xref: /linux/fs/binfmt_elf.c (revision d524dac9279b6a41ffdf7ff7958c577f2e387db6)
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 <linux/coredump.h>
35 #include <asm/uaccess.h>
36 #include <asm/param.h>
37 #include <asm/page.h>
38 
39 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
40 static int load_elf_library(struct file *);
41 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
42 				int, int, unsigned long);
43 
44 /*
45  * If we don't support core dumping, then supply a NULL so we
46  * don't even try.
47  */
48 #ifdef CONFIG_ELF_CORE
49 static int elf_core_dump(struct coredump_params *cprm);
50 #else
51 #define elf_core_dump	NULL
52 #endif
53 
54 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
55 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
56 #else
57 #define ELF_MIN_ALIGN	PAGE_SIZE
58 #endif
59 
60 #ifndef ELF_CORE_EFLAGS
61 #define ELF_CORE_EFLAGS	0
62 #endif
63 
64 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
65 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
66 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
67 
68 static struct linux_binfmt elf_format = {
69 	.module		= THIS_MODULE,
70 	.load_binary	= load_elf_binary,
71 	.load_shlib	= load_elf_library,
72 	.core_dump	= elf_core_dump,
73 	.min_coredump	= ELF_EXEC_PAGESIZE,
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 static unsigned long elf_map(struct file *filep, unsigned long addr,
319 		struct elf_phdr *eppnt, int prot, int type,
320 		unsigned long total_size)
321 {
322 	unsigned long map_addr;
323 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
324 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
325 	addr = ELF_PAGESTART(addr);
326 	size = ELF_PAGEALIGN(size);
327 
328 	/* mmap() will return -EINVAL if given a zero size, but a
329 	 * segment with zero filesize is perfectly valid */
330 	if (!size)
331 		return addr;
332 
333 	down_write(&current->mm->mmap_sem);
334 	/*
335 	* total_size is the size of the ELF (interpreter) image.
336 	* The _first_ mmap needs to know the full size, otherwise
337 	* randomization might put this image into an overlapping
338 	* position with the ELF binary image. (since size < total_size)
339 	* So we first map the 'big' image - and unmap the remainder at
340 	* the end. (which unmap is needed for ELF images with holes.)
341 	*/
342 	if (total_size) {
343 		total_size = ELF_PAGEALIGN(total_size);
344 		map_addr = do_mmap(filep, addr, total_size, prot, type, off);
345 		if (!BAD_ADDR(map_addr))
346 			do_munmap(current->mm, map_addr+size, total_size-size);
347 	} else
348 		map_addr = do_mmap(filep, addr, size, prot, type, off);
349 
350 	up_write(&current->mm->mmap_sem);
351 	return(map_addr);
352 }
353 
354 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
355 {
356 	int i, first_idx = -1, last_idx = -1;
357 
358 	for (i = 0; i < nr; i++) {
359 		if (cmds[i].p_type == PT_LOAD) {
360 			last_idx = i;
361 			if (first_idx == -1)
362 				first_idx = i;
363 		}
364 	}
365 	if (first_idx == -1)
366 		return 0;
367 
368 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
369 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
370 }
371 
372 
373 /* This is much more generalized than the library routine read function,
374    so we keep this separate.  Technically the library read function
375    is only provided so that we can read a.out libraries that have
376    an ELF header */
377 
378 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
379 		struct file *interpreter, unsigned long *interp_map_addr,
380 		unsigned long no_base)
381 {
382 	struct elf_phdr *elf_phdata;
383 	struct elf_phdr *eppnt;
384 	unsigned long load_addr = 0;
385 	int load_addr_set = 0;
386 	unsigned long last_bss = 0, elf_bss = 0;
387 	unsigned long error = ~0UL;
388 	unsigned long total_size;
389 	int retval, i, size;
390 
391 	/* First of all, some simple consistency checks */
392 	if (interp_elf_ex->e_type != ET_EXEC &&
393 	    interp_elf_ex->e_type != ET_DYN)
394 		goto out;
395 	if (!elf_check_arch(interp_elf_ex))
396 		goto out;
397 	if (!interpreter->f_op || !interpreter->f_op->mmap)
398 		goto out;
399 
400 	/*
401 	 * If the size of this structure has changed, then punt, since
402 	 * we will be doing the wrong thing.
403 	 */
404 	if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
405 		goto out;
406 	if (interp_elf_ex->e_phnum < 1 ||
407 		interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
408 		goto out;
409 
410 	/* Now read in all of the header information */
411 	size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
412 	if (size > ELF_MIN_ALIGN)
413 		goto out;
414 	elf_phdata = kmalloc(size, GFP_KERNEL);
415 	if (!elf_phdata)
416 		goto out;
417 
418 	retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
419 			     (char *)elf_phdata, size);
420 	error = -EIO;
421 	if (retval != size) {
422 		if (retval < 0)
423 			error = retval;
424 		goto out_close;
425 	}
426 
427 	total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
428 	if (!total_size) {
429 		error = -EINVAL;
430 		goto out_close;
431 	}
432 
433 	eppnt = elf_phdata;
434 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
435 		if (eppnt->p_type == PT_LOAD) {
436 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
437 			int elf_prot = 0;
438 			unsigned long vaddr = 0;
439 			unsigned long k, map_addr;
440 
441 			if (eppnt->p_flags & PF_R)
442 		    		elf_prot = PROT_READ;
443 			if (eppnt->p_flags & PF_W)
444 				elf_prot |= PROT_WRITE;
445 			if (eppnt->p_flags & PF_X)
446 				elf_prot |= PROT_EXEC;
447 			vaddr = eppnt->p_vaddr;
448 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
449 				elf_type |= MAP_FIXED;
450 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
451 				load_addr = -vaddr;
452 
453 			map_addr = elf_map(interpreter, load_addr + vaddr,
454 					eppnt, elf_prot, elf_type, total_size);
455 			total_size = 0;
456 			if (!*interp_map_addr)
457 				*interp_map_addr = map_addr;
458 			error = map_addr;
459 			if (BAD_ADDR(map_addr))
460 				goto out_close;
461 
462 			if (!load_addr_set &&
463 			    interp_elf_ex->e_type == ET_DYN) {
464 				load_addr = map_addr - ELF_PAGESTART(vaddr);
465 				load_addr_set = 1;
466 			}
467 
468 			/*
469 			 * Check to see if the section's size will overflow the
470 			 * allowed task size. Note that p_filesz must always be
471 			 * <= p_memsize so it's only necessary to check p_memsz.
472 			 */
473 			k = load_addr + eppnt->p_vaddr;
474 			if (BAD_ADDR(k) ||
475 			    eppnt->p_filesz > eppnt->p_memsz ||
476 			    eppnt->p_memsz > TASK_SIZE ||
477 			    TASK_SIZE - eppnt->p_memsz < k) {
478 				error = -ENOMEM;
479 				goto out_close;
480 			}
481 
482 			/*
483 			 * Find the end of the file mapping for this phdr, and
484 			 * keep track of the largest address we see for this.
485 			 */
486 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
487 			if (k > elf_bss)
488 				elf_bss = k;
489 
490 			/*
491 			 * Do the same thing for the memory mapping - between
492 			 * elf_bss and last_bss is the bss section.
493 			 */
494 			k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
495 			if (k > last_bss)
496 				last_bss = k;
497 		}
498 	}
499 
500 	if (last_bss > elf_bss) {
501 		/*
502 		 * Now fill out the bss section.  First pad the last page up
503 		 * to the page boundary, and then perform a mmap to make sure
504 		 * that there are zero-mapped pages up to and including the
505 		 * last bss page.
