to the
996 	 * function descriptor entries when executing dynamically links apps.
997 	 */
998 	ELF_PLAT_INIT(regs, reloc_func_desc);
999 #endif
1000 
1001 	start_thread(regs, elf_entry, bprm->p);
1002 	retval = 0;
1003 out:
1004 	kfree(loc);
1005 out_ret:
1006 	return retval;
1007 
1008 	/* error cleanup */
1009 out_free_dentry:
1010 	allow_write_access(interpreter);
1011 	if (interpreter)
1012 		fput(interpreter);
1013 out_free_interp:
1014 	kfree(elf_interpreter);
1015 out_free_ph:
1016 	kfree(elf_phdata);
1017 	goto out;
1018 }
1019 
1020 /* This is really simpleminded and specialized - we are loading an
1021    a.out library that is given an ELF header. */
1022 static int load_elf_library(struct file *file)
1023 {
1024 	struct elf_phdr *elf_phdata;
1025 	struct elf_phdr *eppnt;
1026 	unsigned long elf_bss, bss, len;
1027 	int retval, error, i, j;
1028 	struct elfhdr elf_ex;
1029 
1030 	error = -ENOEXEC;
1031 	retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1032 	if (retval != sizeof(elf_ex))
1033 		goto out;
1034 
1035 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1036 		goto out;
1037 
1038 	/* First of all, some simple consistency checks */
1039 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1040 	    !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1041 		goto out;
1042 
1043 	/* Now read in all of the header information */
1044 
1045 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1046 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1047 
1048 	error = -ENOMEM;
1049 	elf_phdata = kmalloc(j, GFP_KERNEL);
1050 	if (!elf_phdata)
1051 		goto out;
1052 
1053 	eppnt = elf_phdata;
1054 	error = -ENOEXEC;
1055 	retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1056 	if (retval != j)
1057 		goto out_free_ph;
1058 
1059 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1060 		if ((eppnt + i)->p_type == PT_LOAD)
1061 			j++;
1062 	if (j != 1)
1063 		goto out_free_ph;
1064 
1065 	while (eppnt->p_type != PT_LOAD)
1066 		eppnt++;
1067 
1068 	/* Now use mmap to map the library into memory. */
1069 	down_write(&current->mm->mmap_sem);
1070 	error = do_mmap(file,
1071 			ELF_PAGESTART(eppnt->p_vaddr),
1072 			(eppnt->p_filesz +
1073 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1074 			PROT_READ | PROT_WRITE | PROT_EXEC,
1075 			MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1076 			(eppnt->p_offset -
1077 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1078 	up_write(&current->mm->mmap_sem);
1079 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1080 		goto out_free_ph;
1081 
1082 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1083 	if (padzero(elf_bss)) {
1084 		error = -EFAULT;
1085 		goto out_free_ph;
1086 	}
1087 
1088 	len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1089 			    ELF_MIN_ALIGN - 1);
1090 	bss = eppnt->p_memsz + eppnt->p_vaddr;
1091 	if (bss > len) {
1092 		down_write(&current->mm->mmap_sem);
1093 		do_brk(len, bss - len);
1094 		up_write(&current->mm->mmap_sem);
1095 	}
1096 	error = 0;
1097 
1098 out_free_ph:
1099 	kfree(elf_phdata);
1100 out:
1101 	return error;
1102 }
1103 
1104 #ifdef CONFIG_ELF_CORE
1105 /*
1106  * ELF core dumper
1107  *
1108  * Modelled on fs/exec.c:aout_core_dump()
1109  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1110  */
1111 /*
1112  * These are the only things you should do on a core-file: use only these
1113  * functions to write out all the necessary info.
1114  */
1115 static int dump_write(struct file *file, const void *addr, int nr)
1116 {
1117 	return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1118 }
1119 
1120 static int dump_seek(struct file *file, loff_t off)
1121 {
1122 	if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
1123 		if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
1124 			return 0;
1125 	} else {
1126 		char *buf = (char *)get_zeroed_page(GFP_KERNEL);
1127 		if (!buf)
1128 			return 0;
1129 		while (off > 0) {
1130 			unsigned long n = off;
1131 			if (n > PAGE_SIZE)
1132 				n = PAGE_SIZE;
1133 			if (!dump_write(file, buf, n))
1134 				return 0;
1135 			off -= n;
1136 		}
1137 		free_page((unsigned long)buf);
1138 	}
1139 	return 1;
1140 }
1141 
1142 /*
1143  * Decide what to dump of a segment, part, all or none.
