xref: /linux/fs/binfmt_elf.c (revision 17afab1de42236ee2f6235f4383cc6f3f13f8a10)
1 /*
2  * linux/fs/binfmt_elf.c
3  *
4  * These are the functions used to load ELF format executables as used
5  * on SVr4 machines.  Information on the format may be found in the book
6  * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
7  * Tools".
8  *
9  * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
10  */
11 
12 #include <linux/module.h>
13 #include <linux/kernel.h>
14 #include <linux/fs.h>
15 #include <linux/mm.h>
16 #include <linux/mman.h>
17 #include <linux/errno.h>
18 #include <linux/signal.h>
19 #include <linux/binfmts.h>
20 #include <linux/string.h>
21 #include <linux/file.h>
22 #include <linux/slab.h>
23 #include <linux/personality.h>
24 #include <linux/elfcore.h>
25 #include <linux/init.h>
26 #include <linux/highuid.h>
27 #include <linux/compiler.h>
28 #include <linux/highmem.h>
29 #include <linux/pagemap.h>
30 #include <linux/vmalloc.h>
31 #include <linux/security.h>
32 #include <linux/random.h>
33 #include <linux/elf.h>
34 #include <linux/utsname.h>
35 #include <linux/coredump.h>
36 #include <linux/sched.h>
37 #include <asm/uaccess.h>
38 #include <asm/param.h>
39 #include <asm/page.h>
40 
41 #ifndef user_long_t
42 #define user_long_t long
43 #endif
44 #ifndef user_siginfo_t
45 #define user_siginfo_t siginfo_t
46 #endif
47 
48 static int load_elf_binary(struct linux_binprm *bprm);
49 static int load_elf_library(struct file *);
50 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
51 				int, int, unsigned long);
52 
53 /*
54  * If we don't support core dumping, then supply a NULL so we
55  * don't even try.
56  */
57 #ifdef CONFIG_ELF_CORE
58 static int elf_core_dump(struct coredump_params *cprm);
59 #else
60 #define elf_core_dump	NULL
61 #endif
62 
63 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
64 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
65 #else
66 #define ELF_MIN_ALIGN	PAGE_SIZE
67 #endif
68 
69 #ifndef ELF_CORE_EFLAGS
70 #define ELF_CORE_EFLAGS	0
71 #endif
72 
73 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
74 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
75 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
76 
77 static struct linux_binfmt elf_format = {
78 	.module		= THIS_MODULE,
79 	.load_binary	= load_elf_binary,
80 	.load_shlib	= load_elf_library,
81 	.core_dump	= elf_core_dump,
82 	.min_coredump	= ELF_EXEC_PAGESIZE,
83 };
84 
85 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
86 
87 static int set_brk(unsigned long start, unsigned long end)
88 {
89 	start = ELF_PAGEALIGN(start);
90 	end = ELF_PAGEALIGN(end);
91 	if (end > start) {
92 		unsigned long addr;
93 		addr = vm_brk(start, end - start);
94 		if (BAD_ADDR(addr))
95 			return addr;
96 	}
97 	current->mm->start_brk = current->mm->brk = end;
98 	return 0;
99 }
100 
101 /* We need to explicitly zero any fractional pages
102    after the data section (i.e. bss).  This would
103    contain the junk from the file that should not
104    be in memory
105  */
106 static int padzero(unsigned long elf_bss)
107 {
108 	unsigned long nbyte;
109 
110 	nbyte = ELF_PAGEOFFSET(elf_bss);
111 	if (nbyte) {
112 		nbyte = ELF_MIN_ALIGN - nbyte;
113 		if (clear_user((void __user *) elf_bss, nbyte))
114 			return -EFAULT;
115 	}
116 	return 0;
117 }
118 
119 /* Let's use some macros to make this stack manipulation a little clearer */
120 #ifdef CONFIG_STACK_GROWSUP
121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
122 #define STACK_ROUND(sp, items) \
123 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
124 #define STACK_ALLOC(sp, len) ({ \
125 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
126 	old_sp; })
127 #else
128 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
129 #define STACK_ROUND(sp, items) \
130 	(((unsigned long) (sp - items)) &~ 15UL)
131 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
132 #endif
133 
134 #ifndef ELF_BASE_PLATFORM
135 /*
136  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
137  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
138  * will be copied to the user stack in the same manner as AT_PLATFORM.
139  */
140 #define ELF_BASE_PLATFORM NULL
141 #endif
142 
143 static int
144 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
145 		unsigned long load_addr, unsigned long interp_load_addr)
146 {
147 	unsigned long p = bprm->p;
148 	int argc = bprm->argc;
149 	int envc = bprm->envc;
150 	elf_addr_t __user *argv;
151 	elf_addr_t __user *envp;
152 	elf_addr_t __user *sp;
153 	elf_addr_t __user *u_platform;
154 	elf_addr_t __user *u_base_platform;
155 	elf_addr_t __user *u_rand_bytes;
156 	const char *k_platform = ELF_PLATFORM;
157 	const char *k_base_platform = ELF_BASE_PLATFORM;
158 	unsigned char k_rand_bytes[16];
159 	int items;
160 	elf_addr_t *elf_info;
161 	int ei_index = 0;
162 	const struct cred *cred = current_cred();
163 	struct vm_area_struct *vma;
164 
165 	/*
166 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
167 	 * evictions by the processes running on the same package. One
168 	 * thing we can do is to shuffle the initial stack for them.
169 	 */
170 
171 	p = arch_align_stack(p);
172 
173 	/*
174 	 * If this architecture has a platform capability string, copy it
175 	 * to userspace.  In some cases (Sparc), this info is impossible
176 	 * for userspace to get any other way, in others (i386) it is
177 	 * merely difficult.
178 	 */
179 	u_platform = NULL;
180 	if (k_platform) {
181 		size_t len = strlen(k_platform) + 1;
182 
183 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
184 		if (__copy_to_user(u_platform, k_platform, len))
185 			return -EFAULT;
186 	}
187 
188 	/*
189 	 * If this architecture has a "base" platform capability
190 	 * string, copy it to userspace.
191 	 */
192 	u_base_platform = NULL;
193 	if (k_base_platform) {
194 		size_t len = strlen(k_base_platform) + 1;
195 
196 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
197 		if (__copy_to_user(u_base_platform, k_base_platform, len))
198 			return -EFAULT;
199 	}
200 
201 	/*
202 	 * Generate 16 random bytes for userspace PRNG seeding.
203 	 */
204 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
205 	u_rand_bytes = (elf_addr_t __user *)
206 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
207 	if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
208 		return -EFAULT;
209 
210 	/* Create the ELF interpreter info */
211 	elf_info = (elf_addr_t *)current->mm->saved_auxv;
212 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
213 #define NEW_AUX_ENT(id, val) \
214 	do { \
215 		elf_info[ei_index++] = id; \
216 		elf_info[ei_index++] = val; \
217 	} while (0)
218 
219 #ifdef ARCH_DLINFO
220 	/*
221 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
222 	 * AUXV.
223 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
224 	 * ARCH_DLINFO changes
225 	 */
226 	ARCH_DLINFO;
227 #endif
228 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
229 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
230 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
231 	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
232 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
233 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
234 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
235 	NEW_AUX_ENT(AT_FLAGS, 0);
236 	NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
237 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
238 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
239 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
240 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
241  	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
242 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
243 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
244 	if (k_platform) {
245 		NEW_AUX_ENT(AT_PLATFORM,
246 			    (elf_addr_t)(unsigned long)u_platform);
247 	}
248 	if (k_base_platform) {
249 		NEW_AUX_ENT(AT_BASE_PLATFORM,
250 			    (elf_addr_t)(unsigned long)u_base_platform);
251 	}
252 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
253 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
254 	}
255 #undef NEW_AUX_ENT
256 	/* AT_NULL is zero; clear the rest too */
257 	memset(&elf_info[ei_index], 0,
258 	       sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
259 
260 	/* And advance past the AT_NULL entry.  */
261 	ei_index += 2;
262 
263 	sp = STACK_ADD(p, ei_index);
264 
265 	items = (argc + 1) + (envc + 1) + 1;
266 	bprm->p = STACK_ROUND(sp, items);
267 
268 	/* Point sp at the lowest address on the stack */
269 #ifdef CONFIG_STACK_GROWSUP
270 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
271 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
272 #else
273 	sp = (elf_addr_t __user *)bprm->p;
274 #endif
275 
276 
277 	/*
278 	 * Grow the stack manually; some architectures have a limit on how
279 	 * far ahead a user-space access may be in order to grow the stack.
