xref: /openbmc/linux/fs/binfmt_elf.c (revision afc98d90)
1 /*
2  * linux/fs/binfmt_elf.c
3  *
4  * These are the functions used to load ELF format executables as used
5  * on SVr4 machines.  Information on the format may be found in the book
6  * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
7  * Tools".
8  *
9  * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
10  */
11 
12 #include <linux/module.h>
13 #include <linux/kernel.h>
14 #include <linux/fs.h>
15 #include <linux/mm.h>
16 #include <linux/mman.h>
17 #include <linux/errno.h>
18 #include <linux/signal.h>
19 #include <linux/binfmts.h>
20 #include <linux/string.h>
21 #include <linux/file.h>
22 #include <linux/slab.h>
23 #include <linux/personality.h>
24 #include <linux/elfcore.h>
25 #include <linux/init.h>
26 #include <linux/highuid.h>
27 #include <linux/compiler.h>
28 #include <linux/highmem.h>
29 #include <linux/pagemap.h>
30 #include <linux/vmalloc.h>
31 #include <linux/security.h>
32 #include <linux/random.h>
33 #include <linux/elf.h>
34 #include <linux/utsname.h>
35 #include <linux/coredump.h>
36 #include <linux/sched.h>
37 #include <asm/uaccess.h>
38 #include <asm/param.h>
39 #include <asm/page.h>
40 
41 #ifndef user_long_t
42 #define user_long_t long
43 #endif
44 #ifndef user_siginfo_t
45 #define user_siginfo_t siginfo_t
46 #endif
47 
48 static int load_elf_binary(struct linux_binprm *bprm);
49 static int load_elf_library(struct file *);
50 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
51 				int, int, unsigned long);
52 
53 /*
54  * If we don't support core dumping, then supply a NULL so we
55  * don't even try.
56  */
57 #ifdef CONFIG_ELF_CORE
58 static int elf_core_dump(struct coredump_params *cprm);
59 #else
60 #define elf_core_dump	NULL
61 #endif
62 
63 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
64 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
65 #else
66 #define ELF_MIN_ALIGN	PAGE_SIZE
67 #endif
68 
69 #ifndef ELF_CORE_EFLAGS
70 #define ELF_CORE_EFLAGS	0
71 #endif
72 
73 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
74 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
75 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
76 
77 static struct linux_binfmt elf_format = {
78 	.module		= THIS_MODULE,
79 	.load_binary	= load_elf_binary,
80 	.load_shlib	= load_elf_library,
81 	.core_dump	= elf_core_dump,
82 	.min_coredump	= ELF_EXEC_PAGESIZE,
83 };
84 
85 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
86 
87 static int set_brk(unsigned long start, unsigned long end)
88 {
89 	start = ELF_PAGEALIGN(start);
90 	end = ELF_PAGEALIGN(end);
91 	if (end > start) {
92 		unsigned long addr;
93 		addr = vm_brk(start, end - start);
94 		if (BAD_ADDR(addr))
95 			return addr;
96 	}
97 	current->mm->start_brk = current->mm->brk = end;
98 	return 0;
99 }
100 
101 /* We need to explicitly zero any fractional pages
102    after the data section (i.e. bss).  This would
103    contain the junk from the file that should not
104    be in memory
105  */
106 static int padzero(unsigned long elf_bss)
107 {
108 	unsigned long nbyte;
109 
110 	nbyte = ELF_PAGEOFFSET(elf_bss);
111 	if (nbyte) {
112 		nbyte = ELF_MIN_ALIGN - nbyte;
113 		if (clear_user((void __user *) elf_bss, nbyte))
114 			return -EFAULT;
115 	}
116 	return 0;
117 }
118 
119 /* Let's use some macros to make this stack manipulation a little clearer */
120 #ifdef CONFIG_STACK_GROWSUP
121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
122 #define STACK_ROUND(sp, items) \
123 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
124 #define STACK_ALLOC(sp, len) ({ \
125 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
126 	old_sp; })
127 #else
128 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
129 #define STACK_ROUND(sp, items) \
130 	(((unsigned long) (sp - items)) &~ 15UL)
131 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
132 #endif
133 
134 #ifndef ELF_BASE_PLATFORM
135 /*
136  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
137  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
138  * will be copied to the user stack in the same manner as AT_PLATFORM.
139  */
140 #define ELF_BASE_PLATFORM NULL
141 #endif
142 
143 static int
144 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
145 		unsigned long load_addr, unsigned long interp_load_addr)
146 {
147 	unsigned long p = bprm->p;
148 	int argc = bprm->argc;
149 	int envc = bprm->envc;
150 	elf_addr_t __user *argv;
151 	elf_addr_t __user *envp;
152 	elf_addr_t __user *sp;
153 	elf_addr_t __user *u_platform;
154 	elf_addr_t __user *u_base_platform;
155 	elf_addr_t __user *u_rand_bytes;
156 	const char *k_platform = ELF_PLATFORM;
157 	const char *k_base_platform = ELF_BASE_PLATFORM;
158 	unsigned char k_rand_bytes[16];
159 	int items;
160 	elf_addr_t *elf_info;
161 	int ei_index = 0;
162 	const struct cred *cred = current_cred();
163 	struct vm_area_struct *vma;
164 
165 	/*
166 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
167 	 * evictions by the processes running on the same package. One
168 	 * thing we can do is to shuffle the initial stack for them.
169 	 */
170 
171 	p = arch_align_stack(p);
172 
173 	/*
174 	 * If this architecture has a platform capability string, copy it
175 	 * to userspace.  In some cases (Sparc), this info is impossible
176 	 * for userspace to get any other way, in others (i386) it is
177 	 * merely difficult.
178 	 */
179 	u_platform = NULL;
180 	if (k_platform) {
181 		size_t len = strlen(k_platform) + 1;
182 
183 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
184 		if (__copy_to_user(u_platform, k_platform, len))
185 			return -EFAULT;
186 	}
187 
188 	/*
189 	 * If this architecture has a "base" platform capability
190 	 * string, copy it to userspace.
191 	 */
192 	u_base_platform = NULL;
193 	if (k_base_platform) {
194 		size_t len = strlen(k_base_platform) + 1;
195 
196 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
197 		if (__copy_to_user(u_base_platform, k_base_platform, len))
198 			return -EFAULT;
199 	}
200 
201 	/*
202 	 * Generate 16 random bytes for userspace PRNG seeding.
203 	 */
204 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
205 	u_rand_bytes = (elf_addr_t __user *)
206 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
207 	if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
208 		return -EFAULT;
209 
210 	/* Create the ELF interpreter info */
211 	elf_info = (elf_addr_t *)current->mm->saved_auxv;
212 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
213 #define NEW_AUX_ENT(id, val) \
214 	do { \
215 		elf_info[ei_index++] = id; \
216 		elf_info[ei_index++] = val; \
217 	} while (0)
218 
219 #ifdef ARCH_DLINFO
220 	/*
221 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
222 	 * AUXV.
223 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
224 	 * ARCH_DLINFO changes
225 	 */
226 	ARCH_DLINFO;
227 #endif
228 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
229 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
230 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
231 	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
232 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
233 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
234 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
235 	NEW_AUX_ENT(AT_FLAGS, 0);
236 	NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
237 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
238 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
239 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
240 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
241  	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
242 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
243 #ifdef ELF_HWCAP2
244 	NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
245 #endif
246 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
247 	if (k_platform) {
248 		NEW_AUX_ENT(AT_PLATFORM,
249 			    (elf_addr_t)(unsigned long)u_platform);
250 	}
251 	if (k_base_platform) {
252 		NEW_AUX_ENT(AT_BASE_PLATFORM,
253 			    (elf_addr_t)(unsigned long)u_base_platform);
254 	}
255 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
256 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
257 	}
258 #undef NEW_AUX_ENT
259 	/* AT_NULL is zero; clear the rest too */
260 	memset(&elf_info[ei_index], 0,
261 	       sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
262 
263 	/* And advance past the AT_NULL entry.  */
264 	ei_index += 2;
265 
266 	sp = STACK_ADD(p, ei_index);
267 
268 	items = (argc + 1) + (envc + 1) + 1;
269 	bprm->p = STACK_ROUND(sp, items);
270 
271 	/* Point sp at the lowest address on the stack */
272 #ifdef CONFIG_STACK_GROWSUP
273 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
274 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
275 #else
276 	sp = (elf_addr_t __user *)bprm->p;
277 #endif
278 
279 
280 	/*
281 	 * Grow the stack manually; some architectures have a limit on how
282 	 * far ahead a user-space access may be in order to grow the stack.