506 		 */
507 		if (padzero(elf_bss)) {
508 			error = -EFAULT;
509 			goto out_close;
510 		}
511 
512 		/* What we have mapped so far */
513 		elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
514 
515 		/* Map the last of the bss segment */
516 		down_write(&current->mm->mmap_sem);
517 		error = do_brk(elf_bss, last_bss - elf_bss);
518 		up_write(&current->mm->mmap_sem);
519 		if (BAD_ADDR(error))
520 			goto out_close;
521 	}
522 
523 	error = load_addr;
524 
525 out_close:
526 	kfree(elf_phdata);
527 out:
528 	return error;
529 }
530 
531 /*
532  * These are the functions used to load ELF style executables and shared
533  * libraries.  There is no binary dependent code anywhere else.
534  */
535 
536 #define INTERPRETER_NONE 0
537 #define INTERPRETER_ELF 2
538 
539 #ifndef STACK_RND_MASK
540 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))	/* 8MB of VA */
541 #endif
542 
543 static unsigned long randomize_stack_top(unsigned long stack_top)
544 {
545 	unsigned int random_variable = 0;
546 
547 	if ((current->flags & PF_RANDOMIZE) &&
548 		!(current->personality & ADDR_NO_RANDOMIZE)) {
549 		random_variable = get_random_int() & STACK_RND_MASK;
550 		random_variable <<= PAGE_SHIFT;
551 	}
552 #ifdef CONFIG_STACK_GROWSUP
553 	return PAGE_ALIGN(stack_top) + random_variable;
554 #else
555 	return PAGE_ALIGN(stack_top) - random_variable;
556 #endif
557 }
558 
559 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
560 {
561 	struct file *interpreter = NULL; /* to shut gcc up */
562  	unsigned long load_addr = 0, load_bias = 0;
563 	int load_addr_set = 0;
564 	char * elf_interpreter = NULL;
565 	unsigned long error;
566 	struct elf_phdr *elf_ppnt, *elf_phdata;
567 	unsigned long elf_bss, elf_brk;
568 	int retval, i;
569 	unsigned int size;
570 	unsigned long elf_entry;
571 	unsigned long interp_load_addr = 0;
572 	unsigned long start_code, end_code, start_data, end_data;
573 	unsigned long reloc_func_desc = 0;
574 	int executable_stack = EXSTACK_DEFAULT;
575 	unsigned long def_flags = 0;
576 	struct {
577 		struct elfhdr elf_ex;
578 		struct elfhdr interp_elf_ex;
579 	} *loc;
580 
581 	loc = kmalloc(sizeof(*loc), GFP_KERNEL);
582 	if (!loc) {
583 		retval = -ENOMEM;
584 		goto out_ret;
585 	}
586 
587 	/* Get the exec-header */
588 	loc->elf_ex = *((struct elfhdr *)bprm->buf);
589 
590 	retval = -ENOEXEC;
591 	/* First of all, some simple consistency checks */
592 	if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
593 		goto out;
594 
595 	if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
596 		goto out;
597 	if (!elf_check_arch(&loc->elf_ex))
598 		goto out;
599 	if (!bprm->file->f_op || !bprm->file->f_op->mmap)
600 		goto out;
601 
602 	/* Now read in all of the header information */
603 	if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
604 		goto out;
605 	if (loc->elf_ex.e_phnum < 1 ||
606 	 	loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
607 		goto out;
608 	size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
609 	retval = -ENOMEM;
610 	elf_phdata = kmalloc(size, GFP_KERNEL);
611 	if (!elf_phdata)
612 		goto out;
613 
614 	retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
615 			     (char *)elf_phdata, size);
616 	if (retval != size) {
617 		if (retval >= 0)
618 			retval = -EIO;
619 		goto out_free_ph;
620 	}
621 
622 	elf_ppnt = elf_phdata;
623 	elf_bss = 0;
624 	elf_brk = 0;
625 
626 	start_code = ~0UL;
627 	end_code = 0;
628 	start_data = 0;
629 	end_data = 0;
630 
631 	for (i = 0; i < loc->elf_ex.e_phnum; i++) {
632 		if (elf_ppnt->p_type == PT_INTERP) {
633 			/* This is the program interpreter used for
634 			 * shared libraries - for now assume that this
635 			 * is an a.out format binary
636 			 */
637 			retval = -ENOEXEC;
638 			if (elf_ppnt->p_filesz > PATH_MAX ||
639 			    elf_ppnt->p_filesz < 2)
640 				goto out_free_ph;
641 
642 			retval = -ENOMEM;
643 			elf_interpreter = kmalloc(elf_ppnt->p_filesz,
644 						  GFP_KERNEL);
645 			if (!elf_interpreter)
646 				goto out_free_ph;
647 
648 			retval = kernel_read(bprm->file, elf_ppnt->p_offset,
649 					     elf_interpreter,
650 					     elf_ppnt->p_filesz);
651 			if (retval != elf_ppnt->p_filesz) {
652 				if (retval >= 0)
653 					retval = -EIO;
654 				goto out_free_interp;
655 			}
656 			/* make sure path is NULL terminated */
657 			retval = -ENOEXEC;
658 			if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
659 				goto out_free_interp;
660 
661 			interpreter = open_exec(elf_interpreter);
662 			retval = PTR_ERR(interpreter);
663 			if (IS_ERR(interpreter))
664 				goto out_free_interp;
665 
666 			/*
667 			 * If the binary is not readable then enforce
668 			 * mm->dumpable = 0 regardless of the interpreter's
669 			 * permissions.
