xref: /openbmc/linux/fs/binfmt_elf.c (revision 930beb5a)
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/vmalloc.h>
31 #include <linux/security.h>
32 #include <linux/random.h>
33 #include <linux/elf.h>
34 #include <linux/utsname.h>
35 #include <linux/coredump.h>
36 #include <linux/sched.h>
37 #include <asm/uaccess.h>
38 #include <asm/param.h>
39 #include <asm/page.h>
40 
41 #ifndef user_long_t
42 #define user_long_t long
43 #endif
44 #ifndef user_siginfo_t
45 #define user_siginfo_t siginfo_t
46 #endif
47 
48 static int load_elf_binary(struct linux_binprm *bprm);
49 static int load_elf_library(struct file *);
50 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
51 				int, int, unsigned long);
52 
53 /*
54  * If we don't support core dumping, then supply a NULL so we
55  * don't even try.
56  */
57 #ifdef CONFIG_ELF_CORE
58 static int elf_core_dump(struct coredump_params *cprm);
59 #else
60 #define elf_core_dump	NULL
61 #endif
62 
63 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
64 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
65 #else
66 #define ELF_MIN_ALIGN	PAGE_SIZE
67 #endif
68 
69 #ifndef ELF_CORE_EFLAGS
70 #define ELF_CORE_EFLAGS	0
71 #endif
72 
73 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
74 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
75 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
76 
77 static struct linux_binfmt elf_format = {
78 	.module		= THIS_MODULE,
79 	.load_binary	= load_elf_binary,
80 	.load_shlib	= load_elf_library,
81 	.core_dump	= elf_core_dump,
82 	.min_coredump	= ELF_EXEC_PAGESIZE,
83 };
84 
85 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
86 
87 static int set_brk(unsigned long start, unsigned long end)
88 {
89 	start = ELF_PAGEALIGN(start);
90 	end = ELF_PAGEALIGN(end);
91 	if (end > start) {
92 		unsigned long addr;
93 		addr = vm_brk(start, end - start);
94 		if (BAD_ADDR(addr))
95 			return addr;
96 	}
97 	current->mm->start_brk = current->mm->brk = end;
98 	return 0;
99 }
100 
101 /* We need to explicitly zero any fractional pages
102    after the data section (i.e. bss).  This would
103    contain the junk from the file that should not
104    be in memory
105  */
106 static int padzero(unsigned long elf_bss)
107 {
108 	unsigned long nbyte;
109 
110 	nbyte = ELF_PAGEOFFSET(elf_bss);
111 	if (nbyte) {
112 		nbyte = ELF_MIN_ALIGN - nbyte;
113 		if (clear_user((void __user *) elf_bss, nbyte))
114 			return -EFAULT;
115 	}
116 	return 0;
117 }
118 
119 /* Let's use some macros to make this stack manipulation a little clearer */
120 #ifdef CONFIG_STACK_GROWSUP
121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
122 #define STACK_ROUND(sp, items) \
123 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
124 #define STACK_ALLOC(sp, len) ({ \
125 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
126 	old_sp; })
127 #else
128 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
129 #define STACK_ROUND(sp, items) \
130 	(((unsigned long) (sp - items)) &~ 15UL)
131 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
132 #endif
133 
134 #ifndef ELF_BASE_PLATFORM
135 /*
136  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
137  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
138  * will be copied to the user stack in the same manner as AT_PLATFORM.
139  */
140 #define ELF_BASE_PLATFORM NULL
141 #endif
142 
143 static int
144 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
145 		unsigned long load_addr, unsigned long interp_load_addr)
146 {
147 	unsigned long p = bprm->p;
148 	int argc = bprm->argc;
149 	int envc = bprm->envc;
150 	elf_addr_t __user *argv;
151 	elf_addr_t __user *envp;
152 	elf_addr_t __user *sp;
153 	elf_addr_t __user *u_platform;
154 	elf_addr_t __user *u_base_platform;
155 	elf_addr_t __user *u_rand_bytes;
156 	const char *k_platform = ELF_PLATFORM;
157 	const char *k_base_platform = ELF_BASE_PLATFORM;
158 	unsigned char k_rand_bytes[16];
159 	int items;
160 	elf_addr_t *elf_info;
161 	int ei_index = 0;
162 	const struct cred *cred = current_cred();
163 	struct vm_area_struct *vma;
164 
165 	/*
166 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
167 	 * evictions by the processes running on the same package. One
168 	 * thing we can do is to shuffle the initial stack for them.
169 	 */
170 
171 	p = arch_align_stack(p);
172 
173 	/*
174 	 * If this architecture has a platform capability string, copy it
175 	 * to userspace.  In some cases (Sparc), this info is impossible
176 	 * for userspace to get any other way, in others (i386) it is
177 	 * merely difficult.
178 	 */
179 	u_platform = NULL;
180 	if (k_platform) {
181 		size_t len = strlen(k_platform) + 1;
182 
183 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
184 		if (__copy_to_user(u_platform, k_platform, len))
185 			return -EFAULT;
186 	}
187 
188 	/*
189 	 * If this architecture has a "base" platform capability
190 	 * string, copy it to userspace.
191 	 */
192 	u_base_platform = NULL;
193 	if (k_base_platform) {
194 		size_t len = strlen(k_base_platform) + 1;
195 
196 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
197 		if (__copy_to_user(u_base_platform, k_base_platform, len))
198 			return -EFAULT;
199 	}
200 
201 	/*
202 	 * Generate 16 random bytes for userspace PRNG seeding.
203 	 */
204 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
205 	u_rand_bytes = (elf_addr_t __user *)
206 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
207 	if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
208 		return -EFAULT;
209 
210 	/* Create the ELF interpreter info */
211 	elf_info = (elf_addr_t *)current->mm->saved_auxv;
212 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
213 #define NEW_AUX_ENT(id, val) \
214 	do { \
215 		elf_info[ei_index++] = id; \
216 		elf_info[ei_index++] = val; \
217 	} while (0)
218 
219 #ifdef ARCH_DLINFO
220 	/*
221 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
222 	 * AUXV.
223 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
224 	 * ARCH_DLINFO changes
225 	 */
226 	ARCH_DLINFO;
227 #endif
228 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
229 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
230 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
231 	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
232 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
233 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
234 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
235 	NEW_AUX_ENT(AT_FLAGS, 0);
236 	NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
237 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
238 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
239 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
240 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
241  	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
242 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
243 #ifdef ELF_HWCAP2
244 	NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
245 #endif
246 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
247 	if (k_platform) {
248 		NEW_AUX_ENT(AT_PLATFORM,
249 			    (elf_addr_t)(unsigned long)u_platform);
250 	}
251 	if (k_base_platform) {
252 		NEW_AUX_ENT(AT_BASE_PLATFORM,
253 			    (elf_addr_t)(unsigned long)u_base_platform);
254 	}
255 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
256 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
257 	}
258 #undef NEW_AUX_ENT
259 	/* AT_NULL is zero; clear the rest too */
260 	memset(&elf_info[ei_index], 0,
261 	       sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
262 
263 	/* And advance past the AT_NULL entry.  */
264 	ei_index += 2;
265 
266 	sp = STACK_ADD(p, ei_index);
267 
268 	items = (argc + 1) + (envc + 1) + 1;
269 	bprm->p = STACK_ROUND(sp, items);
270 
271 	/* Point sp at the lowest address on the stack */
272 #ifdef CONFIG_STACK_GROWSUP
273 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
274 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
275 #else
276 	sp = (elf_addr_t __user *)bprm->p;
277 #endif
278 
279 
280 	/*
281 	 * Grow the stack manually; some architectures have a limit on how
282 	 * far ahead a user-space access may be in order to grow the stack.
283 	 */
284 	vma = find_extend_vma(current->mm, bprm->p);
285 	if (!vma)
286 		return -EFAULT;
287 
288 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
289 	if (__put_user(argc, sp++))
290 		return -EFAULT;
291 	argv = sp;
292 	envp = argv + argc + 1;
293 
294 	/* Populate argv and envp */
295 	p = current->mm->arg_end = current->mm->arg_start;
296 	while (argc-- > 0) {
297 		size_t len;
298 		if (__put_user((elf_addr_t)p, argv++))
299 			return -EFAULT;
300 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
301 		if (!len || len > MAX_ARG_STRLEN)
302 			return -EINVAL;
303 		p += len;
304 	}
305 	if (__put_user(0, argv))
306 		return -EFAULT;
307 	current->mm->arg_end = current->mm->env_start = p;
308 	while (envc-- > 0) {
309 		size_t len;
310 		if (__put_user((elf_addr_t)p, envp++))
311 			return -EFAULT;
312 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
313 		if (!len || len > MAX_ARG_STRLEN)
314 			return -EINVAL;
315 		p += len;
316 	}
317 	if (__put_user(0, envp))
318 		return -EFAULT;
319 	current->mm->env_end = p;
320 
321 	/* Put the elf_info on the stack in the right place.  */
322 	sp = (elf_addr_t __user *)envp + 1;
323 	if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
324 		return -EFAULT;
325 	return 0;
326 }
327 
328 #ifndef elf_map
329 
330 static unsigned long elf_map(struct file *filep, unsigned long addr,
331 		struct elf_phdr *eppnt, int prot, int type,
332 		unsigned long total_size)
333 {
334 	unsigned long map_addr;
335 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
336 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
337 	addr = ELF_PAGESTART(addr);
338 	size = ELF_PAGEALIGN(size);
339 
340 	/* mmap() will return -EINVAL if given a zero size, but a
341 	 * segment with zero filesize is perfectly valid */
342 	if (!size)
343 		return addr;
344 
345 	/*
346 	* total_size is the size of the ELF (interpreter) image.
