xref: /openbmc/linux/fs/binfmt_elf.c (revision 8fdff1dc)
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 <asm/uaccess.h>
37 #include <asm/param.h>
38 #include <asm/page.h>
39 
40 #ifndef user_long_t
41 #define user_long_t long
42 #endif
43 #ifndef user_siginfo_t
44 #define user_siginfo_t siginfo_t
45 #endif
46 
47 static int load_elf_binary(struct linux_binprm *bprm);
48 static int load_elf_library(struct file *);
49 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
50 				int, int, unsigned long);
51 
52 /*
53  * If we don't support core dumping, then supply a NULL so we
54  * don't even try.
55  */
56 #ifdef CONFIG_ELF_CORE
57 static int elf_core_dump(struct coredump_params *cprm);
58 #else
59 #define elf_core_dump	NULL
60 #endif
61 
62 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
63 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
64 #else
65 #define ELF_MIN_ALIGN	PAGE_SIZE
66 #endif
67 
68 #ifndef ELF_CORE_EFLAGS
69 #define ELF_CORE_EFLAGS	0
70 #endif
71 
72 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
73 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
74 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
75 
76 static struct linux_binfmt elf_format = {
77 	.module		= THIS_MODULE,
78 	.load_binary	= load_elf_binary,
79 	.load_shlib	= load_elf_library,
80 	.core_dump	= elf_core_dump,
81 	.min_coredump	= ELF_EXEC_PAGESIZE,
82 };
83 
84 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
85 
86 static int set_brk(unsigned long start, unsigned long end)
87 {
88 	start = ELF_PAGEALIGN(start);
89 	end = ELF_PAGEALIGN(end);
90 	if (end > start) {
91 		unsigned long addr;
92 		addr = vm_brk(start, end - start);
93 		if (BAD_ADDR(addr))
94 			return addr;
95 	}
96 	current->mm->start_brk = current->mm->brk = end;
97 	return 0;
98 }
99 
100 /* We need to explicitly zero any fractional pages
101    after the data section (i.e. bss).  This would
102    contain the junk from the file that should not
103    be in memory
104  */
105 static int padzero(unsigned long elf_bss)
106 {
107 	unsigned long nbyte;
108 
109 	nbyte = ELF_PAGEOFFSET(elf_bss);
110 	if (nbyte) {
111 		nbyte = ELF_MIN_ALIGN - nbyte;
112 		if (clear_user((void __user *) elf_bss, nbyte))
113 			return -EFAULT;
114 	}
115 	return 0;
116 }
117 
118 /* Let's use some macros to make this stack manipulation a little clearer */
119 #ifdef CONFIG_STACK_GROWSUP
120 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
121 #define STACK_ROUND(sp, items) \
122 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
123 #define STACK_ALLOC(sp, len) ({ \
124 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
125 	old_sp; })
126 #else
127 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
128 #define STACK_ROUND(sp, items) \
129 	(((unsigned long) (sp - items)) &~ 15UL)
130 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
131 #endif
132 
133 #ifndef ELF_BASE_PLATFORM
134 /*
135  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
136  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
137  * will be copied to the user stack in the same manner as AT_PLATFORM.
138  */
139 #define ELF_BASE_PLATFORM NULL
140 #endif
141 
142 static int
143 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
144 		unsigned long load_addr, unsigned long interp_load_addr)
145 {
146 	unsigned long p = bprm->p;
147 	int argc = bprm->argc;
148 	int envc = bprm->envc;
149 	elf_addr_t __user *argv;
150 	elf_addr_t __user *envp;
151 	elf_addr_t __user *sp;
152 	elf_addr_t __user *u_platform;
153 	elf_addr_t __user *u_base_platform;
154 	elf_addr_t __user *u_rand_bytes;
155 	const char *k_platform = ELF_PLATFORM;
156 	const char *k_base_platform = ELF_BASE_PLATFORM;
157 	unsigned char k_rand_bytes[16];
158 	int items;
159 	elf_addr_t *elf_info;
160 	int ei_index = 0;
161 	const struct cred *cred = current_cred();
162 	struct vm_area_struct *vma;
163 
164 	/*
165 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
166 	 * evictions by the processes running on the same package. One
167 	 * thing we can do is to shuffle the initial stack for them.
168 	 */
169 
170 	p = arch_align_stack(p);
171 
172 	/*
173 	 * If this architecture has a platform capability string, copy it
174 	 * to userspace.  In some cases (Sparc), this info is impossible
175 	 * for userspace to get any other way, in others (i386) it is
176 	 * merely difficult.
177 	 */
178 	u_platform = NULL;
179 	if (k_platform) {
180 		size_t len = strlen(k_platform) + 1;
181 
182 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
183 		if (__copy_to_user(u_platform, k_platform, len))
184 			return -EFAULT;
185 	}
186 
187 	/*
188 	 * If this architecture has a "base" platform capability
189 	 * string, copy it to userspace.
190 	 */
191 	u_base_platform = NULL;
192 	if (k_base_platform) {
193 		size_t len = strlen(k_base_platform) + 1;
194 
195 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
196 		if (__copy_to_user(u_base_platform, k_base_platform, len))
197 			return -EFAULT;
198 	}
199 
200 	/*
201 	 * Generate 16 random bytes for userspace PRNG seeding.
202 	 */
203 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
204 	u_rand_bytes = (elf_addr_t __user *)
205 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
206 	if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
207 		return -EFAULT;
208 
209 	/* Create the ELF interpreter info */
210 	elf_info = (elf_addr_t *)current->mm->saved_auxv;
211 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
212 #define NEW_AUX_ENT(id, val) \
213 	do { \
214 		elf_info[ei_index++] = id; \
215 		elf_info[ei_index++] = val; \
216 	} while (0)
217 
218 #ifdef ARCH_DLINFO
219 	/*
220 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
221 	 * AUXV.
222 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
223 	 * ARCH_DLINFO changes
224 	 */
225 	ARCH_DLINFO;
226 #endif
227 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
228 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
229 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
230 	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
231 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
232 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
233 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
234 	NEW_AUX_ENT(AT_FLAGS, 0);
235 	NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
236 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
237 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
238 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
239 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
240  	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
241 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
242 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
243 	if (k_platform) {
244 		NEW_AUX_ENT(AT_PLATFORM,
245 			    (elf_addr_t)(unsigned long)u_platform);
246 	}
247 	if (k_base_platform) {
248 		NEW_AUX_ENT(AT_BASE_PLATFORM,
249 			    (elf_addr_t)(unsigned long)u_base_platform);
250 	}
251 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
252 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
253 	}
254 #undef NEW_AUX_ENT
255 	/* AT_NULL is zero; clear the rest too */
256 	memset(&elf_info[ei_index], 0,
257 	       sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
258 
259 	/* And advance past the AT_NULL entry.  */
260 	ei_index += 2;
261 
262 	sp = STACK_ADD(p, ei_index);
263 
264 	items = (argc + 1) + (envc + 1) + 1;
265 	bprm->p = STACK_ROUND(sp, items);
266 
267 	/* Point sp at the lowest address on the stack */
268 #ifdef CONFIG_STACK_GROWSUP
269 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
270 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
271 #else
272 	sp = (elf_addr_t __user *)bprm->p;
273 #endif
274 
275 
276 	/*
277 	 * Grow the stack manually; some architectures have a limit on how
278 	 * far ahead a user-space access may be in order to grow the stack.
279 	 */
280 	vma = find_extend_vma(current->mm, bprm->p);
281 	if (!vma)
282 		return -EFAULT;
283 
284 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
285 	if (__put_user(argc, sp++))
286 		return -EFAULT;
287 	argv = sp;
288 	envp = argv + argc + 1;
289 
290 	/* Populate argv and envp */
291 	p = current->mm->arg_end = current->mm->arg_start;
292 	while (argc-- > 0) {
293 		size_t len;
294 		if (__put_user((elf_addr_t)p, argv++))
295 			return -EFAULT;
296 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
297 		if (!len || len > MAX_ARG_STRLEN)
298 			return -EINVAL;
299 		p += len;
300 	}
301 	if (__put_user(0, argv))
302 		return -EFAULT;
303 	current->mm->arg_end = current->mm->env_start = p;
304 	while (envc-- > 0) {
305 		size_t len;
306 		if (__put_user((elf_addr_t)p, envp++))
307 			return -EFAULT;
308 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
309 		if (!len || len > MAX_ARG_STRLEN)
310 			return -EINVAL;
311 		p += len;
312 	}
313 	if (__put_user(0, envp))
314 		return -EFAULT;
315 	current->mm->env_end = p;
316 
317 	/* Put the elf_info on the stack in the right place.  */
318 	sp = (elf_addr_t __user *)envp + 1;
319 	if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
320 		return -EFAULT;
321 	return 0;
322 }
323 
324 static unsigned long elf_map(struct file *filep, unsigned long addr,
325 		struct elf_phdr *eppnt, int prot, int type,
326 		unsigned long total_size)
327 {
328 	unsigned long map_addr;
329 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
330 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
331 	addr = ELF_PAGESTART(addr);
332 	size = ELF_PAGEALIGN(size);
333 
334 	/* mmap() will return -EINVAL if given a zero size, but a
335 	 * segment with zero filesize is perfectly valid */
336 	if (!size)
337 		return addr;
338 
339 	/*
340 	* total_size is the size of the ELF (interpreter) image.
