xref: /openbmc/linux/fs/binfmt_elf.c (revision d2168146)
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 unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
50 				int, int, unsigned long);
51 
52 #ifdef CONFIG_USELIB
53 static int load_elf_library(struct file *);
54 #else
55 #define load_elf_library NULL
56 #endif
57 
58 /*
59  * If we don't support core dumping, then supply a NULL so we
60  * don't even try.
61  */
62 #ifdef CONFIG_ELF_CORE
63 static int elf_core_dump(struct coredump_params *cprm);
64 #else
65 #define elf_core_dump	NULL
66 #endif
67 
68 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
69 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
70 #else
71 #define ELF_MIN_ALIGN	PAGE_SIZE
72 #endif
73 
74 #ifndef ELF_CORE_EFLAGS
75 #define ELF_CORE_EFLAGS	0
76 #endif
77 
78 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
79 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
80 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
81 
82 static struct linux_binfmt elf_format = {
83 	.module		= THIS_MODULE,
84 	.load_binary	= load_elf_binary,
85 	.load_shlib	= load_elf_library,
86 	.core_dump	= elf_core_dump,
87 	.min_coredump	= ELF_EXEC_PAGESIZE,
88 };
89 
90 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
91 
92 static int set_brk(unsigned long start, unsigned long end)
93 {
94 	start = ELF_PAGEALIGN(start);
95 	end = ELF_PAGEALIGN(end);
96 	if (end > start) {
97 		unsigned long addr;
98 		addr = vm_brk(start, end - start);
99 		if (BAD_ADDR(addr))
100 			return addr;
101 	}
102 	current->mm->start_brk = current->mm->brk = end;
103 	return 0;
104 }
105 
106 /* We need to explicitly zero any fractional pages
107    after the data section (i.e. bss).  This would
108    contain the junk from the file that should not
109    be in memory
110  */
111 static int padzero(unsigned long elf_bss)
112 {
113 	unsigned long nbyte;
114 
115 	nbyte = ELF_PAGEOFFSET(elf_bss);
116 	if (nbyte) {
117 		nbyte = ELF_MIN_ALIGN - nbyte;
118 		if (clear_user((void __user *) elf_bss, nbyte))
119 			return -EFAULT;
120 	}
121 	return 0;
122 }
123 
124 /* Let's use some macros to make this stack manipulation a little clearer */
125 #ifdef CONFIG_STACK_GROWSUP
126 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
127 #define STACK_ROUND(sp, items) \
128 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
129 #define STACK_ALLOC(sp, len) ({ \
130 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
131 	old_sp; })
132 #else
133 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
134 #define STACK_ROUND(sp, items) \
135 	(((unsigned long) (sp - items)) &~ 15UL)
136 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
137 #endif
138 
139 #ifndef ELF_BASE_PLATFORM
140 /*
141  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
142  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
143  * will be copied to the user stack in the same manner as AT_PLATFORM.
144  */
145 #define ELF_BASE_PLATFORM NULL
146 #endif
147 
148 static int
149 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
150 		unsigned long load_addr, unsigned long interp_load_addr)
151 {
152 	unsigned long p = bprm->p;
153 	int argc = bprm->argc;
154 	int envc = bprm->envc;
155 	elf_addr_t __user *argv;
156 	elf_addr_t __user *envp;
157 	elf_addr_t __user *sp;
158 	elf_addr_t __user *u_platform;
159 	elf_addr_t __user *u_base_platform;
160 	elf_addr_t __user *u_rand_bytes;
161 	const char *k_platform = ELF_PLATFORM;
162 	const char *k_base_platform = ELF_BASE_PLATFORM;
163 	unsigned char k_rand_bytes[16];
164 	int items;
165 	elf_addr_t *elf_info;
166 	int ei_index = 0;
167 	const struct cred *cred = current_cred();
168 	struct vm_area_struct *vma;
169 
170 	/*
171 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
172 	 * evictions by the processes running on the same package. One
173 	 * thing we can do is to shuffle the initial stack for them.
174 	 */
175 
176 	p = arch_align_stack(p);
177 
178 	/*
179 	 * If this architecture has a platform capability string, copy it
180 	 * to userspace.  In some cases (Sparc), this info is impossible
181 	 * for userspace to get any other way, in others (i386) it is
182 	 * merely difficult.
183 	 */
184 	u_platform = NULL;
185 	if (k_platform) {
186 		size_t len = strlen(k_platform) + 1;
187 
188 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
189 		if (__copy_to_user(u_platform, k_platform, len))
190 			return -EFAULT;
191 	}
192 
193 	/*
194 	 * If this architecture has a "base" platform capability
195 	 * string, copy it to userspace.
196 	 */
197 	u_base_platform = NULL;
198 	if (k_base_platform) {
199 		size_t len = strlen(k_base_platform) + 1;
200 
201 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
202 		if (__copy_to_user(u_base_platform, k_base_platform, len))
203 			return -EFAULT;
204 	}
205 
206 	/*
207 	 * Generate 16 random bytes for userspace PRNG seeding.
208 	 */
209 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
210 	u_rand_bytes = (elf_addr_t __user *)
211 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
212 	if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
213 		return -EFAULT;
214 
215 	/* Create the ELF interpreter info */
216 	elf_info = (elf_addr_t *)current->mm->saved_auxv;
217 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
218 #define NEW_AUX_ENT(id, val) \
219 	do { \
220 		elf_info[ei_index++] = id; \
221 		elf_info[ei_index++] = val; \
222 	} while (0)
223 
224 #ifdef ARCH_DLINFO
225 	/*
226 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
227 	 * AUXV.
228 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
229 	 * ARCH_DLINFO changes
230 	 */
231 	ARCH_DLINFO;
232 #endif
233 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
234 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
235 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
236 	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
237 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
238 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
239 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
240 	NEW_AUX_ENT(AT_FLAGS, 0);
241 	NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
242 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
243 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
244 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
245 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
246  	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
247 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
248 #ifdef ELF_HWCAP2
249 	NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
250 #endif
251 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
252 	if (k_platform) {
253 		NEW_AUX_ENT(AT_PLATFORM,
254 			    (elf_addr_t)(unsigned long)u_platform);
255 	}
256 	if (k_base_platform) {
257 		NEW_AUX_ENT(AT_BASE_PLATFORM,
258 			    (elf_addr_t)(unsigned long)u_base_platform);
259 	}
260 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
261 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
262 	}
263 #undef NEW_AUX_ENT
264 	/* AT_NULL is zero; clear the rest too */
265 	memset(&elf_info[ei_index], 0,
266 	       sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
267 
268 	/* And advance past the AT_NULL entry.  */
269 	ei_index += 2;
270 
271 	sp = STACK_ADD(p, ei_index);
272 
273 	items = (argc + 1) + (envc + 1) + 1;
274 	bprm->p = STACK_ROUND(sp, items);
275 
276 	/* Point sp at the lowest address on the stack */
277 #ifdef CONFIG_STACK_GROWSUP
278 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
279 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
280 #else
281 	sp = (elf_addr_t __user *)bprm->p;
282 #endif
283 
284 
285 	/*
286 	 * Grow the stack manually; some architectures have a limit on how
287 	 * far ahead a user-space access may be in order to grow the stack.
288 	 */
289 	vma = find_extend_vma(current->mm, bprm->p);
290 	if (!vma)
291 		return -EFAULT;
292 
293 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
294 	if (__put_user(argc, sp++))
295 		return -EFAULT;
296 	argv = sp;
297 	envp = argv + argc + 1;
298 
299 	/* Populate argv and envp */
300 	p = current->mm->arg_end = current->mm->arg_start;
301 	while (argc-- > 0) {
302 		size_t len;
303 		if (__put_user((elf_addr_t)p, argv++))
304 			return -EFAULT;
305 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
306 		if (!len || len > MAX_ARG_STRLEN)
307 			return -EINVAL;
308 		p += len;
309 	}
310 	if (__put_user(0, argv))
311 		return -EFAULT;
312 	current->mm->arg_end = current->mm->env_start = p;
313 	while (envc-- > 0) {
314 		size_t len;
315 		if (__put_user((elf_addr_t)p, envp++))
316 			return -EFAULT;
317 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
318 		if (!len || len > MAX_ARG_STRLEN)
319 			return -EINVAL;
320 		p += len;
321 	}
322 	if (__put_user(0, envp))
323 		return -EFAULT;
324 	current->mm->env_end = p;
325 
326 	/* Put the elf_info on the stack in the right place.  */
327 	sp = (elf_addr_t __user *)envp + 1;
328 	if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
329 		return -EFAULT;
330 	return 0;
331 }
332 
333 #ifndef elf_map
334 
335 static unsigned long elf_map(struct file *filep, unsigned long addr,
336 		struct elf_phdr *eppnt, int prot, int type,
337 		unsigned long total_size)
338 {
339 	unsigned long map_addr;
340 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
341 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
342 	addr = ELF_PAGESTART(addr);
343 	size = ELF_PAGEALIGN(size);
344 
345 	/* mmap() will return -EINVAL if given a zero size, but a
346 	 * segment with zero filesize is perfectly valid */
347 	if (!size)
348 		return addr;
349 
350 	/*
351 	* total_size is the size of the ELF (interpreter) image.