347 	* The _first_ mmap needs to know the full size, otherwise
348 	* randomization might put this image into an overlapping
349 	* position with the ELF binary image. (since size < total_size)
350 	* So we first map the 'big' image - and unmap the remainder at
351 	* the end. (which unmap is needed for ELF images with holes.)
352 	*/
353 	if (total_size) {
354 		total_size = ELF_PAGEALIGN(total_size);
355 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
356 		if (!BAD_ADDR(map_addr))
357 			vm_munmap(map_addr+size, total_size-size);
358 	} else
359 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
360 
361 	return(map_addr);
362 }
363 
364 #endif /* !elf_map */
365 
366 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
367 {
368 	int i, first_idx = -1, last_idx = -1;
369 
370 	for (i = 0; i < nr; i++) {
371 		if (cmds[i].p_type == PT_LOAD) {
372 			last_idx = i;
373 			if (first_idx == -1)
374 				first_idx = i;
375 		}
376 	}
377 	if (first_idx == -1)
378 		return 0;
379 
380 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
381 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
382 }
383 
384 
385 /* This is much more generalized than the library routine read function,
386    so we keep this separate.  Technically the library read function
387    is only provided so that we can read a.out libraries that have
388    an ELF header */
389 
390 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
391 		struct file *interpreter, unsigned long *interp_map_addr,
392 		unsigned long no_base)
393 {
394 	struct elf_phdr *elf_phdata;
395 	struct elf_phdr *eppnt;
396 	unsigned long load_addr = 0;
397 	int load_addr_set = 0;
398 	unsigned long last_bss = 0, elf_bss = 0;
399 	unsigned long error = ~0UL;
400 	unsigned long total_size;
401 	int retval, i, size;
402 
403 	/* First of all, some simple consistency checks */
404 	if (interp_elf_ex->e_type != ET_EXEC &&
405 	    interp_elf_ex->e_type != ET_DYN)
406 		goto out;
407 	if (!elf_check_arch(interp_elf_ex))
408 		goto out;
409 	if (!interpreter->f_op->mmap)
410 		goto out;
411 
412 	/*
413 	 * If the size of this structure has changed, then punt, since
414 	 * we will be doing the wrong thing.
415 	 */
416 	if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
417 		goto out;
418 	if (interp_elf_ex->e_phnum < 1 ||
419 		interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
420 		goto out;
421 
422 	/* Now read in all of the header information */
423 	size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
424 	if (size > ELF_MIN_ALIGN)
425 		goto out;
426 	elf_phdata = kmalloc(size, GFP_KERNEL);
427 	if (!elf_phdata)
428 		goto out;
429 
430 	retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
431 			     (char *)elf_phdata, size);
432 	error = -EIO;
433 	if (retval != size) {
434 		if (retval < 0)
435 			error = retval;
436 		goto out_close;
437 	}
438 
439 	total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
440 	if (!total_size) {
441 		error = -EINVAL;
442 		goto out_close;
443 	}
444 
445 	eppnt = elf_phdata;
446 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
447 		if (eppnt->p_type == PT_LOAD) {
448 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
449 			int elf_prot = 0;
450 			unsigned long vaddr = 0;
451 			unsigned long k, map_addr;
452 
453 			if (eppnt->p_flags & PF_R)
454 		    		elf_prot = PROT_READ;
455 			if (eppnt->p_flags & PF_W)
456 				elf_prot |= PROT_WRITE;
457 			if (eppnt->p_flags & PF_X)
458 				elf_prot |= PROT_EXEC;
459 			vaddr = eppnt->p_vaddr;
460 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
461 				elf_type |= MAP_FIXED;
462 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
463 				load_addr = -vaddr;
464 
465 			map_addr = elf_map(interpreter, load_addr + vaddr,
466 					eppnt, elf_prot, elf_type, total_size);
467 			total_size = 0;
468 			if (!*interp_map_addr)
469 				*interp_map_addr = map_addr;
470 			error = map_addr;
471 			if (BAD_ADDR(map_addr))
472 				goto out_close;
473 
474 			if (!load_addr_set &&
475 			    interp_elf_ex->e_type == ET_DYN) {
476 				load_addr = map_addr - ELF_PAGESTART(vaddr);
477 				load_addr_set = 1;
478 			}
479 
480 			/*
481 			 * Check to see if the section's size will overflow the
482 			 * allowed task size. Note that p_filesz must always be
483 			 * <= p_memsize so it's only necessary to check p_memsz.
484 			 */
485 			k = load_addr + eppnt->p_vaddr;
486 			if (BAD_ADDR(k) ||
487 			    eppnt->p_filesz > eppnt->p_memsz ||
488 			    eppnt->p_memsz > TASK_SIZE ||
489 			    TASK_SIZE - eppnt->p_memsz < k) {
490 				error = -ENOMEM;
491 				goto out_close;
492 			}
493 
494 			/*
495 			 * Find the end of the file mapping for this phdr, and
496 			 * keep track of the largest address we see for this.
497 			 */
498 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
499 			if (k > elf_bss)
500 				elf_bss = k;
501 
502 			/*
503 			 * Do the same thing for the memory mapping - between
504 			 * elf_bss and last_bss is the bss section.
505 			 */
506 			k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
507 			if (k > last_bss)
508 				last_bss = k;
509 		}
510 	}
511 
512 	if (last_bss > elf_bss) {
513 		/*
514 		 * Now fill out the bss section.  First pad the last page up
515 		 * to the page boundary, and then perform a mmap to make sure
516 		 * that there are zero-mapped pages up to and including the
517 		 * last bss page.
518 		 */
519 		if (padzero(elf_bss)) {
520 			error = -EFAULT;
521 			goto out_close;
522 		}
523 
524 		/* What we have mapped so far */
525 		elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
526 
527 		/* Map the last of the bss segment */
528 		error = vm_brk(elf_bss, last_bss - elf_bss);
529 		if (BAD_ADDR(error))
530 			goto out_close;
531 	}
532 
533 	error = load_addr;
534 
535 out_close:
536 	kfree(elf_phdata);
537 out:
538 	return error;
539 }
540 
541 /*
542  * These are the functions used to load ELF style executables and shared
543  * libraries.  There is no binary dependent code anywhere else.