670 			 */
671 			if (file_permission(interpreter, MAY_READ) < 0)
672 				bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
673 
674 			retval = kernel_read(interpreter, 0, bprm->buf,
675 					     BINPRM_BUF_SIZE);
676 			if (retval != BINPRM_BUF_SIZE) {
677 				if (retval >= 0)
678 					retval = -EIO;
679 				goto out_free_dentry;
680 			}
681 
682 			/* Get the exec headers */
683 			loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
684 			break;
685 		}
686 		elf_ppnt++;
687 	}
688 
689 	elf_ppnt = elf_phdata;
690 	for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
691 		if (elf_ppnt->p_type == PT_GNU_STACK) {
692 			if (elf_ppnt->p_flags & PF_X)
693 				executable_stack = EXSTACK_ENABLE_X;
694 			else
695 				executable_stack = EXSTACK_DISABLE_X;
696 			break;
697 		}
698 
699 	/* Some simple consistency checks for the interpreter */
700 	if (elf_interpreter) {
701 		retval = -ELIBBAD;
702 		/* Not an ELF interpreter */
703 		if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
704 			goto out_free_dentry;
705 		/* Verify the interpreter has a valid arch */
706 		if (!elf_check_arch(&loc->interp_elf_ex))
707 			goto out_free_dentry;
708 	}
709 
710 	/* Flush all traces of the currently running executable */
711 	retval = flush_old_exec(bprm);
712 	if (retval)
713 		goto out_free_dentry;
714 
715 	/* OK, This is the point of no return */
716 	current->flags &= ~PF_FORKNOEXEC;
717 	current->mm->def_flags = def_flags;
718 
719 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
720 	   may depend on the personality.  */
721 	SET_PERSONALITY(loc->elf_ex);
722 	if (elf_read_implies_exec(loc->elf_ex, executable_stack))
723 		current->personality |= READ_IMPLIES_EXEC;
724 
725 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
726 		current->flags |= PF_RANDOMIZE;
727 
728 	setup_new_exec(bprm);
729 
730 	/* Do this so that we can load the interpreter, if need be.  We will
731 	   change some of these later */
732 	current->mm->free_area_cache = current->mm->mmap_base;
733 	current->mm->cached_hole_size = 0;
734 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
735 				 executable_stack);
736 	if (retval < 0) {
737 		send_sig(SIGKILL, current, 0);
738 		goto out_free_dentry;
739 	}
740 
741 	current->mm->start_stack = bprm->p;
742 
743 	/* Now we do a little grungy work by mmapping the ELF image into
744 	   the correct location in memory. */
745 	for(i = 0, elf_ppnt = elf_phdata;
746 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
747 		int elf_prot = 0, elf_flags;
748 		unsigned long k, vaddr;
749 
750 		if (elf_ppnt->p_type != PT_LOAD)
751 			continue;
752 
753 		if (unlikely (elf_brk > elf_bss)) {
754 			unsigned long nbyte;
755 
756 			/* There was a PT_LOAD segment with p_memsz > p_filesz
757 			   before this one. Map anonymous pages, if needed,
758 			   and clear the area.  */
759 			retval = set_brk(elf_bss + load_bias,
760 					 elf_brk + load_bias);
761 			if (retval) {
762 				send_sig(SIGKILL, current, 0);
763 				goto out_free_dentry;
764 			}
765 			nbyte = ELF_PAGEOFFSET(elf_bss);
766 			if (nbyte) {
767 				nbyte = ELF_MIN_ALIGN - nbyte;
768 				if (nbyte > elf_brk - elf_bss)
769 					nbyte = elf_brk - elf_bss;
770 				if (clear_user((void __user *)elf_bss +
771 							load_bias, nbyte)) {
772 					/*
773 					 * This bss-zeroing can fail if the ELF
774 					 * file specifies odd protections. So
775 					 * we don't check the return value
776 					 */
777 				}
778 			}
779 		}
780 
781 		if (elf_ppnt->p_flags & PF_R)
782 			elf_prot |= PROT_READ;
783 		if (elf_ppnt->p_flags & PF_W)
784 			elf_prot |= PROT_WRITE;
785 		if (elf_ppnt->p_flags & PF_X)
786 			elf_prot |= PROT_EXEC;
787 
788 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
789 
790 		vaddr = elf_ppnt->p_vaddr;
791 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
792 			elf_flags |= MAP_FIXED;
793 		} else if (loc->elf_ex.e_type == ET_DYN) {
794 			/* Try and get dynamic programs out of the way of the
795 			 * default mmap base, as well as whatever program they
796 			 * might try to exec.  This is because the brk will
797 			 * follow the loader, and is not movable.  */
798 #if defined(CONFIG_X86) || defined(CONFIG_ARM)
799 			load_bias = 0;
800 #else
801 			load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
802 #endif
803 		}
804 
805 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
806 				elf_prot, elf_flags, 0);
807 		if (BAD_ADDR(error)) {
808 			send_sig(SIGKILL, current, 0);
809 			retval = IS_ERR((void *)error) ?
810 				PTR_ERR((void*)error) : -EINVAL;
811 			goto out_free_dentry;
812 		}
813 
814 		if (!load_addr_set) {
815 			load_addr_set = 1;
816 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
817 			if (loc->elf_ex.e_type == ET_DYN) {
818 				load_bias += error -
819 				             ELF_PAGESTART(load_bias + vaddr);
820 				load_addr += load_bias;
821 				reloc_func_desc = load_bias;
822 			}
823 		}
824 		k = elf_ppnt->p_vaddr;
825 		if (k < start_code)
826 			start_code = k;
827 		if (start_data < k)
828 			start_data = k;
829 
830 		/*
831 		 * Check to see if the section's size will overflow the
832 		 * allowed task size. Note that p_filesz must always be
833 		 * <= p_memsz so it is only necessary to check p_memsz.
834 		 */
835 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
836 		    elf_ppnt->p_memsz > TASK_SIZE ||
837 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
838 			/* set_brk can never work. Avoid overflows. */
839 			send_sig(SIGKILL, current, 0);
840 			retval = -EINVAL;
841 			goto out_free_dentry;
842 		}
843 
844 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
845 
846 		if (k > elf_bss)
847 			elf_bss = k;
848 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
849 			end_code = k;
850 		if (end_data < k)
851 			end_data = k;
852 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
853 		if (k > elf_brk)
854 			elf_brk = k;
855 	}
856 
857 	loc->elf_ex.e_entry += load_bias;
858 	elf_bss += load_bias;
859 	elf_brk += load_bias;
860 	start_code += load_bias;
861 	end_code += load_bias;
862 	start_data += load_bias;
863 	end_data += load_bias;
864 
865 	/* Calling set_brk effectively mmaps the pages that we need
866 	 * for the bss and break sections.  We must do this before
867 	 * mapping in the interpreter, to make sure it doesn't wind
868 	 * up getting placed where the bss needs to go.
869 	 */
870 	retval = set_brk(elf_bss, elf_brk);
871 	if (retval) {
872 		send_sig(SIGKILL, current, 0);
873 		goto out_free_dentry;
874 	}
875 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
876 		send_sig(SIGSEGV, current, 0);
877 		retval = -EFAULT; /* Nobody gets to see this, but.. */
878 		goto out_free_dentry;
879 	}
880 
881 	if (elf_interpreter) {
882 		unsigned long uninitialized_var(interp_map_addr);
883 
884 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
885 					    interpreter,
886 					    &interp_map_addr,
887 					    load_bias);
888 		if (!IS_ERR((void *)elf_entry)) {
889 			/*
890 			 * load_elf_interp() returns relocation
891 			 * adjustment
892 			 */
893 			interp_load_addr = elf_entry;
894 			elf_entry += loc->interp_elf_ex.e_entry;
895 		}
896 		if (BAD_ADDR(elf_entry)) {
897 			force_sig(SIGSEGV, current);
898 			retval = IS_ERR((void *)elf_entry) ?
899 					(int)elf_entry : -EINVAL;
900 			goto out_free_dentry;
901 		}
902 		reloc_func_desc = interp_load_addr;
903 
904 		allow_write_access(interpreter);
905 		fput(interpreter);
906 		kfree(elf_interpreter);
907 	} else {
908 		elf_entry = loc->elf_ex.e_entry;
909 		if (BAD_ADDR(elf_entry)) {
910 			force_sig(SIGSEGV, current);
911 			retval = -EINVAL;
912 			goto out_free_dentry;
913 		}
914 	}
915 
916 	kfree(elf_phdata);
917 
918 	set_binfmt(&elf_format);
919 
920 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
921 	retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
922 	if (retval < 0) {
923 		send_sig(SIGKILL, current, 0);
924 		goto out;
925 	}
926 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
927 
928 	install_exec_creds(bprm);
929 	current->flags &= ~PF_FORKNOEXEC;
930 	retval = create_elf_tables(bprm, &loc->elf_ex,
931 			  load_addr, interp_load_addr);
932 	if (retval < 0) {
933 		send_sig(SIGKILL, current, 0);
934 		goto out;
935 	}
936 	/* N.B. passed_fileno might not be initialized? */
937 	current->mm->end_code = end_code;
938 	current->mm->start_code = start_code;
939 	current->mm->start_data = start_data;
940 	current->mm->end_data = end_data;
941 	current->mm->start_stack = bprm->p;
942 
943 #ifdef arch_randomize_brk
944 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1))
945 		current->mm->brk = current->mm->start_brk =
946 			arch_randomize_brk(current->mm);
947 #endif
948 
949 	if (current->personality & MMAP_PAGE_ZERO) {
950 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
951 		   and some applications "depend" upon this behavior.
