xref: /openbmc/linux/fs/binfmt_elf.c (revision 3c6a73cc)
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 		goto out_free_dentry;
743 
744 	current->mm->start_stack = bprm->p;
745 
746 	/* Now we do a little grungy work by mmapping the ELF image into
747 	   the correct location in memory. */
748 	for(i = 0, elf_ppnt = elf_phdata;
749 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
750 		int elf_prot = 0, elf_flags;
751 		unsigned long k, vaddr;
752 
753 		if (elf_ppnt->p_type != PT_LOAD)
754 			continue;
755 
756 		if (unlikely (elf_brk > elf_bss)) {
757 			unsigned long nbyte;
758 
759 			/* There was a PT_LOAD segment with p_memsz > p_filesz
760 			   before this one. Map anonymous pages, if needed,
761 			   and clear the area.  */
762 			retval = set_brk(elf_bss + load_bias,
763 					 elf_brk + load_bias);
764 			if (retval)
765 				goto out_free_dentry;
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 or explicit setting of
802 			 * personality flags.
803 			 * If that is the case, retain the original non-zero
804 			 * load_bias value in order to establish proper
805 			 * non-randomized mappings.
806 			 */
807 			if (current->flags & PF_RANDOMIZE)
808 				load_bias = 0;
809 			else
810 				load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
811 #else
812 			load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
813 #endif
814 		}
815 
816 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
817 				elf_prot, elf_flags, 0);
818 		if (BAD_ADDR(error)) {
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 			retval = -EINVAL;
850 			goto out_free_dentry;
851 		}
852 
853 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
854 
855 		if (k > elf_bss)
856 			elf_bss = k;
857 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
858 			end_code = k;
859 		if (end_data < k)
860 			end_data = k;
861 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
862 		if (k > elf_brk)
863 			elf_brk = k;
864 	}
865 
866 	loc->elf_ex.e_entry += load_bias;
867 	elf_bss += load_bias;
868 	elf_brk += load_bias;
869 	start_code += load_bias;
870 	end_code += load_bias;
871 	start_data += load_bias;
872 	end_data += load_bias;
873 
874 	/* Calling set_brk effectively mmaps the pages that we need
875 	 * for the bss and break sections.  We must do this before
876 	 * mapping in the interpreter, to make sure it doesn't wind
877 	 * up getting placed where the bss needs to go.
878 	 */
879 	retval = set_brk(elf_bss, elf_brk);
880 	if (retval)
881 		goto out_free_dentry;
882 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
883 		retval = -EFAULT; /* Nobody gets to see this, but.. */
884 		goto out_free_dentry;
885 	}
886 
887 	if (elf_interpreter) {
888 		unsigned long interp_map_addr = 0;
889 
890 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
891 					    interpreter,
892 					    &interp_map_addr,
893 					    load_bias);
894 		if (!IS_ERR((void *)elf_entry)) {
895 			/*
896 			 * load_elf_interp() returns relocation
897 			 * adjustment
898 			 */
899 			interp_load_addr = elf_entry;
900 			elf_entry += loc->interp_elf_ex.e_entry;
901 		}
902 		if (BAD_ADDR(elf_entry)) {
903 			retval = IS_ERR((void *)elf_entry) ?
904 					(int)elf_entry : -EINVAL;
905 			goto out_free_dentry;
906 		}
907 		reloc_func_desc = interp_load_addr;
908 
909 		allow_write_access(interpreter);
910 		fput(interpreter);
911 		kfree(elf_interpreter);
912 	} else {
913 		elf_entry = loc->elf_ex.e_entry;
914 		if (BAD_ADDR(elf_entry)) {
915 			retval = -EINVAL;
916 			goto out_free_dentry;
917 		}
918 	}
919 
920 	kfree(elf_phdata);
921 
922 	set_binfmt(&elf_format);
923 
924 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
925 	retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
926 	if (retval < 0)
927 		goto out;
928 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
929 
930 	install_exec_creds(bprm);
931 	retval = create_elf_tables(bprm, &loc->elf_ex,
932 			  load_addr, interp_load_addr);
933 	if (retval < 0)
934 		goto out;
935 	/* N.B. passed_fileno might not be initialized? */
936 	current->mm->end_code = end_code;
937 	current->mm->start_code = start_code;
938 	current->mm->start_data = start_data;
939 	current->mm->end_data = end_data;
940 	current->mm->start_stack = bprm->p;
941 
942 #ifdef arch_randomize_brk
943 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
944 		current->mm->brk = current->mm->start_brk =
945 			arch_randomize_brk(current->mm);
946 #ifdef CONFIG_COMPAT_BRK
947 		current->brk_randomized = 1;
948 #endif
949 	}
950 #endif
951 
952 	if (current->personality & MMAP_PAGE_ZERO) {
953 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
954 		   and some applications "depend" upon this behavior.
955 		   Since we do not have the power to recompile these, we
956 		   emulate the SVr4 behavior. Sigh. */
957 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
958 				MAP_FIXED | MAP_PRIVATE, 0);
959 	}
960 
961 #ifdef ELF_PLAT_INIT
962 	/*
963 	 * The ABI may specify that certain registers be set up in special
964 	 * ways (on i386 %edx is the address of a DT_FINI function, for
965 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
966 	 * that the e_entry field is the address of the function descriptor
967 	 * for the startup routine, rather than the address of the startup
968 	 * routine itself.  This macro performs whatever initialization to
969 	 * the regs structure is required as well as any relocations to the
970 	 * function descriptor entries when executing dynamically links apps.