544  */
545 
546 #define INTERPRETER_NONE 0
547 #define INTERPRETER_ELF 2
548 
549 #ifndef STACK_RND_MASK
550 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))	/* 8MB of VA */
551 #endif
552 
553 static unsigned long randomize_stack_top(unsigned long stack_top)
554 {
555 	unsigned int random_variable = 0;
556 
557 	if ((current->flags & PF_RANDOMIZE) &&
558 		!(current->personality & ADDR_NO_RANDOMIZE)) {
559 		random_variable = get_random_int() & STACK_RND_MASK;
560 		random_variable <<= PAGE_SHIFT;
561 	}
562 #ifdef CONFIG_STACK_GROWSUP
563 	return PAGE_ALIGN(stack_top) + random_variable;
564 #else
565 	return PAGE_ALIGN(stack_top) - random_variable;
566 #endif
567 }
568 
569 static int load_elf_binary(struct linux_binprm *bprm)
570 {
571 	struct file *interpreter = NULL; /* to shut gcc up */
572  	unsigned long load_addr = 0, load_bias = 0;
573 	int load_addr_set = 0;
574 	char * elf_interpreter = NULL;
575 	unsigned long error;
576 	struct elf_phdr *elf_ppnt, *elf_phdata;
577 	unsigned long elf_bss, elf_brk;
578 	int retval, i;
579 	unsigned int size;
580 	unsigned long elf_entry;
581 	unsigned long interp_load_addr = 0;
582 	unsigned long start_code, end_code, start_data, end_data;
583 	unsigned long reloc_func_desc __maybe_unused = 0;
584 	int executable_stack = EXSTACK_DEFAULT;
585 	unsigned long def_flags = 0;
586 	struct pt_regs *regs = current_pt_regs();
587 	struct {
588 		struct elfhdr elf_ex;
589 		struct elfhdr interp_elf_ex;
590 	} *loc;
591 
592 	loc = kmalloc(sizeof(*loc), GFP_KERNEL);
593 	if (!loc) {
594 		retval = -ENOMEM;
595 		goto out_ret;
596 	}
597 
598 	/* Get the exec-header */
599 	loc->elf_ex = *((struct elfhdr *)bprm->buf);
600 
601 	retval = -ENOEXEC;
602 	/* First of all, some simple consistency checks */
603 	if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
604 		goto out;
605 
606 	if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
607 		goto out;
608 	if (!elf_check_arch(&loc->elf_ex))
609 		goto out;
610 	if (!bprm->file->f_op->mmap)
611 		goto out;
612 
613 	/* Now read in all of the header information */
614 	if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
615 		goto out;
616 	if (loc->elf_ex.e_phnum < 1 ||
617 	 	loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
618 		goto out;
619 	size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
620 	retval = -ENOMEM;
621 	elf_phdata = kmalloc(size, GFP_KERNEL);
622 	if (!elf_phdata)
623 		goto out;
624 
625 	retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
626 			     (char *)elf_phdata, size);
627 	if (retval != size) {
628 		if (retval >= 0)
629 			retval = -EIO;
630 		goto out_free_ph;
631 	}
632 
633 	elf_ppnt = elf_phdata;
634 	elf_bss = 0;
635 	elf_brk = 0;
636 
637 	start_code = ~0UL;
638 	end_code = 0;
639 	start_data = 0;
640 	end_data = 0;
641 
642 	for (i = 0; i < loc->elf_ex.e_phnum; i++) {
643 		if (elf_ppnt->p_type == PT_INTERP) {
644 			/* This is the program interpreter used for
645 			 * shared libraries - for now assume that this
646 			 * is an a.out format binary
647 			 */
648 			retval = -ENOEXEC;
649 			if (elf_ppnt->p_filesz > PATH_MAX ||
650 			    elf_ppnt->p_filesz < 2)
651 				goto out_free_ph;
652 
653 			retval = -ENOMEM;
654 			elf_interpreter = kmalloc(elf_ppnt->p_filesz,
655 						  GFP_KERNEL);
656 			if (!elf_interpreter)
657 				goto out_free_ph;
658 
659 			retval = kernel_read(bprm->file, elf_ppnt->p_offset,
660 					     elf_interpreter,
661 					     elf_ppnt->p_filesz);
662 			if (retval != elf_ppnt->p_filesz) {
663 				if (retval >= 0)
664 					retval = -EIO;
665 				goto out_free_interp;
666 			}
667 			/* make sure path is NULL terminated */
668 			retval = -ENOEXEC;
669 			if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
670 				goto out_free_interp;
671 
672 			interpreter = open_exec(elf_interpreter);
673 			retval = PTR_ERR(interpreter);
674 			if (IS_ERR(interpreter))
675 				goto out_free_interp;
676 
677 			/*
678 			 * If the binary is not readable then enforce
679 			 * mm->dumpable = 0 regardless of the interpreter's
680 			 * permissions.
681 			 */
682 			would_dump(bprm, interpreter);
683 
684 			retval = kernel_read(interpreter, 0, bprm->buf,
685 					     BINPRM_BUF_SIZE);
686 			if (retval != BINPRM_BUF_SIZE) {
687 				if (retval >= 0)
688 					retval = -EIO;
689 				goto out_free_dentry;
690 			}
691 
692 			/* Get the exec headers */
693 			loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
694 			break;
695 		}
696 		elf_ppnt++;
697 	}
698 
699 	elf_ppnt = elf_phdata;
700 	for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
701 		if (elf_ppnt->p_type == PT_GNU_STACK) {
702 			if (elf_ppnt->p_flags & PF_X)
703 				executable_stack = EXSTACK_ENABLE_X;
704 			else
705 				executable_stack = EXSTACK_DISABLE_X;
706 			break;
707 		}
708 
709 	/* Some simple consistency checks for the interpreter */
710 	if (elf_interpreter) {
711 		retval = -ELIBBAD;
712 		/* Not an ELF interpreter */
713 		if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
714 			goto out_free_dentry;
715 		/* Verify the interpreter has a valid arch */
716 		if (!elf_check_arch(&loc->interp_elf_ex))
717 			goto out_free_dentry;
718 	}
719 
720 	/* Flush all traces of the currently running executable */
721 	retval = flush_old_exec(bprm);
722 	if (retval)
723 		goto out_free_dentry;
724 
725 	/* OK, This is the point of no return */
726 	current->mm->def_flags = def_flags;
727 
728 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
729 	   may depend on the personality.  */
730 	SET_PERSONALITY(loc->elf_ex);
731 	if (elf_read_implies_exec(loc->elf_ex, executable_stack))
732 		current->personality |= READ_IMPLIES_EXEC;
733 
734 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
735 		current->flags |= PF_RANDOMIZE;
736 
737 	setup_new_exec(bprm);
738 
739 	/* Do this so that we can load the interpreter, if need be.  We will
740 	   change some of these later */
741 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
742 				 executable_stack);
743 	if (retval < 0) {
744 		send_sig(SIGKILL, current, 0);
745 		goto out_free_dentry;
746 	}
747 
748 	current->mm->start_stack = bprm->p;
749 
750 	/* Now we do a little grungy work by mmapping the ELF image into
751 	   the correct location in memory. */
752 	for(i = 0, elf_ppnt = elf_phdata;
753 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
754 		int elf_prot = 0, elf_flags;
755 		unsigned long k, vaddr;
756 
757 		if (elf_ppnt->p_type != PT_LOAD)
758 			continue;
759 
760 		if (unlikely (elf_brk > elf_bss)) {
761 			unsigned long nbyte;
762 
763 			/* There was a PT_LOAD segment with p_memsz > p_filesz
764 			   before this one. Map anonymous pages, if needed,
765 			   and clear the area.  */
766 			retval = set_brk(elf_bss + load_bias,
767 					 elf_brk + load_bias);
768 			if (retval) {
769 				send_sig(SIGKILL, current, 0);
770 				goto out_free_dentry;
771 			}
772 			nbyte = ELF_PAGEOFFSET(elf_bss);
773 			if (nbyte) {
774 				nbyte = ELF_MIN_ALIGN - nbyte;
775 				if (nbyte > elf_brk - elf_bss)
776 					nbyte = elf_brk - elf_bss;
777 				if (clear_user((void __user *)elf_bss +
778 							load_bias, nbyte)) {
779 					/*
780 					 * This bss-zeroing can fail if the ELF
781 					 * file specifies odd protections. So
782 					 * we don't check the return value
783 					 */
784 				}
785 			}
786 		}
787 
788 		if (elf_ppnt->p_flags & PF_R)
789 			elf_prot |= PROT_READ;
790 		if (elf_ppnt->p_flags & PF_W)
791 			elf_prot |= PROT_WRITE;
792 		if (elf_ppnt->p_flags & PF_X)
793 			elf_prot |= PROT_EXEC;
794 
795 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
796 
797 		vaddr = elf_ppnt->p_vaddr;
798 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
799 			elf_flags |= MAP_FIXED;
800 		} else if (loc->elf_ex.e_type == ET_DYN) {
801 			/* Try and get dynamic programs out of the way of the
802 			 * default mmap base, as well as whatever program they
803 			 * might try to exec.  This is because the brk will
804 			 * follow the loader, and is not movable.  */
805 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
806 			/* Memory randomization might have been switched off
807 			 * in runtime via sysctl or explicit setting of
808 			 * personality flags.
809 			 * If that is the case, retain the original non-zero
810 			 * load_bias value in order to establish proper
811 			 * non-randomized mappings.
812 			 */
813 			if (current->flags & PF_RANDOMIZE)
814 				load_bias = 0;
815 			else
816 				load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
817 #else
818 			load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
819 #endif
820 		}
821 
822 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
823 				elf_prot, elf_flags, 0);
824 		if (BAD_ADDR(error)) {
825 			send_sig(SIGKILL, current, 0);
826 			retval = IS_ERR((void *)error) ?
827 				PTR_ERR((void*)error) : -EINVAL;
828 			goto out_free_dentry;
829 		}
830 
831 		if (!load_addr_set) {
832 			load_addr_set = 1;
833 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
834 			if (loc->elf_ex.e_type == ET_DYN) {
835 				load_bias += error -
836 				             ELF_PAGESTART(load_bias + vaddr);
837 				load_addr += load_bias;
838 				reloc_func_desc = load_bias;
839 			}
840 		}
841 		k = elf_ppnt->p_vaddr;
842 		if (k < start_code)
843 			start_code = k;
844 		if (start_data < k)
845 			start_data = k;
846 
847 		/*
848 		 * Check to see if the section's size will overflow the
849 		 * allowed task size. Note that p_filesz must always be
850 		 * <= p_memsz so it is only necessary to check p_memsz.