341 	* The _first_ mmap needs to know the full size, otherwise
342 	* randomization might put this image into an overlapping
343 	* position with the ELF binary image. (since size < total_size)
344 	* So we first map the 'big' image - and unmap the remainder at
345 	* the end. (which unmap is needed for ELF images with holes.)
346 	*/
347 	if (total_size) {
348 		total_size = ELF_PAGEALIGN(total_size);
349 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
350 		if (!BAD_ADDR(map_addr))
351 			vm_munmap(map_addr+size, total_size-size);
352 	} else
353 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
354 
355 	return(map_addr);
356 }
357 
358 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
359 {
360 	int i, first_idx = -1, last_idx = -1;
361 
362 	for (i = 0; i < nr; i++) {
363 		if (cmds[i].p_type == PT_LOAD) {
364 			last_idx = i;
365 			if (first_idx == -1)
366 				first_idx = i;
367 		}
368 	}
369 	if (first_idx == -1)
370 		return 0;
371 
372 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
373 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
374 }
375 
376 
377 /* This is much more generalized than the library routine read function,
378    so we keep this separate.  Technically the library read function
379    is only provided so that we can read a.out libraries that have
380    an ELF header */
381 
382 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
383 		struct file *interpreter, unsigned long *interp_map_addr,
384 		unsigned long no_base)
385 {
386 	struct elf_phdr *elf_phdata;
387 	struct elf_phdr *eppnt;
388 	unsigned long load_addr = 0;
389 	int load_addr_set = 0;
390 	unsigned long last_bss = 0, elf_bss = 0;
391 	unsigned long error = ~0UL;
392 	unsigned long total_size;
393 	int retval, i, size;
394 
395 	/* First of all, some simple consistency checks */
396 	if (interp_elf_ex->e_type != ET_EXEC &&
397 	    interp_elf_ex->e_type != ET_DYN)
398 		goto out;
399 	if (!elf_check_arch(interp_elf_ex))
400 		goto out;
401 	if (!interpreter->f_op || !interpreter->f_op->mmap)
402 		goto out;
403 
404 	/*
405 	 * If the size of this structure has changed, then punt, since
406 	 * we will be doing the wrong thing.
407 	 */
408 	if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
409 		goto out;
410 	if (interp_elf_ex->e_phnum < 1 ||
411 		interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
412 		goto out;
413 
414 	/* Now read in all of the header information */
415 	size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
416 	if (size > ELF_MIN_ALIGN)
417 		goto out;
418 	elf_phdata = kmalloc(size, GFP_KERNEL);
419 	if (!elf_phdata)
420 		goto out;
421 
422 	retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
423 			     (char *)elf_phdata, size);
424 	error = -EIO;
425 	if (retval != size) {
426 		if (retval < 0)
427 			error = retval;
428 		goto out_close;
429 	}
430 
431 	total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
432 	if (!total_size) {
433 		error = -EINVAL;
434 		goto out_close;
435 	}
436 
437 	eppnt = elf_phdata;
438 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
439 		if (eppnt->p_type == PT_LOAD) {
440 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
441 			int elf_prot = 0;
442 			unsigned long vaddr = 0;
443 			unsigned long k, map_addr;
444 
445 			if (eppnt->p_flags & PF_R)
446 		    		elf_prot = PROT_READ;
447 			if (eppnt->p_flags & PF_W)
448 				elf_prot |= PROT_WRITE;
449 			if (eppnt->p_flags & PF_X)
450 				elf_prot |= PROT_EXEC;
451 			vaddr = eppnt->p_vaddr;
452 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
453 				elf_type |= MAP_FIXED;
454 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
455 				load_addr = -vaddr;
456 
457 			map_addr = elf_map(interpreter, load_addr + vaddr,
458 					eppnt, elf_prot, elf_type, total_size);
459 			total_size = 0;
460 			if (!*interp_map_addr)
461 				*interp_map_addr = map_addr;
462 			error = map_addr;
463 			if (BAD_ADDR(map_addr))
464 				goto out_close;
465 
466 			if (!load_addr_set &&
467 			    interp_elf_ex->e_type == ET_DYN) {
468 				load_addr = map_addr - ELF_PAGESTART(vaddr);
469 				load_addr_set = 1;
470 			}
471 
472 			/*
473 			 * Check to see if the section's size will overflow the
474 			 * allowed task size. Note that p_filesz must always be
475 			 * <= p_memsize so it's only necessary to check p_memsz.
476 			 */
477 			k = load_addr + eppnt->p_vaddr;
478 			if (BAD_ADDR(k) ||
479 			    eppnt->p_filesz > eppnt->p_memsz ||
480 			    eppnt->p_memsz > TASK_SIZE ||
481 			    TASK_SIZE - eppnt->p_memsz < k) {
482 				error = -ENOMEM;
483 				goto out_close;
484 			}
485 
486 			/*
487 			 * Find the end of the file mapping for this phdr, and
488 			 * keep track of the largest address we see for this.
489 			 */
490 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
491 			if (k > elf_bss)
492 				elf_bss = k;
493 
494 			/*
495 			 * Do the same thing for the memory mapping - between
496 			 * elf_bss and last_bss is the bss section.
497 			 */
498 			k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
499 			if (k > last_bss)
500 				last_bss = k;
501 		}
502 	}
503 
504 	if (last_bss > elf_bss) {
505 		/*
506 		 * Now fill out the bss section.  First pad the last page up
507 		 * to the page boundary, and then perform a mmap to make sure
508 		 * that there are zero-mapped pages up to and including the
509 		 * last bss page.
510 		 */
511 		if (padzero(elf_bss)) {
512 			error = -EFAULT;
513 			goto out_close;
514 		}
515 
516 		/* What we have mapped so far */
517 		elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
518 
519 		/* Map the last of the bss segment */
520 		error = vm_brk(elf_bss, last_bss - elf_bss);
521 		if (BAD_ADDR(error))
522 			goto out_close;
523 	}
524 
525 	error = load_addr;
526 
527 out_close:
528 	kfree(elf_phdata);
529 out:
530 	return error;
531 }
532 
533 /*
534  * These are the functions used to load ELF style executables and shared
535  * libraries.  There is no binary dependent code anywhere else.