971 	 */
972 	ELF_PLAT_INIT(regs, reloc_func_desc);
973 #endif
974 
975 	start_thread(regs, elf_entry, bprm->p);
976 	retval = 0;
977 out:
978 	kfree(loc);
979 out_ret:
980 	return retval;
981 
982 	/* error cleanup */
983 out_free_dentry:
984 	allow_write_access(interpreter);
985 	if (interpreter)
986 		fput(interpreter);
987 out_free_interp:
988 	kfree(elf_interpreter);
989 out_free_ph:
990 	kfree(elf_phdata);
991 	goto out;
992 }
993 
994 #ifdef CONFIG_USELIB
995 /* This is really simpleminded and specialized - we are loading an
996    a.out library that is given an ELF header. */
997 static int load_elf_library(struct file *file)
998 {
999 	struct elf_phdr *elf_phdata;
1000 	struct elf_phdr *eppnt;
1001 	unsigned long elf_bss, bss, len;
1002 	int retval, error, i, j;
1003 	struct elfhdr elf_ex;
1004 
1005 	error = -ENOEXEC;
1006 	retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1007 	if (retval != sizeof(elf_ex))
1008 		goto out;
1009 
1010 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1011 		goto out;
1012 
1013 	/* First of all, some simple consistency checks */
1014 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1015 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1016 		goto out;
1017 
1018 	/* Now read in all of the header information */
1019 
1020 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1021 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1022 
1023 	error = -ENOMEM;
1024 	elf_phdata = kmalloc(j, GFP_KERNEL);
1025 	if (!elf_phdata)
1026 		goto out;
1027 
1028 	eppnt = elf_phdata;
1029 	error = -ENOEXEC;
1030 	retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1031 	if (retval != j)
1032 		goto out_free_ph;
1033 
1034 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1035 		if ((eppnt + i)->p_type == PT_LOAD)
1036 			j++;
1037 	if (j != 1)
1038 		goto out_free_ph;
1039 
1040 	while (eppnt->p_type != PT_LOAD)
1041 		eppnt++;
1042 
1043 	/* Now use mmap to map the library into memory. */
1044 	error = vm_mmap(file,
1045 			ELF_PAGESTART(eppnt->p_vaddr),
1046 			(eppnt->p_filesz +
1047 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1048 			PROT_READ | PROT_WRITE | PROT_EXEC,
1049 			MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1050 			(eppnt->p_offset -
1051 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1052 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1053 		goto out_free_ph;
1054 
1055 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1056 	if (padzero(elf_bss)) {
1057 		error = -EFAULT;
1058 		goto out_free_ph;
1059 	}
1060 
1061 	len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1062 			    ELF_MIN_ALIGN - 1);
1063 	bss = eppnt->p_memsz + eppnt->p_vaddr;
1064 	if (bss > len)
1065 		vm_brk(len, bss - len);
1066 	error = 0;
1067 
1068 out_free_ph:
1069 	kfree(elf_phdata);
1070 out:
1071 	return error;
1072 }
1073 #endif /* #ifdef CONFIG_USELIB */
1074 
1075 #ifdef CONFIG_ELF_CORE
1076 /*
1077  * ELF core dumper
1078  *
1079  * Modelled on fs/exec.c:aout_core_dump()
1080  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1081  */
1082 
1083 /*
1084  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1085  * that are useful for post-mortem analysis are included in every core dump.
1086  * In that way we ensure that the core dump is fully interpretable later
1087  * without matching up the same kernel and hardware config to see what PC values
1088  * meant. These special mappings include - vDSO, vsyscall, and other
1089  * architecture specific mappings
1090  */
1091 static bool always_dump_vma(struct vm_area_struct *vma)
1092 {
1093 	/* Any vsyscall mappings? */
1094 	if (vma == get_gate_vma(vma->vm_mm))
1095 		return true;
1096 
1097 	/*
1098 	 * Assume that all vmas with a .name op should always be dumped.
1099 	 * If this changes, a new vm_ops field can easily be added.
1100 	 */
1101 	if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1102 		return true;
1103 
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 		return 0;
1136 	}
1137 
1138 	/* Do not dump I/O mapped devices or special mappings */
1139 	if (vma->vm_flags & VM_IO)
1140 		return 0;
1141 
1142 	/* By default, dump shared memory if mapped from an anonymous file. */
1143 	if (vma->vm_flags & VM_SHARED) {
1144 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1145 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1146 			goto whole;
1147 		return 0;
1148 	}
1149 
1150 	/* Dump segments that have been written to.  */
1151 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1152 		goto whole;
1153 	if (vma->vm_file == NULL)
1154 		return 0;
1155 
1156 	if (FILTER(MAPPED_PRIVATE))
1157 		goto whole;
1158 
1159 	/*
1160 	 * If this looks like the beginning of a DSO or executable mapping,
1161 	 * check for an ELF header.  If we find one, dump the first page to
1162 	 * aid in determining what was mapped here.
1163 	 */
1164 	if (FILTER(ELF_HEADERS) &&
1165 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1166 		u32 __user *header = (u32 __user *) vma->vm_start;
1167 		u32 word;
1168 		mm_segment_t fs = get_fs();
1169 		/*
1170 		 * Doing it this way gets the constant folded by GCC.