952 		   Since we do not have the power to recompile these, we
953 		   emulate the SVr4 behavior. Sigh. */
954 		down_write(&current->mm->mmap_sem);
955 		error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
956 				MAP_FIXED | MAP_PRIVATE, 0);
957 		up_write(&current->mm->mmap_sem);
958 	}
959 
960 #ifdef ELF_PLAT_INIT
961 	/*
962 	 * The ABI may specify that certain registers be set up in special
963 	 * ways (on i386 %edx is the address of a DT_FINI function, for
964 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
965 	 * that the e_entry field is the address of the function descriptor
966 	 * for the startup routine, rather than the address of the startup
967 	 * routine itself.  This macro performs whatever initialization to
968 	 * the regs structure is required as well as any relocations to the
969 	 * function descriptor entries when executing dynamically links apps.
970 	 */
971 	ELF_PLAT_INIT(regs, reloc_func_desc);
972 #endif
973 
974 	start_thread(regs, elf_entry, bprm->p);
975 	retval = 0;
976 out:
977 	kfree(loc);
978 out_ret:
979 	return retval;
980 
981 	/* error cleanup */
982 out_free_dentry:
983 	allow_write_access(interpreter);
984 	if (interpreter)
985 		fput(interpreter);
986 out_free_interp:
987 	kfree(elf_interpreter);
988 out_free_ph:
989 	kfree(elf_phdata);
990 	goto out;
991 }
992 
993 /* This is really simpleminded and specialized - we are loading an
994    a.out library that is given an ELF header. */
995 static int load_elf_library(struct file *file)
996 {
997 	struct elf_phdr *elf_phdata;
998 	struct elf_phdr *eppnt;
999 	unsigned long elf_bss, bss, len;
1000 	int retval, error, i, j;
1001 	struct elfhdr elf_ex;
1002 
1003 	error = -ENOEXEC;
1004 	retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1005 	if (retval != sizeof(elf_ex))
1006 		goto out;
1007 
1008 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1009 		goto out;
1010 
1011 	/* First of all, some simple consistency checks */
1012 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1013 	    !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1014 		goto out;
1015 
1016 	/* Now read in all of the header information */
1017 
1018 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1019 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1020 
1021 	error = -ENOMEM;
1022 	elf_phdata = kmalloc(j, GFP_KERNEL);
1023 	if (!elf_phdata)
1024 		goto out;
1025 
1026 	eppnt = elf_phdata;
1027 	error = -ENOEXEC;
1028 	retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1029 	if (retval != j)
1030 		goto out_free_ph;
1031 
1032 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1033 		if ((eppnt + i)->p_type == PT_LOAD)
1034 			j++;
1035 	if (j != 1)
1036 		goto out_free_ph;
1037 
1038 	while (eppnt->p_type != PT_LOAD)
1039 		eppnt++;
1040 
1041 	/* Now use mmap to map the library into memory. */
1042 	down_write(&current->mm->mmap_sem);
1043 	error = do_mmap(file,
1044 			ELF_PAGESTART(eppnt->p_vaddr),
1045 			(eppnt->p_filesz +
1046 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1047 			PROT_READ | PROT_WRITE | PROT_EXEC,
1048 			MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1049 			(eppnt->p_offset -
1050 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1051 	up_write(&current->mm->mmap_sem);
1052 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1053 		goto out_free_ph;
1054 
1055 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1056 	if (padzero(elf_bss)) {
1057 		error = -EFAULT;
1058 		goto out_free_ph;
1059 	}
1060 
1061 	len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1062 			    ELF_MIN_ALIGN - 1);
1063 	bss = eppnt->p_memsz + eppnt->p_vaddr;
1064 	if (bss > len) {
1065 		down_write(&current->mm->mmap_sem);
1066 		do_brk(len, bss - len);
1067 		up_write(&current->mm->mmap_sem);
1068 	}
1069 	error = 0;
1070 
1071 out_free_ph:
1072 	kfree(elf_phdata);
1073 out:
1074 	return error;
1075 }
1076 
1077 #ifdef CONFIG_ELF_CORE
1078 /*
1079  * ELF core dumper
1080  *
1081  * Modelled on fs/exec.c:aout_core_dump()
1082  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1083  */
1084 
1085 /*
1086  * Decide what to dump of a segment, part, all or none.
1087  */
1088 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1089 				   unsigned long mm_flags)
1090 {
1091 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1092 
1093 	/* The vma can be set up to tell us the answer directly.  */
1094 	if (vma->vm_flags & VM_ALWAYSDUMP)
1095 		goto whole;
1096 
1097 	/* Hugetlb memory check */
1098 	if (vma->vm_flags & VM_HUGETLB) {
1099 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1100 			goto whole;
1101 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1102 			goto whole;
1103 	}
1104 
1105 	/* Do not dump I/O mapped devices or special mappings */
1106 	if (vma->vm_flags & (VM_IO | VM_RESERVED))
1107 		return 0;
1108 
1109 	/* By default, dump shared memory if mapped from an anonymous file. */
1110 	if (vma->vm_flags & VM_SHARED) {
1111 		if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1112 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1113 			goto whole;
1114 		return 0;
1115 	}
1116 
1117 	/* Dump segments that have been written to.  */
1118 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1119 		goto whole;
1120 	if (vma->vm_file == NULL)
1121 		return 0;
1122 
1123 	if (FILTER(MAPPED_PRIVATE))
1124 		goto whole;
1125 
1126 	/*
1127 	 * If this looks like the beginning of a DSO or executable mapping,
1128 	 * check for an ELF header.  If we find one, dump the first page to
1129 	 * aid in determining what was mapped here.
1130 	 */
1131 	if (FILTER(ELF_HEADERS) &&
1132 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1133 		u32 __user *header = (u32 __user *) vma->vm_start;
1134 		u32 word;
1135 		mm_segment_t fs = get_fs();
1136 		/*
1137 		 * Doing it this way gets the constant folded by GCC.
1138 		 */
1139 		union {
1140 			u32 cmp;
1141 			char elfmag[SELFMAG];
1142 		} magic;
1143 		BUILD_BUG_ON(SELFMAG != sizeof word);
1144 		magic.elfmag[EI_MAG0] = ELFMAG0;
1145 		magic.elfmag[EI_MAG1] = ELFMAG1;
1146 		magic.elfmag[EI_MAG2] = ELFMAG2;
1147 		magic.elfmag[EI_MAG3] = ELFMAG3;
1148 		/*
1149 		 * Switch to the user "segment" for get_user(),
1150 		 * then put back what elf_core_dump() had in place.