283 	 */
284 	vma = find_extend_vma(current->mm, bprm->p);
285 	if (!vma)
286 		return -EFAULT;
287 
288 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
289 	if (__put_user(argc, sp++))
290 		return -EFAULT;
291 	argv = sp;
292 	envp = argv + argc + 1;
293 
294 	/* Populate argv and envp */
295 	p = current->mm->arg_end = current->mm->arg_start;
296 	while (argc-- > 0) {
297 		size_t len;
298 		if (__put_user((elf_addr_t)p, argv++))
299 			return -EFAULT;
300 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
301 		if (!len || len > MAX_ARG_STRLEN)
302 			return -EINVAL;
303 		p += len;
304 	}
305 	if (__put_user(0, argv))
306 		return -EFAULT;
307 	current->mm->arg_end = current->mm->env_start = p;
308 	while (envc-- > 0) {
309 		size_t len;
310 		if (__put_user((elf_addr_t)p, envp++))
311 			return -EFAULT;
312 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
313 		if (!len || len > MAX_ARG_STRLEN)
314 			return -EINVAL;
315 		p += len;
316 	}
317 	if (__put_user(0, envp))
318 		return -EFAULT;
319 	current->mm->env_end = p;
320 
321 	/* Put the elf_info on the stack in the right place.  */
322 	sp = (elf_addr_t __user *)envp + 1;
323 	if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
324 		return -EFAULT;
325 	return 0;
326 }
327 
328 #ifndef elf_map
329 
330 static unsigned long elf_map(struct file *filep, unsigned long addr,
331 		struct elf_phdr *eppnt, int prot, int type,
332 		unsigned long total_size)
333 {
334 	unsigned long map_addr;
335 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
336 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
337 	addr = ELF_PAGESTART(addr);
338 	size = ELF_PAGEALIGN(size);
339 
340 	/* mmap() will return -EINVAL if given a zero size, but a
341 	 * segment with zero filesize is perfectly valid */
342 	if (!size)
343 		return addr;
344 
345 	/*
346 	* total_size is the size of the ELF (interpreter) image.
347 	* The _first_ mmap needs to know the full size, otherwise
348 	* randomization might put this image into an overlapping
349 	* position with the ELF binary image. (since size < total_size)
350 	* So we first map the 'big' image - and unmap the remainder at
351 	* the end. (which unmap is needed for ELF images with holes.)
352 	*/
353 	if (total_size) {
354 		total_size = ELF_PAGEALIGN(total_size);
355 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
356 		if (!BAD_ADDR(map_addr))
357 			vm_munmap(map_addr+size, total_size-size);
358 	} else
359 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
360 
361 	return(map_addr);
362 }
363 
364 #endif /* !elf_map */
365 
366 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
367 {
368 	int i, first_idx = -1, last_idx = -1;
369 
370 	for (i = 0; i < nr; i++) {
371 		if (cmds[i].p_type == PT_LOAD) {
372 			last_idx = i;
373 			if (first_idx == -1)
374 				first_idx = i;
375 		}
376 	}
377 	if (first_idx == -1)
378 		return 0;
379 
380 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
381 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
382 }
383 
384 
385 /* This is much more generalized than the library routine read function,
386    so we keep this separate.  Technically the library read function
387    is only provided so that we can read a.out libraries that have
388    an ELF header */
389 
390 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
391 		struct file *interpreter, unsigned long *interp_map_addr,
392 		unsigned long no_base)
393 {
394 	struct elf_phdr *elf_phdata;
395 	struct elf_phdr *eppnt;
396 	unsigned long load_addr = 0;
397 	int load_addr_set = 0;
398 	unsigned long last_bss = 0, elf_bss = 0;
399 	unsigned long error = ~0UL;
400 	unsigned long total_size;
401 	int retval, i, size;
402 
403 	/* First of all, some simple consistency checks */
404 	if (interp_elf_ex->e_type != ET_EXEC &&
405 	    interp_elf_ex->e_type != ET_DYN)
406 		goto out;
407 	if (!elf_check_arch(interp_elf_ex))
408 		goto out;
409 	if (!interpreter->f_op->mmap)
410 		goto out;
411 
412 	/*
413 	 * If the size of this structure has changed, then punt, since
414 	 * we will be doing the wrong thing.
415 	 */
416 	if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
417 		goto out;
418 	if (interp_elf_ex->e_phnum < 1 ||
419 		interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
420 		goto out;
421 
422 	/* Now read in all of the header information */
423 	size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
424 	if (size > ELF_MIN_ALIGN)
425 		goto out;
426 	elf_phdata = kmalloc(size, GFP_KERNEL);
427 	if (!elf_phdata)
428 		goto out;
429 
430 	retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
431 			     (char *)elf_phdata, size);
432 	error = -EIO;
433 	if (retval != size) {
434 		if (retval < 0)
435 			error = retval;
436 		goto out_close;
437 	}
438 
439 	total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
440 	if (!total_size) {
441 		error = -EINVAL;
442 		goto out_close;
443 	}
444 
445 	eppnt = elf_phdata;
446 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
447 		if (eppnt->p_type == PT_LOAD) {
448 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
449 			int elf_prot = 0;
450 			unsigned long vaddr = 0;
451 			unsigned long k, map_addr;
452 
453 			if (eppnt->p_flags & PF_R)
454 		    		elf_prot = PROT_READ;
455 			if (eppnt->p_flags & PF_W)
456 				elf_prot |= PROT_WRITE;
457 			if (eppnt->p_flags & PF_X)
458 				elf_prot |= PROT_EXEC;
459 			vaddr = eppnt->p_vaddr;
460 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
461 				elf_type |= MAP_FIXED;
462 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
463 				load_addr = -vaddr;
464 
465 			map_addr = elf_map(interpreter, load_addr + vaddr,
466 					eppnt, elf_prot, elf_type, total_size);
467 			total_size = 0;
468 			if (!*interp_map_addr)
469 				*interp_map_addr = map_addr;
470 			error = map_addr;
471 			if (BAD_ADDR(map_addr))
472 				goto out_close;
473 
474 			if (!load_addr_set &&
475 			    interp_elf_ex->e_type == ET_DYN) {
476 				load_addr = map_addr - ELF_PAGESTART(vaddr);
477 				load_addr_set = 1;
478 			}
479 
480 			/*
481 			 * Check to see if the section's size will overflow the
482 			 * allowed task size. Note that p_filesz must always be
483 			 * <= p_memsize so it's only necessary to check p_memsz.
484 			 */
485 			k = load_addr + eppnt->p_vaddr;
486 			if (BAD_ADDR(k) ||
487 			    eppnt->p_filesz > eppnt->p_memsz ||
488 			    eppnt->p_memsz > TASK_SIZE ||
489 			    TASK_SIZE - eppnt->p_memsz < k) {
490 				error = -ENOMEM;
491 				goto out_close;
492 			}
493 
494 			/*
495 			 * Find the end of the file mapping for this phdr, and
496 			 * keep track of the largest address we see for this.
497 			 */
498 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
499 			if (k > elf_bss)
500 				elf_bss = k;
501 
502 			/*
503 			 * Do the same thing for the memory mapping - between
504 			 * elf_bss and last_bss is the bss section.
505 			 */
506 			k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
507 			if (k > last_bss)
508 				last_bss = k;
509 		}
510 	}
511 
512 	if (last_bss > elf_bss) {
513 		/*
514 		 * Now fill out the bss section.  First pad the last page up
515 		 * to the page boundary, and then perform a mmap to make sure
516 		 * that there are zero-mapped pages up to and including the
517 		 * last bss page.