536  */
537 
538 #define INTERPRETER_NONE 0
539 #define INTERPRETER_ELF 2
540 
541 #ifndef STACK_RND_MASK
542 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))	/* 8MB of VA */
543 #endif
544 
545 static unsigned long randomize_stack_top(unsigned long stack_top)
546 {
547 	unsigned int random_variable = 0;
548 
549 	if ((current->flags & PF_RANDOMIZE) &&
550 		!(current->personality & ADDR_NO_RANDOMIZE)) {
551 		random_variable = get_random_int() & STACK_RND_MASK;
552 		random_variable <<= PAGE_SHIFT;
553 	}
554 #ifdef CONFIG_STACK_GROWSUP
555 	return PAGE_ALIGN(stack_top) + random_variable;
556 #else
557 	return PAGE_ALIGN(stack_top) - random_variable;
558 #endif
559 }
560 
561 static int load_elf_binary(struct linux_binprm *bprm)
562 {
563 	struct file *interpreter = NULL; /* to shut gcc up */
564  	unsigned long load_addr = 0, load_bias = 0;
565 	int load_addr_set = 0;
566 	char * elf_interpreter = NULL;
567 	unsigned long error;
568 	struct elf_phdr *elf_ppnt, *elf_phdata;
569 	unsigned long elf_bss, elf_brk;
570 	int retval, i;
571 	unsigned int size;
572 	unsigned long elf_entry;
573 	unsigned long interp_load_addr = 0;
574 	unsigned long start_code, end_code, start_data, end_data;
575 	unsigned long reloc_func_desc __maybe_unused = 0;
576 	int executable_stack = EXSTACK_DEFAULT;
577 	unsigned long def_flags = 0;
578 	struct pt_regs *regs = current_pt_regs();
579 	struct {
580 		struct elfhdr elf_ex;
581 		struct elfhdr interp_elf_ex;
582 	} *loc;
583 
584 	loc = kmalloc(sizeof(*loc), GFP_KERNEL);
585 	if (!loc) {
586 		retval = -ENOMEM;
587 		goto out_ret;
588 	}
589 
590 	/* Get the exec-header */
591 	loc->elf_ex = *((struct elfhdr *)bprm->buf);
592 
593 	retval = -ENOEXEC;
594 	/* First of all, some simple consistency checks */
595 	if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
596 		goto out;
597 
598 	if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
599 		goto out;
600 	if (!elf_check_arch(&loc->elf_ex))
601 		goto out;
602 	if (!bprm->file->f_op || !bprm->file->f_op->mmap)
603 		goto out;
604 
605 	/* Now read in all of the header information */
606 	if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
607 		goto out;
608 	if (loc->elf_ex.e_phnum < 1 ||
609 	 	loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
610 		goto out;
611 	size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
612 	retval = -ENOMEM;
613 	elf_phdata = kmalloc(size, GFP_KERNEL);
614 	if (!elf_phdata)
615 		goto out;
616 
617 	retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
618 			     (char *)elf_phdata, size);
619 	if (retval != size) {
620 		if (retval >= 0)
621 			retval = -EIO;
622 		goto out_free_ph;
623 	}
624 
625 	elf_ppnt = elf_phdata;
626 	elf_bss = 0;
627 	elf_brk = 0;
628 
629 	start_code = ~0UL;
630 	end_code = 0;
631 	start_data = 0;
632 	end_data = 0;
633 
634 	for (i = 0; i < loc->elf_ex.e_phnum; i++) {
635 		if (elf_ppnt->p_type == PT_INTERP) {
636 			/* This is the program interpreter used for
637 			 * shared libraries - for now assume that this
638 			 * is an a.out format binary
639 			 */
640 			retval = -ENOEXEC;
641 			if (elf_ppnt->p_filesz > PATH_MAX ||
642 			    elf_ppnt->p_filesz < 2)
643 				goto out_free_ph;
644 
645 			retval = -ENOMEM;
646 			elf_interpreter = kmalloc(elf_ppnt->p_filesz,
647 						  GFP_KERNEL);
648 			if (!elf_interpreter)
649 				goto out_free_ph;
650 
651 			retval = kernel_read(bprm->file, elf_ppnt->p_offset,
652 					     elf_interpreter,
653 					     elf_ppnt->p_filesz);
654 			if (retval != elf_ppnt->p_filesz) {
655 				if (retval >= 0)
656 					retval = -EIO;
657 				goto out_free_interp;
658 			}
659 			/* make sure path is NULL terminated */
660 			retval = -ENOEXEC;
661 			if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
662 				goto out_free_interp;
663 
664 			interpreter = open_exec(elf_interpreter);
665 			retval = PTR_ERR(interpreter);
666 			if (IS_ERR(interpreter))
667 				goto out_free_interp;
668 
669 			/*
670 			 * If the binary is not readable then enforce
671 			 * mm->dumpable = 0 regardless of the interpreter's
672 			 * permissions.
673 			 */
674 			would_dump(bprm, interpreter);
675 
676 			retval = kernel_read(interpreter, 0, bprm->buf,
677 					     BINPRM_BUF_SIZE);
678 			if (retval != BINPRM_BUF_SIZE) {
679 				if (retval >= 0)
680 					retval = -EIO;
681 				goto out_free_dentry;
682 			}
683 
684 			/* Get the exec headers */
685 			loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
686 			break;
687 		}
688 		elf_ppnt++;
689 	}
690 
691 	elf_ppnt = elf_phdata;
692 	for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
693 		if (elf_ppnt->p_type == PT_GNU_STACK) {
694 			if (elf_ppnt->p_flags & PF_X)
695 				executable_stack = EXSTACK_ENABLE_X;
696 			else
697 				executable_stack = EXSTACK_DISABLE_X;
698 			break;
699 		}
700 
701 	/* Some simple consistency checks for the interpreter */
702 	if (elf_interpreter) {
703 		retval = -ELIBBAD;
704 		/* Not an ELF interpreter */
705 		if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
706 			goto out_free_dentry;
707 		/* Verify the interpreter has a valid arch */
708 		if (!elf_check_arch(&loc->interp_elf_ex))
709 			goto out_free_dentry;
710 	}
711 
712 	/* Flush all traces of the currently running executable */
713 	retval = flush_old_exec(bprm);
714 	if (retval)
715 		goto out_free_dentry;
716 
717 	/* OK, This is the point of no return */
718 	current->mm->def_flags = def_flags;
719 
720 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
721 	   may depend on the personality.  */
722 	SET_PERSONALITY(loc->elf_ex);
723 	if (elf_read_implies_exec(loc->elf_ex, executable_stack))
724 		current->personality |= READ_IMPLIES_EXEC;
725 
726 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
727 		current->flags |= PF_RANDOMIZE;
728 
729 	setup_new_exec(bprm);
730 
731 	/* Do this so that we can load the interpreter, if need be.  We will
732 	   change some of these later */
733 	current->mm->free_area_cache = current->mm->mmap_base;
734 	current->mm->cached_hole_size = 0;
735 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
736 				 executable_stack);
737 	if (retval < 0) {
738 		send_sig(SIGKILL, current, 0);
739 		goto out_free_dentry;
740 	}
741 
742 	current->mm->start_stack = bprm->p;
743 
744 	/* Now we do a little grungy work by mmapping the ELF image into
745 	   the correct location in memory. */
746 	for(i = 0, elf_ppnt = elf_phdata;
747 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
748 		int elf_prot = 0, elf_flags;
749 		unsigned long k, vaddr;
750 
751 		if (elf_ppnt->p_type != PT_LOAD)
752 			continue;
753 
754 		if (unlikely (elf_brk > elf_bss)) {
755 			unsigned long nbyte;
756 
757 			/* There was a PT_LOAD segment with p_memsz > p_filesz
758 			   before this one. Map anonymous pages, if needed,
759 			   and clear the area.  */
760 			retval = set_brk(elf_bss + load_bias,
761 					 elf_brk + load_bias);
762 			if (retval) {
763 				send_sig(SIGKILL, current, 0);
764 				goto out_free_dentry;
765 			}
766 			nbyte = ELF_PAGEOFFSET(elf_bss);
767 			if (nbyte) {
768 				nbyte = ELF_MIN_ALIGN - nbyte;
769 				if (nbyte > elf_brk - elf_bss)
770 					nbyte = elf_brk - elf_bss;
771 				if (clear_user((void __user *)elf_bss +
772 							load_bias, nbyte)) {
773 					/*
774 					 * This bss-zeroing can fail if the ELF
775 					 * file specifies odd protections. So
776 					 * we don't check the return value
777 					 */
778 				}
779 			}
780 		}
781 
782 		if (elf_ppnt->p_flags & PF_R)
783 			elf_prot |= PROT_READ;
784 		if (elf_ppnt->p_flags & PF_W)
785 			elf_prot |= PROT_WRITE;
786 		if (elf_ppnt->p_flags & PF_X)
787 			elf_prot |= PROT_EXEC;
788 
789 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
790 
791 		vaddr = elf_ppnt->p_vaddr;
792 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
793 			elf_flags |= MAP_FIXED;
794 		} else if (loc->elf_ex.e_type == ET_DYN) {
795 			/* Try and get dynamic programs out of the way of the
796 			 * default mmap base, as well as whatever program they
797 			 * might try to exec.  This is because the brk will
798 			 * follow the loader, and is not movable.  */
799 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
800 			/* Memory randomization might have been switched off
801 			 * in runtime via sysctl.
802 			 * If that is the case, retain the original non-zero
803 			 * load_bias value in order to establish proper
804 			 * non-randomized mappings.
805 			 */
806 			if (current->flags & PF_RANDOMIZE)
807 				load_bias = 0;
808 			else
809 				load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
810 #else
811 			load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
812 #endif
813 		}
814 
815 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
816 				elf_prot, elf_flags, 0);
817 		if (BAD_ADDR(error)) {
818 			send_sig(SIGKILL, current, 0);
819 			retval = IS_ERR((void *)error) ?
820 				PTR_ERR((void*)error) : -EINVAL;
821 			goto out_free_dentry;
822 		}
823 
824 		if (!load_addr_set) {
825 			load_addr_set = 1;
826 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
827 			if (loc->elf_ex.e_type == ET_DYN) {
828 				load_bias += error -
829 				             ELF_PAGESTART(load_bias + vaddr);
830 				load_addr += load_bias;
831 				reloc_func_desc = load_bias;
832 			}
833 		}
834 		k = elf_ppnt->p_vaddr;
835 		if (k < start_code)
836 			start_code = k;
837 		if (start_data < k)
838 			start_data = k;
839 
840 		/*
841 		 * Check to see if the section's size will overflow the
842 		 * allowed task size. Note that p_filesz must always be
843 		 * <= p_memsz so it is only necessary to check p_memsz.