1151 		 */
1152 		set_fs(USER_DS);
1153 		if (unlikely(get_user(word, header)))
1154 			word = 0;
1155 		set_fs(fs);
1156 		if (word == magic.cmp)
1157 			return PAGE_SIZE;
1158 	}
1159 
1160 #undef	FILTER
1161 
1162 	return 0;
1163 
1164 whole:
1165 	return vma->vm_end - vma->vm_start;
1166 }
1167 
1168 /* An ELF note in memory */
1169 struct memelfnote
1170 {
1171 	const char *name;
1172 	int type;
1173 	unsigned int datasz;
1174 	void *data;
1175 };
1176 
1177 static int notesize(struct memelfnote *en)
1178 {
1179 	int sz;
1180 
1181 	sz = sizeof(struct elf_note);
1182 	sz += roundup(strlen(en->name) + 1, 4);
1183 	sz += roundup(en->datasz, 4);
1184 
1185 	return sz;
1186 }
1187 
1188 #define DUMP_WRITE(addr, nr, foffset)	\
1189 	do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1190 
1191 static int alignfile(struct file *file, loff_t *foffset)
1192 {
1193 	static const char buf[4] = { 0, };
1194 	DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1195 	return 1;
1196 }
1197 
1198 static int writenote(struct memelfnote *men, struct file *file,
1199 			loff_t *foffset)
1200 {
1201 	struct elf_note en;
1202 	en.n_namesz = strlen(men->name) + 1;
1203 	en.n_descsz = men->datasz;
1204 	en.n_type = men->type;
1205 
1206 	DUMP_WRITE(&en, sizeof(en), foffset);
1207 	DUMP_WRITE(men->name, en.n_namesz, foffset);
1208 	if (!alignfile(file, foffset))
1209 		return 0;
1210 	DUMP_WRITE(men->data, men->datasz, foffset);
1211 	if (!alignfile(file, foffset))
1212 		return 0;
1213 
1214 	return 1;
1215 }
1216 #undef DUMP_WRITE
1217 
1218 static void fill_elf_header(struct elfhdr *elf, int segs,
1219 			    u16 machine, u32 flags, u8 osabi)
1220 {
1221 	memset(elf, 0, sizeof(*elf));
1222 
1223 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1224 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1225 	elf->e_ident[EI_DATA] = ELF_DATA;
1226 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1227 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1228 
1229 	elf->e_type = ET_CORE;
1230 	elf->e_machine = machine;
1231 	elf->e_version = EV_CURRENT;
1232 	elf->e_phoff = sizeof(struct elfhdr);
1233 	elf->e_flags = flags;
1234 	elf->e_ehsize = sizeof(struct elfhdr);
1235 	elf->e_phentsize = sizeof(struct elf_phdr);
1236 	elf->e_phnum = segs;
1237 
1238 	return;
1239 }
1240 
1241 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1242 {
1243 	phdr->p_type = PT_NOTE;
1244 	phdr->p_offset = offset;
1245 	phdr->p_vaddr = 0;
1246 	phdr->p_paddr = 0;
1247 	phdr->p_filesz = sz;
1248 	phdr->p_memsz = 0;
1249 	phdr->p_flags = 0;
1250 	phdr->p_align = 0;
1251 	return;
1252 }
1253 
1254 static void fill_note(struct memelfnote *note, const char *name, int type,
1255 		unsigned int sz, void *data)
1256 {
1257 	note->name = name;
1258 	note->type = type;
1259 	note->datasz = sz;
1260 	note->data = data;
1261 	return;
1262 }
1263 
1264 /*
1265  * fill up all the fields in prstatus from the given task struct, except
1266  * registers which need to be filled up separately.
1267  */
1268 static void fill_prstatus(struct elf_prstatus *prstatus,
1269 		struct task_struct *p, long signr)
1270 {
1271 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1272 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1273 	prstatus->pr_sighold = p->blocked.sig[0];
1274 	rcu_read_lock();
1275 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1276 	rcu_read_unlock();
1277 	prstatus->pr_pid = task_pid_vnr(p);
1278 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1279 	prstatus->pr_sid = task_session_vnr(p);
1280 	if (thread_group_leader(p)) {
1281 		struct task_cputime cputime;
1282 
1283 		/*
1284 		 * This is the record for the group leader.  It shows the
1285 		 * group-wide total, not its individual thread total.
1286 		 */
1287 		thread_group_cputime(p, &cputime);
1288 		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1289 		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1290 	} else {
1291 		cputime_to_timeval(p->utime, &prstatus->pr_utime);
1292 		cputime_to_timeval(p->stime, &prstatus->pr_stime);
1293 	}
1294 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1295 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1296 }
1297 
1298 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1299 		       struct mm_struct *mm)
1300 {
1301 	const struct cred *cred;
1302 	unsigned int i, len;
1303 
1304 	/* first copy the parameters from user space */
1305 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1306 
1307 	len = mm->arg_end - mm->arg_start;
1308 	if (len >= ELF_PRARGSZ)
1309 		len = ELF_PRARGSZ-1;
1310 	if (copy_from_user(&psinfo->pr_psargs,
1311 		           (const char __user *)mm->arg_start, len))
1312 		return -EFAULT;
1313 	for(i = 0; i < len; i++)
1314 		if (psinfo->pr_psargs[i] == 0)
1315 			psinfo->pr_psargs[i] = ' ';
1316 	psinfo->pr_psargs[len] = 0;
1317 
1318 	rcu_read_lock();
1319 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1320 	rcu_read_unlock();
1321 	psinfo->pr_pid = task_pid_vnr(p);
1322 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1323 	psinfo->pr_sid = task_session_vnr(p);
1324 
1325 	i = p->state ? ffz(~p->state) + 1 : 0;
1326 	psinfo->pr_state = i;
1327 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1328 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1329 	psinfo->pr_nice = task_nice(p);
1330 	psinfo->pr_flag = p->flags;
1331 	rcu_read_lock();
1332 	cred = __task_cred(p);
1333 	SET_UID(psinfo->pr_uid, cred->uid);
1334 	SET_GID(psinfo->pr_gid, cred->gid);
1335 	rcu_read_unlock();
1336 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1337 
1338 	return 0;
1339 }
1340 
1341 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1342 {
1343 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1344 	int i = 0;
1345 	do
1346 		i += 2;
1347 	while (auxv[i - 2] != AT_NULL);
1348 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1349 }
1350 
1351 #ifdef CORE_DUMP_USE_REGSET
1352 #include <linux/regset.h>
1353 
1354 struct elf_thread_core_info {
1355 	struct elf_thread_core_info *next;
1356 	struct task_struct *task;
1357 	struct elf_prstatus prstatus;
1358 	struct memelfnote notes[0];
1359 };
1360 
1361 struct elf_note_info {
1362 	struct elf_thread_core_info *thread;
1363 	struct memelfnote psinfo;
1364 	struct memelfnote auxv;
1365 	size_t size;
1366 	int thread_notes;
1367 };
1368 
1369 /*
1370  * When a regset has a writeback hook, we call it on each thread before
1371  * dumping user memory.  On register window machines, this makes sure the
1372  * user memory backing the register data is up to date before we read it.
1373  */
1374 static void do_thread_regset_writeback(struct task_struct *task,
1375 				       const struct user_regset *regset)
1376 {
1377 	if (regset->writeback)
1378 		regset->writeback(task, regset, 1);
1379 }
1380 
1381 static int fill_thread_core_info(struct elf_thread_core_info *t,
1382 				 const struct user_regset_view *view,
1383 				 long signr, size_t *total)
1384 {
1385 	unsigned int i;
1386 
1387 	/*
1388 	 * NT_PRSTATUS is the one special case, because the regset data
1389 	 * goes into the pr_reg field inside the note contents, rather
1390 	 * than being the whole note contents.  We fill the reset in here.