518 		 */
519 		if (padzero(elf_bss)) {
520 			error = -EFAULT;
521 			goto out_close;
522 		}
523 
524 		/* What we have mapped so far */
525 		elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
526 
527 		/* Map the last of the bss segment */
528 		error = vm_brk(elf_bss, last_bss - elf_bss);
529 		if (BAD_ADDR(error))
530 			goto out_close;
531 	}
532 
533 	error = load_addr;
534 
535 out_close:
536 	kfree(elf_phdata);
537 out:
538 	return error;
539 }
540 
541 /*
542  * These are the functions used to load ELF style executables and shared
543  * libraries.  There is no binary dependent code anywhere else.
544  */
545 
546 #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->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 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
739 				 executable_stack);
740 	if (retval < 0) {
741 		send_sig(SIGKILL, current, 0);
742 		goto out_free_dentry;
743 	}
744 
745 	current->mm->start_stack = bprm->p;
746 
747 	/* Now we do a little grungy work by mmapping the ELF image into
748 	   the correct location in memory. */
749 	for(i = 0, elf_ppnt = elf_phdata;
750 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
751 		int elf_prot = 0, elf_flags;
752 		unsigned long k, vaddr;
753 
754 		if (elf_ppnt->p_type != PT_LOAD)
755 			continue;
756 
757 		if (unlikely (elf_brk > elf_bss)) {
758 			unsigned long nbyte;
759 
760 			/* There was a PT_LOAD segment with p_memsz > p_filesz
761 			   before this one. Map anonymous pages, if needed,
762 			   and clear the area.  */
763 			retval = set_brk(elf_bss + load_bias,
764 					 elf_brk + load_bias);
765 			if (retval) {
766 				send_sig(SIGKILL, current, 0);
767 				goto out_free_dentry;
768 			}
769 			nbyte = ELF_PAGEOFFSET(elf_bss);
770 			if (nbyte) {
771 				nbyte = ELF_MIN_ALIGN - nbyte;
772 				if (nbyte > elf_brk - elf_bss)
773 					nbyte = elf_brk - elf_bss;
774 				if (clear_user((void __user *)elf_bss +
775 							load_bias, nbyte)) {
776 					/*
777 					 * This bss-zeroing can fail if the ELF
778 					 * file specifies odd protections. So
779 					 * we don't check the return value
780 					 */
781 				}
782 			}
783 		}
784 
785 		if (elf_ppnt->p_flags & PF_R)
786 			elf_prot |= PROT_READ;
787 		if (elf_ppnt->p_flags & PF_W)
788 			elf_prot |= PROT_WRITE;
789 		if (elf_ppnt->p_flags & PF_X)
790 			elf_prot |= PROT_EXEC;
791 
792 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
793 
794 		vaddr = elf_ppnt->p_vaddr;
795 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
796 			elf_flags |= MAP_FIXED;
797 		} else if (loc->elf_ex.e_type == ET_DYN) {
798 			/* Try and get dynamic programs out of the way of the
799 			 * default mmap base, as well as whatever program they
800 			 * might try to exec.  This is because the brk will
801 			 * follow the loader, and is not movable.  */
802 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
803 			/* Memory randomization might have been switched off
804 			 * in runtime via sysctl or explicit setting of
805 			 * personality flags.
806 			 * If that is the case, retain the original non-zero
807 			 * load_bias value in order to establish proper
808 			 * non-randomized mappings.
809 			 */
810 			if (current->flags & PF_RANDOMIZE)
811 				load_bias = 0;
812 			else
813 				load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
814 #else
815 			load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
816 #endif
817 		}
818 
819 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
820 				elf_prot, elf_flags, 0);
821 		if (BAD_ADDR(error)) {
822 			send_sig(SIGKILL, current, 0);
823 			retval = IS_ERR((void *)error) ?
824 				PTR_ERR((void*)error) : -EINVAL;
825 			goto out_free_dentry;
826 		}
827 
828 		if (!load_addr_set) {
829 			load_addr_set = 1;
830 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
831 			if (loc->elf_ex.e_type == ET_DYN) {
832 				load_bias += error -
833 				             ELF_PAGESTART(load_bias + vaddr);
834 				load_addr += load_bias;
835 				reloc_func_desc = load_bias;
836 			}
837 		}
838 		k = elf_ppnt->p_vaddr;
839 		if (k < start_code)
840 			start_code = k;
841 		if (start_data < k)
842 			start_data = k;
843 
844 		/*
845 		 * Check to see if the section's size will overflow the
846 		 * allowed task size. Note that p_filesz must always be
847 		 * <= p_memsz so it is only necessary to check p_memsz.
848 		 */
849 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
850 		    elf_ppnt->p_memsz > TASK_SIZE ||
851 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
852 			/* set_brk can never work. Avoid overflows. */
853 			send_sig(SIGKILL, current, 0);
854 			retval = -EINVAL;
855 			goto out_free_dentry;
856 		}
857 
858 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
859 
860 		if (k > elf_bss)
861 			elf_bss = k;
862 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
863 			end_code = k;
864 		if (end_data < k)
865 			end_data = k;
866 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
867 		if (k > elf_brk)
868 			elf_brk = k;
869 	}
870 
871 	loc->elf_ex.e_entry += load_bias;
872 	elf_bss += load_bias;
873 	elf_brk += load_bias;
874 	start_code += load_bias;
875 	end_code += load_bias;
876 	start_data += load_bias;
877 	end_data += load_bias;
878 
879 	/* Calling set_brk effectively mmaps the pages that we need
880 	 * for the bss and break sections.  We must do this before
881 	 * mapping in the interpreter, to make sure it doesn't wind
882 	 * up getting placed where the bss needs to go.
883 	 */
884 	retval = set_brk(elf_bss, elf_brk);
885 	if (retval) {
886 		send_sig(SIGKILL, current, 0);
887 		goto out_free_dentry;
888 	}
889 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
890 		send_sig(SIGSEGV, current, 0);
891 		retval = -EFAULT; /* Nobody gets to see this, but.. */
892 		goto out_free_dentry;
893 	}
894 
895 	if (elf_interpreter) {
896 		unsigned long interp_map_addr = 0;
897 
898 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
899 					    interpreter,
900 					    &interp_map_addr,
901 					    load_bias);
902 		if (!IS_ERR((void *)elf_entry)) {
903 			/*
904 			 * load_elf_interp() returns relocation
905 			 * adjustment
906 			 */
907 			interp_load_addr = elf_entry;
908 			elf_entry += loc->interp_elf_ex.e_entry;
909 		}
910 		if (BAD_ADDR(elf_entry)) {
911 			force_sig(SIGSEGV, current);
912 			retval = IS_ERR((void *)elf_entry) ?
913 					(int)elf_entry : -EINVAL;
914 			goto out_free_dentry;
915 		}
916 		reloc_func_desc = interp_load_addr;
917 
918 		allow_write_access(interpreter);
919 		fput(interpreter);
920 		kfree(elf_interpreter);
921 	} else {
922 		elf_entry = loc->elf_ex.e_entry;
923 		if (BAD_ADDR(elf_entry)) {
924 			force_sig(SIGSEGV, current);
925 			retval = -EINVAL;
926 			goto out_free_dentry;
927 		}
928 	}
929 
930 	kfree(elf_phdata);
931 
932 	set_binfmt(&elf_format);
933 
934 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
935 	retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
936 	if (retval < 0) {
937 		send_sig(SIGKILL, current, 0);
938 		goto out;
939 	}
940 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
941 
942 	install_exec_creds(bprm);
943 	retval = create_elf_tables(bprm, &loc->elf_ex,
944 			  load_addr, interp_load_addr);
945 	if (retval < 0) {
946 		send_sig(SIGKILL, current, 0);
947 		goto out;
948 	}
949 	/* N.B. passed_fileno might not be initialized? */
950 	current->mm->end_code = end_code;
951 	current->mm->start_code = start_code;
952 	current->mm->start_data = start_data;
953 	current->mm->end_data = end_data;
954 	current->mm->start_stack = bprm->p;
955 
956 #ifdef arch_randomize_brk
957 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
958 		current->mm->brk = current->mm->start_brk =
959 			arch_randomize_brk(current->mm);
960 #ifdef CONFIG_COMPAT_BRK
961 		current->brk_randomized = 1;
962 #endif
963 	}
964 #endif
965 
966 	if (current->personality & MMAP_PAGE_ZERO) {
967 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
968 		   and some applications "depend" upon this behavior.