1144  */
1145 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1146 				   unsigned long mm_flags)
1147 {
1148 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1149 
1150 	/* The vma can be set up to tell us the answer directly.  */
1151 	if (vma->vm_flags & VM_ALWAYSDUMP)
1152 		goto whole;
1153 
1154 	/* Hugetlb memory check */
1155 	if (vma->vm_flags & VM_HUGETLB) {
1156 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1157 			goto whole;
1158 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1159 			goto whole;
1160 	}
1161 
1162 	/* Do not dump I/O mapped devices or special mappings */
1163 	if (vma->vm_flags & (VM_IO | VM_RESERVED))
1164 		return 0;
1165 
1166 	/* By default, dump shared memory if mapped from an anonymous file. */
1167 	if (vma->vm_flags & VM_SHARED) {
1168 		if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1169 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1170 			goto whole;
1171 		return 0;
1172 	}
1173 
1174 	/* Dump segments that have been written to.  */
1175 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1176 		goto whole;
1177 	if (vma->vm_file == NULL)
1178 		return 0;
1179 
1180 	if (FILTER(MAPPED_PRIVATE))
1181 		goto whole;
1182 
1183 	/*
1184 	 * If this looks like the beginning of a DSO or executable mapping,
1185 	 * check for an ELF header.  If we find one, dump the first page to
1186 	 * aid in determining what was mapped here.
1187 	 */
1188 	if (FILTER(ELF_HEADERS) &&
1189 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1190 		u32 __user *header = (u32 __user *) vma->vm_start;
1191 		u32 word;
1192 		mm_segment_t fs = get_fs();
1193 		/*
1194 		 * Doing it this way gets the constant folded by GCC.
1195 		 */
1196 		union {
1197 			u32 cmp;
1198 			char elfmag[SELFMAG];
1199 		} magic;
1200 		BUILD_BUG_ON(SELFMAG != sizeof word);
1201 		magic.elfmag[EI_MAG0] = ELFMAG0;
1202 		magic.elfmag[EI_MAG1] = ELFMAG1;
1203 		magic.elfmag[EI_MAG2] = ELFMAG2;
1204 		magic.elfmag[EI_MAG3] = ELFMAG3;
1205 		/*
1206 		 * Switch to the user "segment" for get_user(),
1207 		 * then put back what elf_core_dump() had in place.
1208 		 */
1209 		set_fs(USER_DS);
1210 		if (unlikely(get_user(word, header)))
1211 			word = 0;
1212 		set_fs(fs);
1213 		if (word == magic.cmp)
1214 			return PAGE_SIZE;
1215 	}
1216 
1217 #undef	FILTER
1218 
1219 	return 0;
1220 
1221 whole:
1222 	return vma->vm_end - vma->vm_start;
1223 }
1224 
1225 /* An ELF note in memory */
1226 struct memelfnote
1227 {
1228 	const char *name;
1229 	int type;
1230 	unsigned int datasz;
1231 	void *data;
1232 };
1233 
1234 static int notesize(struct memelfnote *en)
1235 {
1236 	int sz;
1237 
1238 	sz = sizeof(struct elf_note);
1239 	sz += roundup(strlen(en->name) + 1, 4);
1240 	sz += roundup(en->datasz, 4);
1241 
1242 	return sz;
1243 }
1244 
1245 #define DUMP_WRITE(addr, nr, foffset)	\
1246 	do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1247 
1248 static int alignfile(struct file *file, loff_t *foffset)
1249 {
1250 	static const char buf[4] = { 0, };
1251 	DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1252 	return 1;
1253 }
1254 
1255 static int writenote(struct memelfnote *men, struct file *file,
1256 			loff_t *foffset)
1257 {
1258 	struct elf_note en;
1259 	en.n_namesz = strlen(men->name) + 1;
1260 	en.n_descsz = men->datasz;
1261 	en.n_type = men->type;
1262 
1263 	DUMP_WRITE(&en, sizeof(en), foffset);
1264 	DUMP_WRITE(men->name, en.n_namesz, foffset);
1265 	if (!alignfile(file, foffset))
1266 		return 0;
1267 	DUMP_WRITE(men->data, men->datasz, foffset);
1268 	if (!alignfile(file, foffset))
1269 		return 0;
1270 
1271 	return 1;
1272 }
1273 #undef DUMP_WRITE
1274 
1275 #define DUMP_WRITE(addr, nr)				\
1276 	if ((size += (nr)) > cprm->limit ||		\
1277 	    !dump_write(cprm->file, (addr), (nr)))	\
1278 		goto end_coredump;
1279 
1280 static void fill_elf_header(struct elfhdr *elf, int segs,
1281 			    u16 machine, u32 flags, u8 osabi)
1282 {
1283 	memset(elf, 0, sizeof(*elf));
1284 
1285 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1286 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1287 	elf->e_ident[EI_DATA] = ELF_DATA;
1288 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1289 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1290 
1291 	elf->e_type = ET_CORE;
1292 	elf->e_machine = machine;
1293 	elf->e_version = EV_CURRENT;
1294 	elf->e_phoff = sizeof(struct elfhdr);
1295 	elf->e_flags = flags;