1391 	 * We assume that regset 0 is NT_PRSTATUS.
1392 	 */
1393 	fill_prstatus(&t->prstatus, t->task, signr);
1394 	(void) view->regsets[0].get(t->task, &view->regsets[0],
1395 				    0, sizeof(t->prstatus.pr_reg),
1396 				    &t->prstatus.pr_reg, NULL);
1397 
1398 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1399 		  sizeof(t->prstatus), &t->prstatus);
1400 	*total += notesize(&t->notes[0]);
1401 
1402 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1403 
1404 	/*
1405 	 * Each other regset might generate a note too.  For each regset
1406 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1407 	 * all zero and we'll know to skip writing it later.
1408 	 */
1409 	for (i = 1; i < view->n; ++i) {
1410 		const struct user_regset *regset = &view->regsets[i];
1411 		do_thread_regset_writeback(t->task, regset);
1412 		if (regset->core_note_type &&
1413 		    (!regset->active || regset->active(t->task, regset))) {
1414 			int ret;
1415 			size_t size = regset->n * regset->size;
1416 			void *data = kmalloc(size, GFP_KERNEL);
1417 			if (unlikely(!data))
1418 				return 0;
1419 			ret = regset->get(t->task, regset,
1420 					  0, size, data, NULL);
1421 			if (unlikely(ret))
1422 				kfree(data);
1423 			else {
1424 				if (regset->core_note_type != NT_PRFPREG)
1425 					fill_note(&t->notes[i], "LINUX",
1426 						  regset->core_note_type,
1427 						  size, data);
1428 				else {
1429 					t->prstatus.pr_fpvalid = 1;
1430 					fill_note(&t->notes[i], "CORE",
1431 						  NT_PRFPREG, size, data);
1432 				}
1433 				*total += notesize(&t->notes[i]);
1434 			}
1435 		}
1436 	}
1437 
1438 	return 1;
1439 }
1440 
1441 static int fill_note_info(struct elfhdr *elf, int phdrs,
1442 			  struct elf_note_info *info,
1443 			  long signr, struct pt_regs *regs)
1444 {
1445 	struct task_struct *dump_task = current;
1446 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1447 	struct elf_thread_core_info *t;
1448 	struct elf_prpsinfo *psinfo;
1449 	struct core_thread *ct;
1450 	unsigned int i;
1451 
1452 	info->size = 0;
1453 	info->thread = NULL;
1454 
1455 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1456 	if (psinfo == NULL)
1457 		return 0;
1458 
1459 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1460 
1461 	/*
1462 	 * Figure out how many notes we're going to need for each thread.
1463 	 */
1464 	info->thread_notes = 0;
1465 	for (i = 0; i < view->n; ++i)
1466 		if (view->regsets[i].core_note_type != 0)
1467 			++info->thread_notes;
1468 
1469 	/*
1470 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1471 	 * since it is our one special case.
1472 	 */
1473 	if (unlikely(info->thread_notes == 0) ||
1474 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1475 		WARN_ON(1);
1476 		return 0;
1477 	}
1478 
1479 	/*
1480 	 * Initialize the ELF file header.
1481 	 */
1482 	fill_elf_header(elf, phdrs,
1483 			view->e_machine, view->e_flags, view->ei_osabi);
1484 
1485 	/*
1486 	 * Allocate a structure for each thread.
1487 	 */
1488 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1489 		t = kzalloc(offsetof(struct elf_thread_core_info,
1490 				     notes[info->thread_notes]),
1491 			    GFP_KERNEL);
1492 		if (unlikely(!t))
1493 			return 0;
1494 
1495 		t->task = ct->task;
1496 		if (ct->task == dump_task || !info->thread) {
1497 			t->next = info->thread;
1498 			info->thread = t;
1499 		} else {
1500 			/*
1501 			 * Make sure to keep the original task at
1502 			 * the head of the list.
1503 			 */
1504 			t->next = info->thread->next;
1505 			info->thread->next = t;
1506 		}
1507 	}
1508 
1509 	/*
1510 	 * Now fill in each thread's information.
1511 	 */
1512 	for (t = info->thread; t != NULL; t = t->next)
1513 		if (!fill_thread_core_info(t, view, signr, &info->size))
1514 			return 0;
1515 
1516 	/*
1517 	 * Fill in the two process-wide notes.
1518 	 */
1519 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1520 	info->size += notesize(&info->psinfo);
1521 
1522 	fill_auxv_note(&info->auxv, current->mm);
1523 	info->size += notesize(&info->auxv);
1524 
1525 	return 1;
1526 }
1527 
1528 static size_t get_note_info_size(struct elf_note_info *info)
1529 {
1530 	return info->size;
1531 }
1532 
1533 /*
1534  * Write all the notes for each thread.  When writing the first thread, the
1535  * process-wide notes are interleaved after the first thread-specific note.
1536  */
1537 static int write_note_info(struct elf_note_info *info,
1538 			   struct file *file, loff_t *foffset)
1539 {
1540 	bool first = 1;
1541 	struct elf_thread_core_info *t = info->thread;
1542 
1543 	do {
1544 		int i;
1545 
1546 		if (!writenote(&t->notes[0], file, foffset))
1547 			return 0;
1548 
1549 		if (first && !writenote(&info->psinfo, file, foffset))
1550 			return 0;
1551 		if (first && !writenote(&info->auxv, file, foffset))
1552 			return 0;
1553 
1554 		for (i = 1; i < info->thread_notes; ++i)
1555 			if (t->notes[i].data &&
1556 			    !writenote(&t->notes[i], file, foffset))
1557 				return 0;
1558 
1559 		first = 0;
1560 		t = t->next;
1561 	} while (t);
1562 
1563 	return 1;
1564 }
1565 
1566 static void free_note_info(struct elf_note_info *info)
1567 {
1568 	struct elf_thread_core_info *threads = info->thread;
1569 	while (threads) {
1570 		unsigned int i;
1571 		struct elf_thread_core_info *t = threads;
1572 		threads = t->next;
1573 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1574 		for (i = 1; i < info->thread_notes; ++i)
1575 			kfree(t->notes[i].data);
1576 		kfree(t);
1577 	}
1578 	kfree(info->psinfo.data);
1579 }
1580 
1581 #else
1582 
1583 /* Here is the structure in which status of each thread is captured. */
1584 struct elf_thread_status
1585 {
1586 	struct list_head list;
1587 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1588 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1589 	struct task_struct *thread;
1590 #ifdef ELF_CORE_COPY_XFPREGS
1591 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1592 #endif
1593 	struct memelfnote notes[3];
1594 	int num_notes;
1595 };
1596 
1597 /*
1598  * In order to add the specific thread information for the elf file format,
1599  * we need to keep a linked list of every threads pr_status and then create
1600  * a single section for them in the final core file.