969 		   Since we do not have the power to recompile these, we
970 		   emulate the SVr4 behavior. Sigh. */
971 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
972 				MAP_FIXED | MAP_PRIVATE, 0);
973 	}
974 
975 #ifdef ELF_PLAT_INIT
976 	/*
977 	 * The ABI may specify that certain registers be set up in special
978 	 * ways (on i386 %edx is the address of a DT_FINI function, for
979 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
980 	 * that the e_entry field is the address of the function descriptor
981 	 * for the startup routine, rather than the address of the startup
982 	 * routine itself.  This macro performs whatever initialization to
983 	 * the regs structure is required as well as any relocations to the
984 	 * function descriptor entries when executing dynamically links apps.
985 	 */
986 	ELF_PLAT_INIT(regs, reloc_func_desc);
987 #endif
988 
989 	start_thread(regs, elf_entry, bprm->p);
990 	retval = 0;
991 out:
992 	kfree(loc);
993 out_ret:
994 	return retval;
995 
996 	/* error cleanup */
997 out_free_dentry:
998 	allow_write_access(interpreter);
999 	if (interpreter)
1000 		fput(interpreter);
1001 out_free_interp:
1002 	kfree(elf_interpreter);
1003 out_free_ph:
1004 	kfree(elf_phdata);
1005 	goto out;
1006 }
1007 
1008 /* This is really simpleminded and specialized - we are loading an
1009    a.out library that is given an ELF header. */
1010 static int load_elf_library(struct file *file)
1011 {
1012 	struct elf_phdr *elf_phdata;
1013 	struct elf_phdr *eppnt;
1014 	unsigned long elf_bss, bss, len;
1015 	int retval, error, i, j;
1016 	struct elfhdr elf_ex;
1017 
1018 	error = -ENOEXEC;
1019 	retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1020 	if (retval != sizeof(elf_ex))
1021 		goto out;
1022 
1023 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1024 		goto out;
1025 
1026 	/* First of all, some simple consistency checks */
1027 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1028 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1029 		goto out;
1030 
1031 	/* Now read in all of the header information */
1032 
1033 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1034 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1035 
1036 	error = -ENOMEM;
1037 	elf_phdata = kmalloc(j, GFP_KERNEL);
1038 	if (!elf_phdata)
1039 		goto out;
1040 
1041 	eppnt = elf_phdata;
1042 	error = -ENOEXEC;
1043 	retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1044 	if (retval != j)
1045 		goto out_free_ph;
1046 
1047 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1048 		if ((eppnt + i)->p_type == PT_LOAD)
1049 			j++;
1050 	if (j != 1)
1051 		goto out_free_ph;
1052 
1053 	while (eppnt->p_type != PT_LOAD)
1054 		eppnt++;
1055 
1056 	/* Now use mmap to map the library into memory. */
1057 	error = vm_mmap(file,
1058 			ELF_PAGESTART(eppnt->p_vaddr),
1059 			(eppnt->p_filesz +
1060 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1061 			PROT_READ | PROT_WRITE | PROT_EXEC,
1062 			MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1063 			(eppnt->p_offset -
1064 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1065 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1066 		goto out_free_ph;
1067 
1068 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1069 	if (padzero(elf_bss)) {
1070 		error = -EFAULT;
1071 		goto out_free_ph;
1072 	}
1073 
1074 	len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1075 			    ELF_MIN_ALIGN - 1);
1076 	bss = eppnt->p_memsz + eppnt->p_vaddr;
1077 	if (bss > len)
1078 		vm_brk(len, bss - len);
1079 	error = 0;
1080 
1081 out_free_ph:
1082 	kfree(elf_phdata);
1083 out:
1084 	return error;
1085 }
1086 
1087 #ifdef CONFIG_ELF_CORE
1088 /*
1089  * ELF core dumper
1090  *
1091  * Modelled on fs/exec.c:aout_core_dump()
1092  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1093  */
1094 
1095 /*
1096  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1097  * that are useful for post-mortem analysis are included in every core dump.
1098  * In that way we ensure that the core dump is fully interpretable later
1099  * without matching up the same kernel and hardware config to see what PC values
1100  * meant. These special mappings include - vDSO, vsyscall, and other
1101  * architecture specific mappings
1102  */
1103 static bool always_dump_vma(struct vm_area_struct *vma)
1104 {
1105 	/* Any vsyscall mappings? */
1106 	if (vma == get_gate_vma(vma->vm_mm))
1107 		return true;
1108 	/*
1109 	 * arch_vma_name() returns non-NULL for special architecture mappings,
1110 	 * such as vDSO sections.
1111 	 */
1112 	if (arch_vma_name(vma))
1113 		return true;
1114 
1115 	return false;
1116 }
1117 
1118 /*
1119  * Decide what to dump of a segment, part, all or none.
1120  */
1121 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1122 				   unsigned long mm_flags)
1123 {
1124 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1125 
1126 	/* always dump the vdso and vsyscall sections */
1127 	if (always_dump_vma(vma))
1128 		goto whole;
1129 
1130 	if (vma->vm_flags & VM_DONTDUMP)
1131 		return 0;
1132 
1133 	/* Hugetlb memory check */
1134 	if (vma->vm_flags & VM_HUGETLB) {
1135 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1136 			goto whole;
1137 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1138 			goto whole;
1139 		return 0;
1140 	}
1141 
1142 	/* Do not dump I/O mapped devices or special mappings */
1143 	if (vma->vm_flags & VM_IO)
1144 		return 0;
1145 
1146 	/* By default, dump shared memory if mapped from an anonymous file. */
1147 	if (vma->vm_flags & VM_SHARED) {
1148 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1149 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1150 			goto whole;
1151 		return 0;
1152 	}
1153 
1154 	/* Dump segments that have been written to.  */
1155 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1156 		goto whole;
1157 	if (vma->vm_file == NULL)
1158 		return 0;
1159 
1160 	if (FILTER(MAPPED_PRIVATE))
1161 		goto whole;
1162 
1163 	/*
1164 	 * If this looks like the beginning of a DSO or executable mapping,
1165 	 * check for an ELF header.  If we find one, dump the first page to
1166 	 * aid in determining what was mapped here.
1167 	 */
1168 	if (FILTER(ELF_HEADERS) &&
1169 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1170 		u32 __user *header = (u32 __user *) vma->vm_start;
1171 		u32 word;
1172 		mm_segment_t fs = get_fs();
1173 		/*
1174 		 * Doing it this way gets the constant folded by GCC.
1175 		 */
1176 		union {
1177 			u32 cmp;
1178 			char elfmag[SELFMAG];
1179 		} magic;
1180 		BUILD_BUG_ON(SELFMAG != sizeof word);
1181 		magic.elfmag[EI_MAG0] = ELFMAG0;
1182 		magic.elfmag[EI_MAG1] = ELFMAG1;
1183 		magic.elfmag[EI_MAG2] = ELFMAG2;
1184 		magic.elfmag[EI_MAG3] = ELFMAG3;
1185 		/*
1186 		 * Switch to the user "segment" for get_user(),
1187 		 * then put back what elf_core_dump() had in place.