280 	 */
281 	vma = find_extend_vma(current->mm, bprm->p);
282 	if (!vma)
283 		return -EFAULT;
284 
285 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
286 	if (__put_user(argc, sp++))
287 		return -EFAULT;
288 	argv = sp;
289 	envp = argv + argc + 1;
290 
291 	/* Populate argv and envp */
292 	p = current->mm->arg_end = current->mm->arg_start;
293 	while (argc-- > 0) {
294 		size_t len;
295 		if (__put_user((elf_addr_t)p, argv++))
296 			return -EFAULT;
297 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
298 		if (!len || len > MAX_ARG_STRLEN)
299 			return -EINVAL;
300 		p += len;
301 	}
302 	if (__put_user(0, argv))
303 		return -EFAULT;
304 	current->mm->arg_end = current->mm->env_start = p;
305 	while (envc-- > 0) {
306 		size_t len;
307 		if (__put_user((elf_addr_t)p, envp++))
308 			return -EFAULT;
309 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
310 		if (!len || len > MAX_ARG_STRLEN)
311 			return -EINVAL;
312 		p += len;
313 	}
314 	if (__put_user(0, envp))
315 		return -EFAULT;
316 	current->mm->env_end = p;
317 
318 	/* Put the elf_info on the stack in the right place.  */
319 	sp = (elf_addr_t __user *)envp + 1;
320 	if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
321 		return -EFAULT;
322 	return 0;
323 }
324 
325 #ifndef elf_map
326 
327 static unsigned long elf_map(struct file *filep, unsigned long addr,
328 		struct elf_phdr *eppnt, int prot, int type,
329 		unsigned long total_size)
330 {
331 	unsigned long map_addr;
332 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
333 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
334 	addr = ELF_PAGESTART(addr);
335 	size = ELF_PAGEALIGN(size);
336 
337 	/* mmap() will return -EINVAL if given a zero size, but a
338 	 * segment with zero filesize is perfectly valid */
339 	if (!size)
340 		return addr;
341 
342 	/*
343 	* total_size is the size of the ELF (interpreter) image.
344 	* The _first_ mmap needs to know the full size, otherwise
345 	* randomization might put this image into an overlapping
346 	* position with the ELF binary image. (since size < total_size)
347 	* So we first map the 'big' image - and unmap the remainder at
348 	* the end. (which unmap is needed for ELF images with holes.)
349 	*/
350 	if (total_size) {
351 		total_size = ELF_PAGEALIGN(total_size);
352 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
353 		if (!BAD_ADDR(map_addr))
354 			vm_munmap(map_addr+size, total_size-size);
355 	} else
356 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
357 
358 	return(map_addr);
359 }
360 
361 #endif /* !elf_map */
362 
363 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
364 {
365 	int i, first_idx = -1, last_idx = -1;
366 
367 	for (i = 0; i < nr; i++) {
368 		if (cmds[i].p_type == PT_LOAD) {
369 			last_idx = i;
370 			if (first_idx == -1)
371 				first_idx = i;
372 		}
373 	}
374 	if (first_idx == -1)
375 		return 0;
376 
377 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
378 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
379 }
380 
381 
382 /* This is much more generalized than the library routine read function,
383    so we keep this separate.  Technically the library read function
384    is only provided so that we can read a.out libraries that have
385    an ELF header */
386 
387 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
388 		struct file *interpreter, unsigned long *interp_map_addr,
389 		unsigned long no_base)
390 {
391 	struct elf_phdr *elf_phdata;
392 	struct elf_phdr *eppnt;
393 	unsigned long load_addr = 0;
394 	int load_addr_set = 0;
395 	unsigned long last_bss = 0, elf_bss = 0;
396 	unsigned long error = ~0UL;
397 	unsigned long total_size;
398 	int retval, i, size;
399 
400 	/* First of all, some simple consistency checks */
401 	if (interp_elf_ex->e_type != ET_EXEC &&
402 	    interp_elf_ex->e_type != ET_DYN)
403 		goto out;
404 	if (!elf_check_arch(interp_elf_ex))
405 		goto out;
406 	if (!interpreter->f_op || !interpreter->f_op->mmap)
407 		goto out;
408 
409 	/*
410 	 * If the size of this structure has changed, then punt, since
411 	 * we will be doing the wrong thing.
412 	 */
413 	if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
414 		goto out;
415 	if (interp_elf_ex->e_phnum < 1 ||
416 		interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
417 		goto out;
418 
419 	/* Now read in all of the header information */
420 	size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
421 	if (size > ELF_MIN_ALIGN)
422 		goto out;
423 	elf_phdata = kmalloc(size, GFP_KERNEL);
424 	if (!elf_phdata)
425 		goto out;
426 
427 	retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
428 			     (char *)elf_phdata, size);
429 	error = -EIO;
430 	if (retval != size) {
431 		if (retval < 0)
432 			error = retval;
433 		goto out_close;
434 	}
435 
436 	total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
437 	if (!total_size) {
438 		error = -EINVAL;
439 		goto out_close;
440 	}
441 
442 	eppnt = elf_phdata;
443 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
444 		if (eppnt->p_type == PT_LOAD) {
445 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
446 			int elf_prot = 0;
447 			unsigned long vaddr = 0;
448 			unsigned long k, map_addr;
449 
450 			if (eppnt->p_flags & PF_R)
451 		    		elf_prot = PROT_READ;
452 			if (eppnt->p_flags & PF_W)
453 				elf_prot |= PROT_WRITE;
454 			if (eppnt->p_flags & PF_X)
455 				elf_prot |= PROT_EXEC;
456 			vaddr = eppnt->p_vaddr;
457 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
458 				elf_type |= MAP_FIXED;
459 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
460 				load_addr = -vaddr;
461 
462 			map_addr = elf_map(interpreter, load_addr + vaddr,
463 					eppnt, elf_prot, elf_type, total_size);
464 			total_size = 0;
465 			if (!*interp_map_addr)
466 				*interp_map_addr = map_addr;
467 			error = map_addr;
468 			if (BAD_ADDR(map_addr))
469 				goto out_close;
470 
471 			if (!load_addr_set &&
472 			    interp_elf_ex->e_type == ET_DYN) {
473 				load_addr = map_addr - ELF_PAGESTART(vaddr);
474 				load_addr_set = 1;
475 			}
476 
477 			/*
478 			 * Check to see if the section's size will overflow the
479 			 * allowed task size. Note that p_filesz must always be
480 			 * <= p_memsize so it's only necessary to check p_memsz.
481 			 */
482 			k = load_addr + eppnt->p_vaddr;
483 			if (BAD_ADDR(k) ||
484 			    eppnt->p_filesz > eppnt->p_memsz ||
485 			    eppnt->p_memsz > TASK_SIZE ||
486 			    TASK_SIZE - eppnt->p_memsz < k) {
487 				error = -ENOMEM;
488 				goto out_close;
489 			}
490 
491 			/*
492 			 * Find the end of the file mapping for this phdr, and
493 			 * keep track of the largest address we see for this.
494 			 */
495 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
496 			if (k > elf_bss)
497 				elf_bss = k;
498 
499 			/*
500 			 * Do the same thing for the memory mapping - between
501 			 * elf_bss and last_bss is the bss section.
502 			 */
503 			k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
504 			if (k > last_bss)
505 				last_bss = k;
506 		}
507 	}
508 
509 	if (last_bss > elf_bss) {
510 		/*
511 		 * Now fill out the bss section.  First pad the last page up
512 		 * to the page boundary, and then perform a mmap to make sure
513 		 * that there are zero-mapped pages up to and including the
514 		 * last bss page.
515 		 */
516 		if (padzero(elf_bss)) {
517 			error = -EFAULT;
518 			goto out_close;
519 		}
520 
521 		/* What we have mapped so far */
522 		elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
523 
524 		/* Map the last of the bss segment */
525 		error = vm_brk(elf_bss, last_bss - elf_bss);
526 		if (BAD_ADDR(error))
527 			goto out_close;
528 	}
529 
530 	error = load_addr;
531 
532 out_close:
533 	kfree(elf_phdata);
534 out:
535 	return error;
536 }
537 
538 /*
539  * These are the functions used to load ELF style executables and shared
540  * libraries.  There is no binary dependent code anywhere else.