851 		 */
852 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
853 		    elf_ppnt->p_memsz > TASK_SIZE ||
854 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
855 			/* set_brk can never work. Avoid overflows. */
856 			send_sig(SIGKILL, current, 0);
857 			retval = -EINVAL;
858 			goto out_free_dentry;
859 		}
860 
861 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
862 
863 		if (k > elf_bss)
864 			elf_bss = k;
865 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
866 			end_code = k;
867 		if (end_data < k)
868 			end_data = k;
869 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
870 		if (k > elf_brk)
871 			elf_brk = k;
872 	}
873 
874 	loc->elf_ex.e_entry += load_bias;
875 	elf_bss += load_bias;
876 	elf_brk += load_bias;
877 	start_code += load_bias;
878 	end_code += load_bias;
879 	start_data += load_bias;
880 	end_data += load_bias;
881 
882 	/* Calling set_brk effectively mmaps the pages that we need
883 	 * for the bss and break sections.  We must do this before
884 	 * mapping in the interpreter, to make sure it doesn't wind
885 	 * up getting placed where the bss needs to go.
886 	 */
887 	retval = set_brk(elf_bss, elf_brk);
888 	if (retval) {
889 		send_sig(SIGKILL, current, 0);
890 		goto out_free_dentry;
891 	}
892 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
893 		send_sig(SIGSEGV, current, 0);
894 		retval = -EFAULT; /* Nobody gets to see this, but.. */
895 		goto out_free_dentry;
896 	}
897 
898 	if (elf_interpreter) {
899 		unsigned long interp_map_addr = 0;
900 
901 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
902 					    interpreter,
903 					    &interp_map_addr,
904 					    load_bias);
905 		if (!IS_ERR((void *)elf_entry)) {
906 			/*
907 			 * load_elf_interp() returns relocation
908 			 * adjustment
909 			 */
910 			interp_load_addr = elf_entry;
911 			elf_entry += loc->interp_elf_ex.e_entry;
912 		}
913 		if (BAD_ADDR(elf_entry)) {
914 			force_sig(SIGSEGV, current);
915 			retval = IS_ERR((void *)elf_entry) ?
916 					(int)elf_entry : -EINVAL;
917 			goto out_free_dentry;
918 		}
919 		reloc_func_desc = interp_load_addr;
920 
921 		allow_write_access(interpreter);
922 		fput(interpreter);
923 		kfree(elf_interpreter);
924 	} else {
925 		elf_entry = loc->elf_ex.e_entry;
926 		if (BAD_ADDR(elf_entry)) {
927 			force_sig(SIGSEGV, current);
928 			retval = -EINVAL;
929 			goto out_free_dentry;
930 		}
931 	}
932 
933 	kfree(elf_phdata);
934 
935 	set_binfmt(&elf_format);
936 
937 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
938 	retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
939 	if (retval < 0) {
940 		send_sig(SIGKILL, current, 0);
941 		goto out;
942 	}
943 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
944 
945 	install_exec_creds(bprm);
946 	retval = create_elf_tables(bprm, &loc->elf_ex,
947 			  load_addr, interp_load_addr);
948 	if (retval < 0) {
949 		send_sig(SIGKILL, current, 0);
950 		goto out;
951 	}
952 	/* N.B. passed_fileno might not be initialized? */
953 	current->mm->end_code = end_code;
954 	current->mm->start_code = start_code;
955 	current->mm->start_data = start_data;
956 	current->mm->end_data = end_data;
957 	current->mm->start_stack = bprm->p;
958 
959 #ifdef arch_randomize_brk
960 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
961 		current->mm->brk = current->mm->start_brk =
962 			arch_randomize_brk(current->mm);
963 #ifdef CONFIG_COMPAT_BRK
964 		current->brk_randomized = 1;
965 #endif
966 	}
967 #endif
968 
969 	if (current->personality & MMAP_PAGE_ZERO) {
970 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
971 		   and some applications "depend" upon this behavior.
972 		   Since we do not have the power to recompile these, we
973 		   emulate the SVr4 behavior. Sigh. */
974 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
975 				MAP_FIXED | MAP_PRIVATE, 0);
976 	}
977 
978 #ifdef ELF_PLAT_INIT
979 	/*
980 	 * The ABI may specify that certain registers be set up in special
981 	 * ways (on i386 %edx is the address of a DT_FINI function, for
982 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
983 	 * that the e_entry field is the address of the function descriptor
984 	 * for the startup routine, rather than the address of the startup
985 	 * routine itself.  This macro performs whatever initialization to
986 	 * the regs structure is required as well as any relocations to the
987 	 * function descriptor entries when executing dynamically links apps.
988 	 */
989 	ELF_PLAT_INIT(regs, reloc_func_desc);
990 #endif
991 
992 	start_thread(regs, elf_entry, bprm->p);
993 	retval = 0;
994 out:
995 	kfree(loc);
996 out_ret:
997 	return retval;
998 
999 	/* error cleanup */
1000 out_free_dentry:
1001 	allow_write_access(interpreter);
1002 	if (interpreter)
1003 		fput(interpreter);
1004 out_free_interp:
1005 	kfree(elf_interpreter);
1006 out_free_ph:
1007 	kfree(elf_phdata);
1008 	goto out;
1009 }
1010 
1011 /* This is really simpleminded and specialized - we are loading an
1012    a.out library that is given an ELF header. */
1013 static int load_elf_library(struct file *file)
1014 {
1015 	struct elf_phdr *elf_phdata;
1016 	struct elf_phdr *eppnt;
1017 	unsigned long elf_bss, bss, len;
1018 	int retval, error, i, j;
1019 	struct elfhdr elf_ex;
1020 
1021 	error = -ENOEXEC;
1022 	retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1023 	if (retval != sizeof(elf_ex))
1024 		goto out;
1025 
1026 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1027 		goto out;
1028 
1029 	/* First of all, some simple consistency checks */
1030 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1031 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1032 		goto out;
1033 
1034 	/* Now read in all of the header information */
1035 
1036 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1037 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1038 
1039 	error = -ENOMEM;
1040 	elf_phdata = kmalloc(j, GFP_KERNEL);
1041 	if (!elf_phdata)
1042 		goto out;
1043 
1044 	eppnt = elf_phdata;
1045 	error = -ENOEXEC;
1046 	retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1047 	if (retval != j)
1048 		goto out_free_ph;
1049 
1050 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1051 		if ((eppnt + i)->p_type == PT_LOAD)
1052 			j++;
1053 	if (j != 1)
1054 		goto out_free_ph;
1055 
1056 	while (eppnt->p_type != PT_LOAD)
1057 		eppnt++;
1058 
1059 	/* Now use mmap to map the library into memory. */
1060 	error = vm_mmap(file,
1061 			ELF_PAGESTART(eppnt->p_vaddr),
1062 			(eppnt->p_filesz +
1063 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1064 			PROT_READ | PROT_WRITE | PROT_EXEC,
1065 			MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1066 			(eppnt->p_offset -
1067 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1068 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1069 		goto out_free_ph;
1070 
1071 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1072 	if (padzero(elf_bss)) {
1073 		error = -EFAULT;
1074 		goto out_free_ph;
1075 	}
1076 
1077 	len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1078 			    ELF_MIN_ALIGN - 1);
1079 	bss = eppnt->p_memsz + eppnt->p_vaddr;
1080 	if (bss > len)
1081 		vm_brk(len, bss - len);
1082 	error = 0;
1083 
1084 out_free_ph:
1085 	kfree(elf_phdata);
1086 out:
1087 	return error;
1088 }
1089 
1090 #ifdef CONFIG_ELF_CORE
1091 /*
1092  * ELF core dumper
1093  *
1094  * Modelled on fs/exec.c:aout_core_dump()
1095  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1096  */
1097 
1098 /*
1099  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1100  * that are useful for post-mortem analysis are included in every core dump.
1101  * In that way we ensure that the core dump is fully interpretable later
1102  * without matching up the same kernel and hardware config to see what PC values
1103  * meant. These special mappings include - vDSO, vsyscall, and other
1104  * architecture specific mappings
1105  */
1106 static bool always_dump_vma(struct vm_area_struct *vma)
1107 {
1108 	/* Any vsyscall mappings? */
1109 	if (vma == get_gate_vma(vma->vm_mm))
1110 		return true;
1111 	/*
1112 	 * arch_vma_name() returns non-NULL for special architecture mappings,
1113 	 * such as vDSO sections.
