xref: /linux/fs/binfmt_elf.c (revision 3f07c0144132e4f59d88055ac8ff3e691a5fa2b8)
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/elf-randomize.h>
35 #include <linux/utsname.h>
36 #include <linux/coredump.h>
37 #include <linux/sched.h>
38 #include <linux/sched/coredump.h>
39 #include <linux/dax.h>
40 #include <linux/uaccess.h>
41 #include <asm/param.h>
42 #include <asm/page.h>
43 
44 #ifndef user_long_t
45 #define user_long_t long
46 #endif
47 #ifndef user_siginfo_t
48 #define user_siginfo_t siginfo_t
49 #endif
50 
51 static int load_elf_binary(struct linux_binprm *bprm);
52 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
53 				int, int, unsigned long);
54 
55 #ifdef CONFIG_USELIB
56 static int load_elf_library(struct file *);
57 #else
58 #define load_elf_library NULL
59 #endif
60 
61 /*
62  * If we don't support core dumping, then supply a NULL so we
63  * don't even try.
64  */
65 #ifdef CONFIG_ELF_CORE
66 static int elf_core_dump(struct coredump_params *cprm);
67 #else
68 #define elf_core_dump	NULL
69 #endif
70 
71 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
72 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
73 #else
74 #define ELF_MIN_ALIGN	PAGE_SIZE
75 #endif
76 
77 #ifndef ELF_CORE_EFLAGS
78 #define ELF_CORE_EFLAGS	0
79 #endif
80 
81 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
82 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
83 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
84 
85 static struct linux_binfmt elf_format = {
86 	.module		= THIS_MODULE,
87 	.load_binary	= load_elf_binary,
88 	.load_shlib	= load_elf_library,
89 	.core_dump	= elf_core_dump,
90 	.min_coredump	= ELF_EXEC_PAGESIZE,
91 };
92 
93 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
94 
95 static int set_brk(unsigned long start, unsigned long end, int prot)
96 {
97 	start = ELF_PAGEALIGN(start);
98 	end = ELF_PAGEALIGN(end);
99 	if (end > start) {
100 		/*
101 		 * Map the last of the bss segment.
102 		 * If the header is requesting these pages to be
103 		 * executable, honour that (ppc32 needs this).
104 		 */
105 		int error = vm_brk_flags(start, end - start,
106 				prot & PROT_EXEC ? VM_EXEC : 0);
107 		if (error)
108 			return error;
109 	}
110 	current->mm->start_brk = current->mm->brk = end;
111 	return 0;
112 }
113 
114 /* We need to explicitly zero any fractional pages
115    after the data section (i.e. bss).  This would
116    contain the junk from the file that should not
117    be in memory
118  */
119 static int padzero(unsigned long elf_bss)
120 {
121 	unsigned long nbyte;
122 
123 	nbyte = ELF_PAGEOFFSET(elf_bss);
124 	if (nbyte) {
125 		nbyte = ELF_MIN_ALIGN - nbyte;
126 		if (clear_user((void __user *) elf_bss, nbyte))
127 			return -EFAULT;
128 	}
129 	return 0;
130 }
131 
132 /* Let's use some macros to make this stack manipulation a little clearer */
133 #ifdef CONFIG_STACK_GROWSUP
134 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
135 #define STACK_ROUND(sp, items) \
136 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
137 #define STACK_ALLOC(sp, len) ({ \
138 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
139 	old_sp; })
140 #else
141 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
142 #define STACK_ROUND(sp, items) \
143 	(((unsigned long) (sp - items)) &~ 15UL)
144 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
145 #endif
146 
147 #ifndef ELF_BASE_PLATFORM
148 /*
149  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
150  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
151  * will be copied to the user stack in the same manner as AT_PLATFORM.
152  */
153 #define ELF_BASE_PLATFORM NULL
154 #endif
155 
156 static int
157 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
158 		unsigned long load_addr, unsigned long interp_load_addr)
159 {
160 	unsigned long p = bprm->p;
161 	int argc = bprm->argc;
162 	int envc = bprm->envc;
163 	elf_addr_t __user *argv;
164 	elf_addr_t __user *envp;
165 	elf_addr_t __user *sp;
166 	elf_addr_t __user *u_platform;
167 	elf_addr_t __user *u_base_platform;
168 	elf_addr_t __user *u_rand_bytes;
169 	const char *k_platform = ELF_PLATFORM;
170 	const char *k_base_platform = ELF_BASE_PLATFORM;
171 	unsigned char k_rand_bytes[16];
172 	int items;
173 	elf_addr_t *elf_info;
174 	int ei_index = 0;
175 	const struct cred *cred = current_cred();
176 	struct vm_area_struct *vma;
177 
178 	/*
179 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
180 	 * evictions by the processes running on the same package. One
181 	 * thing we can do is to shuffle the initial stack for them.