541  */
542 
543 #define INTERPRETER_NONE 0
544 #define INTERPRETER_ELF 2
545 
546 #ifndef STACK_RND_MASK
547 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))	/* 8MB of VA */
548 #endif
549 
550 static unsigned long randomize_stack_top(unsigned long stack_top)
551 {
552 	unsigned int random_variable = 0;
553 
554 	if ((current->flags & PF_RANDOMIZE) &&
555 		!(current->personality & ADDR_NO_RANDOMIZE)) {
556 		random_variable = get_random_int() & STACK_RND_MASK;
557 		random_variable <<= PAGE_SHIFT;
558 	}
559 #ifdef CONFIG_STACK_GROWSUP
560 	return PAGE_ALIGN(stack_top) + random_variable;
561 #else
562 	return PAGE_ALIGN(stack_top) - random_variable;
563 #endif
564 }
565 
566 static int load_elf_binary(struct linux_binprm *bprm)
567 {
568 	struct file *interpreter = NULL; /* to shut gcc up */
569  	unsigned long load_addr = 0, load_bias = 0;
570 	int load_addr_set = 0;
571 	char * elf_interpreter = NULL;
572 	unsigned long error;
573 	struct elf_phdr *elf_ppnt, *elf_phdata;
574 	unsigned long elf_bss, elf_brk;
575 	int retval, i;
576 	unsigned int size;
577 	unsigned long elf_entry;
578 	unsigned long interp_load_addr = 0;
579 	unsigned long start_code, end_code, start_data, end_data;
580 	unsigned long reloc_func_desc __maybe_unused = 0;
581 	int executable_stack = EXSTACK_DEFAULT;
582 	unsigned long def_flags = 0;
583 	struct pt_regs *regs = current_pt_regs();
584 	struct {
585 		struct elfhdr elf_ex;
586 		struct elfhdr interp_elf_ex;
587 	} *loc;
588 
589 	loc = kmalloc(sizeof(*loc), GFP_KERNEL);
590 	if (!loc) {
591 		retval = -ENOMEM;
592 		goto out_ret;
593 	}
594 
595 	/* Get the exec-header */
596 	loc->elf_ex = *((struct elfhdr *)bprm->buf);
597 
598 	retval = -ENOEXEC;
599 	/* First of all, some simple consistency checks */
600 	if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
601 		goto out;
602 
603 	if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
604 		goto out;
605 	if (!elf_check_arch(&loc->elf_ex))
606 		goto out;
607 	if (!bprm->file->f_op || !bprm->file->f_op->mmap)
608 		goto out;
609 
610 	/* Now read in all of the header information */
611 	if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
612 		goto out;
613 	if (loc->elf_ex.e_phnum < 1 ||
614 	 	loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
615 		goto out;
616 	size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
617 	retval = -ENOMEM;
618 	elf_phdata = kmalloc(size, GFP_KERNEL);
619 	if (!elf_phdata)
620 		goto out;
621 
622 	retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
623 			     (char *)elf_phdata, size);
624 	if (retval != size) {
625 		if (retval >= 0)
626 			retval = -EIO;
627 		goto out_free_ph;
628 	}
629 
630 	elf_ppnt = elf_phdata;
631 	elf_bss = 0;
632 	elf_brk = 0;
633 
634 	start_code = ~0UL;
635 	end_code = 0;
636 	start_data = 0;
637 	end_data = 0;
638 
639 	for (i = 0; i < loc->elf_ex.e_phnum; i++) {
640 		if (elf_ppnt->p_type == PT_INTERP) {
641 			/* This is the program interpreter used for
642 			 * shared libraries - for now assume that this
643 			 * is an a.out format binary
644 			 */
645 			retval = -ENOEXEC;
646 			if (elf_ppnt->p_filesz > PATH_MAX ||
647 			    elf_ppnt->p_filesz < 2)
648 				goto out_free_ph;
649 
650 			retval = -ENOMEM;
651 			elf_interpreter = kmalloc(elf_ppnt->p_filesz,
652 						  GFP_KERNEL);
653 			if (!elf_interpreter)
654 				goto out_free_ph;
655 
656 			retval = kernel_read(bprm->file, elf_ppnt->p_offset,
657 					     elf_interpreter,
658 					     elf_ppnt->p_filesz);
659 			if (retval != elf_ppnt->p_filesz) {
660 				if (retval >= 0)
661 					retval = -EIO;
662 				goto out_free_interp;
663 			}
664 			/* make sure path is NULL terminated */
665 			retval = -ENOEXEC;
666 			if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
667 				goto out_free_interp;
668 
669 			interpreter = open_exec(elf_interpreter);
670 			retval = PTR_ERR(interpreter);
671 			if (IS_ERR(interpreter))
672 				goto out_free_interp;
673 
674 			/*
675 			 * If the binary is not readable then enforce
676 			 * mm->dumpable = 0 regardless of the interpreter's
677 			 * permissions.
678 			 */
679 			would_dump(bprm, interpreter);
680 
681 			retval = kernel_read(interpreter, 0, bprm->buf,
682 					     BINPRM_BUF_SIZE);
683 			if (retval != BINPRM_BUF_SIZE) {
684 				if (retval >= 0)
685 					retval = -EIO;
686 				goto out_free_dentry;
687 			}
688 
689 			/* Get the exec headers */
690 			loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
691 			break;
692 		}
693 		elf_ppnt++;
694 	}
695 
696 	elf_ppnt = elf_phdata;
697 	for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
698 		if (elf_ppnt->p_type == PT_GNU_STACK) {
699 			if (elf_ppnt->p_flags & PF_X)
700 				executable_stack = EXSTACK_ENABLE_X;
701 			else
702 				executable_stack = EXSTACK_DISABLE_X;
703 			break;
704 		}
705 
706 	/* Some simple consistency checks for the interpreter */
707 	if (elf_interpreter) {
708 		retval = -ELIBBAD;
709 		/* Not an ELF interpreter */
710 		if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
711 			goto out_free_dentry;
712 		/* Verify the interpreter has a valid arch */
713 		if (!elf_check_arch(&loc->interp_elf_ex))
714 			goto out_free_dentry;
715 	}
716 
717 	/* Flush all traces of the currently running executable */
718 	retval = flush_old_exec(bprm);
719 	if (retval)
720 		goto out_free_dentry;
721 
722 	/* OK, This is the point of no return */
723 	current->mm->def_flags = def_flags;
724 
725 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
726 	   may depend on the personality.  */
727 	SET_PERSONALITY(loc->elf_ex);
728 	if (elf_read_implies_exec(loc->elf_ex, executable_stack))
729 		current->personality |= READ_IMPLIES_EXEC;
730 
731 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
732 		current->flags |= PF_RANDOMIZE;
733 
734 	setup_new_exec(bprm);
735 
736 	/* Do this so that we can load the interpreter, if need be.  We will
737 	   change some of these later */
738 	current->mm->free_area_cache = current->mm->mmap_base;
739 	current->mm->cached_hole_size = 0;
740 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
741 				 executable_stack);
742 	if (retval < 0) {
743 		send_sig(SIGKILL, current, 0);
744 		goto out_free_dentry;
745 	}
746 
747 	current->mm->start_stack = bprm->p;
748 
749 	/* Now we do a little grungy work by mmapping the ELF image into
750 	   the correct location in memory. */
751 	for(i = 0, elf_ppnt = elf_phdata;
752 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
753 		int elf_prot = 0, elf_flags;
754 		unsigned long k, vaddr;
755 
756 		if (elf_ppnt->p_type != PT_LOAD)
757 			continue;
758 
759 		if (unlikely (elf_brk > elf_bss)) {
760 			unsigned long nbyte;
761 
762 			/* There was a PT_LOAD segment with p_memsz > p_filesz
763 			   before this one. Map anonymous pages, if needed,
764 			   and clear the area.  */
765 			retval = set_brk(elf_bss + load_bias,
766 					 elf_brk + load_bias);
767 			if (retval) {
768 				send_sig(SIGKILL, current, 0);
769 				goto out_free_dentry;
770 			}
771 			nbyte = ELF_PAGEOFFSET(elf_bss);
772 			if (nbyte) {
773 				nbyte = ELF_MIN_ALIGN - nbyte;
774 				if (nbyte > elf_brk - elf_bss)
775 					nbyte = elf_brk - elf_bss;
776 				if (clear_user((void __user *)elf_bss +
777 							load_bias, nbyte)) {
778 					/*
779 					 * This bss-zeroing can fail if the ELF
780 					 * file specifies odd protections. So
781 					 * we don't check the return value
782 					 */
783 				}
784 			}
785 		}
786 
787 		if (elf_ppnt->p_flags & PF_R)
788 			elf_prot |= PROT_READ;
789 		if (elf_ppnt->p_flags & PF_W)
790 			elf_prot |= PROT_WRITE;
791 		if (elf_ppnt->p_flags & PF_X)
792 			elf_prot |= PROT_EXEC;
793 
794 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
795 
796 		vaddr = elf_ppnt->p_vaddr;
797 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
798 			elf_flags |= MAP_FIXED;
799 		} else if (loc->elf_ex.e_type == ET_DYN) {
800 			/* Try and get dynamic programs out of the way of the
801 			 * default mmap base, as well as whatever program they
802 			 * might try to exec.  This is because the brk will
803 			 * follow the loader, and is not movable.  */
804 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
805 			/* Memory randomization might have been switched off
806 			 * in runtime via sysctl or explicit setting of
807 			 * personality flags.