1114 	 */
1115 	if (arch_vma_name(vma))
1116 		return true;
1117 
1118 	return false;
1119 }
1120 
1121 /*
1122  * Decide what to dump of a segment, part, all or none.
1123  */
1124 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1125 				   unsigned long mm_flags)
1126 {
1127 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1128 
1129 	/* always dump the vdso and vsyscall sections */
1130 	if (always_dump_vma(vma))
1131 		goto whole;
1132 
1133 	if (vma->vm_flags & VM_DONTDUMP)
1134 		return 0;
1135 
1136 	/* Hugetlb memory check */
1137 	if (vma->vm_flags & VM_HUGETLB) {
1138 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1139 			goto whole;
1140 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1141 			goto whole;
1142 		return 0;
1143 	}
1144 
1145 	/* Do not dump I/O mapped devices or special mappings */
1146 	if (vma->vm_flags & VM_IO)
1147 		return 0;
1148 
1149 	/* By default, dump shared memory if mapped from an anonymous file. */
1150 	if (vma->vm_flags & VM_SHARED) {
1151 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1152 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1153 			goto whole;
1154 		return 0;
1155 	}
1156 
1157 	/* Dump segments that have been written to.  */
1158 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1159 		goto whole;
1160 	if (vma->vm_file == NULL)
1161 		return 0;
1162 
1163 	if (FILTER(MAPPED_PRIVATE))
1164 		goto whole;
1165 
1166 	/*
1167 	 * If this looks like the beginning of a DSO or executable mapping,
1168 	 * check for an ELF header.  If we find one, dump the first page to
1169 	 * aid in determining what was mapped here.
1170 	 */
1171 	if (FILTER(ELF_HEADERS) &&
1172 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1173 		u32 __user *header = (u32 __user *) vma->vm_start;
1174 		u32 word;
1175 		mm_segment_t fs = get_fs();
1176 		/*
1177 		 * Doing it this way gets the constant folded by GCC.
1178 		 */
1179 		union {
1180 			u32 cmp;
1181 			char elfmag[SELFMAG];
1182 		} magic;
1183 		BUILD_BUG_ON(SELFMAG != sizeof word);
1184 		magic.elfmag[EI_MAG0] = ELFMAG0;
1185 		magic.elfmag[EI_MAG1] = ELFMAG1;
1186 		magic.elfmag[EI_MAG2] = ELFMAG2;
1187 		magic.elfmag[EI_MAG3] = ELFMAG3;
1188 		/*
1189 		 * Switch to the user "segment" for get_user(),
1190 		 * then put back what elf_core_dump() had in place.
1191 		 */
1192 		set_fs(USER_DS);
1193 		if (unlikely(get_user(word, header)))
1194 			word = 0;
1195 		set_fs(fs);
1196 		if (word == magic.cmp)
1197 			return PAGE_SIZE;
1198 	}
1199 
1200 #undef	FILTER
1201 
1202 	return 0;
1203 
1204 whole:
1205 	return vma->vm_end - vma->vm_start;
1206 }
1207 
1208 /* An ELF note in memory */
1209 struct memelfnote
1210 {
1211 	const char *name;
1212 	int type;
1213 	unsigned int datasz;
1214 	void *data;
1215 };
1216 
1217 static int notesize(struct memelfnote *en)
1218 {
1219 	int sz;
1220 
1221 	sz = sizeof(struct elf_note);
1222 	sz += roundup(strlen(en->name) + 1, 4);
1223 	sz += roundup(en->datasz, 4);
1224 
1225 	return sz;
1226 }
1227 
1228 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1229 {
1230 	struct elf_note en;
1231 	en.n_namesz = strlen(men->name) + 1;
1232 	en.n_descsz = men->datasz;
1233 	en.n_type = men->type;
1234 
1235 	return dump_emit(cprm, &en, sizeof(en)) &&
1236 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1237 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1238 }
1239 
1240 static void fill_elf_header(struct elfhdr *elf, int segs,
1241 			    u16 machine, u32 flags)
1242 {
1243 	memset(elf, 0, sizeof(*elf));
1244 
1245 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1246 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1247 	elf->e_ident[EI_DATA] = ELF_DATA;
1248 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1249 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1250 
1251 	elf->e_type = ET_CORE;
1252 	elf->e_machine = machine;
1253 	elf->e_version = EV_CURRENT;
1254 	elf->e_phoff = sizeof(struct elfhdr);
1255 	elf->e_flags = flags;
1256 	elf->e_ehsize = sizeof(struct elfhdr);
1257 	elf->e_phentsize = sizeof(struct elf_phdr);
1258 	elf->e_phnum = segs;
1259 
1260 	return;
1261 }
1262 
1263 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1264 {
1265 	phdr->p_type = PT_NOTE;
1266 	phdr->p_offset = offset;
1267 	phdr->p_vaddr = 0;
1268 	phdr->p_paddr = 0;
1269 	phdr->p_filesz = sz;
1270 	phdr->p_memsz = 0;
1271 	phdr->p_flags = 0;
1272 	phdr->p_align = 0;
1273 	return;
1274 }
1275 
1276 static void fill_note(struct memelfnote *note, const char *name, int type,
1277 		unsigned int sz, void *data)
1278 {
1279 	note->name = name;
1280 	note->type = type;
1281 	note->datasz = sz;
1282 	note->data = data;
1283 	return;
1284 }
1285 
1286 /*
1287  * fill up all the fields in prstatus from the given task struct, except
1288  * registers which need to be filled up separately.
1289  */
1290 static void fill_prstatus(struct elf_prstatus *prstatus,
1291 		struct task_struct *p, long signr)
1292 {
1293 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1294 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1295 	prstatus->pr_sighold = p->blocked.sig[0];
1296 	rcu_read_lock();
1297 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1298 	rcu_read_unlock();
1299 	prstatus->pr_pid = task_pid_vnr(p);
1300 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1301 	prstatus->pr_sid = task_session_vnr(p);
1302 	if (thread_group_leader(p)) {
1303 		struct task_cputime cputime;
1304 
1305 		/*
1306 		 * This is the record for the group leader.  It shows the
1307 		 * group-wide total, not its individual thread total.
1308 		 */
1309 		thread_group_cputime(p, &cputime);
1310 		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1311 		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1312 	} else {
1313 		cputime_t utime, stime;
1314 
1315 		task_cputime(p, &utime, &stime);
1316 		cputime_to_timeval(utime, &prstatus->pr_utime);
1317 		cputime_to_timeval(stime, &prstatus->pr_stime);
1318 	}
1319 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1320 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1321 }
1322 
1323 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1324 		       struct mm_struct *mm)
1325 {
1326 	const struct cred *cred;
1327 	unsigned int i, len;
1328 
1329 	/* first copy the parameters from user space */
1330 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1331 
1332 	len = mm->arg_end - mm->arg_start;
1333 	if (len >= ELF_PRARGSZ)
1334 		len = ELF_PRARGSZ-1;
1335 	if (copy_from_user(&psinfo->pr_psargs,
1336 		           (const char __user *)mm->arg_start, len))
1337 		return -EFAULT;
1338 	for(i = 0; i < len; i++)
1339 		if (psinfo->pr_psargs[i] == 0)
1340 			psinfo->pr_psargs[i] = ' ';
1341 	psinfo->pr_psargs[len] = 0;
1342 
1343 	rcu_read_lock();
1344 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1345 	rcu_read_unlock();
1346 	psinfo->pr_pid = task_pid_vnr(p);
1347 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1348 	psinfo->pr_sid = task_session_vnr(p);
1349 
1350 	i = p->state ? ffz(~p->state) + 1 : 0;
1351 	psinfo->pr_state = i;
1352 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1353 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1354 	psinfo->pr_nice = task_nice(p);
1355 	psinfo->pr_flag = p->flags;
1356 	rcu_read_lock();
1357 	cred = __task_cred(p);
1358 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1359 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1360 	rcu_read_unlock();
1361 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1362 
1363 	return 0;
1364 }
1365 
1366 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1367 {
1368 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1369 	int i = 0;
1370 	do
1371 		i += 2;
1372 	while (auxv[i - 2] != AT_NULL);
1373 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1374 }
1375 
1376 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1377 		const siginfo_t *siginfo)
1378 {
1379 	mm_segment_t old_fs = get_fs();
1380 	set_fs(KERNEL_DS);
1381 	copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1382 	set_fs(old_fs);
1383 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1384 }
1385 
1386 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1387 /*
1388  * Format of NT_FILE note:
1389  *
1390  * long count     -- how many files are mapped
1391  * long page_size -- units for file_ofs
1392  * array of [COUNT] elements of
1393  *   long start
1394  *   long end
1395  *   long file_ofs
1396  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1397  */
1398 static int fill_files_note(struct memelfnote *note)
1399 {
1400 	struct vm_area_struct *vma;
1401 	unsigned count, size, names_ofs, remaining, n;
1402 	user_long_t *data;
1403 	user_long_t *start_end_ofs;
1404 	char *name_base, *name_curpos;
1405 
1406 	/* *Estimated* file count and total data size needed */
1407 	count = current->mm->map_count;
1408 	size = count * 64;
1409 
1410 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1411  alloc:
1412 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1413 		return -EINVAL;
1414 	size = round_up(size, PAGE_SIZE);
1415 	data = vmalloc(size);
1416 	if (!data)
1417 		return -ENOMEM;
1418 
1419 	start_end_ofs = data + 2;
1420 	name_base = name_curpos = ((char *)data) + names_ofs;
1421 	remaining = size - names_ofs;
1422 	count = 0;
1423 	for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1424 		struct file *file;
1425 		const char *filename;
1426 
1427 		file = vma->vm_file;
1428 		if (!file)
1429 			continue;
1430 		filename = d_path(&file->f_path, name_curpos, remaining);
1431 		if (IS_ERR(filename)) {
1432 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1433 				vfree(data);
1434 				size = size * 5 / 4;
1435 				goto alloc;
1436 			}
1437 			continue;
1438 		}
1439 
1440 		/* d_path() fills at the end, move name down */
1441 		/* n = strlen(filename) + 1: */
1442 		n = (name_curpos + remaining) - filename;
1443 		remaining = filename - name_curpos;
1444 		memmove(name_curpos, filename, n);
1445 		name_curpos += n;
1446 
1447 		*start_end_ofs++ = vma->vm_start;
1448 		*start_end_ofs++ = vma->vm_end;
1449 		*start_end_ofs++ = vma->vm_pgoff;
1450 		count++;
1451 	}
1452 
1453 	/* Now we know exact count of files, can store it */
1454 	data[0] = count;
1455 	data[1] = PAGE_SIZE;
1456 	/*
1457 	 * Count usually is less than current->mm->map_count,
1458 	 * we need to move filenames down.