844 		 */
845 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
846 		    elf_ppnt->p_memsz > TASK_SIZE ||
847 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
848 			/* set_brk can never work. Avoid overflows. */
849 			send_sig(SIGKILL, current, 0);
850 			retval = -EINVAL;
851 			goto out_free_dentry;
852 		}
853 
854 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
855 
856 		if (k > elf_bss)
857 			elf_bss = k;
858 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
859 			end_code = k;
860 		if (end_data < k)
861 			end_data = k;
862 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
863 		if (k > elf_brk)
864 			elf_brk = k;
865 	}
866 
867 	loc->elf_ex.e_entry += load_bias;
868 	elf_bss += load_bias;
869 	elf_brk += load_bias;
870 	start_code += load_bias;
871 	end_code += load_bias;
872 	start_data += load_bias;
873 	end_data += load_bias;
874 
875 	/* Calling set_brk effectively mmaps the pages that we need
876 	 * for the bss and break sections.  We must do this before
877 	 * mapping in the interpreter, to make sure it doesn't wind
878 	 * up getting placed where the bss needs to go.
879 	 */
880 	retval = set_brk(elf_bss, elf_brk);
881 	if (retval) {
882 		send_sig(SIGKILL, current, 0);
883 		goto out_free_dentry;
884 	}
885 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
886 		send_sig(SIGSEGV, current, 0);
887 		retval = -EFAULT; /* Nobody gets to see this, but.. */
888 		goto out_free_dentry;
889 	}
890 
891 	if (elf_interpreter) {
892 		unsigned long interp_map_addr = 0;
893 
894 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
895 					    interpreter,
896 					    &interp_map_addr,
897 					    load_bias);
898 		if (!IS_ERR((void *)elf_entry)) {
899 			/*
900 			 * load_elf_interp() returns relocation
901 			 * adjustment
902 			 */
903 			interp_load_addr = elf_entry;
904 			elf_entry += loc->interp_elf_ex.e_entry;
905 		}
906 		if (BAD_ADDR(elf_entry)) {
907 			force_sig(SIGSEGV, current);
908 			retval = IS_ERR((void *)elf_entry) ?
909 					(int)elf_entry : -EINVAL;
910 			goto out_free_dentry;
911 		}
912 		reloc_func_desc = interp_load_addr;
913 
914 		allow_write_access(interpreter);
915 		fput(interpreter);
916 		kfree(elf_interpreter);
917 	} else {
918 		elf_entry = loc->elf_ex.e_entry;
919 		if (BAD_ADDR(elf_entry)) {
920 			force_sig(SIGSEGV, current);
921 			retval = -EINVAL;
922 			goto out_free_dentry;
923 		}
924 	}
925 
926 	kfree(elf_phdata);
927 
928 	set_binfmt(&elf_format);
929 
930 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
931 	retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
932 	if (retval < 0) {
933 		send_sig(SIGKILL, current, 0);
934 		goto out;
935 	}
936 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
937 
938 	install_exec_creds(bprm);
939 	retval = create_elf_tables(bprm, &loc->elf_ex,
940 			  load_addr, interp_load_addr);
941 	if (retval < 0) {
942 		send_sig(SIGKILL, current, 0);
943 		goto out;
944 	}
945 	/* N.B. passed_fileno might not be initialized? */
946 	current->mm->end_code = end_code;
947 	current->mm->start_code = start_code;
948 	current->mm->start_data = start_data;
949 	current->mm->end_data = end_data;
950 	current->mm->start_stack = bprm->p;
951 
952 #ifdef arch_randomize_brk
953 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
954 		current->mm->brk = current->mm->start_brk =
955 			arch_randomize_brk(current->mm);
956 #ifdef CONFIG_COMPAT_BRK
957 		current->brk_randomized = 1;
958 #endif
959 	}
960 #endif
961 
962 	if (current->personality & MMAP_PAGE_ZERO) {
963 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
964 		   and some applications "depend" upon this behavior.
965 		   Since we do not have the power to recompile these, we
966 		   emulate the SVr4 behavior. Sigh. */
967 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
968 				MAP_FIXED | MAP_PRIVATE, 0);
969 	}
970 
971 #ifdef ELF_PLAT_INIT
972 	/*
973 	 * The ABI may specify that certain registers be set up in special
974 	 * ways (on i386 %edx is the address of a DT_FINI function, for
975 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
976 	 * that the e_entry field is the address of the function descriptor
977 	 * for the startup routine, rather than the address of the startup
978 	 * routine itself.  This macro performs whatever initialization to
979 	 * the regs structure is required as well as any relocations to the
980 	 * function descriptor entries when executing dynamically links apps.
981 	 */
982 	ELF_PLAT_INIT(regs, reloc_func_desc);
983 #endif
984 
985 	start_thread(regs, elf_entry, bprm->p);
986 	retval = 0;
987 out:
988 	kfree(loc);
989 out_ret:
990 	return retval;
991 
992 	/* error cleanup */
993 out_free_dentry:
994 	allow_write_access(interpreter);
995 	if (interpreter)
996 		fput(interpreter);
997 out_free_interp:
998 	kfree(elf_interpreter);
999 out_free_ph:
1000 	kfree(elf_phdata);
1001 	goto out;
1002 }
1003 
1004 /* This is really simpleminded and specialized - we are loading an
1005    a.out library that is given an ELF header. */
1006 static int load_elf_library(struct file *file)
1007 {
1008 	struct elf_phdr *elf_phdata;
1009 	struct elf_phdr *eppnt;
1010 	unsigned long elf_bss, bss, len;
1011 	int retval, error, i, j;
1012 	struct elfhdr elf_ex;
1013 
1014 	error = -ENOEXEC;
1015 	retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1016 	if (retval != sizeof(elf_ex))
1017 		goto out;
1018 
1019 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1020 		goto out;
1021 
1022 	/* First of all, some simple consistency checks */
1023 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1024 	    !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1025 		goto out;
1026 
1027 	/* Now read in all of the header information */
1028 
1029 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1030 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1031 
1032 	error = -ENOMEM;
1033 	elf_phdata = kmalloc(j, GFP_KERNEL);
1034 	if (!elf_phdata)
1035 		goto out;
1036 
1037 	eppnt = elf_phdata;
1038 	error = -ENOEXEC;
1039 	retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1040 	if (retval != j)
1041 		goto out_free_ph;
1042 
1043 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1044 		if ((eppnt + i)->p_type == PT_LOAD)
1045 			j++;
1046 	if (j != 1)
1047 		goto out_free_ph;
1048 
1049 	while (eppnt->p_type != PT_LOAD)
1050 		eppnt++;
1051 
1052 	/* Now use mmap to map the library into memory. */
1053 	error = vm_mmap(file,
1054 			ELF_PAGESTART(eppnt->p_vaddr),
1055 			(eppnt->p_filesz +
1056 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1057 			PROT_READ | PROT_WRITE | PROT_EXEC,
1058 			MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1059 			(eppnt->p_offset -
1060 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1061 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1062 		goto out_free_ph;
1063 
1064 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1065 	if (padzero(elf_bss)) {
1066 		error = -EFAULT;
1067 		goto out_free_ph;
1068 	}
1069 
1070 	len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1071 			    ELF_MIN_ALIGN - 1);
1072 	bss = eppnt->p_memsz + eppnt->p_vaddr;
1073 	if (bss > len)
1074 		vm_brk(len, bss - len);
1075 	error = 0;
1076 
1077 out_free_ph:
1078 	kfree(elf_phdata);
1079 out:
1080 	return error;
1081 }
1082 
1083 #ifdef CONFIG_ELF_CORE
1084 /*
1085  * ELF core dumper
1086  *
1087  * Modelled on fs/exec.c:aout_core_dump()
1088  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1089  */
1090 
1091 /*
1092  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1093  * that are useful for post-mortem analysis are included in every core dump.
1094  * In that way we ensure that the core dump is fully interpretable later
1095  * without matching up the same kernel and hardware config to see what PC values
1096  * meant. These special mappings include - vDSO, vsyscall, and other
1097  * architecture specific mappings
1098  */
1099 static bool always_dump_vma(struct vm_area_struct *vma)
1100 {
1101 	/* Any vsyscall mappings? */
1102 	if (vma == get_gate_vma(vma->vm_mm))
1103 		return true;
1104 	/*
1105 	 * arch_vma_name() returns non-NULL for special architecture mappings,
1106 	 * such as vDSO sections.
1107 	 */
1108 	if (arch_vma_name(vma))
1109 		return true;
1110 
1111 	return false;
1112 }
1113 
1114 /*
1115  * Decide what to dump of a segment, part, all or none.