1188 		 */
1189 		set_fs(USER_DS);
1190 		if (unlikely(get_user(word, header)))
1191 			word = 0;
1192 		set_fs(fs);
1193 		if (word == magic.cmp)
1194 			return PAGE_SIZE;
1195 	}
1196 
1197 #undef	FILTER
1198 
1199 	return 0;
1200 
1201 whole:
1202 	return vma->vm_end - vma->vm_start;
1203 }
1204 
1205 /* An ELF note in memory */
1206 struct memelfnote
1207 {
1208 	const char *name;
1209 	int type;
1210 	unsigned int datasz;
1211 	void *data;
1212 };
1213 
1214 static int notesize(struct memelfnote *en)
1215 {
1216 	int sz;
1217 
1218 	sz = sizeof(struct elf_note);
1219 	sz += roundup(strlen(en->name) + 1, 4);
1220 	sz += roundup(en->datasz, 4);
1221 
1222 	return sz;
1223 }
1224 
1225 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1226 {
1227 	struct elf_note en;
1228 	en.n_namesz = strlen(men->name) + 1;
1229 	en.n_descsz = men->datasz;
1230 	en.n_type = men->type;
1231 
1232 	return dump_emit(cprm, &en, sizeof(en)) &&
1233 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1234 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1235 }
1236 
1237 static void fill_elf_header(struct elfhdr *elf, int segs,
1238 			    u16 machine, u32 flags)
1239 {
1240 	memset(elf, 0, sizeof(*elf));
1241 
1242 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1243 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1244 	elf->e_ident[EI_DATA] = ELF_DATA;
1245 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1246 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1247 
1248 	elf->e_type = ET_CORE;
1249 	elf->e_machine = machine;
1250 	elf->e_version = EV_CURRENT;
1251 	elf->e_phoff = sizeof(struct elfhdr);
1252 	elf->e_flags = flags;
1253 	elf->e_ehsize = sizeof(struct elfhdr);
1254 	elf->e_phentsize = sizeof(struct elf_phdr);
1255 	elf->e_phnum = segs;
1256 
1257 	return;
1258 }
1259 
1260 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1261 {
1262 	phdr->p_type = PT_NOTE;
1263 	phdr->p_offset = offset;
1264 	phdr->p_vaddr = 0;
1265 	phdr->p_paddr = 0;
1266 	phdr->p_filesz = sz;
1267 	phdr->p_memsz = 0;
1268 	phdr->p_flags = 0;
1269 	phdr->p_align = 0;
1270 	return;
1271 }
1272 
1273 static void fill_note(struct memelfnote *note, const char *name, int type,
1274 		unsigned int sz, void *data)
1275 {
1276 	note->name = name;
1277 	note->type = type;
1278 	note->datasz = sz;
1279 	note->data = data;
1280 	return;
1281 }
1282 
1283 /*
1284  * fill up all the fields in prstatus from the given task struct, except
1285  * registers which need to be filled up separately.
1286  */
1287 static void fill_prstatus(struct elf_prstatus *prstatus,
1288 		struct task_struct *p, long signr)
1289 {
1290 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1291 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1292 	prstatus->pr_sighold = p->blocked.sig[0];
1293 	rcu_read_lock();
1294 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1295 	rcu_read_unlock();
1296 	prstatus->pr_pid = task_pid_vnr(p);
1297 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1298 	prstatus->pr_sid = task_session_vnr(p);
1299 	if (thread_group_leader(p)) {
1300 		struct task_cputime cputime;
1301 
1302 		/*
1303 		 * This is the record for the group leader.  It shows the
1304 		 * group-wide total, not its individual thread total.
1305 		 */
1306 		thread_group_cputime(p, &cputime);
1307 		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1308 		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1309 	} else {
1310 		cputime_t utime, stime;
1311 
1312 		task_cputime(p, &utime, &stime);
1313 		cputime_to_timeval(utime, &prstatus->pr_utime);
1314 		cputime_to_timeval(stime, &prstatus->pr_stime);
1315 	}
1316 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1317 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1318 }
1319 
1320 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1321 		       struct mm_struct *mm)
1322 {
1323 	const struct cred *cred;
1324 	unsigned int i, len;
1325 
1326 	/* first copy the parameters from user space */
1327 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1328 
1329 	len = mm->arg_end - mm->arg_start;
1330 	if (len >= ELF_PRARGSZ)
1331 		len = ELF_PRARGSZ-1;
1332 	if (copy_from_user(&psinfo->pr_psargs,
1333 		           (const char __user *)mm->arg_start, len))
1334 		return -EFAULT;
1335 	for(i = 0; i < len; i++)
1336 		if (psinfo->pr_psargs[i] == 0)
1337 			psinfo->pr_psargs[i] = ' ';
1338 	psinfo->pr_psargs[len] = 0;
1339 
1340 	rcu_read_lock();
1341 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1342 	rcu_read_unlock();
1343 	psinfo->pr_pid = task_pid_vnr(p);
1344 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1345 	psinfo->pr_sid = task_session_vnr(p);
1346 
1347 	i = p->state ? ffz(~p->state) + 1 : 0;
1348 	psinfo->pr_state = i;
1349 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1350 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1351 	psinfo->pr_nice = task_nice(p);
1352 	psinfo->pr_flag = p->flags;
1353 	rcu_read_lock();
1354 	cred = __task_cred(p);
1355 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1356 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1357 	rcu_read_unlock();
1358 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1359 
1360 	return 0;
1361 }
1362 
1363 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1364 {
1365 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1366 	int i = 0;
1367 	do
1368 		i += 2;
1369 	while (auxv[i - 2] != AT_NULL);
1370 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1371 }
1372 
1373 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1374 		const siginfo_t *siginfo)
1375 {
1376 	mm_segment_t old_fs = get_fs();
1377 	set_fs(KERNEL_DS);
1378 	copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1379 	set_fs(old_fs);
1380 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1381 }
1382 
1383 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1384 /*
1385  * Format of NT_FILE note:
1386  *
1387  * long count     -- how many files are mapped
1388  * long page_size -- units for file_ofs
1389  * array of [COUNT] elements of
1390  *   long start
1391  *   long end
1392  *   long file_ofs
1393  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1394  */
1395 static int fill_files_note(struct memelfnote *note)
1396 {
1397 	struct vm_area_struct *vma;
1398 	unsigned count, size, names_ofs, remaining, n;
1399 	user_long_t *data;
1400 	user_long_t *start_end_ofs;
1401 	char *name_base, *name_curpos;
1402 
1403 	/* *Estimated* file count and total data size needed */
1404 	count = current->mm->map_count;
1405 	size = count * 64;
1406 
1407 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1408  alloc:
1409 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1410 		return -EINVAL;
1411 	size = round_up(size, PAGE_SIZE);
1412 	data = vmalloc(size);
1413 	if (!data)
1414 		return -ENOMEM;
1415 
1416 	start_end_ofs = data + 2;
1417 	name_base = name_curpos = ((char *)data) + names_ofs;
1418 	remaining = size - names_ofs;
1419 	count = 0;
1420 	for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1421 		struct file *file;
1422 		const char *filename;
1423 
1424 		file = vma->vm_file;
1425 		if (!file)
1426 			continue;
1427 		filename = d_path(&file->f_path, name_curpos, remaining);
1428 		if (IS_ERR(filename)) {
1429 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1430 				vfree(data);
1431 				size = size * 5 / 4;
1432 				goto alloc;
1433 			}
1434 			continue;
1435 		}
1436 
1437 		/* d_path() fills at the end, move name down */
1438 		/* n = strlen(filename) + 1: */
1439 		n = (name_curpos + remaining) - filename;
1440 		remaining = filename - name_curpos;
1441 		memmove(name_curpos, filename, n);
1442 		name_curpos += n;
1443 
1444 		*start_end_ofs++ = vma->vm_start;
1445 		*start_end_ofs++ = vma->vm_end;
1446 		*start_end_ofs++ = vma->vm_pgoff;
1447 		count++;
1448 	}
1449 
1450 	/* Now we know exact count of files, can store it */
1451 	data[0] = count;
1452 	data[1] = PAGE_SIZE;
1453 	/*
1454 	 * Count usually is less than current->mm->map_count,
1455 	 * we need to move filenames down.