1459 	 */
1460 	n = current->mm->map_count - count;
1461 	if (n != 0) {
1462 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1463 		memmove(name_base - shift_bytes, name_base,
1464 			name_curpos - name_base);
1465 		name_curpos -= shift_bytes;
1466 	}
1467 
1468 	size = name_curpos - (char *)data;
1469 	fill_note(note, "CORE", NT_FILE, size, data);
1470 	return 0;
1471 }
1472 
1473 #ifdef CORE_DUMP_USE_REGSET
1474 #include <linux/regset.h>
1475 
1476 struct elf_thread_core_info {
1477 	struct elf_thread_core_info *next;
1478 	struct task_struct *task;
1479 	struct elf_prstatus prstatus;
1480 	struct memelfnote notes[0];
1481 };
1482 
1483 struct elf_note_info {
1484 	struct elf_thread_core_info *thread;
1485 	struct memelfnote psinfo;
1486 	struct memelfnote signote;
1487 	struct memelfnote auxv;
1488 	struct memelfnote files;
1489 	user_siginfo_t csigdata;
1490 	size_t size;
1491 	int thread_notes;
1492 };
1493 
1494 /*
1495  * When a regset has a writeback hook, we call it on each thread before
1496  * dumping user memory.  On register window machines, this makes sure the
1497  * user memory backing the register data is up to date before we read it.
1498  */
1499 static void do_thread_regset_writeback(struct task_struct *task,
1500 				       const struct user_regset *regset)
1501 {
1502 	if (regset->writeback)
1503 		regset->writeback(task, regset, 1);
1504 }
1505 
1506 #ifndef PR_REG_SIZE
1507 #define PR_REG_SIZE(S) sizeof(S)
1508 #endif
1509 
1510 #ifndef PRSTATUS_SIZE
1511 #define PRSTATUS_SIZE(S) sizeof(S)
1512 #endif
1513 
1514 #ifndef PR_REG_PTR
1515 #define PR_REG_PTR(S) (&((S)->pr_reg))
1516 #endif
1517 
1518 #ifndef SET_PR_FPVALID
1519 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1520 #endif
1521 
1522 static int fill_thread_core_info(struct elf_thread_core_info *t,
1523 				 const struct user_regset_view *view,
1524 				 long signr, size_t *total)
1525 {
1526 	unsigned int i;
1527 
1528 	/*
1529 	 * NT_PRSTATUS is the one special case, because the regset data
1530 	 * goes into the pr_reg field inside the note contents, rather
1531 	 * than being the whole note contents.  We fill the reset in here.
1532 	 * We assume that regset 0 is NT_PRSTATUS.
1533 	 */
1534 	fill_prstatus(&t->prstatus, t->task, signr);
1535 	(void) view->regsets[0].get(t->task, &view->regsets[0],
1536 				    0, PR_REG_SIZE(t->prstatus.pr_reg),
1537 				    PR_REG_PTR(&t->prstatus), NULL);
1538 
1539 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1540 		  PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1541 	*total += notesize(&t->notes[0]);
1542 
1543 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1544 
1545 	/*
1546 	 * Each other regset might generate a note too.  For each regset
1547 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1548 	 * all zero and we'll know to skip writing it later.
1549 	 */
1550 	for (i = 1; i < view->n; ++i) {
1551 		const struct user_regset *regset = &view->regsets[i];
1552 		do_thread_regset_writeback(t->task, regset);
1553 		if (regset->core_note_type && regset->get &&
1554 		    (!regset->active || regset->active(t->task, regset))) {
1555 			int ret;
1556 			size_t size = regset->n * regset->size;
1557 			void *data = kmalloc(size, GFP_KERNEL);
1558 			if (unlikely(!data))
1559 				return 0;
1560 			ret = regset->get(t->task, regset,
1561 					  0, size, data, NULL);
1562 			if (unlikely(ret))
1563 				kfree(data);
1564 			else {
1565 				if (regset->core_note_type != NT_PRFPREG)
1566 					fill_note(&t->notes[i], "LINUX",
1567 						  regset->core_note_type,
1568 						  size, data);
1569 				else {
1570 					SET_PR_FPVALID(&t->prstatus, 1);
1571 					fill_note(&t->notes[i], "CORE",
1572 						  NT_PRFPREG, size, data);
1573 				}
1574 				*total += notesize(&t->notes[i]);
1575 			}
1576 		}
1577 	}
1578 
1579 	return 1;
1580 }
1581 
1582 static int fill_note_info(struct elfhdr *elf, int phdrs,
1583 			  struct elf_note_info *info,
1584 			  const siginfo_t *siginfo, struct pt_regs *regs)
1585 {
1586 	struct task_struct *dump_task = current;
1587 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1588 	struct elf_thread_core_info *t;
1589 	struct elf_prpsinfo *psinfo;
1590 	struct core_thread *ct;
1591 	unsigned int i;
1592 
1593 	info->size = 0;
1594 	info->thread = NULL;
1595 
1596 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1597 	if (psinfo == NULL) {
1598 		info->psinfo.data = NULL; /* So we don't free this wrongly */
1599 		return 0;
1600 	}
1601 
1602 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1603 
1604 	/*
1605 	 * Figure out how many notes we're going to need for each thread.
1606 	 */
1607 	info->thread_notes = 0;
1608 	for (i = 0; i < view->n; ++i)
1609 		if (view->regsets[i].core_note_type != 0)
1610 			++info->thread_notes;
1611 
1612 	/*
1613 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1614 	 * since it is our one special case.
1615 	 */
1616 	if (unlikely(info->thread_notes == 0) ||
1617 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1618 		WARN_ON(1);
1619 		return 0;
1620 	}
1621 
1622 	/*
1623 	 * Initialize the ELF file header.
1624 	 */
1625 	fill_elf_header(elf, phdrs,
1626 			view->e_machine, view->e_flags);
1627 
1628 	/*
1629 	 * Allocate a structure for each thread.
1630 	 */
1631 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1632 		t = kzalloc(offsetof(struct elf_thread_core_info,
1633 				     notes[info->thread_notes]),
1634 			    GFP_KERNEL);
1635 		if (unlikely(!t))
1636 			return 0;
1637 
1638 		t->task = ct->task;
1639 		if (ct->task == dump_task || !info->thread) {
1640 			t->next = info->thread;
1641 			info->thread = t;
1642 		} else {
1643 			/*
1644 			 * Make sure to keep the original task at
1645 			 * the head of the list.
1646 			 */
1647 			t->next = info->thread->next;
1648 			info->thread->next = t;
1649 		}
1650 	}
1651 
1652 	/*
1653 	 * Now fill in each thread's information.
1654 	 */
1655 	for (t = info->thread; t != NULL; t = t->next)
1656 		if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1657 			return 0;
1658 
1659 	/*
1660 	 * Fill in the two process-wide notes.