1296 	elf->e_ehsize = sizeof(struct elfhdr);
1297 	elf->e_phentsize = sizeof(struct elf_phdr);
1298 	elf->e_phnum = segs;
1299 
1300 	return;
1301 }
1302 
1303 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1304 {
1305 	phdr->p_type = PT_NOTE;
1306 	phdr->p_offset = offset;
1307 	phdr->p_vaddr = 0;
1308 	phdr->p_paddr = 0;
1309 	phdr->p_filesz = sz;
1310 	phdr->p_memsz = 0;
1311 	phdr->p_flags = 0;
1312 	phdr->p_align = 0;
1313 	return;
1314 }
1315 
1316 static void fill_note(struct memelfnote *note, const char *name, int type,
1317 		unsigned int sz, void *data)
1318 {
1319 	note->name = name;
1320 	note->type = type;
1321 	note->datasz = sz;
1322 	note->data = data;
1323 	return;
1324 }
1325 
1326 /*
1327  * fill up all the fields in prstatus from the given task struct, except
1328  * registers which need to be filled up separately.
1329  */
1330 static void fill_prstatus(struct elf_prstatus *prstatus,
1331 		struct task_struct *p, long signr)
1332 {
1333 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1334 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1335 	prstatus->pr_sighold = p->blocked.sig[0];
1336 	rcu_read_lock();
1337 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1338 	rcu_read_unlock();
1339 	prstatus->pr_pid = task_pid_vnr(p);
1340 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1341 	prstatus->pr_sid = task_session_vnr(p);
1342 	if (thread_group_leader(p)) {
1343 		struct task_cputime cputime;
1344 
1345 		/*
1346 		 * This is the record for the group leader.  It shows the
1347 		 * group-wide total, not its individual thread total.
1348 		 */
1349 		thread_group_cputime(p, &cputime);
1350 		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1351 		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1352 	} else {
1353 		cputime_to_timeval(p->utime, &prstatus->pr_utime);
1354 		cputime_to_timeval(p->stime, &prstatus->pr_stime);
1355 	}
1356 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1357 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1358 }
1359 
1360 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1361 		       struct mm_struct *mm)
1362 {
1363 	const struct cred *cred;
1364 	unsigned int i, len;
1365 
1366 	/* first copy the parameters from user space */
1367 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1368 
1369 	len = mm->arg_end - mm->arg_start;
1370 	if (len >= ELF_PRARGSZ)
1371 		len = ELF_PRARGSZ-1;
1372 	if (copy_from_user(&psinfo->pr_psargs,
1373 		           (const char __user *)mm->arg_start, len))
1374 		return -EFAULT;
1375 	for(i = 0; i < len; i++)
1376 		if (psinfo->pr_psargs[i] == 0)
1377 			psinfo->pr_psargs[i] = ' ';
1378 	psinfo->pr_psargs[len] = 0;
1379 
1380 	rcu_read_lock();
1381 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1382 	rcu_read_unlock();
1383 	psinfo->pr_pid = task_pid_vnr(p);
1384 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1385 	psinfo->pr_sid = task_session_vnr(p);
1386 
1387 	i = p->state ? ffz(~p->state) + 1 : 0;
1388 	psinfo->pr_state = i;
1389 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1390 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1391 	psinfo->pr_nice = task_nice(p);
1392 	psinfo->pr_flag = p->flags;
1393 	rcu_read_lock();
1394 	cred = __task_cred(p);
1395 	SET_UID(psinfo->pr_uid, cred->uid);
1396 	SET_GID(psinfo->pr_gid, cred->gid);
1397 	rcu_read_unlock();
1398 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1399 
1400 	return 0;
1401 }
1402 
1403 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1404 {
1405 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1406 	int i = 0;
1407 	do
1408 		i += 2;
1409 	while (auxv[i - 2] != AT_NULL);
1410 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1411 }
1412 
1413 #ifdef CORE_DUMP_USE_REGSET
1414 #include <linux/regset.h>
1415 
1416 struct elf_thread_core_info {
1417 	struct elf_thread_core_info *next;
1418 	struct task_struct *task;
1419 	struct elf_prstatus prstatus;
1420 	struct memelfnote notes[0];
1421 };
1422 
1423 struct elf_note_info {
1424 	struct elf_thread_core_info *thread;
1425 	struct memelfnote psinfo;
1426 	struct memelfnote auxv;
1427 	size_t size;
1428 	int thread_notes;
1429 };
1430 
1431 /*
1432  * When a regset has a writeback hook, we call it on each thread before
1433  * dumping user memory.  On register window machines, this makes sure the
1434  * user memory backing the register data is up to date before we read it.