352 	* The _first_ mmap needs to know the full size, otherwise
353 	* randomization might put this image into an overlapping
354 	* position with the ELF binary image. (since size < total_size)
355 	* So we first map the 'big' image - and unmap the remainder at
356 	* the end. (which unmap is needed for ELF images with holes.)
357 	*/
358 	if (total_size) {
359 		total_size = ELF_PAGEALIGN(total_size);
360 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
361 		if (!BAD_ADDR(map_addr))
362 			vm_munmap(map_addr+size, total_size-size);
363 	} else
364 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
365 
366 	return(map_addr);
367 }
368 
369 #endif /* !elf_map */
370 
371 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
372 {
373 	int i, first_idx = -1, last_idx = -1;
374 
375 	for (i = 0; i < nr; i++) {
376 		if (cmds[i].p_type == PT_LOAD) {
377 			last_idx = i;
378 			if (first_idx == -1)
379 				first_idx = i;
380 		}
381 	}
382 	if (first_idx == -1)
383 		return 0;
384 
385 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
386 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
387 }
388 
389 
390 /* This is much more generalized than the library routine read function,
391    so we keep this separate.  Technically the library read function
392    is only provided so that we can read a.out libraries that have
393    an ELF header */
394 
395 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
396 		struct file *interpreter, unsigned long *interp_map_addr,
397 		unsigned long no_base)
398 {
399 	struct elf_phdr *elf_phdata;
400 	struct elf_phdr *eppnt;
401 	unsigned long load_addr = 0;
402 	int load_addr_set = 0;
403 	unsigned long last_bss = 0, elf_bss = 0;
404 	unsigned long error = ~0UL;
405 	unsigned long total_size;
406 	int retval, i, size;
407 
408 	/* First of all, some simple consistency checks */
409 	if (interp_elf_ex->e_type != ET_EXEC &&
410 	    interp_elf_ex->e_type != ET_DYN)
411 		goto out;
412 	if (!elf_check_arch(interp_elf_ex))
413 		goto out;
414 	if (!interpreter->f_op->mmap)
415 		goto out;
416 
417 	/*
418 	 * If the size of this structure has changed, then punt, since
419 	 * we will be doing the wrong thing.
420 	 */
421 	if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
422 		goto out;
423 	if (interp_elf_ex->e_phnum < 1 ||
424 		interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
425 		goto out;
426 
427 	/* Now read in all of the header information */
428 	size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
429 	if (size > ELF_MIN_ALIGN)
430 		goto out;
431 	elf_phdata = kmalloc(size, GFP_KERNEL);
432 	if (!elf_phdata)
433 		goto out;
434 
435 	retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
436 			     (char *)elf_phdata, size);
437 	error = -EIO;
438 	if (retval != size) {
439 		if (retval < 0)
440 			error = retval;
441 		goto out_close;
442 	}
443 
444 	total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
445 	if (!total_size) {
446 		error = -EINVAL;
447 		goto out_close;
448 	}
449 
450 	eppnt = elf_phdata;
451 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
452 		if (eppnt->p_type == PT_LOAD) {
453 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
454 			int elf_prot = 0;
455 			unsigned long vaddr = 0;
456 			unsigned long k, map_addr;
457 
458 			if (eppnt->p_flags & PF_R)
459 		    		elf_prot = PROT_READ;
460 			if (eppnt->p_flags & PF_W)
461 				elf_prot |= PROT_WRITE;
462 			if (eppnt->p_flags & PF_X)
463 				elf_prot |= PROT_EXEC;
464 			vaddr = eppnt->p_vaddr;
465 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
466 				elf_type |= MAP_FIXED;
467 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
468 				load_addr = -vaddr;
469 
470 			map_addr = elf_map(interpreter, load_addr + vaddr,
471 					eppnt, elf_prot, elf_type, total_size);
472 			total_size = 0;
473 			if (!*interp_map_addr)
474 				*interp_map_addr = map_addr;
475 			error = map_addr;
476 			if (BAD_ADDR(map_addr))
477 				goto out_close;
478 
479 			if (!load_addr_set &&
480 			    interp_elf_ex->e_type == ET_DYN) {
481 				load_addr = map_addr - ELF_PAGESTART(vaddr);
482 				load_addr_set = 1;
483 			}
484 
485 			/*
486 			 * Check to see if the section's size will overflow the
487 			 * allowed task size. Note that p_filesz must always be
488 			 * <= p_memsize so it's only necessary to check p_memsz.
489 			 */
490 			k = load_addr + eppnt->p_vaddr;
491 			if (BAD_ADDR(k) ||
492 			    eppnt->p_filesz > eppnt->p_memsz ||
493 			    eppnt->p_memsz > TASK_SIZE ||
494 			    TASK_SIZE - eppnt->p_memsz < k) {
495 				error = -ENOMEM;
496 				goto out_close;
497 			}
498 
499 			/*
500 			 * Find the end of the file mapping for this phdr, and
501 			 * keep track of the largest address we see for this.
502 			 */
503 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
504 			if (k > elf_bss)
505 				elf_bss = k;
506 
507 			/*
508 			 * Do the same thing for the memory mapping - between
509 			 * elf_bss and last_bss is the bss section.
510 			 */
511 			k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
512 			if (k > last_bss)
513 				last_bss = k;
514 		}
515 	}
516 
517 	if (last_bss > elf_bss) {
518 		/*
519 		 * Now fill out the bss section.  First pad the last page up
520 		 * to the page boundary, and then perform a mmap to make sure
521 		 * that there are zero-mapped pages up to and including the
522 		 * last bss page.
523 		 */
524 		if (padzero(elf_bss)) {
525 			error = -EFAULT;
526 			goto out_close;
527 		}
528 
529 		/* What we have mapped so far */
530 		elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
531 
532 		/* Map the last of the bss segment */
533 		error = vm_brk(elf_bss, last_bss - elf_bss);
534 		if (BAD_ADDR(error))
535 			goto out_close;
536 	}
537 
538 	error = load_addr;
539 
540 out_close:
541 	kfree(elf_phdata);
542 out:
543 	return error;
544 }
545 
546 /*
547  * These are the functions used to load ELF style executables and shared
548  * libraries.  There is no binary dependent code anywhere else.