1171 		 */
1172 		union {
1173 			u32 cmp;
1174 			char elfmag[SELFMAG];
1175 		} magic;
1176 		BUILD_BUG_ON(SELFMAG != sizeof word);
1177 		magic.elfmag[EI_MAG0] = ELFMAG0;
1178 		magic.elfmag[EI_MAG1] = ELFMAG1;
1179 		magic.elfmag[EI_MAG2] = ELFMAG2;
1180 		magic.elfmag[EI_MAG3] = ELFMAG3;
1181 		/*
1182 		 * Switch to the user "segment" for get_user(),
1183 		 * then put back what elf_core_dump() had in place.
1184 		 */
1185 		set_fs(USER_DS);
1186 		if (unlikely(get_user(word, header)))
1187 			word = 0;
1188 		set_fs(fs);
1189 		if (word == magic.cmp)
1190 			return PAGE_SIZE;
1191 	}
1192 
1193 #undef	FILTER
1194 
1195 	return 0;
1196 
1197 whole:
1198 	return vma->vm_end - vma->vm_start;
1199 }
1200 
1201 /* An ELF note in memory */
1202 struct memelfnote
1203 {
1204 	const char *name;
1205 	int type;
1206 	unsigned int datasz;
1207 	void *data;
1208 };
1209 
1210 static int notesize(struct memelfnote *en)
1211 {
1212 	int sz;
1213 
1214 	sz = sizeof(struct elf_note);
1215 	sz += roundup(strlen(en->name) + 1, 4);
1216 	sz += roundup(en->datasz, 4);
1217 
1218 	return sz;
1219 }
1220 
1221 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1222 {
1223 	struct elf_note en;
1224 	en.n_namesz = strlen(men->name) + 1;
1225 	en.n_descsz = men->datasz;
1226 	en.n_type = men->type;
1227 
1228 	return dump_emit(cprm, &en, sizeof(en)) &&
1229 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1230 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1231 }
1232 
1233 static void fill_elf_header(struct elfhdr *elf, int segs,
1234 			    u16 machine, u32 flags)
1235 {
1236 	memset(elf, 0, sizeof(*elf));
1237 
1238 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1239 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1240 	elf->e_ident[EI_DATA] = ELF_DATA;
1241 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1242 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1243 
1244 	elf->e_type = ET_CORE;
1245 	elf->e_machine = machine;
1246 	elf->e_version = EV_CURRENT;
1247 	elf->e_phoff = sizeof(struct elfhdr);
1248 	elf->e_flags = flags;
1249 	elf->e_ehsize = sizeof(struct elfhdr);
1250 	elf->e_phentsize = sizeof(struct elf_phdr);
1251 	elf->e_phnum = segs;
1252 
1253 	return;
1254 }
1255 
1256 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1257 {
1258 	phdr->p_type = PT_NOTE;
1259 	phdr->p_offset = offset;
1260 	phdr->p_vaddr = 0;
1261 	phdr->p_paddr = 0;
1262 	phdr->p_filesz = sz;
1263 	phdr->p_memsz = 0;
1264 	phdr->p_flags = 0;
1265 	phdr->p_align = 0;
1266 	return;
1267 }
1268 
1269 static void fill_note(struct memelfnote *note, const char *name, int type,
1270 		unsigned int sz, void *data)
1271 {
1272 	note->name = name;
1273 	note->type = type;
1274 	note->datasz = sz;
1275 	note->data = data;
1276 	return;
1277 }
1278 
1279 /*
1280  * fill up all the fields in prstatus from the given task struct, except
1281  * registers which need to be filled up separately.
1282  */
1283 static void fill_prstatus(struct elf_prstatus *prstatus,
1284 		struct task_struct *p, long signr)
1285 {
1286 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1287 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1288 	prstatus->pr_sighold = p->blocked.sig[0];
1289 	rcu_read_lock();
1290 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1291 	rcu_read_unlock();
1292 	prstatus->pr_pid = task_pid_vnr(p);
1293 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1294 	prstatus->pr_sid = task_session_vnr(p);
1295 	if (thread_group_leader(p)) {
1296 		struct task_cputime cputime;
1297 
1298 		/*
1299 		 * This is the record for the group leader.  It shows the
1300 		 * group-wide total, not its individual thread total.