1435  */
1436 static void do_thread_regset_writeback(struct task_struct *task,
1437 				       const struct user_regset *regset)
1438 {
1439 	if (regset->writeback)
1440 		regset->writeback(task, regset, 1);
1441 }
1442 
1443 static int fill_thread_core_info(struct elf_thread_core_info *t,
1444 				 const struct user_regset_view *view,
1445 				 long signr, size_t *total)
1446 {
1447 	unsigned int i;
1448 
1449 	/*
1450 	 * NT_PRSTATUS is the one special case, because the regset data
1451 	 * goes into the pr_reg field inside the note contents, rather
1452 	 * than being the whole note contents.  We fill the reset in here.
1453 	 * We assume that regset 0 is NT_PRSTATUS.
1454 	 */
1455 	fill_prstatus(&t->prstatus, t->task, signr);
1456 	(void) view->regsets[0].get(t->task, &view->regsets[0],
1457 				    0, sizeof(t->prstatus.pr_reg),
1458 				    &t->prstatus.pr_reg, NULL);
1459 
1460 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1461 		  sizeof(t->prstatus), &t->prstatus);
1462 	*total += notesize(&t->notes[0]);
1463 
1464 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1465 
1466 	/*
1467 	 * Each other regset might generate a note too.  For each regset
1468 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1469 	 * all zero and we'll know to skip writing it later.
1470 	 */
1471 	for (i = 1; i < view->n; ++i) {
1472 		const struct user_regset *regset = &view->regsets[i];
1473 		do_thread_regset_writeback(t->task, regset);
1474 		if (regset->core_note_type &&
1475 		    (!regset->active || regset->active(t->task, regset))) {
1476 			int ret;
1477 			size_t size = regset->n * regset->size;
1478 			void *data = kmalloc(size, GFP_KERNEL);
1479 			if (unlikely(!data))
1480 				return 0;
1481 			ret = regset->get(t->task, regset,
1482 					  0, size, data, NULL);
1483 			if (unlikely(ret))
1484 				kfree(data);
1485 			else {
1486 				if (regset->core_note_type != NT_PRFPREG)
1487 					fill_note(&t->notes[i], "LINUX",
1488 						  regset->core_note_type,
1489 						  size, data);
1490 				else {
1491 					t->prstatus.pr_fpvalid = 1;
1492 					fill_note(&t->notes[i], "CORE",
1493 						  NT_PRFPREG, size, data);
1494 				}
1495 				*total += notesize(&t->notes[i]);
1496 			}
1497 		}
1498 	}
1499 
1500 	return 1;
1501 }
1502 
1503 static int fill_note_info(struct elfhdr *elf, int phdrs,
1504 			  struct elf_note_info *info,
1505 			  long signr, struct pt_regs *regs)
1506 {
1507 	struct task_struct *dump_task = current;
1508 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1509 	struct elf_thread_core_info *t;
1510 	struct elf_prpsinfo *psinfo;
1511 	struct core_thread *ct;
1512 	unsigned int i;
1513 
1514 	info->size = 0;
1515 	info->thread = NULL;
1516 
1517 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1518 	if (psinfo == NULL)
1519 		return 0;
1520 
1521 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1522 
1523 	/*
1524 	 * Figure out how many notes we're going to need for each thread.
1525 	 */
1526 	info->thread_notes = 0;
1527 	for (i = 0; i < view->n; ++i)
1528 		if (view->regsets[i].core_note_type != 0)
1529 			++info->thread_notes;
1530 
1531 	/*
1532 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1533 	 * since it is our one special case.