1456 	 */
1457 	n = current->mm->map_count - count;
1458 	if (n != 0) {
1459 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1460 		memmove(name_base - shift_bytes, name_base,
1461 			name_curpos - name_base);
1462 		name_curpos -= shift_bytes;
1463 	}
1464 
1465 	size = name_curpos - (char *)data;
1466 	fill_note(note, "CORE", NT_FILE, size, data);
1467 	return 0;
1468 }
1469 
1470 #ifdef CORE_DUMP_USE_REGSET
1471 #include <linux/regset.h>
1472 
1473 struct elf_thread_core_info {
1474 	struct elf_thread_core_info *next;
1475 	struct task_struct *task;
1476 	struct elf_prstatus prstatus;
1477 	struct memelfnote notes[0];
1478 };
1479 
1480 struct elf_note_info {
1481 	struct elf_thread_core_info *thread;
1482 	struct memelfnote psinfo;
1483 	struct memelfnote signote;
1484 	struct memelfnote auxv;
1485 	struct memelfnote files;
1486 	user_siginfo_t csigdata;
1487 	size_t size;
1488 	int thread_notes;
1489 };
1490 
1491 /*
1492  * When a regset has a writeback hook, we call it on each thread before
1493  * dumping user memory.  On register window machines, this makes sure the
1494  * user memory backing the register data is up to date before we read it.
1495  */
1496 static void do_thread_regset_writeback(struct task_struct *task,
1497 				       const struct user_regset *regset)
1498 {
1499 	if (regset->writeback)
1500 		regset->writeback(task, regset, 1);
1501 }
1502 
1503 #ifndef PR_REG_SIZE
1504 #define PR_REG_SIZE(S) sizeof(S)
1505 #endif
1506 
1507 #ifndef PRSTATUS_SIZE
1508 #define PRSTATUS_SIZE(S) sizeof(S)
1509 #endif
1510 
1511 #ifndef PR_REG_PTR
1512 #define PR_REG_PTR(S) (&((S)->pr_reg))
1513 #endif
1514 
1515 #ifndef SET_PR_FPVALID
1516 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1517 #endif
1518 
1519 static int fill_thread_core_info(struct elf_thread_core_info *t,
1520 				 const struct user_regset_view *view,
1521 				 long signr, size_t *total)
1522 {
1523 	unsigned int i;
1524 
1525 	/*
1526 	 * NT_PRSTATUS is the one special case, because the regset data
1527 	 * goes into the pr_reg field inside the note contents, rather
1528 	 * than being the whole note contents.  We fill the reset in here.
1529 	 * We assume that regset 0 is NT_PRSTATUS.
1530 	 */
1531 	fill_prstatus(&t->prstatus, t->task, signr);
1532 	(void) view->regsets[0].get(t->task, &view->regsets[0],
1533 				    0, PR_REG_SIZE(t->prstatus.pr_reg),
1534 				    PR_REG_PTR(&t->prstatus), NULL);
1535 
1536 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1537 		  PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1538 	*total += notesize(&t->notes[0]);
1539 
1540 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1541 
1542 	/*
1543 	 * Each other regset might generate a note too.  For each regset
1544 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1545 	 * all zero and we'll know to skip writing it later.
1546 	 */
1547 	for (i = 1; i < view->n; ++i) {
1548 		const struct user_regset *regset = &view->regsets[i];
1549 		do_thread_regset_writeback(t->task, regset);
1550 		if (regset->core_note_type && regset->get &&
1551 		    (!regset->active || regset->active(t->task, regset))) {
1552 			int ret;
1553 			size_t size = regset->n * regset->size;
1554 			void *data = kmalloc(size, GFP_KERNEL);
1555 			if (unlikely(!data))
1556 				return 0;
1557 			ret = regset->get(t->task, regset,
1558 					  0, size, data, NULL);
1559 			if (unlikely(ret))
1560 				kfree(data);
1561 			else {
1562 				if (regset->core_note_type != NT_PRFPREG)
1563 					fill_note(&t->notes[i], "LINUX",
1564 						  regset->core_note_type,
1565 						  size, data);
1566 				else {
1567 					SET_PR_FPVALID(&t->prstatus, 1);
1568 					fill_note(&t->notes[i], "CORE",
1569 						  NT_PRFPREG, size, data);
1570 				}
1571 				*total += notesize(&t->notes[i]);
1572 			}
1573 		}
1574 	}
1575 
1576 	return 1;
1577 }
1578 
1579 static int fill_note_info(struct elfhdr *elf, int phdrs,
1580 			  struct elf_note_info *info,
1581 			  const siginfo_t *siginfo, struct pt_regs *regs)
1582 {
1583 	struct task_struct *dump_task = current;
1584 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1585 	struct elf_thread_core_info *t;
1586 	struct elf_prpsinfo *psinfo;
1587 	struct core_thread *ct;
1588 	unsigned int i;
1589 
1590 	info->size = 0;
1591 	info->thread = NULL;
1592 
1593 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1594 	if (psinfo == NULL) {
1595 		info->psinfo.data = NULL; /* So we don't free this wrongly */
1596 		return 0;
1597 	}
1598 
1599 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1600 
1601 	/*
1602 	 * Figure out how many notes we're going to need for each thread.
1603 	 */
1604 	info->thread_notes = 0;
1605 	for (i = 0; i < view->n; ++i)
1606 		if (view->regsets[i].core_note_type != 0)
1607 			++info->thread_notes;
1608 
1609 	/*
1610 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1611 	 * since it is our one special case.
1612 	 */
1613 	if (unlikely(info->thread_notes == 0) ||
1614 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1615 		WARN_ON(1);
1616 		return 0;
1617 	}
1618 
1619 	/*
1620 	 * Initialize the ELF file header.
1621 	 */
1622 	fill_elf_header(elf, phdrs,
1623 			view->e_machine, view->e_flags);
1624 
1625 	/*
1626 	 * Allocate a structure for each thread.
1627 	 */
1628 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1629 		t = kzalloc(offsetof(struct elf_thread_core_info,
1630 				     notes[info->thread_notes]),
1631 			    GFP_KERNEL);
1632 		if (unlikely(!t))
1633 			return 0;
1634 
1635 		t->task = ct->task;
1636 		if (ct->task == dump_task || !info->thread) {
1637 			t->next = info->thread;
1638 			info->thread = t;
1639 		} else {
1640 			/*
1641 			 * Make sure to keep the original task at
1642 			 * the head of the list.
1643 			 */
1644 			t->next = info->thread->next;
1645 			info->thread->next = t;
1646 		}
1647 	}
1648 
1649 	/*
1650 	 * Now fill in each thread's information.
1651 	 */
1652 	for (t = info->thread; t != NULL; t = t->next)
1653 		if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1654 			return 0;
1655 
1656 	/*
1657 	 * Fill in the two process-wide notes.
1658 	 */
1659 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1660 	info->size += notesize(&info->psinfo);
1661 
1662 	fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1663 	info->size += notesize(&info->signote);
1664 
1665 	fill_auxv_note(&info->auxv, current->mm);
1666 	info->size += notesize(&info->auxv);
1667 
1668 	if (fill_files_note(&info->files) == 0)
1669 		info->size += notesize(&info->files);
1670 
1671 	return 1;
1672 }
1673 
1674 static size_t get_note_info_size(struct elf_note_info *info)
1675 {
1676 	return info->size;
1677 }
1678 
1679 /*
1680  * Write all the notes for each thread.  When writing the first thread, the
1681  * process-wide notes are interleaved after the first thread-specific note.