1116  */
1117 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1118 				   unsigned long mm_flags)
1119 {
1120 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1121 
1122 	/* always dump the vdso and vsyscall sections */
1123 	if (always_dump_vma(vma))
1124 		goto whole;
1125 
1126 	if (vma->vm_flags & VM_DONTDUMP)
1127 		return 0;
1128 
1129 	/* Hugetlb memory check */
1130 	if (vma->vm_flags & VM_HUGETLB) {
1131 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1132 			goto whole;
1133 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1134 			goto whole;
1135 	}
1136 
1137 	/* Do not dump I/O mapped devices or special mappings */
1138 	if (vma->vm_flags & VM_IO)
1139 		return 0;
1140 
1141 	/* By default, dump shared memory if mapped from an anonymous file. */
1142 	if (vma->vm_flags & VM_SHARED) {
1143 		if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1144 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1145 			goto whole;
1146 		return 0;
1147 	}
1148 
1149 	/* Dump segments that have been written to.  */
1150 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1151 		goto whole;
1152 	if (vma->vm_file == NULL)
1153 		return 0;
1154 
1155 	if (FILTER(MAPPED_PRIVATE))
1156 		goto whole;
1157 
1158 	/*
1159 	 * If this looks like the beginning of a DSO or executable mapping,
1160 	 * check for an ELF header.  If we find one, dump the first page to
1161 	 * aid in determining what was mapped here.
1162 	 */
1163 	if (FILTER(ELF_HEADERS) &&
1164 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1165 		u32 __user *header = (u32 __user *) vma->vm_start;
1166 		u32 word;
1167 		mm_segment_t fs = get_fs();
1168 		/*
1169 		 * Doing it this way gets the constant folded by GCC.
1170 		 */
1171 		union {
1172 			u32 cmp;
1173 			char elfmag[SELFMAG];
1174 		} magic;
1175 		BUILD_BUG_ON(SELFMAG != sizeof word);
1176 		magic.elfmag[EI_MAG0] = ELFMAG0;
1177 		magic.elfmag[EI_MAG1] = ELFMAG1;
1178 		magic.elfmag[EI_MAG2] = ELFMAG2;
1179 		magic.elfmag[EI_MAG3] = ELFMAG3;
1180 		/*
1181 		 * Switch to the user "segment" for get_user(),
1182 		 * then put back what elf_core_dump() had in place.
1183 		 */
1184 		set_fs(USER_DS);
1185 		if (unlikely(get_user(word, header)))
1186 			word = 0;
1187 		set_fs(fs);
1188 		if (word == magic.cmp)
1189 			return PAGE_SIZE;
1190 	}
1191 
1192 #undef	FILTER
1193 
1194 	return 0;
1195 
1196 whole:
1197 	return vma->vm_end - vma->vm_start;
1198 }
1199 
1200 /* An ELF note in memory */
1201 struct memelfnote
1202 {
1203 	const char *name;
1204 	int type;
1205 	unsigned int datasz;
1206 	void *data;
1207 };
1208 
1209 static int notesize(struct memelfnote *en)
1210 {
1211 	int sz;
1212 
1213 	sz = sizeof(struct elf_note);
1214 	sz += roundup(strlen(en->name) + 1, 4);
1215 	sz += roundup(en->datasz, 4);
1216 
1217 	return sz;
1218 }
1219 
1220 #define DUMP_WRITE(addr, nr, foffset)	\
1221 	do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1222 
1223 static int alignfile(struct file *file, loff_t *foffset)
1224 {
1225 	static const char buf[4] = { 0, };
1226 	DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1227 	return 1;
1228 }
1229 
1230 static int writenote(struct memelfnote *men, struct file *file,
1231 			loff_t *foffset)
1232 {
1233 	struct elf_note en;
1234 	en.n_namesz = strlen(men->name) + 1;
1235 	en.n_descsz = men->datasz;
1236 	en.n_type = men->type;
1237 
1238 	DUMP_WRITE(&en, sizeof(en), foffset);
1239 	DUMP_WRITE(men->name, en.n_namesz, foffset);
1240 	if (!alignfile(file, foffset))
1241 		return 0;
1242 	DUMP_WRITE(men->data, men->datasz, foffset);
1243 	if (!alignfile(file, foffset))
1244 		return 0;
1245 
1246 	return 1;
1247 }
1248 #undef DUMP_WRITE
1249 
1250 static void fill_elf_header(struct elfhdr *elf, int segs,
1251 			    u16 machine, u32 flags, u8 osabi)
1252 {
1253 	memset(elf, 0, sizeof(*elf));
1254 
1255 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1256 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1257 	elf->e_ident[EI_DATA] = ELF_DATA;
1258 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1259 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1260 
1261 	elf->e_type = ET_CORE;
1262 	elf->e_machine = machine;
1263 	elf->e_version = EV_CURRENT;
1264 	elf->e_phoff = sizeof(struct elfhdr);
1265 	elf->e_flags = flags;
1266 	elf->e_ehsize = sizeof(struct elfhdr);
1267 	elf->e_phentsize = sizeof(struct elf_phdr);
1268 	elf->e_phnum = segs;
1269 
1270 	return;
1271 }
1272 
1273 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1274 {
1275 	phdr->p_type = PT_NOTE;
1276 	phdr->p_offset = offset;
1277 	phdr->p_vaddr = 0;
1278 	phdr->p_paddr = 0;
1279 	phdr->p_filesz = sz;
1280 	phdr->p_memsz = 0;
1281 	phdr->p_flags = 0;
1282 	phdr->p_align = 0;
1283 	return;
1284 }
1285 
1286 static void fill_note(struct memelfnote *note, const char *name, int type,
1287 		unsigned int sz, void *data)
1288 {
1289 	note->name = name;
1290 	note->type = type;
1291 	note->datasz = sz;
1292 	note->data = data;
1293 	return;
1294 }
1295 
1296 /*
1297  * fill up all the fields in prstatus from the given task struct, except
1298  * registers which need to be filled up separately.
1299  */
1300 static void fill_prstatus(struct elf_prstatus *prstatus,
1301 		struct task_struct *p, long signr)
1302 {
1303 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1304 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1305 	prstatus->pr_sighold = p->blocked.sig[0];
1306 	rcu_read_lock();
1307 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1308 	rcu_read_unlock();
1309 	prstatus->pr_pid = task_pid_vnr(p);
1310 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1311 	prstatus->pr_sid = task_session_vnr(p);
1312 	if (thread_group_leader(p)) {
1313 		struct task_cputime cputime;
1314 
1315 		/*
1316 		 * This is the record for the group leader.  It shows the
1317 		 * group-wide total, not its individual thread total.
1318 		 */
1319 		thread_group_cputime(p, &cputime);
1320 		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1321 		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1322 	} else {
1323 		cputime_to_timeval(p->utime, &prstatus->pr_utime);
1324 		cputime_to_timeval(p->stime, &prstatus->pr_stime);
1325 	}
1326 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1327 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1328 }
1329 
1330 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1331 		       struct mm_struct *mm)
1332 {
1333 	const struct cred *cred;
1334 	unsigned int i, len;
1335 
1336 	/* first copy the parameters from user space */
1337 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1338 
1339 	len = mm->arg_end - mm->arg_start;
1340 	if (len >= ELF_PRARGSZ)
1341 		len = ELF_PRARGSZ-1;
1342 	if (copy_from_user(&psinfo->pr_psargs,
1343 		           (const char __user *)mm->arg_start, len))
1344 		return -EFAULT;
1345 	for(i = 0; i < len; i++)
1346 		if (psinfo->pr_psargs[i] == 0)
1347 			psinfo->pr_psargs[i] = ' ';
1348 	psinfo->pr_psargs[len] = 0;
1349 
1350 	rcu_read_lock();
1351 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1352 	rcu_read_unlock();
1353 	psinfo->pr_pid = task_pid_vnr(p);
1354 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1355 	psinfo->pr_sid = task_session_vnr(p);
1356 
1357 	i = p->state ? ffz(~p->state) + 1 : 0;
1358 	psinfo->pr_state = i;
1359 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1360 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1361 	psinfo->pr_nice = task_nice(p);
1362 	psinfo->pr_flag = p->flags;
1363 	rcu_read_lock();
1364 	cred = __task_cred(p);
1365 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1366 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1367 	rcu_read_unlock();
1368 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1369 
1370 	return 0;
1371 }
1372 
1373 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1374 {
1375 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1376 	int i = 0;
1377 	do
1378 		i += 2;
1379 	while (auxv[i - 2] != AT_NULL);
1380 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1381 }
1382 
1383 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1384 		siginfo_t *siginfo)
1385 {
1386 	mm_segment_t old_fs = get_fs();
1387 	set_fs(KERNEL_DS);
1388 	copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1389 	set_fs(old_fs);
1390 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1391 }
1392 
1393 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1394 /*
1395  * Format of NT_FILE note:
1396  *
1397  * long count     -- how many files are mapped
1398  * long page_size -- units for file_ofs
1399  * array of [COUNT] elements of
1400  *   long start
1401  *   long end
1402  *   long file_ofs
1403  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1404  */
1405 static void fill_files_note(struct memelfnote *note)
1406 {
1407 	struct vm_area_struct *vma;
1408 	unsigned count, size, names_ofs, remaining, n;
1409 	user_long_t *data;
1410 	user_long_t *start_end_ofs;
1411 	char *name_base, *name_curpos;
1412 
1413 	/* *Estimated* file count and total data size needed */
1414 	count = current->mm->map_count;
1415 	size = count * 64;
1416 
1417 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1418  alloc:
1419 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1420 		goto err;
1421 	size = round_up(size, PAGE_SIZE);
1422 	data = vmalloc(size);
1423 	if (!data)
1424 		goto err;
1425 
1426 	start_end_ofs = data + 2;
1427 	name_base = name_curpos = ((char *)data) + names_ofs;
1428 	remaining = size - names_ofs;
1429 	count = 0;
1430 	for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1431 		struct file *file;
1432 		const char *filename;
1433 
1434 		file = vma->vm_file;
1435 		if (!file)
1436 			continue;
1437 		filename = d_path(&file->f_path, name_curpos, remaining);
1438 		if (IS_ERR(filename)) {
1439 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1440 				vfree(data);
1441 				size = size * 5 / 4;
1442 				goto alloc;
1443 			}
1444 			continue;
1445 		}
1446 
1447 		/* d_path() fills at the end, move name down */
1448 		/* n = strlen(filename) + 1: */
1449 		n = (name_curpos + remaining) - filename;
1450 		remaining = filename - name_curpos;
1451 		memmove(name_curpos, filename, n);
1452 		name_curpos += n;
1453 
1454 		*start_end_ofs++ = vma->vm_start;
1455 		*start_end_ofs++ = vma->vm_end;
1456 		*start_end_ofs++ = vma->vm_pgoff;
1457 		count++;
1458 	}
1459 
1460 	/* Now we know exact count of files, can store it */
1461 	data[0] = count;
1462 	data[1] = PAGE_SIZE;
1463 	/*
1464 	 * Count usually is less than current->mm->map_count,
1465 	 * we need to move filenames down.