808 			 * If that is the case, retain the original non-zero
809 			 * load_bias value in order to establish proper
810 			 * non-randomized mappings.
811 			 */
812 			if (current->flags & PF_RANDOMIZE)
813 				load_bias = 0;
814 			else
815 				load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
816 #else
817 			load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
818 #endif
819 		}
820 
821 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
822 				elf_prot, elf_flags, 0);
823 		if (BAD_ADDR(error)) {
824 			send_sig(SIGKILL, current, 0);
825 			retval = IS_ERR((void *)error) ?
826 				PTR_ERR((void*)error) : -EINVAL;
827 			goto out_free_dentry;
828 		}
829 
830 		if (!load_addr_set) {
831 			load_addr_set = 1;
832 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
833 			if (loc->elf_ex.e_type == ET_DYN) {
834 				load_bias += error -
835 				             ELF_PAGESTART(load_bias + vaddr);
836 				load_addr += load_bias;
837 				reloc_func_desc = load_bias;
838 			}
839 		}
840 		k = elf_ppnt->p_vaddr;
841 		if (k < start_code)
842 			start_code = k;
843 		if (start_data < k)
844 			start_data = k;
845 
846 		/*
847 		 * Check to see if the section's size will overflow the
848 		 * allowed task size. Note that p_filesz must always be
849 		 * <= p_memsz so it is only necessary to check p_memsz.
850 		 */
851 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
852 		    elf_ppnt->p_memsz > TASK_SIZE ||
853 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
854 			/* set_brk can never work. Avoid overflows. */
855 			send_sig(SIGKILL, current, 0);
856 			retval = -EINVAL;
857 			goto out_free_dentry;
858 		}
859 
860 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
861 
862 		if (k > elf_bss)
863 			elf_bss = k;
864 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
865 			end_code = k;
866 		if (end_data < k)
867 			end_data = k;
868 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
869 		if (k > elf_brk)
870 			elf_brk = k;
871 	}
872 
873 	loc->elf_ex.e_entry += load_bias;
874 	elf_bss += load_bias;
875 	elf_brk += load_bias;
876 	start_code += load_bias;
877 	end_code += load_bias;
878 	start_data += load_bias;
879 	end_data += load_bias;
880 
881 	/* Calling set_brk effectively mmaps the pages that we need
882 	 * for the bss and break sections.  We must do this before
883 	 * mapping in the interpreter, to make sure it doesn't wind
884 	 * up getting placed where the bss needs to go.
885 	 */
886 	retval = set_brk(elf_bss, elf_brk);
887 	if (retval) {
888 		send_sig(SIGKILL, current, 0);
889 		goto out_free_dentry;
890 	}
891 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
892 		send_sig(SIGSEGV, current, 0);
893 		retval = -EFAULT; /* Nobody gets to see this, but.. */
894 		goto out_free_dentry;
895 	}
896 
897 	if (elf_interpreter) {
898 		unsigned long interp_map_addr = 0;
899 
900 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
901 					    interpreter,
902 					    &interp_map_addr,
903 					    load_bias);
904 		if (!IS_ERR((void *)elf_entry)) {
905 			/*
906 			 * load_elf_interp() returns relocation
907 			 * adjustment
908 			 */
909 			interp_load_addr = elf_entry;
910 			elf_entry += loc->interp_elf_ex.e_entry;
911 		}
912 		if (BAD_ADDR(elf_entry)) {
913 			force_sig(SIGSEGV, current);
914 			retval = IS_ERR((void *)elf_entry) ?
915 					(int)elf_entry : -EINVAL;
916 			goto out_free_dentry;
917 		}
918 		reloc_func_desc = interp_load_addr;
919 
920 		allow_write_access(interpreter);
921 		fput(interpreter);
922 		kfree(elf_interpreter);
923 	} else {
924 		elf_entry = loc->elf_ex.e_entry;
925 		if (BAD_ADDR(elf_entry)) {
926 			force_sig(SIGSEGV, current);
927 			retval = -EINVAL;
928 			goto out_free_dentry;
929 		}
930 	}
931 
932 	kfree(elf_phdata);
933 
934 	set_binfmt(&elf_format);
935 
936 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
937 	retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
938 	if (retval < 0) {
939 		send_sig(SIGKILL, current, 0);
940 		goto out;
941 	}
942 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
943 
944 	install_exec_creds(bprm);
945 	retval = create_elf_tables(bprm, &loc->elf_ex,
946 			  load_addr, interp_load_addr);
947 	if (retval < 0) {
948 		send_sig(SIGKILL, current, 0);
949 		goto out;
950 	}
951 	/* N.B. passed_fileno might not be initialized? */
952 	current->mm->end_code = end_code;
953 	current->mm->start_code = start_code;
954 	current->mm->start_data = start_data;
955 	current->mm->end_data = end_data;
956 	current->mm->start_stack = bprm->p;
957 
958 #ifdef arch_randomize_brk
959 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
960 		current->mm->brk = current->mm->start_brk =
961 			arch_randomize_brk(current->mm);
962 #ifdef CONFIG_COMPAT_BRK
963 		current->brk_randomized = 1;
964 #endif
965 	}
966 #endif
967 
968 	if (current->personality & MMAP_PAGE_ZERO) {
969 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
970 		   and some applications "depend" upon this behavior.
971 		   Since we do not have the power to recompile these, we
972 		   emulate the SVr4 behavior. Sigh. */
973 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
974 				MAP_FIXED | MAP_PRIVATE, 0);
975 	}
976 
977 #ifdef ELF_PLAT_INIT
978 	/*
979 	 * The ABI may specify that certain registers be set up in special
980 	 * ways (on i386 %edx is the address of a DT_FINI function, for
981 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
982 	 * that the e_entry field is the address of the function descriptor
983 	 * for the startup routine, rather than the address of the startup
984 	 * routine itself.  This macro performs whatever initialization to
985 	 * the regs structure is required as well as any relocations to the
986 	 * function descriptor entries when executing dynamically links apps.
987 	 */
988 	ELF_PLAT_INIT(regs, reloc_func_desc);
989 #endif
990 
991 	start_thread(regs, elf_entry, bprm->p);
992 	retval = 0;
993 out:
994 	kfree(loc);
995 out_ret:
996 	return retval;
997 
998 	/* error cleanup */
999 out_free_dentry:
1000 	allow_write_access(interpreter);
1001 	if (interpreter)
1002 		fput(interpreter);
1003 out_free_interp:
1004 	kfree(elf_interpreter);
1005 out_free_ph:
1006 	kfree(elf_phdata);
1007 	goto out;
1008 }
1009 
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 || !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 
1089 #ifdef CONFIG_ELF_CORE
1090 /*
1091  * ELF core dumper
1092  *
1093  * Modelled on fs/exec.c:aout_core_dump()
1094  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1095  */
1096 
1097 /*
1098  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1099  * that are useful for post-mortem analysis are included in every core dump.
1100  * In that way we ensure that the core dump is fully interpretable later
1101  * without matching up the same kernel and hardware config to see what PC values
1102  * meant. These special mappings include - vDSO, vsyscall, and other
1103  * architecture specific mappings
1104  */
1105 static bool always_dump_vma(struct vm_area_struct *vma)
1106 {
1107 	/* Any vsyscall mappings? */
1108 	if (vma == get_gate_vma(vma->vm_mm))
1109 		return true;
1110 	/*
1111 	 * arch_vma_name() returns non-NULL for special architecture mappings,
1112 	 * such as vDSO sections.
1113 	 */
1114 	if (arch_vma_name(vma))
1115 		return true;
1116 
1117 	return false;
1118 }
1119 
1120 /*
1121  * Decide what to dump of a segment, part, all or none.
1122  */
1123 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1124 				   unsigned long mm_flags)
1125 {
1126 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1127 
1128 	/* always dump the vdso and vsyscall sections */
1129 	if (always_dump_vma(vma))
1130 		goto whole;
1131 
1132 	if (vma->vm_flags & VM_DONTDUMP)
1133 		return 0;
1134 
1135 	/* Hugetlb memory check */
1136 	if (vma->vm_flags & VM_HUGETLB) {
1137 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1138 			goto whole;
1139 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1140 			goto whole;
1141 		return 0;
1142 	}
1143 
1144 	/* Do not dump I/O mapped devices or special mappings */
1145 	if (vma->vm_flags & VM_IO)
1146 		return 0;
1147 
1148 	/* By default, dump shared memory if mapped from an anonymous file. */
1149 	if (vma->vm_flags & VM_SHARED) {
1150 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1151 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1152 			goto whole;
1153 		return 0;
1154 	}
1155 
1156 	/* Dump segments that have been written to.  */
1157 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1158 		goto whole;
1159 	if (vma->vm_file == NULL)
1160 		return 0;
1161 
1162 	if (FILTER(MAPPED_PRIVATE))
1163 		goto whole;
1164 
1165 	/*
1166 	 * If this looks like the beginning of a DSO or executable mapping,
1167 	 * check for an ELF header.  If we find one, dump the first page to
1168 	 * aid in determining what was mapped here.