549  */
550 
551 #ifndef STACK_RND_MASK
552 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))	/* 8MB of VA */
553 #endif
554 
555 static unsigned long randomize_stack_top(unsigned long stack_top)
556 {
557 	unsigned int random_variable = 0;
558 
559 	if ((current->flags & PF_RANDOMIZE) &&
560 		!(current->personality & ADDR_NO_RANDOMIZE)) {
561 		random_variable = get_random_int() & STACK_RND_MASK;
562 		random_variable <<= PAGE_SHIFT;
563 	}
564 #ifdef CONFIG_STACK_GROWSUP
565 	return PAGE_ALIGN(stack_top) + random_variable;
566 #else
567 	return PAGE_ALIGN(stack_top) - random_variable;
568 #endif
569 }
570 
571 static int load_elf_binary(struct linux_binprm *bprm)
572 {
573 	struct file *interpreter = NULL; /* to shut gcc up */
574  	unsigned long load_addr = 0, load_bias = 0;
575 	int load_addr_set = 0;
576 	char * elf_interpreter = NULL;
577 	unsigned long error;
578 	struct elf_phdr *elf_ppnt, *elf_phdata;
579 	unsigned long elf_bss, elf_brk;
580 	int retval, i;
581 	unsigned int size;
582 	unsigned long elf_entry;
583 	unsigned long interp_load_addr = 0;
584 	unsigned long start_code, end_code, start_data, end_data;
585 	unsigned long reloc_func_desc __maybe_unused = 0;
586 	int executable_stack = EXSTACK_DEFAULT;
587 	struct pt_regs *regs = current_pt_regs();
588 	struct {
589 		struct elfhdr elf_ex;
590 		struct elfhdr interp_elf_ex;
591 	} *loc;
592 
593 	loc = kmalloc(sizeof(*loc), GFP_KERNEL);
594 	if (!loc) {
595 		retval = -ENOMEM;
596 		goto out_ret;
597 	}
598 
599 	/* Get the exec-header */
600 	loc->elf_ex = *((struct elfhdr *)bprm->buf);
601 
602 	retval = -ENOEXEC;
603 	/* First of all, some simple consistency checks */
604 	if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
605 		goto out;
606 
607 	if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
608 		goto out;
609 	if (!elf_check_arch(&loc->elf_ex))
610 		goto out;
611 	if (!bprm->file->f_op->mmap)
612 		goto out;
613 
614 	/* Now read in all of the header information */
615 	if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
616 		goto out;
617 	if (loc->elf_ex.e_phnum < 1 ||
618 	 	loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
619 		goto out;
620 	size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
621 	retval = -ENOMEM;
622 	elf_phdata = kmalloc(size, GFP_KERNEL);
623 	if (!elf_phdata)
624 		goto out;
625 
626 	retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
627 			     (char *)elf_phdata, size);
628 	if (retval != size) {
629 		if (retval >= 0)
630 			retval = -EIO;
631 		goto out_free_ph;
632 	}
633 
634 	elf_ppnt = elf_phdata;
635 	elf_bss = 0;
636 	elf_brk = 0;
637 
638 	start_code = ~0UL;
639 	end_code = 0;
640 	start_data = 0;
641 	end_data = 0;
642 
643 	for (i = 0; i < loc->elf_ex.e_phnum; i++) {
644 		if (elf_ppnt->p_type == PT_INTERP) {
645 			/* This is the program interpreter used for
646 			 * shared libraries - for now assume that this
647 			 * is an a.out format binary
648 			 */
649 			retval = -ENOEXEC;
650 			if (elf_ppnt->p_filesz > PATH_MAX ||
651 			    elf_ppnt->p_filesz < 2)
652 				goto out_free_ph;
653 
654 			retval = -ENOMEM;
655 			elf_interpreter = kmalloc(elf_ppnt->p_filesz,
656 						  GFP_KERNEL);
657 			if (!elf_interpreter)
658 				goto out_free_ph;
659 
660 			retval = kernel_read(bprm->file, elf_ppnt->p_offset,
661 					     elf_interpreter,
662 					     elf_ppnt->p_filesz);
663 			if (retval != elf_ppnt->p_filesz) {
664 				if (retval >= 0)
665 					retval = -EIO;
666 				goto out_free_interp;
667 			}
668 			/* make sure path is NULL terminated */
669 			retval = -ENOEXEC;
670 			if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
671 				goto out_free_interp;
672 
673 			interpreter = open_exec(elf_interpreter);
674 			retval = PTR_ERR(interpreter);
675 			if (IS_ERR(interpreter))
676 				goto out_free_interp;
677 
678 			/*
679 			 * If the binary is not readable then enforce
680 			 * mm->dumpable = 0 regardless of the interpreter's
681 			 * permissions.
682 			 */
683 			would_dump(bprm, interpreter);
684 
685 			retval = kernel_read(interpreter, 0, bprm->buf,
686 					     BINPRM_BUF_SIZE);
687 			if (retval != BINPRM_BUF_SIZE) {
688 				if (retval >= 0)
689 					retval = -EIO;
690 				goto out_free_dentry;
691 			}
692 
693 			/* Get the exec headers */
694 			loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
695 			break;
696 		}
697 		elf_ppnt++;
698 	}
699 
700 	elf_ppnt = elf_phdata;
701 	for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
702 		if (elf_ppnt->p_type == PT_GNU_STACK) {
703 			if (elf_ppnt->p_flags & PF_X)
704 				executable_stack = EXSTACK_ENABLE_X;
705 			else
706 				executable_stack = EXSTACK_DISABLE_X;
707 			break;
708 		}
709 
710 	/* Some simple consistency checks for the interpreter */
711 	if (elf_interpreter) {
712 		retval = -ELIBBAD;
713 		/* Not an ELF interpreter */
714 		if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
715 			goto out_free_dentry;
716 		/* Verify the interpreter has a valid arch */
717 		if (!elf_check_arch(&loc->interp_elf_ex))
718 			goto out_free_dentry;
719 	}
720 
721 	/* Flush all traces of the currently running executable */
722 	retval = flush_old_exec(bprm);
723 	if (retval)
724 		goto out_free_dentry;
725 
726 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
727 	   may depend on the personality.  */
728 	SET_PERSONALITY(loc->elf_ex);
729 	if (elf_read_implies_exec(loc->elf_ex, executable_stack))
730 		current->personality |= READ_IMPLIES_EXEC;
731 
732 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
733 		current->flags |= PF_RANDOMIZE;
734 
735 	setup_new_exec(bprm);
736 
737 	/* Do this so that we can load the interpreter, if need be.  We will
738 	   change some of these later */
739 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
740 				 executable_stack);
741 	if (retval < 0) {
742 		send_sig(SIGKILL, current, 0);
743 		goto out_free_dentry;
744 	}
745 
746 	current->mm->start_stack = bprm->p;
747 
748 	/* Now we do a little grungy work by mmapping the ELF image into
749 	   the correct location in memory. */
750 	for(i = 0, elf_ppnt = elf_phdata;
751 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
752 		int elf_prot = 0, elf_flags;
753 		unsigned long k, vaddr;
754 
755 		if (elf_ppnt->p_type != PT_LOAD)
756 			continue;
757 
758 		if (unlikely (elf_brk > elf_bss)) {
759 			unsigned long nbyte;
760 
761 			/* There was a PT_LOAD segment with p_memsz > p_filesz
762 			   before this one. Map anonymous pages, if needed,
763 			   and clear the area.  */
764 			retval = set_brk(elf_bss + load_bias,
765 					 elf_brk + load_bias);
766 			if (retval) {
767 				send_sig(SIGKILL, current, 0);
768 				goto out_free_dentry;
769 			}
770 			nbyte = ELF_PAGEOFFSET(elf_bss);
771 			if (nbyte) {
772 				nbyte = ELF_MIN_ALIGN - nbyte;
773 				if (nbyte > elf_brk - elf_bss)
774 					nbyte = elf_brk - elf_bss;
775 				if (clear_user((void __user *)elf_bss +
776 							load_bias, nbyte)) {
777 					/*
778 					 * This bss-zeroing can fail if the ELF
779 					 * file specifies odd protections. So
780 					 * we don't check the return value
781 					 */
782 				}
783 			}
784 		}
785 
786 		if (elf_ppnt->p_flags & PF_R)
787 			elf_prot |= PROT_READ;
788 		if (elf_ppnt->p_flags & PF_W)
789 			elf_prot |= PROT_WRITE;
790 		if (elf_ppnt->p_flags & PF_X)
791 			elf_prot |= PROT_EXEC;
792 
793 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
794 
795 		vaddr = elf_ppnt->p_vaddr;
796 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
797 			elf_flags |= MAP_FIXED;
798 		} else if (loc->elf_ex.e_type == ET_DYN) {
799 			/* Try and get dynamic programs out of the way of the
800 			 * default mmap base, as well as whatever program they
801 			 * might try to exec.  This is because the brk will
802 			 * follow the loader, and is not movable.  */
803 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
804 			/* Memory randomization might have been switched off
805 			 * in runtime via sysctl or explicit setting of
806 			 * personality flags.
807 			 * If that is the case, retain the original non-zero
808 			 * load_bias value in order to establish proper
809 			 * non-randomized mappings.
810 			 */
811 			if (current->flags & PF_RANDOMIZE)
812 				load_bias = 0;
813 			else
814 				load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
815 #else
816 			load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
817 #endif
818 		}
819 
820 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
821 				elf_prot, elf_flags, 0);
822 		if (BAD_ADDR(error)) {
823 			send_sig(SIGKILL, current, 0);
824 			retval = IS_ERR((void *)error) ?
825 				PTR_ERR((void*)error) : -EINVAL;
826 			goto out_free_dentry;
827 		}
828 
829 		if (!load_addr_set) {
830 			load_addr_set = 1;
831 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
832 			if (loc->elf_ex.e_type == ET_DYN) {
833 				load_bias += error -
834 				             ELF_PAGESTART(load_bias + vaddr);
835 				load_addr += load_bias;
836 				reloc_func_desc = load_bias;
837 			}
838 		}
839 		k = elf_ppnt->p_vaddr;
840 		if (k < start_code)
841 			start_code = k;
842 		if (start_data < k)
843 			start_data = k;
844 
845 		/*
846 		 * Check to see if the section's size will overflow the
847 		 * allowed task size. Note that p_filesz must always be
848 		 * <= p_memsz so it is only necessary to check p_memsz.