1661 	 */
1662 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1663 	info->size += notesize(&info->psinfo);
1664 
1665 	fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1666 	info->size += notesize(&info->signote);
1667 
1668 	fill_auxv_note(&info->auxv, current->mm);
1669 	info->size += notesize(&info->auxv);
1670 
1671 	if (fill_files_note(&info->files) == 0)
1672 		info->size += notesize(&info->files);
1673 
1674 	return 1;
1675 }
1676 
1677 static size_t get_note_info_size(struct elf_note_info *info)
1678 {
1679 	return info->size;
1680 }
1681 
1682 /*
1683  * Write all the notes for each thread.  When writing the first thread, the
1684  * process-wide notes are interleaved after the first thread-specific note.
1685  */
1686 static int write_note_info(struct elf_note_info *info,
1687 			   struct coredump_params *cprm)
1688 {
1689 	bool first = 1;
1690 	struct elf_thread_core_info *t = info->thread;
1691 
1692 	do {
1693 		int i;
1694 
1695 		if (!writenote(&t->notes[0], cprm))
1696 			return 0;
1697 
1698 		if (first && !writenote(&info->psinfo, cprm))
1699 			return 0;
1700 		if (first && !writenote(&info->signote, cprm))
1701 			return 0;
1702 		if (first && !writenote(&info->auxv, cprm))
1703 			return 0;
1704 		if (first && info->files.data &&
1705 				!writenote(&info->files, cprm))
1706 			return 0;
1707 
1708 		for (i = 1; i < info->thread_notes; ++i)
1709 			if (t->notes[i].data &&
1710 			    !writenote(&t->notes[i], cprm))
1711 				return 0;
1712 
1713 		first = 0;
1714 		t = t->next;
1715 	} while (t);
1716 
1717 	return 1;
1718 }
1719 
1720 static void free_note_info(struct elf_note_info *info)
1721 {
1722 	struct elf_thread_core_info *threads = info->thread;
1723 	while (threads) {
1724 		unsigned int i;
1725 		struct elf_thread_core_info *t = threads;
1726 		threads = t->next;
1727 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1728 		for (i = 1; i < info->thread_notes; ++i)
1729 			kfree(t->notes[i].data);
1730 		kfree(t);
1731 	}
1732 	kfree(info->psinfo.data);
1733 	vfree(info->files.data);
1734 }
1735 
1736 #else
1737 
1738 /* Here is the structure in which status of each thread is captured. */
1739 struct elf_thread_status
1740 {
1741 	struct list_head list;
1742 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1743 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1744 	struct task_struct *thread;
1745 #ifdef ELF_CORE_COPY_XFPREGS
1746 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1747 #endif
1748 	struct memelfnote notes[3];
1749 	int num_notes;
1750 };
1751 
1752 /*
1753  * In order to add the specific thread information for the elf file format,
1754  * we need to keep a linked list of every threads pr_status and then create
1755  * a single section for them in the final core file.
1756  */
1757 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1758 {
1759 	int sz = 0;
1760 	struct task_struct *p = t->thread;
1761 	t->num_notes = 0;
1762 
1763 	fill_prstatus(&t->prstatus, p, signr);
1764 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1765 
1766 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1767 		  &(t->prstatus));
1768 	t->num_notes++;
1769 	sz += notesize(&t->notes[0]);
1770 
1771 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1772 								&t->fpu))) {
1773 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1774 			  &(t->fpu));
1775 		t->num_notes++;
1776 		sz += notesize(&t->notes[1]);
1777 	}
1778 
1779 #ifdef ELF_CORE_COPY_XFPREGS
1780 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1781 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1782 			  sizeof(t->xfpu), &t->xfpu);
1783 		t->num_notes++;
1784 		sz += notesize(&t->notes[2]);
1785 	}
1786 #endif
1787 	return sz;
1788 }
1789 
1790 struct elf_note_info {
1791 	struct memelfnote *notes;
1792 	struct memelfnote *notes_files;
1793 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1794 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1795 	struct list_head thread_list;
1796 	elf_fpregset_t *fpu;
1797 #ifdef ELF_CORE_COPY_XFPREGS
1798 	elf_fpxregset_t *xfpu;
1799 #endif
1800 	user_siginfo_t csigdata;
1801 	int thread_status_size;
1802 	int numnote;
1803 };
1804 
1805 static int elf_note_info_init(struct elf_note_info *info)
1806 {
1807 	memset(info, 0, sizeof(*info));
1808 	INIT_LIST_HEAD(&info->thread_list);
1809 
1810 	/* Allocate space for ELF notes */
1811 	info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1812 	if (!info->notes)
1813 		return 0;
1814 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1815 	if (!info->psinfo)
1816 		return 0;
1817 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1818 	if (!info->prstatus)
1819 		return 0;
1820 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1821 	if (!info->fpu)
1822 		return 0;
1823 #ifdef ELF_CORE_COPY_XFPREGS
1824 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1825 	if (!info->xfpu)
1826 		return 0;
1827 #endif
1828 	return 1;
1829 }
1830 
1831 static int fill_note_info(struct elfhdr *elf, int phdrs,
1832 			  struct elf_note_info *info,
1833 			  const siginfo_t *siginfo, struct pt_regs *regs)
1834 {
1835 	struct list_head *t;
1836 	struct core_thread *ct;
1837 	struct elf_thread_status *ets;
1838 
1839 	if (!elf_note_info_init(info))
1840 		return 0;
1841 
1842 	for (ct = current->mm->core_state->dumper.next;
1843 					ct; ct = ct->next) {
1844 		ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1845 		if (!ets)
1846 			return 0;
1847 
1848 		ets->thread = ct->task;
1849 		list_add(&ets->list, &info->thread_list);
1850 	}
1851 
1852 	list_for_each(t, &info->thread_list) {
1853 		int sz;
1854 
1855 		ets = list_entry(t, struct elf_thread_status, list);
1856 		sz = elf_dump_thread_status(siginfo->si_signo, ets);
1857 		info->thread_status_size += sz;
1858 	}
1859 	/* now collect the dump for the current */
1860 	memset(info->prstatus, 0, sizeof(*info->prstatus));
1861 	fill_prstatus(info->prstatus, current, siginfo->si_signo);
1862 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1863 
1864 	/* Set up header */
1865 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1866 
1867 	/*
1868 	 * Set up the notes in similar form to SVR4 core dumps made
1869 	 * with info from their /proc.
1870 	 */
1871 
1872 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1873 		  sizeof(*info->prstatus), info->prstatus);
1874 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
1875 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1876 		  sizeof(*info->psinfo), info->psinfo);
1877 
1878 	fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1879 	fill_auxv_note(info->notes + 3, current->mm);
1880 	info->numnote = 4;
1881 
1882 	if (fill_files_note(info->notes + info->numnote) == 0) {
1883 		info->notes_files = info->notes + info->numnote;
1884 		info->numnote++;
1885 	}
1886 
1887 	/* Try to dump the FPU. */
1888 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1889 							       info->fpu);
1890 	if (info->prstatus->pr_fpvalid)
1891 		fill_note(info->notes + info->numnote++,
1892 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1893 #ifdef ELF_CORE_COPY_XFPREGS
1894 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
1895 		fill_note(info->notes + info->numnote++,
1896 			  "LINUX", ELF_CORE_XFPREG_TYPE,
1897 			  sizeof(*info->xfpu), info->xfpu);
1898 #endif
1899 
1900 	return 1;
1901 }
1902 
1903 static size_t get_note_info_size(struct elf_note_info *info)
1904 {
1905 	int sz = 0;
1906 	int i;
1907 
1908 	for (i = 0; i < info->numnote; i++)
1909 		sz += notesize(info->notes + i);
1910 
1911 	sz += info->thread_status_size;
1912 
1913 	return sz;
1914 }
1915 
1916 static int write_note_info(struct elf_note_info *info,
1917 			   struct coredump_params *cprm)
1918 {
1919 	int i;
1920 	struct list_head *t;
1921 
1922 	for (i = 0; i < info->numnote; i++)
1923 		if (!writenote(info->notes + i, cprm))
1924 			return 0;
1925 
1926 	/* write out the thread status notes section */
1927 	list_for_each(t, &info->thread_list) {
1928 		struct elf_thread_status *tmp =
1929 				list_entry(t, struct elf_thread_status, list);
1930 
1931 		for (i = 0; i < tmp->num_notes; i++)
1932 			if (!writenote(&tmp->notes[i], cprm))
1933 				return 0;
1934 	}
1935 
1936 	return 1;
1937 }
1938 
1939 static void free_note_info(struct elf_note_info *info)
1940 {
1941 	while (!list_empty(&info->thread_list)) {
1942 		struct list_head *tmp = info->thread_list.next;
1943 		list_del(tmp);
1944 		kfree(list_entry(tmp, struct elf_thread_status, list));
1945 	}
1946 
1947 	/* Free data possibly allocated by fill_files_note(): */
1948 	if (info->notes_files)
1949 		vfree(info->notes_files->data);
1950 
1951 	kfree(info->prstatus);
1952 	kfree(info->psinfo);
1953 	kfree(info->notes);
1954 	kfree(info->fpu);
1955 #ifdef ELF_CORE_COPY_XFPREGS
1956 	kfree(info->xfpu);
1957 #endif
1958 }
1959 
1960 #endif
1961 
1962 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1963 					struct vm_area_struct *gate_vma)
1964 {
1965 	struct vm_area_struct *ret = tsk->mm->mmap;
1966 
1967 	if (ret)
1968 		return ret;
1969 	return gate_vma;
1970 }
1971 /*
1972  * Helper function for iterating across a vma list.  It ensures that the caller
1973  * will visit `gate_vma' prior to terminating the search.