1601  */
1602 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1603 {
1604 	int sz = 0;
1605 	struct task_struct *p = t->thread;
1606 	t->num_notes = 0;
1607 
1608 	fill_prstatus(&t->prstatus, p, signr);
1609 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1610 
1611 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1612 		  &(t->prstatus));
1613 	t->num_notes++;
1614 	sz += notesize(&t->notes[0]);
1615 
1616 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1617 								&t->fpu))) {
1618 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1619 			  &(t->fpu));
1620 		t->num_notes++;
1621 		sz += notesize(&t->notes[1]);
1622 	}
1623 
1624 #ifdef ELF_CORE_COPY_XFPREGS
1625 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1626 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1627 			  sizeof(t->xfpu), &t->xfpu);
1628 		t->num_notes++;
1629 		sz += notesize(&t->notes[2]);
1630 	}
1631 #endif
1632 	return sz;
1633 }
1634 
1635 struct elf_note_info {
1636 	struct memelfnote *notes;
1637 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1638 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1639 	struct list_head thread_list;
1640 	elf_fpregset_t *fpu;
1641 #ifdef ELF_CORE_COPY_XFPREGS
1642 	elf_fpxregset_t *xfpu;
1643 #endif
1644 	int thread_status_size;
1645 	int numnote;
1646 };
1647 
1648 static int elf_note_info_init(struct elf_note_info *info)
1649 {
1650 	memset(info, 0, sizeof(*info));
1651 	INIT_LIST_HEAD(&info->thread_list);
1652 
1653 	/* Allocate space for six ELF notes */
1654 	info->notes = kmalloc(6 * sizeof(struct memelfnote), GFP_KERNEL);
1655 	if (!info->notes)
1656 		return 0;
1657 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1658 	if (!info->psinfo)
1659 		goto notes_free;
1660 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1661 	if (!info->prstatus)
1662 		goto psinfo_free;
1663 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1664 	if (!info->fpu)
1665 		goto prstatus_free;
1666 #ifdef ELF_CORE_COPY_XFPREGS
1667 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1668 	if (!info->xfpu)
1669 		goto fpu_free;
1670 #endif
1671 	return 1;
1672 #ifdef ELF_CORE_COPY_XFPREGS
1673  fpu_free:
1674 	kfree(info->fpu);
1675 #endif
1676  prstatus_free:
1677 	kfree(info->prstatus);
1678  psinfo_free:
1679 	kfree(info->psinfo);
1680  notes_free:
1681 	kfree(info->notes);
1682 	return 0;
1683 }
1684 
1685 static int fill_note_info(struct elfhdr *elf, int phdrs,
1686 			  struct elf_note_info *info,
1687 			  long signr, struct pt_regs *regs)
1688 {
1689 	struct list_head *t;
1690 
1691 	if (!elf_note_info_init(info))
1692 		return 0;
1693 
1694 	if (signr) {
1695 		struct core_thread *ct;
1696 		struct elf_thread_status *ets;
1697 
1698 		for (ct = current->mm->core_state->dumper.next;
1699 						ct; ct = ct->next) {
1700 			ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1701 			if (!ets)
1702 				return 0;
1703 
1704 			ets->thread = ct->task;
1705 			list_add(&ets->list, &info->thread_list);
1706 		}
1707 
1708 		list_for_each(t, &info->thread_list) {
1709 			int sz;
1710 
1711 			ets = list_entry(t, struct elf_thread_status, list);
1712 			sz = elf_dump_thread_status(signr, ets);
1713 			info->thread_status_size += sz;
1714 		}
1715 	}
1716 	/* now collect the dump for the current */
1717 	memset(info->prstatus, 0, sizeof(*info->prstatus));
1718 	fill_prstatus(info->prstatus, current, signr);
1719 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1720 
1721 	/* Set up header */
1722 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1723 
1724 	/*
1725 	 * Set up the notes in similar form to SVR4 core dumps made
1726 	 * with info from their /proc.
1727 	 */
1728 
1729 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1730 		  sizeof(*info->prstatus), info->prstatus);
1731 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
1732 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1733 		  sizeof(*info->psinfo), info->psinfo);
1734 
1735 	info->numnote = 2;
1736 
1737 	fill_auxv_note(&info->notes[info->numnote++], current->mm);
1738 
1739 	/* Try to dump the FPU. */
1740 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1741 							       info->fpu);
1742 	if (info->prstatus->pr_fpvalid)
1743 		fill_note(info->notes + info->numnote++,
1744 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1745 #ifdef ELF_CORE_COPY_XFPREGS
1746 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
1747 		fill_note(info->notes + info->numnote++,
1748 			  "LINUX", ELF_CORE_XFPREG_TYPE,
1749 			  sizeof(*info->xfpu), info->xfpu);
1750 #endif
1751 
1752 	return 1;
1753 }
1754 
1755 static size_t get_note_info_size(struct elf_note_info *info)
1756 {
1757 	int sz = 0;
1758 	int i;
1759 
1760 	for (i = 0; i < info->numnote; i++)
1761 		sz += notesize(info->notes + i);
1762 
1763 	sz += info->thread_status_size;
1764 
1765 	return sz;
1766 }
1767 
1768 static int write_note_info(struct elf_note_info *info,
1769 			   struct file *file, loff_t *foffset)
1770 {
1771 	int i;
1772 	struct list_head *t;
1773 
1774 	for (i = 0; i < info->numnote; i++)
1775 		if (!writenote(info->notes + i, file, foffset))
1776 			return 0;
1777 
1778 	/* write out the thread status notes section */
1779 	list_for_each(t, &info->thread_list) {
1780 		struct elf_thread_status *tmp =
1781 				list_entry(t, struct elf_thread_status, list);
1782 
1783 		for (i = 0; i < tmp->num_notes; i++)
1784 			if (!writenote(&tmp->notes[i], file, foffset))
1785 				return 0;
1786 	}
1787 
1788 	return 1;
1789 }
1790 
1791 static void free_note_info(struct elf_note_info *info)
1792 {
1793 	while (!list_empty(&info->thread_list)) {
1794 		struct list_head *tmp = info->thread_list.next;
1795 		list_del(tmp);
1796 		kfree(list_entry(tmp, struct elf_thread_status, list));
1797 	}
1798 
1799 	kfree(info->prstatus);
1800 	kfree(info->psinfo);
1801 	kfree(info->notes);
1802 	kfree(info->fpu);
1803 #ifdef ELF_CORE_COPY_XFPREGS
1804 	kfree(info->xfpu);
1805 #endif
1806 }
1807 
1808 #endif
1809 
1810 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1811 					struct vm_area_struct *gate_vma)
1812 {
1813 	struct vm_area_struct *ret = tsk->mm->mmap;
1814 
1815 	if (ret)
1816 		return ret;
1817 	return gate_vma;
1818 }
1819 /*
1820  * Helper function for iterating across a vma list.  It ensures that the caller
1821  * will visit `gate_vma' prior to terminating the search.
1822  */
1823 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1824 					struct vm_area_struct *gate_vma)
1825 {
1826 	struct vm_area_struct *ret;
1827 
1828 	ret = this_vma->vm_next;
1829 	if (ret)
1830 		return ret;
1831 	if (this_vma == gate_vma)
1832 		return NULL;
1833 	return gate_vma;
1834 }
1835 
1836 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1837 			     elf_addr_t e_shoff, int segs)
1838 {
1839 	elf->e_shoff = e_shoff;
1840 	elf->e_shentsize = sizeof(*shdr4extnum);
1841 	elf->e_shnum = 1;
1842 	elf->e_shstrndx = SHN_UNDEF;
1843 
1844 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1845 
1846 	shdr4extnum->sh_type = SHT_NULL;
1847 	shdr4extnum->sh_size = elf->e_shnum;
1848 	shdr4extnum->sh_link = elf->e_shstrndx;
1849 	shdr4extnum->sh_info = segs;
1850 }
1851 
1852 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
1853 				     unsigned long mm_flags)
1854 {
1855 	struct vm_area_struct *vma;
1856 	size_t size = 0;
1857 
1858 	for (vma = first_vma(current, gate_vma); vma != NULL;
1859 	     vma = next_vma(vma, gate_vma))
1860 		size += vma_dump_size(vma, mm_flags);
1861 	return size;
1862 }
1863 
1864 /*
1865  * Actual dumper
1866  *
1867  * This is a two-pass process; first we find the offsets of the bits,
1868  * and then they are actually written out.  If we run out of core limit
1869  * we just truncate.