1682  */
1683 static int write_note_info(struct elf_note_info *info,
1684 			   struct coredump_params *cprm)
1685 {
1686 	bool first = 1;
1687 	struct elf_thread_core_info *t = info->thread;
1688 
1689 	do {
1690 		int i;
1691 
1692 		if (!writenote(&t->notes[0], cprm))
1693 			return 0;
1694 
1695 		if (first && !writenote(&info->psinfo, cprm))
1696 			return 0;
1697 		if (first && !writenote(&info->signote, cprm))
1698 			return 0;
1699 		if (first && !writenote(&info->auxv, cprm))
1700 			return 0;
1701 		if (first && info->files.data &&
1702 				!writenote(&info->files, cprm))
1703 			return 0;
1704 
1705 		for (i = 1; i < info->thread_notes; ++i)
1706 			if (t->notes[i].data &&
1707 			    !writenote(&t->notes[i], cprm))
1708 				return 0;
1709 
1710 		first = 0;
1711 		t = t->next;
1712 	} while (t);
1713 
1714 	return 1;
1715 }
1716 
1717 static void free_note_info(struct elf_note_info *info)
1718 {
1719 	struct elf_thread_core_info *threads = info->thread;
1720 	while (threads) {
1721 		unsigned int i;
1722 		struct elf_thread_core_info *t = threads;
1723 		threads = t->next;
1724 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1725 		for (i = 1; i < info->thread_notes; ++i)
1726 			kfree(t->notes[i].data);
1727 		kfree(t);
1728 	}
1729 	kfree(info->psinfo.data);
1730 	vfree(info->files.data);
1731 }
1732 
1733 #else
1734 
1735 /* Here is the structure in which status of each thread is captured. */
1736 struct elf_thread_status
1737 {
1738 	struct list_head list;
1739 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1740 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1741 	struct task_struct *thread;
1742 #ifdef ELF_CORE_COPY_XFPREGS
1743 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1744 #endif
1745 	struct memelfnote notes[3];
1746 	int num_notes;
1747 };
1748 
1749 /*
1750  * In order to add the specific thread information for the elf file format,
1751  * we need to keep a linked list of every threads pr_status and then create
1752  * a single section for them in the final core file.
1753  */
1754 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1755 {
1756 	int sz = 0;
1757 	struct task_struct *p = t->thread;
1758 	t->num_notes = 0;
1759 
1760 	fill_prstatus(&t->prstatus, p, signr);
1761 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1762 
1763 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1764 		  &(t->prstatus));
1765 	t->num_notes++;
1766 	sz += notesize(&t->notes[0]);
1767 
1768 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1769 								&t->fpu))) {
1770 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1771 			  &(t->fpu));
1772 		t->num_notes++;
1773 		sz += notesize(&t->notes[1]);
1774 	}
1775 
1776 #ifdef ELF_CORE_COPY_XFPREGS
1777 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1778 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1779 			  sizeof(t->xfpu), &t->xfpu);
1780 		t->num_notes++;
1781 		sz += notesize(&t->notes[2]);
1782 	}
1783 #endif
1784 	return sz;
1785 }
1786 
1787 struct elf_note_info {
1788 	struct memelfnote *notes;
1789 	struct memelfnote *notes_files;
1790 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1791 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1792 	struct list_head thread_list;
1793 	elf_fpregset_t *fpu;
1794 #ifdef ELF_CORE_COPY_XFPREGS
1795 	elf_fpxregset_t *xfpu;
1796 #endif
1797 	user_siginfo_t csigdata;
1798 	int thread_status_size;
1799 	int numnote;
1800 };
1801 
1802 static int elf_note_info_init(struct elf_note_info *info)
1803 {
1804 	memset(info, 0, sizeof(*info));
1805 	INIT_LIST_HEAD(&info->thread_list);
1806 
1807 	/* Allocate space for ELF notes */
1808 	info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1809 	if (!info->notes)
1810 		return 0;
1811 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1812 	if (!info->psinfo)
1813 		return 0;
1814 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1815 	if (!info->prstatus)
1816 		return 0;
1817 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1818 	if (!info->fpu)
1819 		return 0;
1820 #ifdef ELF_CORE_COPY_XFPREGS
1821 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1822 	if (!info->xfpu)
1823 		return 0;
1824 #endif
1825 	return 1;
1826 }
1827 
1828 static int fill_note_info(struct elfhdr *elf, int phdrs,
1829 			  struct elf_note_info *info,
1830 			  const siginfo_t *siginfo, struct pt_regs *regs)
1831 {
1832 	struct list_head *t;
1833 	struct core_thread *ct;
1834 	struct elf_thread_status *ets;
1835 
1836 	if (!elf_note_info_init(info))
1837 		return 0;
1838 
1839 	for (ct = current->mm->core_state->dumper.next;
1840 					ct; ct = ct->next) {
1841 		ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1842 		if (!ets)
1843 			return 0;
1844 
1845 		ets->thread = ct->task;
1846 		list_add(&ets->list, &info->thread_list);
1847 	}
1848 
1849 	list_for_each(t, &info->thread_list) {
1850 		int sz;
1851 
1852 		ets = list_entry(t, struct elf_thread_status, list);
1853 		sz = elf_dump_thread_status(siginfo->si_signo, ets);
1854 		info->thread_status_size += sz;
1855 	}
1856 	/* now collect the dump for the current */
1857 	memset(info->prstatus, 0, sizeof(*info->prstatus));
1858 	fill_prstatus(info->prstatus, current, siginfo->si_signo);
1859 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1860 
1861 	/* Set up header */
1862 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1863 
1864 	/*
1865 	 * Set up the notes in similar form to SVR4 core dumps made
1866 	 * with info from their /proc.
1867 	 */
1868 
1869 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1870 		  sizeof(*info->prstatus), info->prstatus);
1871 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
1872 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1873 		  sizeof(*info->psinfo), info->psinfo);
1874 
1875 	fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1876 	fill_auxv_note(info->notes + 3, current->mm);
1877 	info->numnote = 4;
1878 
1879 	if (fill_files_note(info->notes + info->numnote) == 0) {
1880 		info->notes_files = info->notes + info->numnote;
1881 		info->numnote++;
1882 	}
1883 
1884 	/* Try to dump the FPU. */
1885 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1886 							       info->fpu);
1887 	if (info->prstatus->pr_fpvalid)
1888 		fill_note(info->notes + info->numnote++,
1889 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1890 #ifdef ELF_CORE_COPY_XFPREGS
1891 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
1892 		fill_note(info->notes + info->numnote++,
1893 			  "LINUX", ELF_CORE_XFPREG_TYPE,
1894 			  sizeof(*info->xfpu), info->xfpu);
1895 #endif
1896 
1897 	return 1;
1898 }
1899 
1900 static size_t get_note_info_size(struct elf_note_info *info)
1901 {
1902 	int sz = 0;
1903 	int i;
1904 
1905 	for (i = 0; i < info->numnote; i++)
1906 		sz += notesize(info->notes + i);
1907 
1908 	sz += info->thread_status_size;
1909 
1910 	return sz;
1911 }
1912 
1913 static int write_note_info(struct elf_note_info *info,
1914 			   struct coredump_params *cprm)
1915 {
1916 	int i;
1917 	struct list_head *t;
1918 
1919 	for (i = 0; i < info->numnote; i++)
1920 		if (!writenote(info->notes + i, cprm))
1921 			return 0;
1922 
1923 	/* write out the thread status notes section */
1924 	list_for_each(t, &info->thread_list) {
1925 		struct elf_thread_status *tmp =
1926 				list_entry(t, struct elf_thread_status, list);
1927 
1928 		for (i = 0; i < tmp->num_notes; i++)
1929 			if (!writenote(&tmp->notes[i], cprm))
1930 				return 0;
1931 	}
1932 
1933 	return 1;
1934 }
1935 
1936 static void free_note_info(struct elf_note_info *info)
1937 {
1938 	while (!list_empty(&info->thread_list)) {
1939 		struct list_head *tmp = info->thread_list.next;
1940 		list_del(tmp);
1941 		kfree(list_entry(tmp, struct elf_thread_status, list));
1942 	}
1943 
1944 	/* Free data possibly allocated by fill_files_note(): */
1945 	if (info->notes_files)
1946 		vfree(info->notes_files->data);
1947 
1948 	kfree(info->prstatus);
1949 	kfree(info->psinfo);
1950 	kfree(info->notes);
1951 	kfree(info->fpu);
1952 #ifdef ELF_CORE_COPY_XFPREGS
1953 	kfree(info->xfpu);
1954 #endif
1955 }
1956 
1957 #endif
1958 
1959 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1960 					struct vm_area_struct *gate_vma)
1961 {
1962 	struct vm_area_struct *ret = tsk->mm->mmap;
1963 
1964 	if (ret)
1965 		return ret;
1966 	return gate_vma;
1967 }
1968 /*
1969  * Helper function for iterating across a vma list.  It ensures that the caller
1970  * will visit `gate_vma' prior to terminating the search.