849 		 */
850 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
851 		    elf_ppnt->p_memsz > TASK_SIZE ||
852 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
853 			/* set_brk can never work. Avoid overflows. */
854 			send_sig(SIGKILL, current, 0);
855 			retval = -EINVAL;
856 			goto out_free_dentry;
857 		}
858 
859 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
860 
861 		if (k > elf_bss)
862 			elf_bss = k;
863 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
864 			end_code = k;
865 		if (end_data < k)
866 			end_data = k;
867 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
868 		if (k > elf_brk)
869 			elf_brk = k;
870 	}
871 
872 	loc->elf_ex.e_entry += load_bias;
873 	elf_bss += load_bias;
874 	elf_brk += load_bias;
875 	start_code += load_bias;
876 	end_code += load_bias;
877 	start_data += load_bias;
878 	end_data += load_bias;
879 
880 	/* Calling set_brk effectively mmaps the pages that we need
881 	 * for the bss and break sections.  We must do this before
882 	 * mapping in the interpreter, to make sure it doesn't wind
883 	 * up getting placed where the bss needs to go.
884 	 */
885 	retval = set_brk(elf_bss, elf_brk);
886 	if (retval) {
887 		send_sig(SIGKILL, current, 0);
888 		goto out_free_dentry;
889 	}
890 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
891 		send_sig(SIGSEGV, current, 0);
892 		retval = -EFAULT; /* Nobody gets to see this, but.. */
893 		goto out_free_dentry;
894 	}
895 
896 	if (elf_interpreter) {
897 		unsigned long interp_map_addr = 0;
898 
899 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
900 					    interpreter,
901 					    &interp_map_addr,
902 					    load_bias);
903 		if (!IS_ERR((void *)elf_entry)) {
904 			/*
905 			 * load_elf_interp() returns relocation
906 			 * adjustment
907 			 */
908 			interp_load_addr = elf_entry;
909 			elf_entry += loc->interp_elf_ex.e_entry;
910 		}
911 		if (BAD_ADDR(elf_entry)) {
912 			force_sig(SIGSEGV, current);
913 			retval = IS_ERR((void *)elf_entry) ?
914 					(int)elf_entry : -EINVAL;
915 			goto out_free_dentry;
916 		}
917 		reloc_func_desc = interp_load_addr;
918 
919 		allow_write_access(interpreter);
920 		fput(interpreter);
921 		kfree(elf_interpreter);
922 	} else {
923 		elf_entry = loc->elf_ex.e_entry;
924 		if (BAD_ADDR(elf_entry)) {
925 			force_sig(SIGSEGV, current);
926 			retval = -EINVAL;
927 			goto out_free_dentry;
928 		}
929 	}
930 
931 	kfree(elf_phdata);
932 
933 	set_binfmt(&elf_format);
934 
935 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
936 	retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
937 	if (retval < 0) {
938 		send_sig(SIGKILL, current, 0);
939 		goto out;
940 	}
941 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
942 
943 	install_exec_creds(bprm);
944 	retval = create_elf_tables(bprm, &loc->elf_ex,
945 			  load_addr, interp_load_addr);
946 	if (retval < 0) {
947 		send_sig(SIGKILL, current, 0);
948 		goto out;
949 	}
950 	/* N.B. passed_fileno might not be initialized? */
951 	current->mm->end_code = end_code;
952 	current->mm->start_code = start_code;
953 	current->mm->start_data = start_data;
954 	current->mm->end_data = end_data;
955 	current->mm->start_stack = bprm->p;
956 
957 #ifdef arch_randomize_brk
958 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
959 		current->mm->brk = current->mm->start_brk =
960 			arch_randomize_brk(current->mm);
961 #ifdef CONFIG_COMPAT_BRK
962 		current->brk_randomized = 1;
963 #endif
964 	}
965 #endif
966 
967 	if (current->personality & MMAP_PAGE_ZERO) {
968 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
969 		   and some applications "depend" upon this behavior.
970 		   Since we do not have the power to recompile these, we
971 		   emulate the SVr4 behavior. Sigh. */
972 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
973 				MAP_FIXED | MAP_PRIVATE, 0);
974 	}
975 
976 #ifdef ELF_PLAT_INIT
977 	/*
978 	 * The ABI may specify that certain registers be set up in special
979 	 * ways (on i386 %edx is the address of a DT_FINI function, for
980 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
981 	 * that the e_entry field is the address of the function descriptor
982 	 * for the startup routine, rather than the address of the startup
983 	 * routine itself.  This macro performs whatever initialization to
984 	 * the regs structure is required as well as any relocations to the
985 	 * function descriptor entries when executing dynamically links apps.
986 	 */
987 	ELF_PLAT_INIT(regs, reloc_func_desc);
988 #endif
989 
990 	start_thread(regs, elf_entry, bprm->p);
991 	retval = 0;
992 out:
993 	kfree(loc);
994 out_ret:
995 	return retval;
996 
997 	/* error cleanup */
998 out_free_dentry:
999 	allow_write_access(interpreter);
1000 	if (interpreter)
1001 		fput(interpreter);
1002 out_free_interp:
1003 	kfree(elf_interpreter);
1004 out_free_ph:
1005 	kfree(elf_phdata);
1006 	goto out;
1007 }
1008 
1009 #ifdef CONFIG_USELIB
1010 /* This is really simpleminded and specialized - we are loading an
1011    a.out library that is given an ELF header. */
1012 static int load_elf_library(struct file *file)
1013 {
1014 	struct elf_phdr *elf_phdata;
1015 	struct elf_phdr *eppnt;
1016 	unsigned long elf_bss, bss, len;
1017 	int retval, error, i, j;
1018 	struct elfhdr elf_ex;
1019 
1020 	error = -ENOEXEC;
1021 	retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1022 	if (retval != sizeof(elf_ex))
1023 		goto out;
1024 
1025 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1026 		goto out;
1027 
1028 	/* First of all, some simple consistency checks */
1029 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1030 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1031 		goto out;
1032 
1033 	/* Now read in all of the header information */
1034 
1035 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1036 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1037 
1038 	error = -ENOMEM;
1039 	elf_phdata = kmalloc(j, GFP_KERNEL);
1040 	if (!elf_phdata)
1041 		goto out;
1042 
1043 	eppnt = elf_phdata;
1044 	error = -ENOEXEC;
1045 	retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1046 	if (retval != j)
1047 		goto out_free_ph;
1048 
1049 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1050 		if ((eppnt + i)->p_type == PT_LOAD)
1051 			j++;
1052 	if (j != 1)
1053 		goto out_free_ph;
1054 
1055 	while (eppnt->p_type != PT_LOAD)
1056 		eppnt++;
1057 
1058 	/* Now use mmap to map the library into memory. */
1059 	error = vm_mmap(file,
1060 			ELF_PAGESTART(eppnt->p_vaddr),
1061 			(eppnt->p_filesz +
1062 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1063 			PROT_READ | PROT_WRITE | PROT_EXEC,
1064 			MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1065 			(eppnt->p_offset -
1066 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1067 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1068 		goto out_free_ph;
1069 
1070 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1071 	if (padzero(elf_bss)) {
1072 		error = -EFAULT;
1073 		goto out_free_ph;
1074 	}
1075 
1076 	len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1077 			    ELF_MIN_ALIGN - 1);
1078 	bss = eppnt->p_memsz + eppnt->p_vaddr;
1079 	if (bss > len)
1080 		vm_brk(len, bss - len);
1081 	error = 0;
1082 
1083 out_free_ph:
1084 	kfree(elf_phdata);
1085 out:
1086 	return error;
1087 }
1088 #endif /* #ifdef CONFIG_USELIB */
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 	/*
1113 	 * Assume that all vmas with a .name op should always be dumped.
1114 	 * If this changes, a new vm_ops field can easily be added.
1115 	 */
1116 	if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1117 		return true;
1118 
1119 	/*
1120 	 * arch_vma_name() returns non-NULL for special architecture mappings,
1121 	 * such as vDSO sections.
1122 	 */
1123 	if (arch_vma_name(vma))
1124 		return true;
1125 
1126 	return false;
1127 }
1128 
1129 /*
1130  * Decide what to dump of a segment, part, all or none.