1301 		 */
1302 		thread_group_cputime(p, &cputime);
1303 		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1304 		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1305 	} else {
1306 		cputime_t utime, stime;
1307 
1308 		task_cputime(p, &utime, &stime);
1309 		cputime_to_timeval(utime, &prstatus->pr_utime);
1310 		cputime_to_timeval(stime, &prstatus->pr_stime);
1311 	}
1312 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1313 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1314 }
1315 
1316 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1317 		       struct mm_struct *mm)
1318 {
1319 	const struct cred *cred;
1320 	unsigned int i, len;
1321 
1322 	/* first copy the parameters from user space */
1323 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1324 
1325 	len = mm->arg_end - mm->arg_start;
1326 	if (len >= ELF_PRARGSZ)
1327 		len = ELF_PRARGSZ-1;
1328 	if (copy_from_user(&psinfo->pr_psargs,
1329 		           (const char __user *)mm->arg_start, len))
1330 		return -EFAULT;
1331 	for(i = 0; i < len; i++)
1332 		if (psinfo->pr_psargs[i] == 0)
1333 			psinfo->pr_psargs[i] = ' ';
1334 	psinfo->pr_psargs[len] = 0;
1335 
1336 	rcu_read_lock();
1337 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1338 	rcu_read_unlock();
1339 	psinfo->pr_pid = task_pid_vnr(p);
1340 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1341 	psinfo->pr_sid = task_session_vnr(p);
1342 
1343 	i = p->state ? ffz(~p->state) + 1 : 0;
1344 	psinfo->pr_state = i;
1345 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1346 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1347 	psinfo->pr_nice = task_nice(p);
1348 	psinfo->pr_flag = p->flags;
1349 	rcu_read_lock();
1350 	cred = __task_cred(p);
1351 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1352 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1353 	rcu_read_unlock();
1354 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1355 
1356 	return 0;
1357 }
1358 
1359 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1360 {
1361 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1362 	int i = 0;
1363 	do
1364 		i += 2;
1365 	while (auxv[i - 2] != AT_NULL);
1366 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1367 }
1368 
1369 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1370 		const siginfo_t *siginfo)
1371 {
1372 	mm_segment_t old_fs = get_fs();
1373 	set_fs(KERNEL_DS);
1374 	copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1375 	set_fs(old_fs);
1376 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1377 }
1378 
1379 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1380 /*
1381  * Format of NT_FILE note:
1382  *
1383  * long count     -- how many files are mapped
1384  * long page_size -- units for file_ofs
1385  * array of [COUNT] elements of
1386  *   long start
1387  *   long end
1388  *   long file_ofs
1389  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1390  */
1391 static int fill_files_note(struct memelfnote *note)
1392 {
1393 	struct vm_area_struct *vma;
1394 	unsigned count, size, names_ofs, remaining, n;
1395 	user_long_t *data;
1396 	user_long_t *start_end_ofs;
1397 	char *name_base, *name_curpos;
1398 
1399 	/* *Estimated* file count and total data size needed */
1400 	count = current->mm->map_count;
1401 	size = count * 64;
1402 
1403 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1404  alloc:
1405 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1406 		return -EINVAL;
1407 	size = round_up(size, PAGE_SIZE);
1408 	data = vmalloc(size);
1409 	if (!data)
1410 		return -ENOMEM;
1411 
1412 	start_end_ofs = data + 2;
1413 	name_base = name_curpos = ((char *)data) + names_ofs;
1414 	remaining = size - names_ofs;
1415 	count = 0;
1416 	for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1417 		struct file *file;
1418 		const char *filename;
1419 
1420 		file = vma->vm_file;
1421 		if (!file)
1422 			continue;
1423 		filename = d_path(&file->f_path, name_curpos, remaining);
1424 		if (IS_ERR(filename)) {
1425 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1426 				vfree(data);
1427 				size = size * 5 / 4;
1428 				goto alloc;
1429 			}
1430 			continue;
1431 		}
1432 
1433 		/* d_path() fills at the end, move name down */
1434 		/* n = strlen(filename) + 1: */
1435 		n = (name_curpos + remaining) - filename;
1436 		remaining = filename - name_curpos;
1437 		memmove(name_curpos, filename, n);
1438 		name_curpos += n;
1439 
1440 		*start_end_ofs++ = vma->vm_start;
1441 		*start_end_ofs++ = vma->vm_end;
1442 		*start_end_ofs++ = vma->vm_pgoff;
1443 		count++;
1444 	}
1445 
1446 	/* Now we know exact count of files, can store it */
1447 	data[0] = count;
1448 	data[1] = PAGE_SIZE;
1449 	/*
1450 	 * Count usually is less than current->mm->map_count,
1451 	 * we need to move filenames down.
1452 	 */
1453 	n = current->mm->map_count - count;
1454 	if (n != 0) {
1455 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1456 		memmove(name_base - shift_bytes, name_base,
1457 			name_curpos - name_base);
1458 		name_curpos -= shift_bytes;
1459 	}
1460 
1461 	size = name_curpos - (char *)data;
1462 	fill_note(note, "CORE", NT_FILE, size, data);
1463 	return 0;
1464 }
1465 
1466 #ifdef CORE_DUMP_USE_REGSET
1467 #include <linux/regset.h>
1468 
1469 struct elf_thread_core_info {
1470 	struct elf_thread_core_info *next;
1471 	struct task_struct *task;
1472 	struct elf_prstatus prstatus;
1473 	struct memelfnote notes[0];
1474 };
1475 
1476 struct elf_note_info {
1477 	struct elf_thread_core_info *thread;
1478 	struct memelfnote psinfo;
1479 	struct memelfnote signote;
1480 	struct memelfnote auxv;
1481 	struct memelfnote files;
1482 	user_siginfo_t csigdata;
1483 	size_t size;
1484 	int thread_notes;
1485 };
1486 
1487 /*
1488  * When a regset has a writeback hook, we call it on each thread before
1489  * dumping user memory.  On register window machines, this makes sure the
1490  * user memory backing the register data is up to date before we read it.
1491  */
1492 static void do_thread_regset_writeback(struct task_struct *task,
1493 				       const struct user_regset *regset)
1494 {
1495 	if (regset->writeback)
1496 		regset->writeback(task, regset, 1);
1497 }
1498 
1499 #ifndef PR_REG_SIZE
1500 #define PR_REG_SIZE(S) sizeof(S)
1501 #endif
1502 
1503 #ifndef PRSTATUS_SIZE
1504 #define PRSTATUS_SIZE(S) sizeof(S)
1505 #endif
1506 
1507 #ifndef PR_REG_PTR
1508 #define PR_REG_PTR(S) (&((S)->pr_reg))
1509 #endif
1510 
1511 #ifndef SET_PR_FPVALID
1512 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1513 #endif
1514 
1515 static int fill_thread_core_info(struct elf_thread_core_info *t,
1516 				 const struct user_regset_view *view,
1517 				 long signr, size_t *total)
1518 {
1519 	unsigned int i;
1520 
1521 	/*
1522 	 * NT_PRSTATUS is the one special case, because the regset data
1523 	 * goes into the pr_reg field inside the note contents, rather
1524 	 * than being the whole note contents.  We fill the reset in here.