1466 	 */
1467 	n = current->mm->map_count - count;
1468 	if (n != 0) {
1469 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1470 		memmove(name_base - shift_bytes, name_base,
1471 			name_curpos - name_base);
1472 		name_curpos -= shift_bytes;
1473 	}
1474 
1475 	size = name_curpos - (char *)data;
1476 	fill_note(note, "CORE", NT_FILE, size, data);
1477  err: ;
1478 }
1479 
1480 #ifdef CORE_DUMP_USE_REGSET
1481 #include <linux/regset.h>
1482 
1483 struct elf_thread_core_info {
1484 	struct elf_thread_core_info *next;
1485 	struct task_struct *task;
1486 	struct elf_prstatus prstatus;
1487 	struct memelfnote notes[0];
1488 };
1489 
1490 struct elf_note_info {
1491 	struct elf_thread_core_info *thread;
1492 	struct memelfnote psinfo;
1493 	struct memelfnote signote;
1494 	struct memelfnote auxv;
1495 	struct memelfnote files;
1496 	user_siginfo_t csigdata;
1497 	size_t size;
1498 	int thread_notes;
1499 };
1500 
1501 /*
1502  * When a regset has a writeback hook, we call it on each thread before
1503  * dumping user memory.  On register window machines, this makes sure the
1504  * user memory backing the register data is up to date before we read it.
1505  */
1506 static void do_thread_regset_writeback(struct task_struct *task,
1507 				       const struct user_regset *regset)
1508 {
1509 	if (regset->writeback)
1510 		regset->writeback(task, regset, 1);
1511 }
1512 
1513 #ifndef PR_REG_SIZE
1514 #define PR_REG_SIZE(S) sizeof(S)
1515 #endif
1516 
1517 #ifndef PRSTATUS_SIZE
1518 #define PRSTATUS_SIZE(S) sizeof(S)
1519 #endif
1520 
1521 #ifndef PR_REG_PTR
1522 #define PR_REG_PTR(S) (&((S)->pr_reg))
1523 #endif
1524 
1525 #ifndef SET_PR_FPVALID
1526 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1527 #endif
1528 
1529 static int fill_thread_core_info(struct elf_thread_core_info *t,
1530 				 const struct user_regset_view *view,
1531 				 long signr, size_t *total)
1532 {
1533 	unsigned int i;
1534 
1535 	/*
1536 	 * NT_PRSTATUS is the one special case, because the regset data
1537 	 * goes into the pr_reg field inside the note contents, rather
1538 	 * than being the whole note contents.  We fill the reset in here.
1539 	 * We assume that regset 0 is NT_PRSTATUS.
1540 	 */
1541 	fill_prstatus(&t->prstatus, t->task, signr);
1542 	(void) view->regsets[0].get(t->task, &view->regsets[0],
1543 				    0, PR_REG_SIZE(t->prstatus.pr_reg),
1544 				    PR_REG_PTR(&t->prstatus), NULL);
1545 
1546 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1547 		  PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1548 	*total += notesize(&t->notes[0]);
1549 
1550 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1551 
1552 	/*
1553 	 * Each other regset might generate a note too.  For each regset
1554 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1555 	 * all zero and we'll know to skip writing it later.
1556 	 */
1557 	for (i = 1; i < view->n; ++i) {
1558 		const struct user_regset *regset = &view->regsets[i];
1559 		do_thread_regset_writeback(t->task, regset);
1560 		if (regset->core_note_type && regset->get &&
1561 		    (!regset->active || regset->active(t->task, regset))) {
1562 			int ret;
1563 			size_t size = regset->n * regset->size;
1564 			void *data = kmalloc(size, GFP_KERNEL);
1565 			if (unlikely(!data))
1566 				return 0;
1567 			ret = regset->get(t->task, regset,
1568 					  0, size, data, NULL);
1569 			if (unlikely(ret))
1570 				kfree(data);
1571 			else {
1572 				if (regset->core_note_type != NT_PRFPREG)
1573 					fill_note(&t->notes[i], "LINUX",
1574 						  regset->core_note_type,
1575 						  size, data);
1576 				else {
1577 					SET_PR_FPVALID(&t->prstatus, 1);
1578 					fill_note(&t->notes[i], "CORE",
1579 						  NT_PRFPREG, size, data);
1580 				}
1581 				*total += notesize(&t->notes[i]);
1582 			}
1583 		}
1584 	}
1585 
1586 	return 1;
1587 }
1588 
1589 static int fill_note_info(struct elfhdr *elf, int phdrs,
1590 			  struct elf_note_info *info,
1591 			  siginfo_t *siginfo, struct pt_regs *regs)
1592 {
1593 	struct task_struct *dump_task = current;
1594 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1595 	struct elf_thread_core_info *t;
1596 	struct elf_prpsinfo *psinfo;
1597 	struct core_thread *ct;
1598 	unsigned int i;
1599 
1600 	info->size = 0;
1601 	info->thread = NULL;
1602 
1603 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1604 	if (psinfo == NULL) {
1605 		info->psinfo.data = NULL; /* So we don't free this wrongly */
1606 		return 0;
1607 	}
1608 
1609 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1610 
1611 	/*
1612 	 * Figure out how many notes we're going to need for each thread.
1613 	 */
1614 	info->thread_notes = 0;
1615 	for (i = 0; i < view->n; ++i)
1616 		if (view->regsets[i].core_note_type != 0)
1617 			++info->thread_notes;
1618 
1619 	/*
1620 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1621 	 * since it is our one special case.
1622 	 */
1623 	if (unlikely(info->thread_notes == 0) ||
1624 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1625 		WARN_ON(1);
1626 		return 0;
1627 	}
1628 
1629 	/*
1630 	 * Initialize the ELF file header.
1631 	 */
1632 	fill_elf_header(elf, phdrs,
1633 			view->e_machine, view->e_flags, view->ei_osabi);
1634 
1635 	/*
1636 	 * Allocate a structure for each thread.
1637 	 */
1638 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1639 		t = kzalloc(offsetof(struct elf_thread_core_info,
1640 				     notes[info->thread_notes]),
1641 			    GFP_KERNEL);
1642 		if (unlikely(!t))
1643 			return 0;
1644 
1645 		t->task = ct->task;
1646 		if (ct->task == dump_task || !info->thread) {
1647 			t->next = info->thread;
1648 			info->thread = t;
1649 		} else {
1650 			/*
1651 			 * Make sure to keep the original task at
1652 			 * the head of the list.
1653 			 */
1654 			t->next = info->thread->next;
1655 			info->thread->next = t;
1656 		}
1657 	}
1658 
1659 	/*
1660 	 * Now fill in each thread's information.
1661 	 */
1662 	for (t = info->thread; t != NULL; t = t->next)
1663 		if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1664 			return 0;
1665 
1666 	/*
1667 	 * Fill in the two process-wide notes.