1131  */
1132 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1133 				   unsigned long mm_flags)
1134 {
1135 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1136 
1137 	/* always dump the vdso and vsyscall sections */
1138 	if (always_dump_vma(vma))
1139 		goto whole;
1140 
1141 	if (vma->vm_flags & VM_DONTDUMP)
1142 		return 0;
1143 
1144 	/* Hugetlb memory check */
1145 	if (vma->vm_flags & VM_HUGETLB) {
1146 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1147 			goto whole;
1148 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1149 			goto whole;
1150 		return 0;
1151 	}
1152 
1153 	/* Do not dump I/O mapped devices or special mappings */
1154 	if (vma->vm_flags & VM_IO)
1155 		return 0;
1156 
1157 	/* By default, dump shared memory if mapped from an anonymous file. */
1158 	if (vma->vm_flags & VM_SHARED) {
1159 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1160 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1161 			goto whole;
1162 		return 0;
1163 	}
1164 
1165 	/* Dump segments that have been written to.  */
1166 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1167 		goto whole;
1168 	if (vma->vm_file == NULL)
1169 		return 0;
1170 
1171 	if (FILTER(MAPPED_PRIVATE))
1172 		goto whole;
1173 
1174 	/*
1175 	 * If this looks like the beginning of a DSO or executable mapping,
1176 	 * check for an ELF header.  If we find one, dump the first page to
1177 	 * aid in determining what was mapped here.
1178 	 */
1179 	if (FILTER(ELF_HEADERS) &&
1180 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1181 		u32 __user *header = (u32 __user *) vma->vm_start;
1182 		u32 word;
1183 		mm_segment_t fs = get_fs();
1184 		/*
1185 		 * Doing it this way gets the constant folded by GCC.
1186 		 */
1187 		union {
1188 			u32 cmp;
1189 			char elfmag[SELFMAG];
1190 		} magic;
1191 		BUILD_BUG_ON(SELFMAG != sizeof word);
1192 		magic.elfmag[EI_MAG0] = ELFMAG0;
1193 		magic.elfmag[EI_MAG1] = ELFMAG1;
1194 		magic.elfmag[EI_MAG2] = ELFMAG2;
1195 		magic.elfmag[EI_MAG3] = ELFMAG3;
1196 		/*
1197 		 * Switch to the user "segment" for get_user(),
1198 		 * then put back what elf_core_dump() had in place.
1199 		 */
1200 		set_fs(USER_DS);
1201 		if (unlikely(get_user(word, header)))
1202 			word = 0;
1203 		set_fs(fs);
1204 		if (word == magic.cmp)
1205 			return PAGE_SIZE;
1206 	}
1207 
1208 #undef	FILTER
1209 
1210 	return 0;
1211 
1212 whole:
1213 	return vma->vm_end - vma->vm_start;
1214 }
1215 
1216 /* An ELF note in memory */
1217 struct memelfnote
1218 {
1219 	const char *name;
1220 	int type;
1221 	unsigned int datasz;
1222 	void *data;
1223 };
1224 
1225 static int notesize(struct memelfnote *en)
1226 {
1227 	int sz;
1228 
1229 	sz = sizeof(struct elf_note);
1230 	sz += roundup(strlen(en->name) + 1, 4);
1231 	sz += roundup(en->datasz, 4);
1232 
1233 	return sz;
1234 }
1235 
1236 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1237 {
1238 	struct elf_note en;
1239 	en.n_namesz = strlen(men->name) + 1;
1240 	en.n_descsz = men->datasz;
1241 	en.n_type = men->type;
1242 
1243 	return dump_emit(cprm, &en, sizeof(en)) &&
1244 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1245 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1246 }
1247 
1248 static void fill_elf_header(struct elfhdr *elf, int segs,
1249 			    u16 machine, u32 flags)
1250 {
1251 	memset(elf, 0, sizeof(*elf));
1252 
1253 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1254 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1255 	elf->e_ident[EI_DATA] = ELF_DATA;
1256 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1257 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1258 
1259 	elf->e_type = ET_CORE;
1260 	elf->e_machine = machine;
1261 	elf->e_version = EV_CURRENT;
1262 	elf->e_phoff = sizeof(struct elfhdr);
1263 	elf->e_flags = flags;
1264 	elf->e_ehsize = sizeof(struct elfhdr);
1265 	elf->e_phentsize = sizeof(struct elf_phdr);
1266 	elf->e_phnum = segs;
1267 
1268 	return;
1269 }
1270 
1271 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1272 {
1273 	phdr->p_type = PT_NOTE;
1274 	phdr->p_offset = offset;
1275 	phdr->p_vaddr = 0;
1276 	phdr->p_paddr = 0;
1277 	phdr->p_filesz = sz;
1278 	phdr->p_memsz = 0;
1279 	phdr->p_flags = 0;
1280 	phdr->p_align = 0;
1281 	return;
1282 }
1283 
1284 static void fill_note(struct memelfnote *note, const char *name, int type,
1285 		unsigned int sz, void *data)
1286 {
1287 	note->name = name;
1288 	note->type = type;
1289 	note->datasz = sz;
1290 	note->data = data;
1291 	return;
1292 }
1293 
1294 /*
1295  * fill up all the fields in prstatus from the given task struct, except
1296  * registers which need to be filled up separately.
1297  */
1298 static void fill_prstatus(struct elf_prstatus *prstatus,
1299 		struct task_struct *p, long signr)
1300 {
1301 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1302 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1303 	prstatus->pr_sighold = p->blocked.sig[0];
1304 	rcu_read_lock();
1305 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1306 	rcu_read_unlock();
1307 	prstatus->pr_pid = task_pid_vnr(p);
1308 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1309 	prstatus->pr_sid = task_session_vnr(p);
1310 	if (thread_group_leader(p)) {
1311 		struct task_cputime cputime;
1312 
1313 		/*
1314 		 * This is the record for the group leader.  It shows the
1315 		 * group-wide total, not its individual thread total.
1316 		 */
1317 		thread_group_cputime(p, &cputime);
1318 		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1319 		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1320 	} else {
1321 		cputime_t utime, stime;
1322 
1323 		task_cputime(p, &utime, &stime);
1324 		cputime_to_timeval(utime, &prstatus->pr_utime);
1325 		cputime_to_timeval(stime, &prstatus->pr_stime);
1326 	}
1327 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1328 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1329 }
1330 
1331 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1332 		       struct mm_struct *mm)
1333 {
1334 	const struct cred *cred;
1335 	unsigned int i, len;
1336 
1337 	/* first copy the parameters from user space */
1338 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1339 
1340 	len = mm->arg_end - mm->arg_start;
1341 	if (len >= ELF_PRARGSZ)
1342 		len = ELF_PRARGSZ-1;
1343 	if (copy_from_user(&psinfo->pr_psargs,
1344 		           (const char __user *)mm->arg_start, len))
1345 		return -EFAULT;
1346 	for(i = 0; i < len; i++)
1347 		if (psinfo->pr_psargs[i] == 0)
1348 			psinfo->pr_psargs[i] = ' ';
1349 	psinfo->pr_psargs[len] = 0;
1350 
1351 	rcu_read_lock();
1352 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1353 	rcu_read_unlock();
1354 	psinfo->pr_pid = task_pid_vnr(p);
1355 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1356 	psinfo->pr_sid = task_session_vnr(p);
1357 
1358 	i = p->state ? ffz(~p->state) + 1 : 0;
1359 	psinfo->pr_state = i;
1360 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1361 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1362 	psinfo->pr_nice = task_nice(p);
1363 	psinfo->pr_flag = p->flags;
1364 	rcu_read_lock();
1365 	cred = __task_cred(p);
1366 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1367 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1368 	rcu_read_unlock();
1369 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1370 
1371 	return 0;
1372 }
1373 
1374 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1375 {
1376 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1377 	int i = 0;
1378 	do
1379 		i += 2;
1380 	while (auxv[i - 2] != AT_NULL);
1381 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1382 }
1383 
1384 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1385 		const siginfo_t *siginfo)
1386 {
1387 	mm_segment_t old_fs = get_fs();
1388 	set_fs(KERNEL_DS);
1389 	copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1390 	set_fs(old_fs);
1391 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1392 }
1393 
1394 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1395 /*
1396  * Format of NT_FILE note:
1397  *
1398  * long count     -- how many files are mapped
1399  * long page_size -- units for file_ofs
1400  * array of [COUNT] elements of
1401  *   long start
1402  *   long end
1403  *   long file_ofs
1404  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1405  */
1406 static int fill_files_note(struct memelfnote *note)
1407 {
1408 	struct vm_area_struct *vma;
1409 	unsigned count, size, names_ofs, remaining, n;
1410 	user_long_t *data;
1411 	user_long_t *start_end_ofs;
1412 	char *name_base, *name_curpos;
1413 
1414 	/* *Estimated* file count and total data size needed */
1415 	count = current->mm->map_count;
1416 	size = count * 64;
1417 
1418 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1419  alloc:
1420 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1421 		return -EINVAL;
1422 	size = round_up(size, PAGE_SIZE);
1423 	data = vmalloc(size);
1424 	if (!data)
1425 		return -ENOMEM;
1426 
1427 	start_end_ofs = data + 2;
1428 	name_base = name_curpos = ((char *)data) + names_ofs;
1429 	remaining = size - names_ofs;
1430 	count = 0;
1431 	for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1432 		struct file *file;
1433 		const char *filename;
1434 
1435 		file = vma->vm_file;
1436 		if (!file)
1437 			continue;
1438 		filename = d_path(&file->f_path, name_curpos, remaining);
1439 		if (IS_ERR(filename)) {
1440 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1441 				vfree(data);
1442 				size = size * 5 / 4;
1443 				goto alloc;
1444 			}
1445 			continue;
1446 		}
1447 
1448 		/* d_path() fills at the end, move name down */
1449 		/* n = strlen(filename) + 1: */
1450 		n = (name_curpos + remaining) - filename;
1451 		remaining = filename - name_curpos;
1452 		memmove(name_curpos, filename, n);
1453 		name_curpos += n;
1454 
1455 		*start_end_ofs++ = vma->vm_start;
1456 		*start_end_ofs++ = vma->vm_end;
1457 		*start_end_ofs++ = vma->vm_pgoff;
1458 		count++;
1459 	}
1460 
1461 	/* Now we know exact count of files, can store it */
1462 	data[0] = count;
1463 	data[1] = PAGE_SIZE;
1464 	/*
1465 	 * Count usually is less than current->mm->map_count,
1466 	 * we need to move filenames down.