1525 	 * We assume that regset 0 is NT_PRSTATUS.
1526 	 */
1527 	fill_prstatus(&t->prstatus, t->task, signr);
1528 	(void) view->regsets[0].get(t->task, &view->regsets[0],
1529 				    0, PR_REG_SIZE(t->prstatus.pr_reg),
1530 				    PR_REG_PTR(&t->prstatus), NULL);
1531 
1532 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1533 		  PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1534 	*total += notesize(&t->notes[0]);
1535 
1536 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1537 
1538 	/*
1539 	 * Each other regset might generate a note too.  For each regset
1540 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1541 	 * all zero and we'll know to skip writing it later.
1542 	 */
1543 	for (i = 1; i < view->n; ++i) {
1544 		const struct user_regset *regset = &view->regsets[i];
1545 		do_thread_regset_writeback(t->task, regset);
1546 		if (regset->core_note_type && regset->get &&
1547 		    (!regset->active || regset->active(t->task, regset))) {
1548 			int ret;
1549 			size_t size = regset->n * regset->size;
1550 			void *data = kmalloc(size, GFP_KERNEL);
1551 			if (unlikely(!data))
1552 				return 0;
1553 			ret = regset->get(t->task, regset,
1554 					  0, size, data, NULL);
1555 			if (unlikely(ret))
1556 				kfree(data);
1557 			else {
1558 				if (regset->core_note_type != NT_PRFPREG)
1559 					fill_note(&t->notes[i], "LINUX",
1560 						  regset->core_note_type,
1561 						  size, data);
1562 				else {
1563 					SET_PR_FPVALID(&t->prstatus, 1);
1564 					fill_note(&t->notes[i], "CORE",
1565 						  NT_PRFPREG, size, data);
1566 				}
1567 				*total += notesize(&t->notes[i]);
1568 			}
1569 		}
1570 	}
1571 
1572 	return 1;
1573 }
1574 
1575 static int fill_note_info(struct elfhdr *elf, int phdrs,
1576 			  struct elf_note_info *info,
1577 			  const siginfo_t *siginfo, struct pt_regs *regs)
1578 {
1579 	struct task_struct *dump_task = current;
1580 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1581 	struct elf_thread_core_info *t;
1582 	struct elf_prpsinfo *psinfo;
1583 	struct core_thread *ct;
1584 	unsigned int i;
1585 
1586 	info->size = 0;
1587 	info->thread = NULL;
1588 
1589 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1590 	if (psinfo == NULL) {
1591 		info->psinfo.data = NULL; /* So we don't free this wrongly */
1592 		return 0;
1593 	}
1594 
1595 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1596 
1597 	/*
1598 	 * Figure out how many notes we're going to need for each thread.
1599 	 */
1600 	info->thread_notes = 0;
1601 	for (i = 0; i < view->n; ++i)
1602 		if (view->regsets[i].core_note_type != 0)
1603 			++info->thread_notes;
1604 
1605 	/*
1606 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1607 	 * since it is our one special case.
1608 	 */
1609 	if (unlikely(info->thread_notes == 0) ||
1610 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1611 		WARN_ON(1);
1612 		return 0;
1613 	}
1614 
1615 	/*
1616 	 * Initialize the ELF file header.
1617 	 */
1618 	fill_elf_header(elf, phdrs,
1619 			view->e_machine, view->e_flags);
1620 
1621 	/*
1622 	 * Allocate a structure for each thread.
1623 	 */
1624 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1625 		t = kzalloc(offsetof(struct elf_thread_core_info,
1626 				     notes[info->thread_notes]),
1627 			    GFP_KERNEL);
1628 		if (unlikely(!t))
1629 			return 0;
1630 
1631 		t->task = ct->task;
1632 		if (ct->task == dump_task || !info->thread) {
1633 			t->next = info->thread;
1634 			info->thread = t;
1635 		} else {
1636 			/*
1637 			 * Make sure to keep the original task at
1638 			 * the head of the list.
1639 			 */
1640 			t->next = info->thread->next;
1641 			info->thread->next = t;
1642 		}
1643 	}
1644 
1645 	/*
1646 	 * Now fill in each thread's information.
1647 	 */
1648 	for (t = info->thread; t != NULL; t = t->next)
1649 		if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1650 			return 0;
1651 
1652 	/*
1653 	 * Fill in the two process-wide notes.
1654 	 */
1655 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1656 	info->size += notesize(&info->psinfo);
1657 
1658 	fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1659 	info->size += notesize(&info->signote);
1660 
1661 	fill_auxv_note(&info->auxv, current->mm);
1662 	info->size += notesize(&info->auxv);
1663 
1664 	if (fill_files_note(&info->files) == 0)
1665 		info->size += notesize(&info->files);
1666 
1667 	return 1;
1668 }
1669 
1670 static size_t get_note_info_size(struct elf_note_info *info)
1671 {
1672 	return info->size;
1673 }
1674 
1675 /*
1676  * Write all the notes for each thread.  When writing the first thread, the
1677  * process-wide notes are interleaved after the first thread-specific note.