1668 	 */
1669 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1670 	info->size += notesize(&info->psinfo);
1671 
1672 	fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1673 	info->size += notesize(&info->signote);
1674 
1675 	fill_auxv_note(&info->auxv, current->mm);
1676 	info->size += notesize(&info->auxv);
1677 
1678 	fill_files_note(&info->files);
1679 	info->size += notesize(&info->files);
1680 
1681 	return 1;
1682 }
1683 
1684 static size_t get_note_info_size(struct elf_note_info *info)
1685 {
1686 	return info->size;
1687 }
1688 
1689 /*
1690  * Write all the notes for each thread.  When writing the first thread, the
1691  * process-wide notes are interleaved after the first thread-specific note.
1692  */
1693 static int write_note_info(struct elf_note_info *info,
1694 			   struct file *file, loff_t *foffset)
1695 {
1696 	bool first = 1;
1697 	struct elf_thread_core_info *t = info->thread;
1698 
1699 	do {
1700 		int i;
1701 
1702 		if (!writenote(&t->notes[0], file, foffset))
1703 			return 0;
1704 
1705 		if (first && !writenote(&info->psinfo, file, foffset))
1706 			return 0;
1707 		if (first && !writenote(&info->signote, file, foffset))
1708 			return 0;
1709 		if (first && !writenote(&info->auxv, file, foffset))
1710 			return 0;
1711 		if (first && !writenote(&info->files, file, foffset))
1712 			return 0;
1713 
1714 		for (i = 1; i < info->thread_notes; ++i)
1715 			if (t->notes[i].data &&
1716 			    !writenote(&t->notes[i], file, foffset))
1717 				return 0;
1718 
1719 		first = 0;
1720 		t = t->next;
1721 	} while (t);
1722 
1723 	return 1;
1724 }
1725 
1726 static void free_note_info(struct elf_note_info *info)
1727 {
1728 	struct elf_thread_core_info *threads = info->thread;
1729 	while (threads) {
1730 		unsigned int i;
1731 		struct elf_thread_core_info *t = threads;
1732 		threads = t->next;
1733 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1734 		for (i = 1; i < info->thread_notes; ++i)
1735 			kfree(t->notes[i].data);
1736 		kfree(t);
1737 	}
1738 	kfree(info->psinfo.data);
1739 	vfree(info->files.data);
1740 }
1741 
1742 #else
1743 
1744 /* Here is the structure in which status of each thread is captured. */
1745 struct elf_thread_status
1746 {
1747 	struct list_head list;
1748 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1749 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1750 	struct task_struct *thread;
1751 #ifdef ELF_CORE_COPY_XFPREGS
1752 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1753 #endif
1754 	struct memelfnote notes[3];
1755 	int num_notes;
1756 };
1757 
1758 /*
1759  * In order to add the specific thread information for the elf file format,
1760  * we need to keep a linked list of every threads pr_status and then create
1761  * a single section for them in the final core file.
1762  */
1763 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1764 {
1765 	int sz = 0;
1766 	struct task_struct *p = t->thread;
1767 	t->num_notes = 0;
1768 
1769 	fill_prstatus(&t->prstatus, p, signr);
1770 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1771 
1772 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1773 		  &(t->prstatus));
1774 	t->num_notes++;
1775 	sz += notesize(&t->notes[0]);
1776 
1777 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1778 								&t->fpu))) {
1779 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1780 			  &(t->fpu));
1781 		t->num_notes++;
1782 		sz += notesize(&t->notes[1]);
1783 	}
1784 
1785 #ifdef ELF_CORE_COPY_XFPREGS
1786 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1787 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1788 			  sizeof(t->xfpu), &t->xfpu);
1789 		t->num_notes++;
1790 		sz += notesize(&t->notes[2]);
1791 	}
1792 #endif
1793 	return sz;
1794 }
1795 
1796 struct elf_note_info {
1797 	struct memelfnote *notes;
1798 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1799 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1800 	struct list_head thread_list;
1801 	elf_fpregset_t *fpu;
1802 #ifdef ELF_CORE_COPY_XFPREGS
1803 	elf_fpxregset_t *xfpu;
1804 #endif
1805 	user_siginfo_t csigdata;
1806 	int thread_status_size;
1807 	int numnote;
1808 };
1809 
1810 static int elf_note_info_init(struct elf_note_info *info)
1811 {
1812 	memset(info, 0, sizeof(*info));
1813 	INIT_LIST_HEAD(&info->thread_list);
1814 
1815 	/* Allocate space for ELF notes */
1816 	info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1817 	if (!info->notes)
1818 		return 0;
1819 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1820 	if (!info->psinfo)
1821 		return 0;
1822 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1823 	if (!info->prstatus)
1824 		return 0;
1825 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1826 	if (!info->fpu)
1827 		return 0;
1828 #ifdef ELF_CORE_COPY_XFPREGS
1829 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1830 	if (!info->xfpu)
1831 		return 0;
1832 #endif
1833 	return 1;
1834 }
1835 
1836 static int fill_note_info(struct elfhdr *elf, int phdrs,
1837 			  struct elf_note_info *info,
1838 			  siginfo_t *siginfo, struct pt_regs *regs)
1839 {
1840 	struct list_head *t;
1841 
1842 	if (!elf_note_info_init(info))
1843 		return 0;
1844 
1845 	if (siginfo->si_signo) {
1846 		struct core_thread *ct;
1847 		struct elf_thread_status *ets;
1848 
1849 		for (ct = current->mm->core_state->dumper.next;
1850 						ct; ct = ct->next) {
1851 			ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1852 			if (!ets)
1853 				return 0;
1854 
1855 			ets->thread = ct->task;
1856 			list_add(&ets->list, &info->thread_list);
1857 		}
1858 
1859 		list_for_each(t, &info->thread_list) {
1860 			int sz;
1861 
1862 			ets = list_entry(t, struct elf_thread_status, list);
1863 			sz = elf_dump_thread_status(siginfo->si_signo, ets);
1864 			info->thread_status_size += sz;
1865 		}
1866 	}
1867 	/* now collect the dump for the current */
1868 	memset(info->prstatus, 0, sizeof(*info->prstatus));
1869 	fill_prstatus(info->prstatus, current, siginfo->si_signo);
1870 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1871 
1872 	/* Set up header */
1873 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1874 
1875 	/*
1876 	 * Set up the notes in similar form to SVR4 core dumps made
1877 	 * with info from their /proc.
1878 	 */
1879 
1880 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1881 		  sizeof(*info->prstatus), info->prstatus);
1882 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
1883 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1884 		  sizeof(*info->psinfo), info->psinfo);
1885 
1886 	fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1887 	fill_auxv_note(info->notes + 3, current->mm);
1888 	fill_files_note(info->notes + 4);
1889 
1890 	info->numnote = 5;
1891 
1892 	/* Try to dump the FPU. */
1893 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1894 							       info->fpu);
1895 	if (info->prstatus->pr_fpvalid)
1896 		fill_note(info->notes + info->numnote++,
1897 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1898 #ifdef ELF_CORE_COPY_XFPREGS
1899 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
1900 		fill_note(info->notes + info->numnote++,
1901 			  "LINUX", ELF_CORE_XFPREG_TYPE,
1902 			  sizeof(*info->xfpu), info->xfpu);
1903 #endif
1904 
1905 	return 1;
1906 }
1907 
1908 static size_t get_note_info_size(struct elf_note_info *info)
1909 {
1910 	int sz = 0;
1911 	int i;
1912 
1913 	for (i = 0; i < info->numnote; i++)
1914 		sz += notesize(info->notes + i);
1915 
1916 	sz += info->thread_status_size;
1917 
1918 	return sz;
1919 }
1920 
1921 static int write_note_info(struct elf_note_info *info,
1922 			   struct file *file, loff_t *foffset)
1923 {
1924 	int i;
1925 	struct list_head *t;
1926 
1927 	for (i = 0; i < info->numnote; i++)
1928 		if (!writenote(info->notes + i, file, foffset))
1929 			return 0;
1930 
1931 	/* write out the thread status notes section */
1932 	list_for_each(t, &info->thread_list) {
1933 		struct elf_thread_status *tmp =
1934 				list_entry(t, struct elf_thread_status, list);
1935 
1936 		for (i = 0; i < tmp->num_notes; i++)
1937 			if (!writenote(&tmp->notes[i], file, foffset))
1938 				return 0;
1939 	}
1940 
1941 	return 1;
1942 }
1943 
1944 static void free_note_info(struct elf_note_info *info)
1945 {
1946 	while (!list_empty(&info->thread_list)) {
1947 		struct list_head *tmp = info->thread_list.next;
1948 		list_del(tmp);
1949 		kfree(list_entry(tmp, struct elf_thread_status, list));
1950 	}
1951 
1952 	/* Free data allocated by fill_files_note(): */
1953 	vfree(info->notes[4].data);
1954 
1955 	kfree(info->prstatus);
1956 	kfree(info->psinfo);
1957 	kfree(info->notes);
1958 	kfree(info->fpu);
1959 #ifdef ELF_CORE_COPY_XFPREGS
1960 	kfree(info->xfpu);
1961 #endif
1962 }
1963 
1964 #endif
1965 
1966 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1967 					struct vm_area_struct *gate_vma)
1968 {
1969 	struct vm_area_struct *ret = tsk->mm->mmap;
1970 
1971 	if (ret)
1972 		return ret;
1973 	return gate_vma;
1974 }
1975 /*
1976  * Helper function for iterating across a vma list.  It ensures that the caller
1977  * will visit `gate_vma' prior to terminating the search.