1971  */
1972 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1973 					struct vm_area_struct *gate_vma)
1974 {
1975 	struct vm_area_struct *ret;
1976 
1977 	ret = this_vma->vm_next;
1978 	if (ret)
1979 		return ret;
1980 	if (this_vma == gate_vma)
1981 		return NULL;
1982 	return gate_vma;
1983 }
1984 
1985 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1986 			     elf_addr_t e_shoff, int segs)
1987 {
1988 	elf->e_shoff = e_shoff;
1989 	elf->e_shentsize = sizeof(*shdr4extnum);
1990 	elf->e_shnum = 1;
1991 	elf->e_shstrndx = SHN_UNDEF;
1992 
1993 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1994 
1995 	shdr4extnum->sh_type = SHT_NULL;
1996 	shdr4extnum->sh_size = elf->e_shnum;
1997 	shdr4extnum->sh_link = elf->e_shstrndx;
1998 	shdr4extnum->sh_info = segs;
1999 }
2000 
2001 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
2002 				     unsigned long mm_flags)
2003 {
2004 	struct vm_area_struct *vma;
2005 	size_t size = 0;
2006 
2007 	for (vma = first_vma(current, gate_vma); vma != NULL;
2008 	     vma = next_vma(vma, gate_vma))
2009 		size += vma_dump_size(vma, mm_flags);
2010 	return size;
2011 }
2012 
2013 /*
2014  * Actual dumper
2015  *
2016  * This is a two-pass process; first we find the offsets of the bits,
2017  * and then they are actually written out.  If we run out of core limit
2018  * we just truncate.
2019  */
2020 static int elf_core_dump(struct coredump_params *cprm)
2021 {
2022 	int has_dumped = 0;
2023 	mm_segment_t fs;
2024 	int segs;
2025 	struct vm_area_struct *vma, *gate_vma;
2026 	struct elfhdr *elf = NULL;
2027 	loff_t offset = 0, dataoff;
2028 	struct elf_note_info info = { };
2029 	struct elf_phdr *phdr4note = NULL;
2030 	struct elf_shdr *shdr4extnum = NULL;
2031 	Elf_Half e_phnum;
2032 	elf_addr_t e_shoff;
2033 
2034 	/*
2035 	 * We no longer stop all VM operations.
2036 	 *
2037 	 * This is because those proceses that could possibly change map_count
2038 	 * or the mmap / vma pages are now blocked in do_exit on current
2039 	 * finishing this core dump.
2040 	 *
2041 	 * Only ptrace can touch these memory addresses, but it doesn't change
2042 	 * the map_count or the pages allocated. So no possibility of crashing
2043 	 * exists while dumping the mm->vm_next areas to the core file.
2044 	 */
2045 
2046 	/* alloc memory for large data structures: too large to be on stack */
2047 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2048 	if (!elf)
2049 		goto out;
2050 	/*
2051 	 * The number of segs are recored into ELF header as 16bit value.
2052 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2053 	 */
2054 	segs = current->mm->map_count;
2055 	segs += elf_core_extra_phdrs();
2056 
2057 	gate_vma = get_gate_vma(current->mm);
2058 	if (gate_vma != NULL)
2059 		segs++;
2060 
2061 	/* for notes section */
2062 	segs++;
2063 
2064 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2065 	 * this, kernel supports extended numbering. Have a look at
2066 	 * include/linux/elf.h for further information. */
2067 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2068 
2069 	/*
2070 	 * Collect all the non-memory information about the process for the
2071 	 * notes.  This also sets up the file header.
2072 	 */
2073 	if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2074 		goto cleanup;
2075 
2076 	has_dumped = 1;
2077 
2078 	fs = get_fs();
2079 	set_fs(KERNEL_DS);
2080 
2081 	offset += sizeof(*elf);				/* Elf header */
2082 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2083 
2084 	/* Write notes phdr entry */
2085 	{
2086 		size_t sz = get_note_info_size(&info);
2087 
2088 		sz += elf_coredump_extra_notes_size();
2089 
2090 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2091 		if (!phdr4note)
2092 			goto end_coredump;
2093 
2094 		fill_elf_note_phdr(phdr4note, sz, offset);
2095 		offset += sz;
2096 	}
2097 
2098 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2099 
2100 	offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2101 	offset += elf_core_extra_data_size();
2102 	e_shoff = offset;
2103 
2104 	if (e_phnum == PN_XNUM) {
2105 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2106 		if (!shdr4extnum)
2107 			goto end_coredump;
2108 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2109 	}
2110 
2111 	offset = dataoff;
2112 
2113 	if (!dump_emit(cprm, elf, sizeof(*elf)))
2114 		goto end_coredump;
2115 
2116 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2117 		goto end_coredump;
2118 
2119 	/* Write program headers for segments dump */
2120 	for (vma = first_vma(current, gate_vma); vma != NULL;
2121 			vma = next_vma(vma, gate_vma)) {
2122 		struct elf_phdr phdr;
2123 
2124 		phdr.p_type = PT_LOAD;
2125 		phdr.p_offset = offset;
2126 		phdr.p_vaddr = vma->vm_start;
2127 		phdr.p_paddr = 0;
2128 		phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2129 		phdr.p_memsz = vma->vm_end - vma->vm_start;
2130 		offset += phdr.p_filesz;
2131 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2132 		if (vma->vm_flags & VM_WRITE)
2133 			phdr.p_flags |= PF_W;
2134 		if (vma->vm_flags & VM_EXEC)
2135 			phdr.p_flags |= PF_X;
2136 		phdr.p_align = ELF_EXEC_PAGESIZE;
2137 
2138 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2139 			goto end_coredump;
2140 	}
2141 
2142 	if (!elf_core_write_extra_phdrs(cprm, offset))
2143 		goto end_coredump;
2144 
2145  	/* write out the notes section */
2146 	if (!write_note_info(&info, cprm))
2147 		goto end_coredump;
2148 
2149 	if (elf_coredump_extra_notes_write(cprm))
2150 		goto end_coredump;
2151 
2152 	/* Align to page */
2153 	if (!dump_skip(cprm, dataoff - cprm->written))
2154 		goto end_coredump;
2155 
2156 	for (vma = first_vma(current, gate_vma); vma != NULL;
2157 			vma = next_vma(vma, gate_vma)) {
2158 		unsigned long addr;
2159 		unsigned long end;
2160 
2161 		end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2162 
2163 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2164 			struct page *page;
2165 			int stop;
2166 
2167 			page = get_dump_page(addr);
2168 			if (page) {
2169 				void *kaddr = kmap(page);
2170 				stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2171 				kunmap(page);
2172 				page_cache_release(page);
2173 			} else
2174 				stop = !dump_skip(cprm, PAGE_SIZE);
2175 			if (stop)
2176 				goto end_coredump;
2177 		}
2178 	}
2179 
2180 	if (!elf_core_write_extra_data(cprm))
2181 		goto end_coredump;
2182 
2183 	if (e_phnum == PN_XNUM) {
2184 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2185 			goto end_coredump;
2186 	}
2187 
2188 end_coredump:
2189 	set_fs(fs);
2190 
2191 cleanup:
2192 	free_note_info(&info);
2193 	kfree(shdr4extnum);
2194 	kfree(phdr4note);
2195 	kfree(elf);
2196 out:
2197 	return has_dumped;
2198 }
2199 
2200 #endif		/* CONFIG_ELF_CORE */
2201 
2202 static int __init init_elf_binfmt(void)
2203 {
2204 	register_binfmt(&elf_format);
2205 	return 0;
2206 }
2207 
2208 static void __exit exit_elf_binfmt(void)
2209 {
2210 	/* Remove the COFF and ELF loaders. */
2211 	unregister_binfmt(&elf_format);
2212 }
2213 
2214 core_initcall(init_elf_binfmt);
2215 module_exit(exit_elf_binfmt);
2216 MODULE_LICENSE("GPL");
2217