1534 	 */
1535 	if (unlikely(info->thread_notes == 0) ||
1536 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1537 		WARN_ON(1);
1538 		return 0;
1539 	}
1540 
1541 	/*
1542 	 * Initialize the ELF file header.
1543 	 */
1544 	fill_elf_header(elf, phdrs,
1545 			view->e_machine, view->e_flags, view->ei_osabi);
1546 
1547 	/*
1548 	 * Allocate a structure for each thread.
1549 	 */
1550 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1551 		t = kzalloc(offsetof(struct elf_thread_core_info,
1552 				     notes[info->thread_notes]),
1553 			    GFP_KERNEL);
1554 		if (unlikely(!t))
1555 			return 0;
1556 
1557 		t->task = ct->task;
1558 		if (ct->task == dump_task || !info->thread) {
1559 			t->next = info->thread;
1560 			info->thread = t;
1561 		} else {
1562 			/*
1563 			 * Make sure to keep the original task at
1564 			 * the head of the list.
1565 			 */
1566 			t->next = info->thread->next;
1567 			info->thread->next = t;
1568 		}
1569 	}
1570 
1571 	/*
1572 	 * Now fill in each thread's information.
1573 	 */
1574 	for (t = info->thread; t != NULL; t = t->next)
1575 		if (!fill_thread_core_info(t, view, signr, &info->size))
1576 			return 0;
1577 
1578 	/*
1579 	 * Fill in the two process-wide notes.
1580 	 */
1581 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1582 	info->size += notesize(&info->psinfo);
1583 
1584 	fill_auxv_note(&info->auxv, current->mm);
1585 	info->size += notesize(&info->auxv);
1586 
1587 	return 1;
1588 }
1589 
1590 static size_t get_note_info_size(struct elf_note_info *info)
1591 {
1592 	return info->size;
1593 }
1594 
1595 /*
1596  * Write all the notes for each thread.  When writing the first thread, the
1597  * process-wide notes are interleaved after the first thread-specific note.
1598  */
1599 static int write_note_info(struct elf_note_info *info,
1600 			   struct file *file, loff_t *foffset)
1601 {
1602 	bool first = 1;
1603 	struct elf_thread_core_info *t = info->thread;
1604 
1605 	do {
1606 		int i;
1607 
1608 		if (!writenote(&t->notes[0], file, foffset))
1609 			return 0;
1610 
1611 		if (first && !writenote(&info->psinfo, file, foffset))
1612 			return 0;
1613 		if (first && !writenote(&info->auxv, file, foffset))
1614 			return 0;
1615 
1616 		for (i = 1; i < info->thread_notes; ++i)
1617 			if (t->notes[i].data &&
1618 			    !writenote(&t->notes[i], file, foffset))
1619 				return 0;
1620 
1621 		first = 0;
1622 		t = t->next;
1623 	} while (t);
1624 
1625 	return 1;
1626 }
1627 
1628 static void free_note_info(struct elf_note_info *info)
1629 {
1630 	struct elf_thread_core_info *threads = info->thread;
1631 	while (threads) {
1632 		unsigned int i;
1633 		struct elf_thread_core_info *t = threads;
1634 		threads = t->next;
1635 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1636 		for (i = 1; i < info->thread_notes; ++i)
1637 			kfree(t->notes[i].data);
1638 		kfree(t);
1639 	}
1640 	kfree(info->psinfo.data);
1641 }
1642 
1643 #else
1644 
1645 /* Here is the structure in which status of each thread is captured. */
1646 struct elf_thread_status
1647 {
1648 	struct list_head list;
1649 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1650 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1651 	struct task_struct *thread;
1652 #ifdef ELF_CORE_COPY_XFPREGS
1653 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1654 #endif
1655 	struct memelfnote notes[3];
1656 	int num_notes;
1657 };
1658 
1659 /*
1660  * In order to add the specific thread information for the elf file format,
1661  * we need to keep a linked list of every threads pr_status and then create
1662  * a single section for them in the final core file.