1169 	 */
1170 	if (FILTER(ELF_HEADERS) &&
1171 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1172 		u32 __user *header = (u32 __user *) vma->vm_start;
1173 		u32 word;
1174 		mm_segment_t fs = get_fs();
1175 		/*
1176 		 * Doing it this way gets the constant folded by GCC.
1177 		 */
1178 		union {
1179 			u32 cmp;
1180 			char elfmag[SELFMAG];
1181 		} magic;
1182 		BUILD_BUG_ON(SELFMAG != sizeof word);
1183 		magic.elfmag[EI_MAG0] = ELFMAG0;
1184 		magic.elfmag[EI_MAG1] = ELFMAG1;
1185 		magic.elfmag[EI_MAG2] = ELFMAG2;
1186 		magic.elfmag[EI_MAG3] = ELFMAG3;
1187 		/*
1188 		 * Switch to the user "segment" for get_user(),
1189 		 * then put back what elf_core_dump() had in place.
1190 		 */
1191 		set_fs(USER_DS);
1192 		if (unlikely(get_user(word, header)))
1193 			word = 0;
1194 		set_fs(fs);
1195 		if (word == magic.cmp)
1196 			return PAGE_SIZE;
1197 	}
1198 
1199 #undef	FILTER
1200 
1201 	return 0;
1202 
1203 whole:
1204 	return vma->vm_end - vma->vm_start;
1205 }
1206 
1207 /* An ELF note in memory */
1208 struct memelfnote
1209 {
1210 	const char *name;
1211 	int type;
1212 	unsigned int datasz;
1213 	void *data;
1214 };
1215 
1216 static int notesize(struct memelfnote *en)
1217 {
1218 	int sz;
1219 
1220 	sz = sizeof(struct elf_note);
1221 	sz += roundup(strlen(en->name) + 1, 4);
1222 	sz += roundup(en->datasz, 4);
1223 
1224 	return sz;
1225 }
1226 
1227 #define DUMP_WRITE(addr, nr, foffset)	\
1228 	do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1229 
1230 static int alignfile(struct file *file, loff_t *foffset)
1231 {
1232 	static const char buf[4] = { 0, };
1233 	DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1234 	return 1;
1235 }
1236 
1237 static int writenote(struct memelfnote *men, struct file *file,
1238 			loff_t *foffset)
1239 {
1240 	struct elf_note en;
1241 	en.n_namesz = strlen(men->name) + 1;
1242 	en.n_descsz = men->datasz;
1243 	en.n_type = men->type;
1244 
1245 	DUMP_WRITE(&en, sizeof(en), foffset);
1246 	DUMP_WRITE(men->name, en.n_namesz, foffset);
1247 	if (!alignfile(file, foffset))
1248 		return 0;
1249 	DUMP_WRITE(men->data, men->datasz, foffset);
1250 	if (!alignfile(file, foffset))
1251 		return 0;
1252 
1253 	return 1;
1254 }
1255 #undef DUMP_WRITE
1256 
1257 static void fill_elf_header(struct elfhdr *elf, int segs,
1258 			    u16 machine, u32 flags)
1259 {
1260 	memset(elf, 0, sizeof(*elf));
1261 
1262 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1263 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1264 	elf->e_ident[EI_DATA] = ELF_DATA;
1265 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1266 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1267 
1268 	elf->e_type = ET_CORE;
1269 	elf->e_machine = machine;
1270 	elf->e_version = EV_CURRENT;
1271 	elf->e_phoff = sizeof(struct elfhdr);
1272 	elf->e_flags = flags;
1273 	elf->e_ehsize = sizeof(struct elfhdr);
1274 	elf->e_phentsize = sizeof(struct elf_phdr);
1275 	elf->e_phnum = segs;
1276 
1277 	return;
1278 }
1279 
1280 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1281 {
1282 	phdr->p_type = PT_NOTE;
1283 	phdr->p_offset = offset;
1284 	phdr->p_vaddr = 0;
1285 	phdr->p_paddr = 0;
1286 	phdr->p_filesz = sz;
1287 	phdr->p_memsz = 0;
1288 	phdr->p_flags = 0;
1289 	phdr->p_align = 0;
1290 	return;
1291 }
1292 
1293 static void fill_note(struct memelfnote *note, const char *name, int type,
1294 		unsigned int sz, void *data)
1295 {
1296 	note->name = name;
1297 	note->type = type;
1298 	note->datasz = sz;
1299 	note->data = data;
1300 	return;
1301 }
1302 
1303 /*
1304  * fill up all the fields in prstatus from the given task struct, except
1305  * registers which need to be filled up separately.
1306  */
1307 static void fill_prstatus(struct elf_prstatus *prstatus,
1308 		struct task_struct *p, long signr)
1309 {
1310 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1311 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1312 	prstatus->pr_sighold = p->blocked.sig[0];
1313 	rcu_read_lock();
1314 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1315 	rcu_read_unlock();
1316 	prstatus->pr_pid = task_pid_vnr(p);
1317 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1318 	prstatus->pr_sid = task_session_vnr(p);
1319 	if (thread_group_leader(p)) {
1320 		struct task_cputime cputime;
1321 
1322 		/*
1323 		 * This is the record for the group leader.  It shows the
1324 		 * group-wide total, not its individual thread total.
1325 		 */
1326 		thread_group_cputime(p, &cputime);
1327 		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1328 		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1329 	} else {
1330 		cputime_t utime, stime;
1331 
1332 		task_cputime(p, &utime, &stime);
1333 		cputime_to_timeval(utime, &prstatus->pr_utime);
1334 		cputime_to_timeval(stime, &prstatus->pr_stime);
1335 	}
1336 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1337 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1338 }
1339 
1340 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1341 		       struct mm_struct *mm)
1342 {
1343 	const struct cred *cred;
1344 	unsigned int i, len;
1345 
1346 	/* first copy the parameters from user space */
1347 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1348 
1349 	len = mm->arg_end - mm->arg_start;
1350 	if (len >= ELF_PRARGSZ)
1351 		len = ELF_PRARGSZ-1;
1352 	if (copy_from_user(&psinfo->pr_psargs,
1353 		           (const char __user *)mm->arg_start, len))
1354 		return -EFAULT;
1355 	for(i = 0; i < len; i++)
1356 		if (psinfo->pr_psargs[i] == 0)
1357 			psinfo->pr_psargs[i] = ' ';
1358 	psinfo->pr_psargs[len] = 0;
1359 
1360 	rcu_read_lock();
1361 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1362 	rcu_read_unlock();
1363 	psinfo->pr_pid = task_pid_vnr(p);
1364 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1365 	psinfo->pr_sid = task_session_vnr(p);
1366 
1367 	i = p->state ? ffz(~p->state) + 1 : 0;
1368 	psinfo->pr_state = i;
1369 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1370 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1371 	psinfo->pr_nice = task_nice(p);
1372 	psinfo->pr_flag = p->flags;
1373 	rcu_read_lock();
1374 	cred = __task_cred(p);
1375 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1376 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1377 	rcu_read_unlock();
1378 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1379 
1380 	return 0;
1381 }
1382 
1383 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1384 {
1385 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1386 	int i = 0;
1387 	do
1388 		i += 2;
1389 	while (auxv[i - 2] != AT_NULL);
1390 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1391 }
1392 
1393 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1394 		siginfo_t *siginfo)
1395 {
1396 	mm_segment_t old_fs = get_fs();
1397 	set_fs(KERNEL_DS);
1398 	copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1399 	set_fs(old_fs);
1400 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1401 }
1402 
1403 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1404 /*
1405  * Format of NT_FILE note:
1406  *
1407  * long count     -- how many files are mapped
1408  * long page_size -- units for file_ofs
1409  * array of [COUNT] elements of
1410  *   long start
1411  *   long end
1412  *   long file_ofs
1413  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1414  */
1415 static void fill_files_note(struct memelfnote *note)
1416 {
1417 	struct vm_area_struct *vma;
1418 	unsigned count, size, names_ofs, remaining, n;
1419 	user_long_t *data;
1420 	user_long_t *start_end_ofs;
1421 	char *name_base, *name_curpos;
1422 
1423 	/* *Estimated* file count and total data size needed */
1424 	count = current->mm->map_count;
1425 	size = count * 64;
1426 
1427 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1428  alloc:
1429 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1430 		goto err;
1431 	size = round_up(size, PAGE_SIZE);
1432 	data = vmalloc(size);
1433 	if (!data)
1434 		goto err;
1435 
1436 	start_end_ofs = data + 2;
1437 	name_base = name_curpos = ((char *)data) + names_ofs;
1438 	remaining = size - names_ofs;
1439 	count = 0;
1440 	for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1441 		struct file *file;
1442 		const char *filename;
1443 
1444 		file = vma->vm_file;
1445 		if (!file)
1446 			continue;
1447 		filename = d_path(&file->f_path, name_curpos, remaining);
1448 		if (IS_ERR(filename)) {
1449 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1450 				vfree(data);
1451 				size = size * 5 / 4;
1452 				goto alloc;
1453 			}
1454 			continue;
1455 		}
1456 
1457 		/* d_path() fills at the end, move name down */
1458 		/* n = strlen(filename) + 1: */
1459 		n = (name_curpos + remaining) - filename;
1460 		remaining = filename - name_curpos;
1461 		memmove(name_curpos, filename, n);
1462 		name_curpos += n;
1463 
1464 		*start_end_ofs++ = vma->vm_start;
1465 		*start_end_ofs++ = vma->vm_end;
1466 		*start_end_ofs++ = vma->vm_pgoff;
1467 		count++;
1468 	}
1469 
1470 	/* Now we know exact count of files, can store it */
1471 	data[0] = count;
1472 	data[1] = PAGE_SIZE;
1473 	/*
1474 	 * Count usually is less than current->mm->map_count,
1475 	 * we need to move filenames down.