1678  */
1679 static int write_note_info(struct elf_note_info *info,
1680 			   struct coredump_params *cprm)
1681 {
1682 	bool first = true;
1683 	struct elf_thread_core_info *t = info->thread;
1684 
1685 	do {
1686 		int i;
1687 
1688 		if (!writenote(&t->notes[0], cprm))
1689 			return 0;
1690 
1691 		if (first && !writenote(&info->psinfo, cprm))
1692 			return 0;
1693 		if (first && !writenote(&info->signote, cprm))
1694 			return 0;
1695 		if (first && !writenote(&info->auxv, cprm))
1696 			return 0;
1697 		if (first && info->files.data &&
1698 				!writenote(&info->files, cprm))
1699 			return 0;
1700 
1701 		for (i = 1; i < info->thread_notes; ++i)
1702 			if (t->notes[i].data &&
1703 			    !writenote(&t->notes[i], cprm))
1704 				return 0;
1705 
1706 		first = false;
1707 		t = t->next;
1708 	} while (t);
1709 
1710 	return 1;
1711 }
1712 
1713 static void free_note_info(struct elf_note_info *info)
1714 {
1715 	struct elf_thread_core_info *threads = info->thread;
1716 	while (threads) {
1717 		unsigned int i;
1718 		struct elf_thread_core_info *t = threads;
1719 		threads = t->next;
1720 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1721 		for (i = 1; i < info->thread_notes; ++i)
1722 			kfree(t->notes[i].data);
1723 		kfree(t);
1724 	}
1725 	kfree(info->psinfo.data);
1726 	vfree(info->files.data);
1727 }
1728 
1729 #else
1730 
1731 /* Here is the structure in which status of each thread is captured. */
1732 struct elf_thread_status
1733 {
1734 	struct list_head list;
1735 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1736 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1737 	struct task_struct *thread;
1738 #ifdef ELF_CORE_COPY_XFPREGS
1739 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1740 #endif
1741 	struct memelfnote notes[3];
1742 	int num_notes;
1743 };
1744 
1745 /*
1746  * In order to add the specific thread information for the elf file format,
1747  * we need to keep a linked list of every threads pr_status and then create
1748  * a single section for them in the final core file.
1749  */
1750 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1751 {
1752 	int sz = 0;
1753 	struct task_struct *p = t->thread;
1754 	t->num_notes = 0;
1755 
1756 	fill_prstatus(&t->prstatus, p, signr);
1757 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1758 
1759 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1760 		  &(t->prstatus));
1761 	t->num_notes++;
1762 	sz += notesize(&t->notes[0]);
1763 
1764 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1765 								&t->fpu))) {
1766 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1767 			  &(t->fpu));
1768 		t->num_notes++;
1769 		sz += notesize(&t->notes[1]);
1770 	}
1771 
1772 #ifdef ELF_CORE_COPY_XFPREGS
1773 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1774 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1775 			  sizeof(t->xfpu), &t->xfpu);
1776 		t->num_notes++;
1777 		sz += notesize(&t->notes[2]);
1778 	}
1779 #endif
1780 	return sz;
1781 }
1782 
1783 struct elf_note_info {
1784 	struct memelfnote *notes;
1785 	struct memelfnote *notes_files;
1786 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1787 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1788 	struct list_head thread_list;
1789 	elf_fpregset_t *fpu;
1790 #ifdef ELF_CORE_COPY_XFPREGS
1791 	elf_fpxregset_t *xfpu;
1792 #endif
1793 	user_siginfo_t csigdata;
1794 	int thread_status_size;
1795 	int numnote;
1796 };
1797 
1798 static int elf_note_info_init(struct elf_note_info *info)
1799 {
1800 	memset(info, 0, sizeof(*info));
1801 	INIT_LIST_HEAD(&info->thread_list);
1802 
1803 	/* Allocate space for ELF notes */
1804 	info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1805 	if (!info->notes)
1806 		return 0;
1807 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1808 	if (!info->psinfo)
1809 		return 0;
1810 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1811 	if (!info->prstatus)
1812 		return 0;
1813 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1814 	if (!info->fpu)
1815 		return 0;
1816 #ifdef ELF_CORE_COPY_XFPREGS
1817 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1818 	if (!info->xfpu)
1819 		return 0;
1820 #endif
1821 	return 1;
1822 }
1823 
1824 static int fill_note_info(struct elfhdr *elf, int phdrs,
1825 			  struct elf_note_info *info,
1826 			  const siginfo_t *siginfo, struct pt_regs *regs)
1827 {
1828 	struct list_head *t;
1829 	struct core_thread *ct;
1830 	struct elf_thread_status *ets;
1831 
1832 	if (!elf_note_info_init(info))
1833 		return 0;
1834 
1835 	for (ct = current->mm->core_state->dumper.next;
1836 					ct; ct = ct->next) {
1837 		ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1838 		if (!ets)
1839 			return 0;
1840 
1841 		ets->thread = ct->task;
1842 		list_add(&ets->list, &info->thread_list);
1843 	}
1844 
1845 	list_for_each(t, &info->thread_list) {
1846 		int sz;
1847 
1848 		ets = list_entry(t, struct elf_thread_status, list);
1849 		sz = elf_dump_thread_status(siginfo->si_signo, ets);
1850 		info->thread_status_size += sz;
1851 	}
1852 	/* now collect the dump for the current */
1853 	memset(info->prstatus, 0, sizeof(*info->prstatus));
1854 	fill_prstatus(info->prstatus, current, siginfo->si_signo);
1855 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1856 
1857 	/* Set up header */
1858 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1859 
1860 	/*
1861 	 * Set up the notes in similar form to SVR4 core dumps made
1862 	 * with info from their /proc.