1663  */
1664 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1665 {
1666 	int sz = 0;
1667 	struct task_struct *p = t->thread;
1668 	t->num_notes = 0;
1669 
1670 	fill_prstatus(&t->prstatus, p, signr);
1671 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1672 
1673 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1674 		  &(t->prstatus));
1675 	t->num_notes++;
1676 	sz += notesize(&t->notes[0]);
1677 
1678 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1679 								&t->fpu))) {
1680 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1681 			  &(t->fpu));
1682 		t->num_notes++;
1683 		sz += notesize(&t->notes[1]);
1684 	}
1685 
1686 #ifdef ELF_CORE_COPY_XFPREGS
1687 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1688 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1689 			  sizeof(t->xfpu), &t->xfpu);
1690 		t->num_notes++;
1691 		sz += notesize(&t->notes[2]);
1692 	}
1693 #endif
1694 	return sz;
1695 }
1696 
1697 struct elf_note_info {
1698 	struct memelfnote *notes;
1699 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1700 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1701 	struct list_head thread_list;
1702 	elf_fpregset_t *fpu;
1703 #ifdef ELF_CORE_COPY_XFPREGS
1704 	elf_fpxregset_t *xfpu;
1705 #endif
1706 	int thread_status_size;
1707 	int numnote;
1708 };
1709 
1710 static int elf_note_info_init(struct elf_note_info *info)
1711 {
1712 	memset(info, 0, sizeof(*info));
1713 	INIT_LIST_HEAD(&info->thread_list);
1714 
1715 	/* Allocate space for six ELF notes */
1716 	info->notes = kmalloc(6 * sizeof(struct memelfnote), GFP_KERNEL);
1717 	if (!info->notes)
1718 		return 0;
1719 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1720 	if (!info->psinfo)
1721 		goto notes_free;
1722 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1723 	if (!info->prstatus)
1724 		goto psinfo_free;
1725 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1726 	if (!info->fpu)
1727 		goto prstatus_free;
1728 #ifdef ELF_CORE_COPY_XFPREGS
1729 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1730 	if (!info->xfpu)
1731 		goto fpu_free;
1732 #endif
1733 	return 1;
1734 #ifdef ELF_CORE_COPY_XFPREGS
1735  fpu_free:
1736 	kfree(info->fpu);
1737 #endif
1738  prstatus_free:
1739 	kfree(info->prstatus);
1740  psinfo_free:
1741 	kfree(info->psinfo);
1742  notes_free:
1743 	kfree(info->notes);
1744 	return 0;
1745 }
1746 
1747 static int fill_note_info(struct elfhdr *elf, int phdrs,
1748 			  struct elf_note_info *info,
1749 			  long signr, struct pt_regs *regs)
1750 {
1751 	struct list_head *t;
1752 
1753 	if (!elf_note_info_init(info))
1754 		return 0;
1755 
1756 	if (signr) {
1757 		struct core_thread *ct;
1758 		struct elf_thread_status *ets;
1759 
1760 		for (ct = current->mm->core_state->dumper.next;
1761 						ct; ct = ct->next) {
1762 			ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1763 			if (!ets)
1764 				return 0;
1765 
1766 			ets->thread = ct->task;
1767 			list_add(&ets->list, &info->thread_list);
1768 		}
1769 
1770 		list_for_each(t, &info->thread_list) {
1771 			int sz;
1772 
1773 			ets = list_entry(t, struct elf_thread_status, list);
1774 			sz = elf_dump_thread_status(signr, ets);
1775 			info->thread_status_size += sz;
1776 		}
1777 	}
1778 	/* now collect the dump for the current */
1779 	memset(info->prstatus, 0, sizeof(*info->prstatus));
1780 	fill_prstatus(info->prstatus, current, signr);
1781 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1782 
1783 	/* Set up header */
1784 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1785 
1786 	/*
1787 	 * Set up the notes in similar form to SVR4 core dumps made
1788 	 * with info from their /proc.