1476 	 */
1477 	n = current->mm->map_count - count;
1478 	if (n != 0) {
1479 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1480 		memmove(name_base - shift_bytes, name_base,
1481 			name_curpos - name_base);
1482 		name_curpos -= shift_bytes;
1483 	}
1484 
1485 	size = name_curpos - (char *)data;
1486 	fill_note(note, "CORE", NT_FILE, size, data);
1487  err: ;
1488 }
1489 
1490 #ifdef CORE_DUMP_USE_REGSET
1491 #include <linux/regset.h>
1492 
1493 struct elf_thread_core_info {
1494 	struct elf_thread_core_info *next;
1495 	struct task_struct *task;
1496 	struct elf_prstatus prstatus;
1497 	struct memelfnote notes[0];
1498 };
1499 
1500 struct elf_note_info {
1501 	struct elf_thread_core_info *thread;
1502 	struct memelfnote psinfo;
1503 	struct memelfnote signote;
1504 	struct memelfnote auxv;
1505 	struct memelfnote files;
1506 	user_siginfo_t csigdata;
1507 	size_t size;
1508 	int thread_notes;
1509 };
1510 
1511 /*
1512  * When a regset has a writeback hook, we call it on each thread before
1513  * dumping user memory.  On register window machines, this makes sure the
1514  * user memory backing the register data is up to date before we read it.
1515  */
1516 static void do_thread_regset_writeback(struct task_struct *task,
1517 				       const struct user_regset *regset)
1518 {
1519 	if (regset->writeback)
1520 		regset->writeback(task, regset, 1);
1521 }
1522 
1523 #ifndef PR_REG_SIZE
1524 #define PR_REG_SIZE(S) sizeof(S)
1525 #endif
1526 
1527 #ifndef PRSTATUS_SIZE
1528 #define PRSTATUS_SIZE(S) sizeof(S)
1529 #endif
1530 
1531 #ifndef PR_REG_PTR
1532 #define PR_REG_PTR(S) (&((S)->pr_reg))
1533 #endif
1534 
1535 #ifndef SET_PR_FPVALID
1536 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1537 #endif
1538 
1539 static int fill_thread_core_info(struct elf_thread_core_info *t,
1540 				 const struct user_regset_view *view,
1541 				 long signr, size_t *total)
1542 {
1543 	unsigned int i;
1544 
1545 	/*
1546 	 * NT_PRSTATUS is the one special case, because the regset data
1547 	 * goes into the pr_reg field inside the note contents, rather
1548 	 * than being the whole note contents.  We fill the reset in here.
1549 	 * We assume that regset 0 is NT_PRSTATUS.
1550 	 */
1551 	fill_prstatus(&t->prstatus, t->task, signr);
1552 	(void) view->regsets[0].get(t->task, &view->regsets[0],
1553 				    0, PR_REG_SIZE(t->prstatus.pr_reg),
1554 				    PR_REG_PTR(&t->prstatus), NULL);
1555 
1556 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1557 		  PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1558 	*total += notesize(&t->notes[0]);
1559 
1560 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1561 
1562 	/*
1563 	 * Each other regset might generate a note too.  For each regset
1564 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1565 	 * all zero and we'll know to skip writing it later.
1566 	 */
1567 	for (i = 1; i < view->n; ++i) {
1568 		const struct user_regset *regset = &view->regsets[i];
1569 		do_thread_regset_writeback(t->task, regset);
1570 		if (regset->core_note_type && regset->get &&
1571 		    (!regset->active || regset->active(t->task, regset))) {
1572 			int ret;
1573 			size_t size = regset->n * regset->size;
1574 			void *data = kmalloc(size, GFP_KERNEL);
1575 			if (unlikely(!data))
1576 				return 0;
1577 			ret = regset->get(t->task, regset,
1578 					  0, size, data, NULL);
1579 			if (unlikely(ret))
1580 				kfree(data);
1581 			else {
1582 				if (regset->core_note_type != NT_PRFPREG)
1583 					fill_note(&t->notes[i], "LINUX",
1584 						  regset->core_note_type,
1585 						  size, data);
1586 				else {
1587 					SET_PR_FPVALID(&t->prstatus, 1);
1588 					fill_note(&t->notes[i], "CORE",
1589 						  NT_PRFPREG, size, data);
1590 				}
1591 				*total += notesize(&t->notes[i]);
1592 			}
1593 		}
1594 	}
1595 
1596 	return 1;
1597 }
1598 
1599 static int fill_note_info(struct elfhdr *elf, int phdrs,
1600 			  struct elf_note_info *info,
1601 			  siginfo_t *siginfo, struct pt_regs *regs)
1602 {
1603 	struct task_struct *dump_task = current;
1604 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1605 	struct elf_thread_core_info *t;
1606 	struct elf_prpsinfo *psinfo;
1607 	struct core_thread *ct;
1608 	unsigned int i;
1609 
1610 	info->size = 0;
1611 	info->thread = NULL;
1612 
1613 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1614 	if (psinfo == NULL) {
1615 		info->psinfo.data = NULL; /* So we don't free this wrongly */
1616 		return 0;
1617 	}
1618 
1619 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1620 
1621 	/*
1622 	 * Figure out how many notes we're going to need for each thread.
1623 	 */
1624 	info->thread_notes = 0;
1625 	for (i = 0; i < view->n; ++i)
1626 		if (view->regsets[i].core_note_type != 0)
1627 			++info->thread_notes;
1628 
1629 	/*
1630 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1631 	 * since it is our one special case.
1632 	 */
1633 	if (unlikely(info->thread_notes == 0) ||
1634 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1635 		WARN_ON(1);
1636 		return 0;
1637 	}
1638 
1639 	/*
1640 	 * Initialize the ELF file header.
1641 	 */
1642 	fill_elf_header(elf, phdrs,
1643 			view->e_machine, view->e_flags);
1644 
1645 	/*
1646 	 * Allocate a structure for each thread.
1647 	 */
1648 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1649 		t = kzalloc(offsetof(struct elf_thread_core_info,
1650 				     notes[info->thread_notes]),
1651 			    GFP_KERNEL);
1652 		if (unlikely(!t))
1653 			return 0;
1654 
1655 		t->task = ct->task;
1656 		if (ct->task == dump_task || !info->thread) {
1657 			t->next = info->thread;
1658 			info->thread = t;
1659 		} else {
1660 			/*
1661 			 * Make sure to keep the original task at
1662 			 * the head of the list.
1663 			 */
1664 			t->next = info->thread->next;
1665 			info->thread->next = t;
1666 		}
1667 	}
1668 
1669 	/*
1670 	 * Now fill in each thread's information.
1671 	 */
1672 	for (t = info->thread; t != NULL; t = t->next)
1673 		if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1674 			return 0;
1675 
1676 	/*
1677 	 * Fill in the two process-wide notes.
1678 	 */
1679 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1680 	info->size += notesize(&info->psinfo);
1681 
1682 	fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1683 	info->size += notesize(&info->signote);
1684 
1685 	fill_auxv_note(&info->auxv, current->mm);
1686 	info->size += notesize(&info->auxv);
1687 
1688 	fill_files_note(&info->files);
1689 	info->size += notesize(&info->files);
1690 
1691 	return 1;
1692 }
1693 
1694 static size_t get_note_info_size(struct elf_note_info *info)
1695 {
1696 	return info->size;
1697 }
1698 
1699 /*
1700  * Write all the notes for each thread.  When writing the first thread, the
1701  * process-wide notes are interleaved after the first thread-specific note.