1863 	 */
1864 
1865 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1866 		  sizeof(*info->prstatus), info->prstatus);
1867 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
1868 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1869 		  sizeof(*info->psinfo), info->psinfo);
1870 
1871 	fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1872 	fill_auxv_note(info->notes + 3, current->mm);
1873 	info->numnote = 4;
1874 
1875 	if (fill_files_note(info->notes + info->numnote) == 0) {
1876 		info->notes_files = info->notes + info->numnote;
1877 		info->numnote++;
1878 	}
1879 
1880 	/* Try to dump the FPU. */
1881 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1882 							       info->fpu);
1883 	if (info->prstatus->pr_fpvalid)
1884 		fill_note(info->notes + info->numnote++,
1885 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1886 #ifdef ELF_CORE_COPY_XFPREGS
1887 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
1888 		fill_note(info->notes + info->numnote++,
1889 			  "LINUX", ELF_CORE_XFPREG_TYPE,
1890 			  sizeof(*info->xfpu), info->xfpu);
1891 #endif
1892 
1893 	return 1;
1894 }
1895 
1896 static size_t get_note_info_size(struct elf_note_info *info)
1897 {
1898 	int sz = 0;
1899 	int i;
1900 
1901 	for (i = 0; i < info->numnote; i++)
1902 		sz += notesize(info->notes + i);
1903 
1904 	sz += info->thread_status_size;
1905 
1906 	return sz;
1907 }
1908 
1909 static int write_note_info(struct elf_note_info *info,
1910 			   struct coredump_params *cprm)
1911 {
1912 	int i;
1913 	struct list_head *t;
1914 
1915 	for (i = 0; i < info->numnote; i++)
1916 		if (!writenote(info->notes + i, cprm))
1917 			return 0;
1918 
1919 	/* write out the thread status notes section */
1920 	list_for_each(t, &info->thread_list) {
1921 		struct elf_thread_status *tmp =
1922 				list_entry(t, struct elf_thread_status, list);
1923 
1924 		for (i = 0; i < tmp->num_notes; i++)
1925 			if (!writenote(&tmp->notes[i], cprm))
1926 				return 0;
1927 	}
1928 
1929 	return 1;
1930 }
1931 
1932 static void free_note_info(struct elf_note_info *info)
1933 {
1934 	while (!list_empty(&info->thread_list)) {
1935 		struct list_head *tmp = info->thread_list.next;
1936 		list_del(tmp);
1937 		kfree(list_entry(tmp, struct elf_thread_status, list));
1938 	}
1939 
1940 	/* Free data possibly allocated by fill_files_note(): */
1941 	if (info->notes_files)
1942 		vfree(info->notes_files->data);
1943 
1944 	kfree(info->prstatus);
1945 	kfree(info->psinfo);
1946 	kfree(info->notes);
1947 	kfree(info->fpu);
1948 #ifdef ELF_CORE_COPY_XFPREGS
1949 	kfree(info->xfpu);
1950 #endif
1951 }
1952 
1953 #endif
1954 
1955 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1956 					struct vm_area_struct *gate_vma)
1957 {
1958 	struct vm_area_struct *ret = tsk->mm->mmap;
1959 
1960 	if (ret)
1961 		return ret;
1962 	return gate_vma;
1963 }
1964 /*
1965  * Helper function for iterating across a vma list.  It ensures that the caller
1966  * will visit `gate_vma' prior to terminating the search.
1967  */
1968 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1969 					struct vm_area_struct *gate_vma)
1970 {
1971 	struct vm_area_struct *ret;
1972 
1973 	ret = this_vma->vm_next;
1974 	if (ret)
1975 		return ret;
1976 	if (this_vma == gate_vma)
1977 		return NULL;
1978 	return gate_vma;
1979 }
1980 
1981 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1982 			     elf_addr_t e_shoff, int segs)
1983 {
1984 	elf->e_shoff = e_shoff;
1985 	elf->e_shentsize = sizeof(*shdr4extnum);
1986 	elf->e_shnum = 1;
1987 	elf->e_shstrndx = SHN_UNDEF;
1988 
1989 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1990 
1991 	shdr4extnum->sh_type = SHT_NULL;
1992 	shdr4extnum->sh_size = elf->e_shnum;
1993 	shdr4extnum->sh_link = elf->e_shstrndx;
1994 	shdr4extnum->sh_info = segs;
1995 }
1996 
1997 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
1998 				     unsigned long mm_flags)
1999 {
2000 	struct vm_area_struct *vma;
2001 	size_t size = 0;
2002 
2003 	for (vma = first_vma(current, gate_vma); vma != NULL;
2004 	     vma = next_vma(vma, gate_vma))
2005 		size += vma_dump_size(vma, mm_flags);
2006 	return size;
2007 }
2008 
2009 /*
2010  * Actual dumper
2011  *
2012  * This is a two-pass process; first we find the offsets of the bits,
2013  * and then they are actually written out.  If we run out of core limit
2014  * we just truncate.
2015  */
2016 static int elf_core_dump(struct coredump_params *cprm)
2017 {
2018 	int has_dumped = 0;
2019 	mm_segment_t fs;
2020 	int segs;
2021 	struct vm_area_struct *vma, *gate_vma;
2022 	struct elfhdr *elf = NULL;
2023 	loff_t offset = 0, dataoff;
2024 	struct elf_note_info info = { };
2025 	struct elf_phdr *phdr4note = NULL;
2026 	struct elf_shdr *shdr4extnum = NULL;
2027 	Elf_Half e_phnum;
2028 	elf_addr_t e_shoff;
2029 
2030 	/*
2031 	 * We no longer stop all VM operations.
2032 	 *
2033 	 * This is because those proceses that could possibly change map_count
2034 	 * or the mmap / vma pages are now blocked in do_exit on current
2035 	 * finishing this core dump.