1789 	 */
1790 
1791 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1792 		  sizeof(*info->prstatus), info->prstatus);
1793 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
1794 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1795 		  sizeof(*info->psinfo), info->psinfo);
1796 
1797 	info->numnote = 2;
1798 
1799 	fill_auxv_note(&info->notes[info->numnote++], current->mm);
1800 
1801 	/* Try to dump the FPU. */
1802 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1803 							       info->fpu);
1804 	if (info->prstatus->pr_fpvalid)
1805 		fill_note(info->notes + info->numnote++,
1806 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1807 #ifdef ELF_CORE_COPY_XFPREGS
1808 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
1809 		fill_note(info->notes + info->numnote++,
1810 			  "LINUX", ELF_CORE_XFPREG_TYPE,
1811 			  sizeof(*info->xfpu), info->xfpu);
1812 #endif
1813 
1814 	return 1;
1815 }
1816 
1817 static size_t get_note_info_size(struct elf_note_info *info)
1818 {
1819 	int sz = 0;
1820 	int i;
1821 
1822 	for (i = 0; i < info->numnote; i++)
1823 		sz += notesize(info->notes + i);
1824 
1825 	sz += info->thread_status_size;
1826 
1827 	return sz;
1828 }
1829 
1830 static int write_note_info(struct elf_note_info *info,
1831 			   struct file *file, loff_t *foffset)
1832 {
1833 	int i;
1834 	struct list_head *t;
1835 
1836 	for (i = 0; i < info->numnote; i++)
1837 		if (!writenote(info->notes + i, file, foffset))
1838 			return 0;
1839 
1840 	/* write out the thread status notes section */
1841 	list_for_each(t, &info->thread_list) {
1842 		struct elf_thread_status *tmp =
1843 				list_entry(t, struct elf_thread_status, list);
1844 
1845 		for (i = 0; i < tmp->num_notes; i++)
1846 			if (!writenote(&tmp->notes[i], file, foffset))
1847 				return 0;
1848 	}
1849 
1850 	return 1;
1851 }
1852 
1853 static void free_note_info(struct elf_note_info *info)
1854 {
1855 	while (!list_empty(&info->thread_list)) {
1856 		struct list_head *tmp = info->thread_list.next;
1857 		list_del(tmp);
1858 		kfree(list_entry(tmp, struct elf_thread_status, list));
1859 	}
1860 
1861 	kfree(info->prstatus);
1862 	kfree(info->psinfo);
1863 	kfree(info->notes);
1864 	kfree(info->fpu);
1865 #ifdef ELF_CORE_COPY_XFPREGS
1866 	kfree(info->xfpu);
1867 #endif
1868 }
1869 
1870 #endif
1871 
1872 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1873 					struct vm_area_struct *gate_vma)
1874 {
1875 	struct vm_area_struct *ret = tsk->mm->mmap;
1876 
1877 	if (ret)
1878 		return ret;
1879 	return gate_vma;
1880 }
1881 /*
1882  * Helper function for iterating across a vma list.  It ensures that the caller
1883  * will visit `gate_vma' prior to terminating the search.
1884  */
1885 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1886 					struct vm_area_struct *gate_vma)
1887 {
1888 	struct vm_area_struct *ret;
1889 
1890 	ret = this_vma->vm_next;
1891 	if (ret)
1892 		return ret;
1893 	if (this_vma == gate_vma)
1894 		return NULL;
1895 	return gate_vma;
1896 }
1897 
1898 /*
1899  * Actual dumper
1900  *
1901  * This is a two-pass process; first we find the offsets of the bits,
1902  * and then they are actually written out.  If we run out of core limit
1903  * we just truncate.
1904  */
1905 static int elf_core_dump(struct coredump_params *cprm)
1906 {
1907 	int has_dumped = 0;
1908 	mm_segment_t fs;
1909 	int segs;
1910 	size_t size = 0;
1911 	struct vm_area_struct *vma, *gate_vma;
1912 	struct elfhdr *elf = NULL;
1913 	loff_t offset = 0, dataoff, foffset;
1914 	unsigned long mm_flags;
1915 	struct elf_note_info info;
1916 
1917 	/*
1918 	 * We no longer stop all VM operations.
1919 	 *
1920 	 * This is because those proceses that could possibly change map_count
1921 	 * or the mmap / vma pages are now blocked in do_exit on current
1922 	 * finishing this core dump.
1923 	 *
1924 	 * Only ptrace can touch these memory addresses, but it doesn't change
1925 	 * the map_count or the pages allocated. So no possibility of crashing
1926 	 * exists while dumping the mm->vm_next areas to the core file.
1927 	 */
1928 
1929 	/* alloc memory for large data structures: too large to be on stack */
1930 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1931 	if (!elf)
1932 		goto out;
1933 	/*
1934 	 * The number of segs are recored into ELF header as 16bit value.
1935 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
1936 	 */
1937 	segs = current->mm->map_count;
1938 #ifdef ELF_CORE_EXTRA_PHDRS
1939 	segs += ELF_CORE_EXTRA_PHDRS;
1940 #endif
1941 
1942 	gate_vma = get_gate_vma(current);
1943 	if (gate_vma != NULL)
1944 		segs++;
1945 
1946 	/*
1947 	 * Collect all the non-memory information about the process for the
1948 	 * notes.  This also sets up the file header.