1702  */
1703 static int write_note_info(struct elf_note_info *info,
1704 			   struct file *file, loff_t *foffset)
1705 {
1706 	bool first = 1;
1707 	struct elf_thread_core_info *t = info->thread;
1708 
1709 	do {
1710 		int i;
1711 
1712 		if (!writenote(&t->notes[0], file, foffset))
1713 			return 0;
1714 
1715 		if (first && !writenote(&info->psinfo, file, foffset))
1716 			return 0;
1717 		if (first && !writenote(&info->signote, file, foffset))
1718 			return 0;
1719 		if (first && !writenote(&info->auxv, file, foffset))
1720 			return 0;
1721 		if (first && !writenote(&info->files, file, foffset))
1722 			return 0;
1723 
1724 		for (i = 1; i < info->thread_notes; ++i)
1725 			if (t->notes[i].data &&
1726 			    !writenote(&t->notes[i], file, foffset))
1727 				return 0;
1728 
1729 		first = 0;
1730 		t = t->next;
1731 	} while (t);
1732 
1733 	return 1;
1734 }
1735 
1736 static void free_note_info(struct elf_note_info *info)
1737 {
1738 	struct elf_thread_core_info *threads = info->thread;
1739 	while (threads) {
1740 		unsigned int i;
1741 		struct elf_thread_core_info *t = threads;
1742 		threads = t->next;
1743 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1744 		for (i = 1; i < info->thread_notes; ++i)
1745 			kfree(t->notes[i].data);
1746 		kfree(t);
1747 	}
1748 	kfree(info->psinfo.data);
1749 	vfree(info->files.data);
1750 }
1751 
1752 #else
1753 
1754 /* Here is the structure in which status of each thread is captured. */
1755 struct elf_thread_status
1756 {
1757 	struct list_head list;
1758 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1759 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1760 	struct task_struct *thread;
1761 #ifdef ELF_CORE_COPY_XFPREGS
1762 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1763 #endif
1764 	struct memelfnote notes[3];
1765 	int num_notes;
1766 };
1767 
1768 /*
1769  * In order to add the specific thread information for the elf file format,
1770  * we need to keep a linked list of every threads pr_status and then create
1771  * a single section for them in the final core file.
1772  */
1773 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1774 {
1775 	int sz = 0;
1776 	struct task_struct *p = t->thread;
1777 	t->num_notes = 0;
1778 
1779 	fill_prstatus(&t->prstatus, p, signr);
1780 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1781 
1782 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1783 		  &(t->prstatus));
1784 	t->num_notes++;
1785 	sz += notesize(&t->notes[0]);
1786 
1787 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1788 								&t->fpu))) {
1789 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1790 			  &(t->fpu));
1791 		t->num_notes++;
1792 		sz += notesize(&t->notes[1]);
1793 	}
1794 
1795 #ifdef ELF_CORE_COPY_XFPREGS
1796 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1797 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1798 			  sizeof(t->xfpu), &t->xfpu);
1799 		t->num_notes++;
1800 		sz += notesize(&t->notes[2]);
1801 	}
1802 #endif
1803 	return sz;
1804 }
1805 
1806 struct elf_note_info {
1807 	struct memelfnote *notes;
1808 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1809 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1810 	struct list_head thread_list;
1811 	elf_fpregset_t *fpu;
1812 #ifdef ELF_CORE_COPY_XFPREGS
1813 	elf_fpxregset_t *xfpu;
1814 #endif
1815 	user_siginfo_t csigdata;
1816 	int thread_status_size;
1817 	int numnote;
1818 };
1819 
1820 static int elf_note_info_init(struct elf_note_info *info)
1821 {
1822 	memset(info, 0, sizeof(*info));
1823 	INIT_LIST_HEAD(&info->thread_list);
1824 
1825 	/* Allocate space for ELF notes */
1826 	info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1827 	if (!info->notes)
1828 		return 0;
1829 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1830 	if (!info->psinfo)
1831 		return 0;
1832 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1833 	if (!info->prstatus)
1834 		return 0;
1835 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1836 	if (!info->fpu)
1837 		return 0;
1838 #ifdef ELF_CORE_COPY_XFPREGS
1839 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1840 	if (!info->xfpu)
1841 		return 0;
1842 #endif
1843 	return 1;
1844 }
1845 
1846 static int fill_note_info(struct elfhdr *elf, int phdrs,
1847 			  struct elf_note_info *info,
1848 			  siginfo_t *siginfo, struct pt_regs *regs)
1849 {
1850 	struct list_head *t;
1851 
1852 	if (!elf_note_info_init(info))
1853 		return 0;
1854 
1855 	if (siginfo->si_signo) {
1856 		struct core_thread *ct;
1857 		struct elf_thread_status *ets;
1858 
1859 		for (ct = current->mm->core_state->dumper.next;
1860 						ct; ct = ct->next) {
1861 			ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1862 			if (!ets)
1863 				return 0;
1864 
1865 			ets->thread = ct->task;
1866 			list_add(&ets->list, &info->thread_list);
1867 		}
1868 
1869 		list_for_each(t, &info->thread_list) {
1870 			int sz;
1871 
1872 			ets = list_entry(t, struct elf_thread_status, list);
1873 			sz = elf_dump_thread_status(siginfo->si_signo, ets);
1874 			info->thread_status_size += sz;
1875 		}
1876 	}
1877 	/* now collect the dump for the current */
1878 	memset(info->prstatus, 0, sizeof(*info->prstatus));
1879 	fill_prstatus(info->prstatus, current, siginfo->si_signo);
1880 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1881 
1882 	/* Set up header */
1883 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1884 
1885 	/*
1886 	 * Set up the notes in similar form to SVR4 core dumps made
1887 	 * with info from their /proc.
1888 	 */
1889 
1890 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1891 		  sizeof(*info->prstatus), info->prstatus);
1892 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
1893 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1894 		  sizeof(*info->psinfo), info->psinfo);
1895 
1896 	fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1897 	fill_auxv_note(info->notes + 3, current->mm);
1898 	fill_files_note(info->notes + 4);
1899 
1900 	info->numnote = 5;
1901 
1902 	/* Try to dump the FPU. */
1903 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1904 							       info->fpu);
1905 	if (info->prstatus->pr_fpvalid)
1906 		fill_note(info->notes + info->numnote++,
1907 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1908 #ifdef ELF_CORE_COPY_XFPREGS
1909 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
1910 		fill_note(info->notes + info->numnote++,
1911 			  "LINUX", ELF_CORE_XFPREG_TYPE,
1912 			  sizeof(*info->xfpu), info->xfpu);
1913 #endif
1914 
1915 	return 1;
1916 }
1917 
1918 static size_t get_note_info_size(struct elf_note_info *info)
1919 {
1920 	int sz = 0;
1921 	int i;
1922 
1923 	for (i = 0; i < info->numnote; i++)
1924 		sz += notesize(info->notes + i);
1925 
1926 	sz += info->thread_status_size;
1927 
1928 	return sz;
1929 }
1930 
1931 static int write_note_info(struct elf_note_info *info,
1932 			   struct file *file, loff_t *foffset)
1933 {
1934 	int i;
1935 	struct list_head *t;
1936 
1937 	for (i = 0; i < info->numnote; i++)
1938 		if (!writenote(info->notes + i, file, foffset))
1939 			return 0;
1940 
1941 	/* write out the thread status notes section */
1942 	list_for_each(t, &info->thread_list) {
1943 		struct elf_thread_status *tmp =
1944 				list_entry(t, struct elf_thread_status, list);
1945 
1946 		for (i = 0; i < tmp->num_notes; i++)
1947 			if (!writenote(&tmp->notes[i], file, foffset))
1948 				return 0;
1949 	}
1950 
1951 	return 1;
1952 }
1953 
1954 static void free_note_info(struct elf_note_info *info)
1955 {
1956 	while (!list_empty(&info->thread_list)) {
1957 		struct list_head *tmp = info->thread_list.next;
1958 		list_del(tmp);
1959 		kfree(list_entry(tmp, struct elf_thread_status, list));
1960 	}
1961 
1962 	/* Free data allocated by fill_files_note(): */
1963 	vfree(info->notes[4].data);
1964 
1965 	kfree(info->prstatus);
1966 	kfree(info->psinfo);
1967 	kfree(info->notes);
1968 	kfree(info->fpu);
1969 #ifdef ELF_CORE_COPY_XFPREGS
1970 	kfree(info->xfpu);
1971 #endif
1972 }
1973 
1974 #endif
1975 
1976 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1977 					struct vm_area_struct *gate_vma)
1978 {
1979 	struct vm_area_struct *ret = tsk->mm->mmap;
1980 
1981 	if (ret)
1982 		return ret;
1983 	return gate_vma;
1984 }
1985 /*
1986  * Helper function for iterating across a vma list.  It ensures that the caller
1987  * will visit `gate_vma' prior to terminating the search.