1467 	 */
1468 	n = current->mm->map_count - count;
1469 	if (n != 0) {
1470 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1471 		memmove(name_base - shift_bytes, name_base,
1472 			name_curpos - name_base);
1473 		name_curpos -= shift_bytes;
1474 	}
1475 
1476 	size = name_curpos - (char *)data;
1477 	fill_note(note, "CORE", NT_FILE, size, data);
1478 	return 0;
1479 }
1480 
1481 #ifdef CORE_DUMP_USE_REGSET
1482 #include <linux/regset.h>
1483 
1484 struct elf_thread_core_info {
1485 	struct elf_thread_core_info *next;
1486 	struct task_struct *task;
1487 	struct elf_prstatus prstatus;
1488 	struct memelfnote notes[0];
1489 };
1490 
1491 struct elf_note_info {
1492 	struct elf_thread_core_info *thread;
1493 	struct memelfnote psinfo;
1494 	struct memelfnote signote;
1495 	struct memelfnote auxv;
1496 	struct memelfnote files;
1497 	user_siginfo_t csigdata;
1498 	size_t size;
1499 	int thread_notes;
1500 };
1501 
1502 /*
1503  * When a regset has a writeback hook, we call it on each thread before
1504  * dumping user memory.  On register window machines, this makes sure the
1505  * user memory backing the register data is up to date before we read it.
1506  */
1507 static void do_thread_regset_writeback(struct task_struct *task,
1508 				       const struct user_regset *regset)
1509 {
1510 	if (regset->writeback)
1511 		regset->writeback(task, regset, 1);
1512 }
1513 
1514 #ifndef PR_REG_SIZE
1515 #define PR_REG_SIZE(S) sizeof(S)
1516 #endif
1517 
1518 #ifndef PRSTATUS_SIZE
1519 #define PRSTATUS_SIZE(S) sizeof(S)
1520 #endif
1521 
1522 #ifndef PR_REG_PTR
1523 #define PR_REG_PTR(S) (&((S)->pr_reg))
1524 #endif
1525 
1526 #ifndef SET_PR_FPVALID
1527 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1528 #endif
1529 
1530 static int fill_thread_core_info(struct elf_thread_core_info *t,
1531 				 const struct user_regset_view *view,
1532 				 long signr, size_t *total)
1533 {
1534 	unsigned int i;
1535 
1536 	/*
1537 	 * NT_PRSTATUS is the one special case, because the regset data
1538 	 * goes into the pr_reg field inside the note contents, rather
1539 	 * than being the whole note contents.  We fill the reset in here.
1540 	 * We assume that regset 0 is NT_PRSTATUS.
1541 	 */
1542 	fill_prstatus(&t->prstatus, t->task, signr);
1543 	(void) view->regsets[0].get(t->task, &view->regsets[0],
1544 				    0, PR_REG_SIZE(t->prstatus.pr_reg),
1545 				    PR_REG_PTR(&t->prstatus), NULL);
1546 
1547 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1548 		  PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1549 	*total += notesize(&t->notes[0]);
1550 
1551 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1552 
1553 	/*
1554 	 * Each other regset might generate a note too.  For each regset
1555 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1556 	 * all zero and we'll know to skip writing it later.
1557 	 */
1558 	for (i = 1; i < view->n; ++i) {
1559 		const struct user_regset *regset = &view->regsets[i];
1560 		do_thread_regset_writeback(t->task, regset);
1561 		if (regset->core_note_type && regset->get &&
1562 		    (!regset->active || regset->active(t->task, regset))) {
1563 			int ret;
1564 			size_t size = regset->n * regset->size;
1565 			void *data = kmalloc(size, GFP_KERNEL);
1566 			if (unlikely(!data))
1567 				return 0;
1568 			ret = regset->get(t->task, regset,
1569 					  0, size, data, NULL);
1570 			if (unlikely(ret))
1571 				kfree(data);
1572 			else {
1573 				if (regset->core_note_type != NT_PRFPREG)
1574 					fill_note(&t->notes[i], "LINUX",
1575 						  regset->core_note_type,
1576 						  size, data);
1577 				else {
1578 					SET_PR_FPVALID(&t->prstatus, 1);
1579 					fill_note(&t->notes[i], "CORE",
1580 						  NT_PRFPREG, size, data);
1581 				}
1582 				*total += notesize(&t->notes[i]);
1583 			}
1584 		}
1585 	}
1586 
1587 	return 1;
1588 }
1589 
1590 static int fill_note_info(struct elfhdr *elf, int phdrs,
1591 			  struct elf_note_info *info,
1592 			  const siginfo_t *siginfo, struct pt_regs *regs)
1593 {
1594 	struct task_struct *dump_task = current;
1595 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1596 	struct elf_thread_core_info *t;
1597 	struct elf_prpsinfo *psinfo;
1598 	struct core_thread *ct;
1599 	unsigned int i;
1600 
1601 	info->size = 0;
1602 	info->thread = NULL;
1603 
1604 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1605 	if (psinfo == NULL) {
1606 		info->psinfo.data = NULL; /* So we don't free this wrongly */
1607 		return 0;
1608 	}
1609 
1610 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1611 
1612 	/*
1613 	 * Figure out how many notes we're going to need for each thread.
1614 	 */
1615 	info->thread_notes = 0;
1616 	for (i = 0; i < view->n; ++i)
1617 		if (view->regsets[i].core_note_type != 0)
1618 			++info->thread_notes;
1619 
1620 	/*
1621 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1622 	 * since it is our one special case.
1623 	 */
1624 	if (unlikely(info->thread_notes == 0) ||
1625 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1626 		WARN_ON(1);
1627 		return 0;
1628 	}
1629 
1630 	/*
1631 	 * Initialize the ELF file header.
1632 	 */
1633 	fill_elf_header(elf, phdrs,
1634 			view->e_machine, view->e_flags);
1635 
1636 	/*
1637 	 * Allocate a structure for each thread.
1638 	 */
1639 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1640 		t = kzalloc(offsetof(struct elf_thread_core_info,
1641 				     notes[info->thread_notes]),
1642 			    GFP_KERNEL);
1643 		if (unlikely(!t))
1644 			return 0;
1645 
1646 		t->task = ct->task;
1647 		if (ct->task == dump_task || !info->thread) {
1648 			t->next = info->thread;
1649 			info->thread = t;
1650 		} else {
1651 			/*
1652 			 * Make sure to keep the original task at
1653 			 * the head of the list.
1654 			 */
1655 			t->next = info->thread->next;
1656 			info->thread->next = t;
1657 		}
1658 	}
1659 
1660 	/*
1661 	 * Now fill in each thread's information.