1870  */
1871 static int elf_core_dump(struct coredump_params *cprm)
1872 {
1873 	int has_dumped = 0;
1874 	mm_segment_t fs;
1875 	int segs;
1876 	size_t size = 0;
1877 	struct vm_area_struct *vma, *gate_vma;
1878 	struct elfhdr *elf = NULL;
1879 	loff_t offset = 0, dataoff, foffset;
1880 	struct elf_note_info info;
1881 	struct elf_phdr *phdr4note = NULL;
1882 	struct elf_shdr *shdr4extnum = NULL;
1883 	Elf_Half e_phnum;
1884 	elf_addr_t e_shoff;
1885 
1886 	/*
1887 	 * We no longer stop all VM operations.
1888 	 *
1889 	 * This is because those proceses that could possibly change map_count
1890 	 * or the mmap / vma pages are now blocked in do_exit on current
1891 	 * finishing this core dump.
1892 	 *
1893 	 * Only ptrace can touch these memory addresses, but it doesn't change
1894 	 * the map_count or the pages allocated. So no possibility of crashing
1895 	 * exists while dumping the mm->vm_next areas to the core file.
1896 	 */
1897 
1898 	/* alloc memory for large data structures: too large to be on stack */
1899 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1900 	if (!elf)
1901 		goto out;
1902 	/*
1903 	 * The number of segs are recored into ELF header as 16bit value.
1904 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
1905 	 */
1906 	segs = current->mm->map_count;
1907 	segs += elf_core_extra_phdrs();
1908 
1909 	gate_vma = get_gate_vma(current);
1910 	if (gate_vma != NULL)
1911 		segs++;
1912 
1913 	/* for notes section */
1914 	segs++;
1915 
1916 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1917 	 * this, kernel supports extended numbering. Have a look at
1918 	 * include/linux/elf.h for further information. */
1919 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1920 
1921 	/*
1922 	 * Collect all the non-memory information about the process for the
1923 	 * notes.  This also sets up the file header.
1924 	 */
1925 	if (!fill_note_info(elf, e_phnum, &info, cprm->signr, cprm->regs))
1926 		goto cleanup;
1927 
1928 	has_dumped = 1;
1929 	current->flags |= PF_DUMPCORE;
1930 
1931 	fs = get_fs();
1932 	set_fs(KERNEL_DS);
1933 
1934 	offset += sizeof(*elf);				/* Elf header */
1935 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1936 	foffset = offset;
1937 
1938 	/* Write notes phdr entry */
1939 	{
1940 		size_t sz = get_note_info_size(&info);
1941 
1942 		sz += elf_coredump_extra_notes_size();
1943 
1944 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1945 		if (!phdr4note)
1946 			goto end_coredump;
1947 
1948 		fill_elf_note_phdr(phdr4note, sz, offset);
1949 		offset += sz;
1950 	}
1951 
1952 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1953 
1954 	offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
1955 	offset += elf_core_extra_data_size();
1956 	e_shoff = offset;
1957 
1958 	if (e_phnum == PN_XNUM) {
1959 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1960 		if (!shdr4extnum)
1961 			goto end_coredump;
1962 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1963 	}
1964 
1965 	offset = dataoff;
1966 
1967 	size += sizeof(*elf);
1968 	if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
1969 		goto end_coredump;
1970 
1971 	size += sizeof(*phdr4note);
1972 	if (size > cprm->limit
1973 	    || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
1974 		goto end_coredump;
1975 
1976 	/* Write program headers for segments dump */
1977 	for (vma = first_vma(current, gate_vma); vma != NULL;
1978 			vma = next_vma(vma, gate_vma)) {
1979 		struct elf_phdr phdr;
1980 
1981 		phdr.p_type = PT_LOAD;
1982 		phdr.p_offset = offset;
1983 		phdr.p_vaddr = vma->vm_start;
1984 		phdr.p_paddr = 0;
1985 		phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
1986 		phdr.p_memsz = vma->vm_end - vma->vm_start;
1987 		offset += phdr.p_filesz;
1988 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1989 		if (vma->vm_flags & VM_WRITE)
1990 			phdr.p_flags |= PF_W;
1991 		if (vma->vm_flags & VM_EXEC)
1992 			phdr.p_flags |= PF_X;
1993 		phdr.p_align = ELF_EXEC_PAGESIZE;
1994 
1995 		size += sizeof(phdr);
1996 		if (size > cprm->limit
1997 		    || !dump_write(cprm->file, &phdr, sizeof(phdr)))
1998 			goto end_coredump;
1999 	}
2000 
2001 	if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2002 		goto end_coredump;
2003 
2004  	/* write out the notes section */
2005 	if (!write_note_info(&info, cprm->file, &foffset))
2006 		goto end_coredump;
2007 
2008 	if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2009 		goto end_coredump;
2010 
2011 	/* Align to page */
2012 	if (!dump_seek(cprm->file, dataoff - foffset))
2013 		goto end_coredump;
2014 
2015 	for (vma = first_vma(current, gate_vma); vma != NULL;
2016 			vma = next_vma(vma, gate_vma)) {
2017 		unsigned long addr;
2018 		unsigned long end;
2019 
2020 		end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2021 
2022 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2023 			struct page *page;
2024 			int stop;
2025 
2026 			page = get_dump_page(addr);
2027 			if (page) {
2028 				void *kaddr = kmap(page);
2029 				stop = ((size += PAGE_SIZE) > cprm->limit) ||
2030 					!dump_write(cprm->file, kaddr,
2031 						    PAGE_SIZE);
2032 				kunmap(page);
2033 				page_cache_release(page);
2034 			} else
2035 				stop = !dump_seek(cprm->file, PAGE_SIZE);
2036 			if (stop)
2037 				goto end_coredump;
2038 		}
2039 	}
2040 
2041 	if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2042 		goto end_coredump;
2043 
2044 	if (e_phnum == PN_XNUM) {
2045 		size += sizeof(*shdr4extnum);
2046 		if (size > cprm->limit
2047 		    || !dump_write(cprm->file, shdr4extnum,
2048 				   sizeof(*shdr4extnum)))
2049 			goto end_coredump;
2050 	}
2051 
2052 end_coredump:
2053 	set_fs(fs);
2054 
2055 cleanup:
2056 	free_note_info(&info);
2057 	kfree(shdr4extnum);
2058 	kfree(phdr4note);
2059 	kfree(elf);
2060 out:
2061 	return has_dumped;
2062 }
2063 
2064 #endif		/* CONFIG_ELF_CORE */
2065 
2066 static int __init init_elf_binfmt(void)
2067 {
2068 	return register_binfmt(&elf_format);
2069 }
2070 
2071 static void __exit exit_elf_binfmt(void)
2072 {
2073 	/* Remove the COFF and ELF loaders. */
2074 	unregister_binfmt(&elf_format);
2075 }
2076 
2077 core_initcall(init_elf_binfmt);
2078 module_exit(exit_elf_binfmt);
2079 MODULE_LICENSE("GPL");
2080