1988  */
1989 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1990 					struct vm_area_struct *gate_vma)
1991 {
1992 	struct vm_area_struct *ret;
1993 
1994 	ret = this_vma->vm_next;
1995 	if (ret)
1996 		return ret;
1997 	if (this_vma == gate_vma)
1998 		return NULL;
1999 	return gate_vma;
2000 }
2001 
2002 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2003 			     elf_addr_t e_shoff, int segs)
2004 {
2005 	elf->e_shoff = e_shoff;
2006 	elf->e_shentsize = sizeof(*shdr4extnum);
2007 	elf->e_shnum = 1;
2008 	elf->e_shstrndx = SHN_UNDEF;
2009 
2010 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2011 
2012 	shdr4extnum->sh_type = SHT_NULL;
2013 	shdr4extnum->sh_size = elf->e_shnum;
2014 	shdr4extnum->sh_link = elf->e_shstrndx;
2015 	shdr4extnum->sh_info = segs;
2016 }
2017 
2018 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
2019 				     unsigned long mm_flags)
2020 {
2021 	struct vm_area_struct *vma;
2022 	size_t size = 0;
2023 
2024 	for (vma = first_vma(current, gate_vma); vma != NULL;
2025 	     vma = next_vma(vma, gate_vma))
2026 		size += vma_dump_size(vma, mm_flags);
2027 	return size;
2028 }
2029 
2030 /*
2031  * Actual dumper
2032  *
2033  * This is a two-pass process; first we find the offsets of the bits,
2034  * and then they are actually written out.  If we run out of core limit
2035  * we just truncate.
2036  */
2037 static int elf_core_dump(struct coredump_params *cprm)
2038 {
2039 	int has_dumped = 0;
2040 	mm_segment_t fs;
2041 	int segs;
2042 	size_t size = 0;
2043 	struct vm_area_struct *vma, *gate_vma;
2044 	struct elfhdr *elf = NULL;
2045 	loff_t offset = 0, dataoff, foffset;
2046 	struct elf_note_info info;
2047 	struct elf_phdr *phdr4note = NULL;
2048 	struct elf_shdr *shdr4extnum = NULL;
2049 	Elf_Half e_phnum;
2050 	elf_addr_t e_shoff;
2051 
2052 	/*
2053 	 * We no longer stop all VM operations.
2054 	 *
2055 	 * This is because those proceses that could possibly change map_count
2056 	 * or the mmap / vma pages are now blocked in do_exit on current
2057 	 * finishing this core dump.
2058 	 *
2059 	 * Only ptrace can touch these memory addresses, but it doesn't change
2060 	 * the map_count or the pages allocated. So no possibility of crashing
2061 	 * exists while dumping the mm->vm_next areas to the core file.
2062 	 */
2063 
2064 	/* alloc memory for large data structures: too large to be on stack */
2065 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2066 	if (!elf)
2067 		goto out;
2068 	/*
2069 	 * The number of segs are recored into ELF header as 16bit value.
2070 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2071 	 */
2072 	segs = current->mm->map_count;
2073 	segs += elf_core_extra_phdrs();
2074 
2075 	gate_vma = get_gate_vma(current->mm);
2076 	if (gate_vma != NULL)
2077 		segs++;
2078 
2079 	/* for notes section */
2080 	segs++;
2081 
2082 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2083 	 * this, kernel supports extended numbering. Have a look at
2084 	 * include/linux/elf.h for further information. */
2085 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2086 
2087 	/*
2088 	 * Collect all the non-memory information about the process for the
2089 	 * notes.  This also sets up the file header.
2090 	 */
2091 	if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2092 		goto cleanup;
2093 
2094 	has_dumped = 1;
2095 	current->flags |= PF_DUMPCORE;
2096 
2097 	fs = get_fs();
2098 	set_fs(KERNEL_DS);
2099 
2100 	offset += sizeof(*elf);				/* Elf header */
2101 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2102 	foffset = offset;
2103 
2104 	/* Write notes phdr entry */
2105 	{
2106 		size_t sz = get_note_info_size(&info);
2107 
2108 		sz += elf_coredump_extra_notes_size();
2109 
2110 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2111 		if (!phdr4note)
2112 			goto end_coredump;
2113 
2114 		fill_elf_note_phdr(phdr4note, sz, offset);
2115 		offset += sz;
2116 	}
2117 
2118 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2119 
2120 	offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2121 	offset += elf_core_extra_data_size();
2122 	e_shoff = offset;
2123 
2124 	if (e_phnum == PN_XNUM) {
2125 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2126 		if (!shdr4extnum)
2127 			goto end_coredump;
2128 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2129 	}
2130 
2131 	offset = dataoff;
2132 
2133 	size += sizeof(*elf);
2134 	if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
2135 		goto end_coredump;
2136 
2137 	size += sizeof(*phdr4note);
2138 	if (size > cprm->limit
2139 	    || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
2140 		goto end_coredump;
2141 
2142 	/* Write program headers for segments dump */
2143 	for (vma = first_vma(current, gate_vma); vma != NULL;
2144 			vma = next_vma(vma, gate_vma)) {
2145 		struct elf_phdr phdr;
2146 
2147 		phdr.p_type = PT_LOAD;
2148 		phdr.p_offset = offset;
2149 		phdr.p_vaddr = vma->vm_start;
2150 		phdr.p_paddr = 0;
2151 		phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2152 		phdr.p_memsz = vma->vm_end - vma->vm_start;
2153 		offset += phdr.p_filesz;
2154 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2155 		if (vma->vm_flags & VM_WRITE)
2156 			phdr.p_flags |= PF_W;
2157 		if (vma->vm_flags & VM_EXEC)
2158 			phdr.p_flags |= PF_X;
2159 		phdr.p_align = ELF_EXEC_PAGESIZE;
2160 
2161 		size += sizeof(phdr);
2162 		if (size > cprm->limit
2163 		    || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2164 			goto end_coredump;
2165 	}
2166 
2167 	if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2168 		goto end_coredump;
2169 
2170  	/* write out the notes section */
2171 	if (!write_note_info(&info, cprm->file, &foffset))
2172 		goto end_coredump;
2173 
2174 	if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2175 		goto end_coredump;
2176 
2177 	/* Align to page */
2178 	if (!dump_seek(cprm->file, dataoff - foffset))
2179 		goto end_coredump;
2180 
2181 	for (vma = first_vma(current, gate_vma); vma != NULL;
2182 			vma = next_vma(vma, gate_vma)) {
2183 		unsigned long addr;
2184 		unsigned long end;
2185 
2186 		end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2187 
2188 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2189 			struct page *page;
2190 			int stop;
2191 
2192 			page = get_dump_page(addr);
2193 			if (page) {
2194 				void *kaddr = kmap(page);
2195 				stop = ((size += PAGE_SIZE) > cprm->limit) ||
2196 					!dump_write(cprm->file, kaddr,
2197 						    PAGE_SIZE);
2198 				kunmap(page);
2199 				page_cache_release(page);
2200 			} else
2201 				stop = !dump_seek(cprm->file, PAGE_SIZE);
2202 			if (stop)
2203 				goto end_coredump;
2204 		}
2205 	}
2206 
2207 	if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2208 		goto end_coredump;
2209 
2210 	if (e_phnum == PN_XNUM) {
2211 		size += sizeof(*shdr4extnum);
2212 		if (size > cprm->limit
2213 		    || !dump_write(cprm->file, shdr4extnum,
2214 				   sizeof(*shdr4extnum)))
2215 			goto end_coredump;
2216 	}
2217 
2218 end_coredump:
2219 	set_fs(fs);
2220 
2221 cleanup:
2222 	free_note_info(&info);
2223 	kfree(shdr4extnum);
2224 	kfree(phdr4note);
2225 	kfree(elf);
2226 out:
2227 	return has_dumped;
2228 }
2229 
2230 #endif		/* CONFIG_ELF_CORE */
2231 
2232 static int __init init_elf_binfmt(void)
2233 {
2234 	register_binfmt(&elf_format);
2235 	return 0;
2236 }
2237 
2238 static void __exit exit_elf_binfmt(void)
2239 {
2240 	/* Remove the COFF and ELF loaders. */
2241 	unregister_binfmt(&elf_format);
2242 }
2243 
2244 core_initcall(init_elf_binfmt);
2245 module_exit(exit_elf_binfmt);
2246 MODULE_LICENSE("GPL");
2247