1662 	 */
1663 	for (t = info->thread; t != NULL; t = t->next)
1664 		if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1665 			return 0;
1666 
1667 	/*
1668 	 * Fill in the two process-wide notes.
1669 	 */
1670 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1671 	info->size += notesize(&info->psinfo);
1672 
1673 	fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1674 	info->size += notesize(&info->signote);
1675 
1676 	fill_auxv_note(&info->auxv, current->mm);
1677 	info->size += notesize(&info->auxv);
1678 
1679 	if (fill_files_note(&info->files) == 0)
1680 		info->size += notesize(&info->files);
1681 
1682 	return 1;
1683 }
1684 
1685 static size_t get_note_info_size(struct elf_note_info *info)
1686 {
1687 	return info->size;
1688 }
1689 
1690 /*
1691  * Write all the notes for each thread.  When writing the first thread, the
1692  * process-wide notes are interleaved after the first thread-specific note.
1693  */
1694 static int write_note_info(struct elf_note_info *info,
1695 			   struct coredump_params *cprm)
1696 {
1697 	bool first = true;
1698 	struct elf_thread_core_info *t = info->thread;
1699 
1700 	do {
1701 		int i;
1702 
1703 		if (!writenote(&t->notes[0], cprm))
1704 			return 0;
1705 
1706 		if (first && !writenote(&info->psinfo, cprm))
1707 			return 0;
1708 		if (first && !writenote(&info->signote, cprm))
1709 			return 0;
1710 		if (first && !writenote(&info->auxv, cprm))
1711 			return 0;
1712 		if (first && info->files.data &&
1713 				!writenote(&info->files, cprm))
1714 			return 0;
1715 
1716 		for (i = 1; i < info->thread_notes; ++i)
1717 			if (t->notes[i].data &&
1718 			    !writenote(&t->notes[i], cprm))
1719 				return 0;
1720 
1721 		first = false;
1722 		t = t->next;
1723 	} while (t);
1724 
1725 	return 1;
1726 }
1727 
1728 static void free_note_info(struct elf_note_info *info)
1729 {
1730 	struct elf_thread_core_info *threads = info->thread;
1731 	while (threads) {
1732 		unsigned int i;
1733 		struct elf_thread_core_info *t = threads;
1734 		threads = t->next;
1735 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1736 		for (i = 1; i < info->thread_notes; ++i)
1737 			kfree(t->notes[i].data);
1738 		kfree(t);
1739 	}
1740 	kfree(info->psinfo.data);
1741 	vfree(info->files.data);
1742 }
1743 
1744 #else
1745 
1746 /* Here is the structure in which status of each thread is captured. */
1747 struct elf_thread_status
1748 {
1749 	struct list_head list;
1750 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1751 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1752 	struct task_struct *thread;
1753 #ifdef ELF_CORE_COPY_XFPREGS
1754 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1755 #endif
1756 	struct memelfnote notes[3];
1757 	int num_notes;
1758 };
1759 
1760 /*
1761  * In order to add the specific thread information for the elf file format,
1762  * we need to keep a linked list of every threads pr_status and then create
1763  * a single section for them in the final core file.
1764  */
1765 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1766 {
1767 	int sz = 0;
1768 	struct task_struct *p = t->thread;
1769 	t->num_notes = 0;
1770 
1771 	fill_prstatus(&t->prstatus, p, signr);
1772 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1773 
1774 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1775 		  &(t->prstatus));
1776 	t->num_notes++;
1777 	sz += notesize(&t->notes[0]);
1778 
1779 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1780 								&t->fpu))) {
1781 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1782 			  &(t->fpu));
1783 		t->num_notes++;
1784 		sz += notesize(&t->notes[1]);
1785 	}
1786 
1787 #ifdef ELF_CORE_COPY_XFPREGS
1788 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1789 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1790 			  sizeof(t->xfpu), &t->xfpu);
1791 		t->num_notes++;
1792 		sz += notesize(&t->notes[2]);
1793 	}
1794 #endif
1795 	return sz;
1796 }
1797 
1798 struct elf_note_info {
1799 	struct memelfnote *notes;
1800 	struct memelfnote *notes_files;
1801 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1802 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1803 	struct list_head thread_list;
1804 	elf_fpregset_t *fpu;
1805 #ifdef ELF_CORE_COPY_XFPREGS
1806 	elf_fpxregset_t *xfpu;
1807 #endif
1808 	user_siginfo_t csigdata;
1809 	int thread_status_size;
1810 	int numnote;
1811 };
1812 
1813 static int elf_note_info_init(struct elf_note_info *info)
1814 {
1815 	memset(info, 0, sizeof(*info));
1816 	INIT_LIST_HEAD(&info->thread_list);
1817 
1818 	/* Allocate space for ELF notes */
1819 	info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1820 	if (!info->notes)
1821 		return 0;
1822 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1823 	if (!info->psinfo)
1824 		return 0;
1825 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1826 	if (!info->prstatus)
1827 		return 0;
1828 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1829 	if (!info->fpu)
1830 		return 0;
1831 #ifdef ELF_CORE_COPY_XFPREGS
1832 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1833 	if (!info->xfpu)
1834 		return 0;
1835 #endif
1836 	return 1;
1837 }
1838 
1839 static int fill_note_info(struct elfhdr *elf, int phdrs,
1840 			  struct elf_note_info *info,
1841 			  const siginfo_t *siginfo, struct pt_regs *regs)
1842 {
1843 	struct list_head *t;
1844 	struct core_thread *ct;
1845 	struct elf_thread_status *ets;
1846 
1847 	if (!elf_note_info_init(info))
1848 		return 0;
1849 
1850 	for (ct = current->mm->core_state->dumper.next;
1851 					ct; ct = ct->next) {
1852 		ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1853 		if (!ets)
1854 			return 0;
1855 
1856 		ets->thread = ct->task;
1857 		list_add(&ets->list, &info->thread_list);
1858 	}
1859 
1860 	list_for_each(t, &info->thread_list) {
1861 		int sz;
1862 
1863 		ets = list_entry(t, struct elf_thread_status, list);
1864 		sz = elf_dump_thread_status(siginfo->si_signo, ets);
1865 		info->thread_status_size += sz;
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);
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 	info->numnote = 4;
1889 
1890 	if (fill_files_note(info->notes + info->numnote) == 0) {
1891 		info->notes_files = info->notes + info->numnote;
1892 		info->numnote++;
1893 	}
1894 
1895 	/* Try to dump the FPU. */
1896 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1897 							       info->fpu);
1898 	if (info->prstatus->pr_fpvalid)
1899 		fill_note(info->notes + info->numnote++,
1900 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1901 #ifdef ELF_CORE_COPY_XFPREGS
1902 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
1903 		fill_note(info->notes + info->numnote++,
1904 			  "LINUX", ELF_CORE_XFPREG_TYPE,
1905 			  sizeof(*info->xfpu), info->xfpu);
1906 #endif
1907 
1908 	return 1;
1909 }
1910 
1911 static size_t get_note_info_size(struct elf_note_info *info)
1912 {
1913 	int sz = 0;
1914 	int i;
1915 
1916 	for (i = 0; i < info->numnote; i++)
1917 		sz += notesize(info->notes + i);
1918 
1919 	sz += info->thread_status_size;
1920 
1921 	return sz;
1922 }
1923 
1924 static int write_note_info(struct elf_note_info *info,
1925 			   struct coredump_params *cprm)
1926 {
1927 	int i;
1928 	struct list_head *t;
1929 
1930 	for (i = 0; i < info->numnote; i++)
1931 		if (!writenote(info->notes + i, cprm))
1932 			return 0;
1933 
1934 	/* write out the thread status notes section */
1935 	list_for_each(t, &info->thread_list) {
1936 		struct elf_thread_status *tmp =
1937 				list_entry(t, struct elf_thread_status, list);
1938 
1939 		for (i = 0; i < tmp->num_notes; i++)
1940 			if (!writenote(&tmp->notes[i], cprm))
1941 				return 0;
1942 	}
1943 
1944 	return 1;
1945 }
1946 
1947 static void free_note_info(struct elf_note_info *info)
1948 {
1949 	while (!list_empty(&info->thread_list)) {
1950 		struct list_head *tmp = info->thread_list.next;
1951 		list_del(tmp);
1952 		kfree(list_entry(tmp, struct elf_thread_status, list));
1953 	}
1954 
1955 	/* Free data possibly allocated by fill_files_note(): */
1956 	if (info->notes_files)
1957 		vfree(info->notes_files->data);
1958 
1959 	kfree(info->prstatus);
1960 	kfree(info->psinfo);
1961 	kfree(info->notes);
1962 	kfree(info->fpu);
1963 #ifdef ELF_CORE_COPY_XFPREGS
1964 	kfree(info->xfpu);
1965 #endif
1966 }
1967 
1968 #endif
1969 
1970 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1971 					struct vm_area_struct *gate_vma)
1972 {
1973 	struct vm_area_struct *ret = tsk->mm->mmap;
1974 
1975 	if (ret)
1976 		return ret;
1977 	return gate_vma;
1978 }
1979 /*
1980  * Helper function for iterating across a vma list.  It ensures that the caller
1981  * will visit `gate_vma' prior to terminating the search.