1974  */
1975 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1976 					struct vm_area_struct *gate_vma)
1977 {
1978 	struct vm_area_struct *ret;
1979 
1980 	ret = this_vma->vm_next;
1981 	if (ret)
1982 		return ret;
1983 	if (this_vma == gate_vma)
1984 		return NULL;
1985 	return gate_vma;
1986 }
1987 
1988 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1989 			     elf_addr_t e_shoff, int segs)
1990 {
1991 	elf->e_shoff = e_shoff;
1992 	elf->e_shentsize = sizeof(*shdr4extnum);
1993 	elf->e_shnum = 1;
1994 	elf->e_shstrndx = SHN_UNDEF;
1995 
1996 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1997 
1998 	shdr4extnum->sh_type = SHT_NULL;
1999 	shdr4extnum->sh_size = elf->e_shnum;
2000 	shdr4extnum->sh_link = elf->e_shstrndx;
2001 	shdr4extnum->sh_info = segs;
2002 }
2003 
2004 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
2005 				     unsigned long mm_flags)
2006 {
2007 	struct vm_area_struct *vma;
2008 	size_t size = 0;
2009 
2010 	for (vma = first_vma(current, gate_vma); vma != NULL;
2011 	     vma = next_vma(vma, gate_vma))
2012 		size += vma_dump_size(vma, mm_flags);
2013 	return size;
2014 }
2015 
2016 /*
2017  * Actual dumper
2018  *
2019  * This is a two-pass process; first we find the offsets of the bits,
2020  * and then they are actually written out.  If we run out of core limit
2021  * we just truncate.
2022  */
2023 static int elf_core_dump(struct coredump_params *cprm)
2024 {
2025 	int has_dumped = 0;
2026 	mm_segment_t fs;
2027 	int segs;
2028 	struct vm_area_struct *vma, *gate_vma;
2029 	struct elfhdr *elf = NULL;
2030 	loff_t offset = 0, dataoff;
2031 	struct elf_note_info info = { };
2032 	struct elf_phdr *phdr4note = NULL;
2033 	struct elf_shdr *shdr4extnum = NULL;
2034 	Elf_Half e_phnum;
2035 	elf_addr_t e_shoff;
2036 
2037 	/*
2038 	 * We no longer stop all VM operations.
2039 	 *
2040 	 * This is because those proceses that could possibly change map_count
2041 	 * or the mmap / vma pages are now blocked in do_exit on current
2042 	 * finishing this core dump.
2043 	 *
2044 	 * Only ptrace can touch these memory addresses, but it doesn't change
2045 	 * the map_count or the pages allocated. So no possibility of crashing
2046 	 * exists while dumping the mm->vm_next areas to the core file.
2047 	 */
2048 
2049 	/* alloc memory for large data structures: too large to be on stack */
2050 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2051 	if (!elf)
2052 		goto out;
2053 	/*
2054 	 * The number of segs are recored into ELF header as 16bit value.
2055 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2056 	 */
2057 	segs = current->mm->map_count;
2058 	segs += elf_core_extra_phdrs();
2059 
2060 	gate_vma = get_gate_vma(current->mm);
2061 	if (gate_vma != NULL)
2062 		segs++;
2063 
2064 	/* for notes section */
2065 	segs++;
2066 
2067 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2068 	 * this, kernel supports extended numbering. Have a look at
2069 	 * include/linux/elf.h for further information. */
2070 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2071 
2072 	/*
2073 	 * Collect all the non-memory information about the process for the
2074 	 * notes.  This also sets up the file header.
2075 	 */
2076 	if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2077 		goto cleanup;
2078 
2079 	has_dumped = 1;
2080 
2081 	fs = get_fs();
2082 	set_fs(KERNEL_DS);
2083 
2084 	offset += sizeof(*elf);				/* Elf header */
2085 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2086 
2087 	/* Write notes phdr entry */
2088 	{
2089 		size_t sz = get_note_info_size(&info);
2090 
2091 		sz += elf_coredump_extra_notes_size();
2092 
2093 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2094 		if (!phdr4note)
2095 			goto end_coredump;
2096 
2097 		fill_elf_note_phdr(phdr4note, sz, offset);
2098 		offset += sz;
2099 	}
2100 
2101 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2102 
2103 	offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2104 	offset += elf_core_extra_data_size();
2105 	e_shoff = offset;
2106 
2107 	if (e_phnum == PN_XNUM) {
2108 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2109 		if (!shdr4extnum)
2110 			goto end_coredump;
2111 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2112 	}
2113 
2114 	offset = dataoff;
2115 
2116 	if (!dump_emit(cprm, elf, sizeof(*elf)))
2117 		goto end_coredump;
2118 
2119 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2120 		goto end_coredump;
2121 
2122 	/* Write program headers for segments dump */
2123 	for (vma = first_vma(current, gate_vma); vma != NULL;
2124 			vma = next_vma(vma, gate_vma)) {
2125 		struct elf_phdr phdr;
2126 
2127 		phdr.p_type = PT_LOAD;
2128 		phdr.p_offset = offset;
2129 		phdr.p_vaddr = vma->vm_start;
2130 		phdr.p_paddr = 0;
2131 		phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2132 		phdr.p_memsz = vma->vm_end - vma->vm_start;
2133 		offset += phdr.p_filesz;
2134 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2135 		if (vma->vm_flags & VM_WRITE)
2136 			phdr.p_flags |= PF_W;
2137 		if (vma->vm_flags & VM_EXEC)
2138 			phdr.p_flags |= PF_X;
2139 		phdr.p_align = ELF_EXEC_PAGESIZE;
2140 
2141 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2142 			goto end_coredump;
2143 	}
2144 
2145 	if (!elf_core_write_extra_phdrs(cprm, offset))
2146 		goto end_coredump;
2147 
2148  	/* write out the notes section */
2149 	if (!write_note_info(&info, cprm))
2150 		goto end_coredump;
2151 
2152 	if (elf_coredump_extra_notes_write(cprm))
2153 		goto end_coredump;
2154 
2155 	/* Align to page */
2156 	if (!dump_skip(cprm, dataoff - cprm->written))
2157 		goto end_coredump;
2158 
2159 	for (vma = first_vma(current, gate_vma); vma != NULL;
2160 			vma = next_vma(vma, gate_vma)) {
2161 		unsigned long addr;
2162 		unsigned long end;
2163 
2164 		end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2165 
2166 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2167 			struct page *page;
2168 			int stop;
2169 
2170 			page = get_dump_page(addr);
2171 			if (page) {
2172 				void *kaddr = kmap(page);
2173 				stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2174 				kunmap(page);
2175 				page_cache_release(page);
2176 			} else
2177 				stop = !dump_skip(cprm, PAGE_SIZE);
2178 			if (stop)
2179 				goto end_coredump;
2180 		}
2181 	}
2182 
2183 	if (!elf_core_write_extra_data(cprm))
2184 		goto end_coredump;
2185 
2186 	if (e_phnum == PN_XNUM) {
2187 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2188 			goto end_coredump;
2189 	}
2190 
2191 end_coredump:
2192 	set_fs(fs);
2193 
2194 cleanup:
2195 	free_note_info(&info);
2196 	kfree(shdr4extnum);
2197 	kfree(phdr4note);
2198 	kfree(elf);
2199 out:
2200 	return has_dumped;
2201 }
2202 
2203 #endif		/* CONFIG_ELF_CORE */
2204 
2205 static int __init init_elf_binfmt(void)
2206 {
2207 	register_binfmt(&elf_format);
2208 	return 0;
2209 }
2210 
2211 static void __exit exit_elf_binfmt(void)
2212 {
2213 	/* Remove the COFF and ELF loaders. */
2214 	unregister_binfmt(&elf_format);
2215 }
2216 
2217 core_initcall(init_elf_binfmt);
2218 module_exit(exit_elf_binfmt);
2219 MODULE_LICENSE("GPL");
2220