2036 	 *
2037 	 * Only ptrace can touch these memory addresses, but it doesn't change
2038 	 * the map_count or the pages allocated. So no possibility of crashing
2039 	 * exists while dumping the mm->vm_next areas to the core file.
2040 	 */
2041 
2042 	/* alloc memory for large data structures: too large to be on stack */
2043 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2044 	if (!elf)
2045 		goto out;
2046 	/*
2047 	 * The number of segs are recored into ELF header as 16bit value.
2048 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2049 	 */
2050 	segs = current->mm->map_count;
2051 	segs += elf_core_extra_phdrs();
2052 
2053 	gate_vma = get_gate_vma(current->mm);
2054 	if (gate_vma != NULL)
2055 		segs++;
2056 
2057 	/* for notes section */
2058 	segs++;
2059 
2060 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2061 	 * this, kernel supports extended numbering. Have a look at
2062 	 * include/linux/elf.h for further information. */
2063 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2064 
2065 	/*
2066 	 * Collect all the non-memory information about the process for the
2067 	 * notes.  This also sets up the file header.
2068 	 */
2069 	if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2070 		goto cleanup;
2071 
2072 	has_dumped = 1;
2073 
2074 	fs = get_fs();
2075 	set_fs(KERNEL_DS);
2076 
2077 	offset += sizeof(*elf);				/* Elf header */
2078 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2079 
2080 	/* Write notes phdr entry */
2081 	{
2082 		size_t sz = get_note_info_size(&info);
2083 
2084 		sz += elf_coredump_extra_notes_size();
2085 
2086 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2087 		if (!phdr4note)
2088 			goto end_coredump;
2089 
2090 		fill_elf_note_phdr(phdr4note, sz, offset);
2091 		offset += sz;
2092 	}
2093 
2094 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2095 
2096 	offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2097 	offset += elf_core_extra_data_size();
2098 	e_shoff = offset;
2099 
2100 	if (e_phnum == PN_XNUM) {
2101 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2102 		if (!shdr4extnum)
2103 			goto end_coredump;
2104 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2105 	}
2106 
2107 	offset = dataoff;
2108 
2109 	if (!dump_emit(cprm, elf, sizeof(*elf)))
2110 		goto end_coredump;
2111 
2112 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2113 		goto end_coredump;
2114 
2115 	/* Write program headers for segments dump */
2116 	for (vma = first_vma(current, gate_vma); vma != NULL;
2117 			vma = next_vma(vma, gate_vma)) {
2118 		struct elf_phdr phdr;
2119 
2120 		phdr.p_type = PT_LOAD;
2121 		phdr.p_offset = offset;
2122 		phdr.p_vaddr = vma->vm_start;
2123 		phdr.p_paddr = 0;
2124 		phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2125 		phdr.p_memsz = vma->vm_end - vma->vm_start;
2126 		offset += phdr.p_filesz;
2127 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2128 		if (vma->vm_flags & VM_WRITE)
2129 			phdr.p_flags |= PF_W;
2130 		if (vma->vm_flags & VM_EXEC)
2131 			phdr.p_flags |= PF_X;
2132 		phdr.p_align = ELF_EXEC_PAGESIZE;
2133 
2134 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2135 			goto end_coredump;
2136 	}
2137 
2138 	if (!elf_core_write_extra_phdrs(cprm, offset))
2139 		goto end_coredump;
2140 
2141  	/* write out the notes section */
2142 	if (!write_note_info(&info, cprm))
2143 		goto end_coredump;
2144 
2145 	if (elf_coredump_extra_notes_write(cprm))
2146 		goto end_coredump;
2147 
2148 	/* Align to page */
2149 	if (!dump_skip(cprm, dataoff - cprm->written))
2150 		goto end_coredump;
2151 
2152 	for (vma = first_vma(current, gate_vma); vma != NULL;
2153 			vma = next_vma(vma, gate_vma)) {
2154 		unsigned long addr;
2155 		unsigned long end;
2156 
2157 		end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2158 
2159 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2160 			struct page *page;
2161 			int stop;
2162 
2163 			page = get_dump_page(addr);
2164 			if (page) {
2165 				void *kaddr = kmap(page);
2166 				stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2167 				kunmap(page);
2168 				page_cache_release(page);
2169 			} else
2170 				stop = !dump_skip(cprm, PAGE_SIZE);
2171 			if (stop)
2172 				goto end_coredump;
2173 		}
2174 	}
2175 
2176 	if (!elf_core_write_extra_data(cprm))
2177 		goto end_coredump;
2178 
2179 	if (e_phnum == PN_XNUM) {
2180 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2181 			goto end_coredump;
2182 	}
2183 
2184 end_coredump:
2185 	set_fs(fs);
2186 
2187 cleanup:
2188 	free_note_info(&info);
2189 	kfree(shdr4extnum);
2190 	kfree(phdr4note);
2191 	kfree(elf);
2192 out:
2193 	return has_dumped;
2194 }
2195 
2196 #endif		/* CONFIG_ELF_CORE */
2197 
2198 static int __init init_elf_binfmt(void)
2199 {
2200 	register_binfmt(&elf_format);
2201 	return 0;
2202 }
2203 
2204 static void __exit exit_elf_binfmt(void)
2205 {
2206 	/* Remove the COFF and ELF loaders. */
2207 	unregister_binfmt(&elf_format);
2208 }
2209 
2210 core_initcall(init_elf_binfmt);
2211 module_exit(exit_elf_binfmt);
2212 MODULE_LICENSE("GPL");
2213