1982  */
1983 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1984 					struct vm_area_struct *gate_vma)
1985 {
1986 	struct vm_area_struct *ret;
1987 
1988 	ret = this_vma->vm_next;
1989 	if (ret)
1990 		return ret;
1991 	if (this_vma == gate_vma)
1992 		return NULL;
1993 	return gate_vma;
1994 }
1995 
1996 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1997 			     elf_addr_t e_shoff, int segs)
1998 {
1999 	elf->e_shoff = e_shoff;
2000 	elf->e_shentsize = sizeof(*shdr4extnum);
2001 	elf->e_shnum = 1;
2002 	elf->e_shstrndx = SHN_UNDEF;
2003 
2004 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2005 
2006 	shdr4extnum->sh_type = SHT_NULL;
2007 	shdr4extnum->sh_size = elf->e_shnum;
2008 	shdr4extnum->sh_link = elf->e_shstrndx;
2009 	shdr4extnum->sh_info = segs;
2010 }
2011 
2012 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
2013 				     unsigned long mm_flags)
2014 {
2015 	struct vm_area_struct *vma;
2016 	size_t size = 0;
2017 
2018 	for (vma = first_vma(current, gate_vma); vma != NULL;
2019 	     vma = next_vma(vma, gate_vma))
2020 		size += vma_dump_size(vma, mm_flags);
2021 	return size;
2022 }
2023 
2024 /*
2025  * Actual dumper
2026  *
2027  * This is a two-pass process; first we find the offsets of the bits,
2028  * and then they are actually written out.  If we run out of core limit
2029  * we just truncate.
2030  */
2031 static int elf_core_dump(struct coredump_params *cprm)
2032 {
2033 	int has_dumped = 0;
2034 	mm_segment_t fs;
2035 	int segs;
2036 	struct vm_area_struct *vma, *gate_vma;
2037 	struct elfhdr *elf = NULL;
2038 	loff_t offset = 0, dataoff;
2039 	struct elf_note_info info = { };
2040 	struct elf_phdr *phdr4note = NULL;
2041 	struct elf_shdr *shdr4extnum = NULL;
2042 	Elf_Half e_phnum;
2043 	elf_addr_t e_shoff;
2044 
2045 	/*
2046 	 * We no longer stop all VM operations.
2047 	 *
2048 	 * This is because those proceses that could possibly change map_count
2049 	 * or the mmap / vma pages are now blocked in do_exit on current
2050 	 * finishing this core dump.
2051 	 *
2052 	 * Only ptrace can touch these memory addresses, but it doesn't change
2053 	 * the map_count or the pages allocated. So no possibility of crashing
2054 	 * exists while dumping the mm->vm_next areas to the core file.
2055 	 */
2056 
2057 	/* alloc memory for large data structures: too large to be on stack */
2058 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2059 	if (!elf)
2060 		goto out;
2061 	/*
2062 	 * The number of segs are recored into ELF header as 16bit value.
2063 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2064 	 */
2065 	segs = current->mm->map_count;
2066 	segs += elf_core_extra_phdrs();
2067 
2068 	gate_vma = get_gate_vma(current->mm);
2069 	if (gate_vma != NULL)
2070 		segs++;
2071 
2072 	/* for notes section */
2073 	segs++;
2074 
2075 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2076 	 * this, kernel supports extended numbering. Have a look at
2077 	 * include/linux/elf.h for further information. */
2078 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2079 
2080 	/*
2081 	 * Collect all the non-memory information about the process for the
2082 	 * notes.  This also sets up the file header.
2083 	 */
2084 	if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2085 		goto cleanup;
2086 
2087 	has_dumped = 1;
2088 
2089 	fs = get_fs();
2090 	set_fs(KERNEL_DS);
2091 
2092 	offset += sizeof(*elf);				/* Elf header */
2093 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2094 
2095 	/* Write notes phdr entry */
2096 	{
2097 		size_t sz = get_note_info_size(&info);
2098 
2099 		sz += elf_coredump_extra_notes_size();
2100 
2101 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2102 		if (!phdr4note)
2103 			goto end_coredump;
2104 
2105 		fill_elf_note_phdr(phdr4note, sz, offset);
2106 		offset += sz;
2107 	}
2108 
2109 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2110 
2111 	offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2112 	offset += elf_core_extra_data_size();
2113 	e_shoff = offset;
2114 
2115 	if (e_phnum == PN_XNUM) {
2116 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2117 		if (!shdr4extnum)
2118 			goto end_coredump;
2119 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2120 	}
2121 
2122 	offset = dataoff;
2123 
2124 	if (!dump_emit(cprm, elf, sizeof(*elf)))
2125 		goto end_coredump;
2126 
2127 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2128 		goto end_coredump;
2129 
2130 	/* Write program headers for segments dump */
2131 	for (vma = first_vma(current, gate_vma); vma != NULL;
2132 			vma = next_vma(vma, gate_vma)) {
2133 		struct elf_phdr phdr;
2134 
2135 		phdr.p_type = PT_LOAD;
2136 		phdr.p_offset = offset;
2137 		phdr.p_vaddr = vma->vm_start;
2138 		phdr.p_paddr = 0;
2139 		phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2140 		phdr.p_memsz = vma->vm_end - vma->vm_start;
2141 		offset += phdr.p_filesz;
2142 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2143 		if (vma->vm_flags & VM_WRITE)
2144 			phdr.p_flags |= PF_W;
2145 		if (vma->vm_flags & VM_EXEC)
2146 			phdr.p_flags |= PF_X;
2147 		phdr.p_align = ELF_EXEC_PAGESIZE;
2148 
2149 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2150 			goto end_coredump;
2151 	}
2152 
2153 	if (!elf_core_write_extra_phdrs(cprm, offset))
2154 		goto end_coredump;
2155 
2156  	/* write out the notes section */
2157 	if (!write_note_info(&info, cprm))
2158 		goto end_coredump;
2159 
2160 	if (elf_coredump_extra_notes_write(cprm))
2161 		goto end_coredump;
2162 
2163 	/* Align to page */
2164 	if (!dump_skip(cprm, dataoff - cprm->written))
2165 		goto end_coredump;
2166 
2167 	for (vma = first_vma(current, gate_vma); vma != NULL;
2168 			vma = next_vma(vma, gate_vma)) {
2169 		unsigned long addr;
2170 		unsigned long end;
2171 
2172 		end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2173 
2174 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2175 			struct page *page;
2176 			int stop;
2177 
2178 			page = get_dump_page(addr);
2179 			if (page) {
2180 				void *kaddr = kmap(page);
2181 				stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2182 				kunmap(page);
2183 				page_cache_release(page);
2184 			} else
2185 				stop = !dump_skip(cprm, PAGE_SIZE);
2186 			if (stop)
2187 				goto end_coredump;
2188 		}
2189 	}
2190 
2191 	if (!elf_core_write_extra_data(cprm))
2192 		goto end_coredump;
2193 
2194 	if (e_phnum == PN_XNUM) {
2195 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2196 			goto end_coredump;
2197 	}
2198 
2199 end_coredump:
2200 	set_fs(fs);
2201 
2202 cleanup:
2203 	free_note_info(&info);
2204 	kfree(shdr4extnum);
2205 	kfree(phdr4note);
2206 	kfree(elf);
2207 out:
2208 	return has_dumped;
2209 }
2210 
2211 #endif		/* CONFIG_ELF_CORE */
2212 
2213 static int __init init_elf_binfmt(void)
2214 {
2215 	register_binfmt(&elf_format);
2216 	return 0;
2217 }
2218 
2219 static void __exit exit_elf_binfmt(void)
2220 {
2221 	/* Remove the COFF and ELF loaders. */
2222 	unregister_binfmt(&elf_format);
2223 }
2224 
2225 core_initcall(init_elf_binfmt);
2226 module_exit(exit_elf_binfmt);
2227 MODULE_LICENSE("GPL");
2228