1978  */
1979 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1980 					struct vm_area_struct *gate_vma)
1981 {
1982 	struct vm_area_struct *ret;
1983 
1984 	ret = this_vma->vm_next;
1985 	if (ret)
1986 		return ret;
1987 	if (this_vma == gate_vma)
1988 		return NULL;
1989 	return gate_vma;
1990 }
1991 
1992 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1993 			     elf_addr_t e_shoff, int segs)
1994 {
1995 	elf->e_shoff = e_shoff;
1996 	elf->e_shentsize = sizeof(*shdr4extnum);
1997 	elf->e_shnum = 1;
1998 	elf->e_shstrndx = SHN_UNDEF;
1999 
2000 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2001 
2002 	shdr4extnum->sh_type = SHT_NULL;
2003 	shdr4extnum->sh_size = elf->e_shnum;
2004 	shdr4extnum->sh_link = elf->e_shstrndx;
2005 	shdr4extnum->sh_info = segs;
2006 }
2007 
2008 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
2009 				     unsigned long mm_flags)
2010 {
2011 	struct vm_area_struct *vma;
2012 	size_t size = 0;
2013 
2014 	for (vma = first_vma(current, gate_vma); vma != NULL;
2015 	     vma = next_vma(vma, gate_vma))
2016 		size += vma_dump_size(vma, mm_flags);
2017 	return size;
2018 }
2019 
2020 /*
2021  * Actual dumper
2022  *
2023  * This is a two-pass process; first we find the offsets of the bits,
2024  * and then they are actually written out.  If we run out of core limit
2025  * we just truncate.
2026  */
2027 static int elf_core_dump(struct coredump_params *cprm)
2028 {
2029 	int has_dumped = 0;
2030 	mm_segment_t fs;
2031 	int segs;
2032 	size_t size = 0;
2033 	struct vm_area_struct *vma, *gate_vma;
2034 	struct elfhdr *elf = NULL;
2035 	loff_t offset = 0, dataoff, foffset;
2036 	struct elf_note_info info;
2037 	struct elf_phdr *phdr4note = NULL;
2038 	struct elf_shdr *shdr4extnum = NULL;
2039 	Elf_Half e_phnum;
2040 	elf_addr_t e_shoff;
2041 
2042 	/*
2043 	 * We no longer stop all VM operations.
2044 	 *
2045 	 * This is because those proceses that could possibly change map_count
2046 	 * or the mmap / vma pages are now blocked in do_exit on current
2047 	 * finishing this core dump.
2048 	 *
2049 	 * Only ptrace can touch these memory addresses, but it doesn't change
2050 	 * the map_count or the pages allocated. So no possibility of crashing
2051 	 * exists while dumping the mm->vm_next areas to the core file.
2052 	 */
2053 
2054 	/* alloc memory for large data structures: too large to be on stack */
2055 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2056 	if (!elf)
2057 		goto out;
2058 	/*
2059 	 * The number of segs are recored into ELF header as 16bit value.
2060 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2061 	 */
2062 	segs = current->mm->map_count;
2063 	segs += elf_core_extra_phdrs();
2064 
2065 	gate_vma = get_gate_vma(current->mm);
2066 	if (gate_vma != NULL)
2067 		segs++;
2068 
2069 	/* for notes section */
2070 	segs++;
2071 
2072 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2073 	 * this, kernel supports extended numbering. Have a look at
2074 	 * include/linux/elf.h for further information. */
2075 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2076 
2077 	/*
2078 	 * Collect all the non-memory information about the process for the
2079 	 * notes.  This also sets up the file header.
2080 	 */
2081 	if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2082 		goto cleanup;
2083 
2084 	has_dumped = 1;
2085 	current->flags |= PF_DUMPCORE;
2086 
2087 	fs = get_fs();
2088 	set_fs(KERNEL_DS);
2089 
2090 	offset += sizeof(*elf);				/* Elf header */
2091 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2092 	foffset = offset;
2093 
2094 	/* Write notes phdr entry */
2095 	{
2096 		size_t sz = get_note_info_size(&info);
2097 
2098 		sz += elf_coredump_extra_notes_size();
2099 
2100 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2101 		if (!phdr4note)
2102 			goto end_coredump;
2103 
2104 		fill_elf_note_phdr(phdr4note, sz, offset);
2105 		offset += sz;
2106 	}
2107 
2108 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2109 
2110 	offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2111 	offset += elf_core_extra_data_size();
2112 	e_shoff = offset;
2113 
2114 	if (e_phnum == PN_XNUM) {
2115 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2116 		if (!shdr4extnum)
2117 			goto end_coredump;
2118 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2119 	}
2120 
2121 	offset = dataoff;
2122 
2123 	size += sizeof(*elf);
2124 	if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
2125 		goto end_coredump;
2126 
2127 	size += sizeof(*phdr4note);
2128 	if (size > cprm->limit
2129 	    || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
2130 		goto end_coredump;
2131 
2132 	/* Write program headers for segments dump */
2133 	for (vma = first_vma(current, gate_vma); vma != NULL;
2134 			vma = next_vma(vma, gate_vma)) {
2135 		struct elf_phdr phdr;
2136 
2137 		phdr.p_type = PT_LOAD;
2138 		phdr.p_offset = offset;
2139 		phdr.p_vaddr = vma->vm_start;
2140 		phdr.p_paddr = 0;
2141 		phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2142 		phdr.p_memsz = vma->vm_end - vma->vm_start;
2143 		offset += phdr.p_filesz;
2144 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2145 		if (vma->vm_flags & VM_WRITE)
2146 			phdr.p_flags |= PF_W;
2147 		if (vma->vm_flags & VM_EXEC)
2148 			phdr.p_flags |= PF_X;
2149 		phdr.p_align = ELF_EXEC_PAGESIZE;
2150 
2151 		size += sizeof(phdr);
2152 		if (size > cprm->limit
2153 		    || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2154 			goto end_coredump;
2155 	}
2156 
2157 	if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2158 		goto end_coredump;
2159 
2160  	/* write out the notes section */
2161 	if (!write_note_info(&info, cprm->file, &foffset))
2162 		goto end_coredump;
2163 
2164 	if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2165 		goto end_coredump;
2166 
2167 	/* Align to page */
2168 	if (!dump_seek(cprm->file, dataoff - foffset))
2169 		goto end_coredump;
2170 
2171 	for (vma = first_vma(current, gate_vma); vma != NULL;
2172 			vma = next_vma(vma, gate_vma)) {
2173 		unsigned long addr;
2174 		unsigned long end;
2175 
2176 		end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2177 
2178 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2179 			struct page *page;
2180 			int stop;
2181 
2182 			page = get_dump_page(addr);
2183 			if (page) {
2184 				void *kaddr = kmap(page);
2185 				stop = ((size += PAGE_SIZE) > cprm->limit) ||
2186 					!dump_write(cprm->file, kaddr,
2187 						    PAGE_SIZE);
2188 				kunmap(page);
2189 				page_cache_release(page);
2190 			} else
2191 				stop = !dump_seek(cprm->file, PAGE_SIZE);
2192 			if (stop)
2193 				goto end_coredump;
2194 		}
2195 	}
2196 
2197 	if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2198 		goto end_coredump;
2199 
2200 	if (e_phnum == PN_XNUM) {
2201 		size += sizeof(*shdr4extnum);
2202 		if (size > cprm->limit
2203 		    || !dump_write(cprm->file, shdr4extnum,
2204 				   sizeof(*shdr4extnum)))
2205 			goto end_coredump;
2206 	}
2207 
2208 end_coredump:
2209 	set_fs(fs);
2210 
2211 cleanup:
2212 	free_note_info(&info);
2213 	kfree(shdr4extnum);
2214 	kfree(phdr4note);
2215 	kfree(elf);
2216 out:
2217 	return has_dumped;
2218 }
2219 
2220 #endif		/* CONFIG_ELF_CORE */
2221 
2222 static int __init init_elf_binfmt(void)
2223 {
2224 	register_binfmt(&elf_format);
2225 	return 0;
2226 }
2227 
2228 static void __exit exit_elf_binfmt(void)
2229 {
2230 	/* Remove the COFF and ELF loaders. */
2231 	unregister_binfmt(&elf_format);
2232 }
2233 
2234 core_initcall(init_elf_binfmt);
2235 module_exit(exit_elf_binfmt);
2236 MODULE_LICENSE("GPL");
2237