1949 	 */
1950 	if (!fill_note_info(elf, segs + 1, /* including notes section */
1951 			    &info, cprm->signr, cprm->regs))
1952 		goto cleanup;
1953 
1954 	has_dumped = 1;
1955 	current->flags |= PF_DUMPCORE;
1956 
1957 	fs = get_fs();
1958 	set_fs(KERNEL_DS);
1959 
1960 	DUMP_WRITE(elf, sizeof(*elf));
1961 	offset += sizeof(*elf);				/* Elf header */
1962 	offset += (segs + 1) * sizeof(struct elf_phdr); /* Program headers */
1963 	foffset = offset;
1964 
1965 	/* Write notes phdr entry */
1966 	{
1967 		struct elf_phdr phdr;
1968 		size_t sz = get_note_info_size(&info);
1969 
1970 		sz += elf_coredump_extra_notes_size();
1971 
1972 		fill_elf_note_phdr(&phdr, sz, offset);
1973 		offset += sz;
1974 		DUMP_WRITE(&phdr, sizeof(phdr));
1975 	}
1976 
1977 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1978 
1979 	/*
1980 	 * We must use the same mm->flags while dumping core to avoid
1981 	 * inconsistency between the program headers and bodies, otherwise an
1982 	 * unusable core file can be generated.
1983 	 */
1984 	mm_flags = current->mm->flags;
1985 
1986 	/* Write program headers for segments dump */
1987 	for (vma = first_vma(current, gate_vma); vma != NULL;
1988 			vma = next_vma(vma, gate_vma)) {
1989 		struct elf_phdr phdr;
1990 
1991 		phdr.p_type = PT_LOAD;
1992 		phdr.p_offset = offset;
1993 		phdr.p_vaddr = vma->vm_start;
1994 		phdr.p_paddr = 0;
1995 		phdr.p_filesz = vma_dump_size(vma, mm_flags);
1996 		phdr.p_memsz = vma->vm_end - vma->vm_start;
1997 		offset += phdr.p_filesz;
1998 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1999 		if (vma->vm_flags & VM_WRITE)
2000 			phdr.p_flags |= PF_W;
2001 		if (vma->vm_flags & VM_EXEC)
2002 			phdr.p_flags |= PF_X;
2003 		phdr.p_align = ELF_EXEC_PAGESIZE;
2004 
2005 		DUMP_WRITE(&phdr, sizeof(phdr));
2006 	}
2007 
2008 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
2009 	ELF_CORE_WRITE_EXTRA_PHDRS;
2010 #endif
2011 
2012  	/* write out the notes section */
2013 	if (!write_note_info(&info, cprm->file, &foffset))
2014 		goto end_coredump;
2015 
2016 	if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2017 		goto end_coredump;
2018 
2019 	/* Align to page */
2020 	if (!dump_seek(cprm->file, dataoff - foffset))
2021 		goto end_coredump;
2022 
2023 	for (vma = first_vma(current, gate_vma); vma != NULL;
2024 			vma = next_vma(vma, gate_vma)) {
2025 		unsigned long addr;
2026 		unsigned long end;
2027 
2028 		end = vma->vm_start + vma_dump_size(vma, mm_flags);
2029 
2030 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2031 			struct page *page;
2032 			int stop;
2033 
2034 			page = get_dump_page(addr);
2035 			if (page) {
2036 				void *kaddr = kmap(page);
2037 				stop = ((size += PAGE_SIZE) > cprm->limit) ||
2038 					!dump_write(cprm->file, kaddr,
2039 						    PAGE_SIZE);
2040 				kunmap(page);
2041 				page_cache_release(page);
2042 			} else
2043 				stop = !dump_seek(cprm->file, PAGE_SIZE);
2044 			if (stop)
2045 				goto end_coredump;
2046 		}
2047 	}
2048 
2049 #ifdef ELF_CORE_WRITE_EXTRA_DATA
2050 	ELF_CORE_WRITE_EXTRA_DATA;
2051 #endif
2052 
2053 end_coredump:
2054 	set_fs(fs);
2055 
2056 cleanup:
2057 	free_note_info(&info);
2058 	kfree(elf);
2059 out:
2060 	return has_dumped;
2061 }
2062 
2063 #endif		/* CONFIG_ELF_CORE */
2064 
2065 static int __init init_elf_binfmt(void)
2066 {
2067 	return register_binfmt(&elf_format);
2068 }
2069 
2070 static void __exit exit_elf_binfmt(void)
2071 {
2072 	/* Remove the COFF and ELF loaders. */
2073 	unregister_binfmt(&elf_format);
2074 }
2075 
2076 core_initcall(init_elf_binfmt);
2077 module_exit(exit_elf_binfmt);
2078 MODULE_LICENSE("GPL");
2079