182 	 */
183 
184 	p = arch_align_stack(p);
185 
186 	/*
187 	 * If this architecture has a platform capability string, copy it
188 	 * to userspace.  In some cases (Sparc), this info is impossible
189 	 * for userspace to get any other way, in others (i386) it is
190 	 * merely difficult.
191 	 */
192 	u_platform = NULL;
193 	if (k_platform) {
194 		size_t len = strlen(k_platform) + 1;
195 
196 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
197 		if (__copy_to_user(u_platform, k_platform, len))
198 			return -EFAULT;
199 	}
200 
201 	/*
202 	 * If this architecture has a "base" platform capability
203 	 * string, copy it to userspace.
204 	 */
205 	u_base_platform = NULL;
206 	if (k_base_platform) {
207 		size_t len = strlen(k_base_platform) + 1;
208 
209 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
210 		if (__copy_to_user(u_base_platform, k_base_platform, len))
211 			return -EFAULT;
212 	}
213 
214 	/*
215 	 * Generate 16 random bytes for userspace PRNG seeding.
216 	 */
217 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
218 	u_rand_bytes = (elf_addr_t __user *)
219 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
220 	if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
221 		return -EFAULT;
222 
223 	/* Create the ELF interpreter info */
224 	elf_info = (elf_addr_t *)current->mm->saved_auxv;
225 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
226 #define NEW_AUX_ENT(id, val) \
227 	do { \
228 		elf_info[ei_index++] = id; \
229 		elf_info[ei_index++] = val; \
230 	} while (0)
231 
232 #ifdef ARCH_DLINFO
233 	/*
234 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
235 	 * AUXV.
236 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
237 	 * ARCH_DLINFO changes
238 	 */
239 	ARCH_DLINFO;
240 #endif
241 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
242 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
243 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
244 	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
245 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
246 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
247 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
248 	NEW_AUX_ENT(AT_FLAGS, 0);
249 	NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
250 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
251 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
252 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
253 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
254  	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
255 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
256 #ifdef ELF_HWCAP2
257 	NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
258 #endif
259 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
260 	if (k_platform) {
261 		NEW_AUX_ENT(AT_PLATFORM,
262 			    (elf_addr_t)(unsigned long)u_platform);
263 	}
264 	if (k_base_platform) {
265 		NEW_AUX_ENT(AT_BASE_PLATFORM,
266 			    (elf_addr_t)(unsigned long)u_base_platform);
267 	}
268 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
269 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
270 	}
271 #undef NEW_AUX_ENT
272 	/* AT_NULL is zero; clear the rest too */
273 	memset(&elf_info[ei_index], 0,
274 	       sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
275 
276 	/* And advance past the AT_NULL entry.  */
277 	ei_index += 2;
278 
279 	sp = STACK_ADD(p, ei_index);
280 
281 	items = (argc + 1) + (envc + 1) + 1;
282 	bprm->p = STACK_ROUND(sp, items);
283 
284 	/* Point sp at the lowest address on the stack */
285 #ifdef CONFIG_STACK_GROWSUP
286 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
287 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
288 #else
289 	sp = (elf_addr_t __user *)bprm->p;
290 #endif
291 
292 
293 	/*
294 	 * Grow the stack manually; some architectures have a limit on how
295 	 * far ahead a user-space access may be in order to grow the stack.
296 	 */
297 	vma = find_extend_vma(current->mm, bprm->p);
298 	if (!vma)
299 		return -EFAULT;
300 
301 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
302 	if (__put_user(argc, sp++))
303 		return -EFAULT;
304 	argv = sp;
305 	envp = argv + argc + 1;
306 
307 	/* Populate argv and envp */
308 	p = current->mm->arg_end = current->mm->arg_start;
309 	while (argc-- > 0) {
310 		size_t len;
311 		if (__put_user((elf_addr_t)p, argv++))
312 			return -EFAULT;
313 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
314 		if (!len || len > MAX_ARG_STRLEN)
315 			return -EINVAL;
316 		p += len;
317 	}
318 	if (__put_user(0, argv))
319 		return -EFAULT;
320 	current->mm->arg_end = current->mm->env_start = p;
321 	while (envc-- > 0) {
322 		size_t len;
323 		if (__put_user((elf_addr_t)p, envp++))
324 			return -EFAULT;
325 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
326 		if (!len || len > MAX_ARG_STRLEN)
327 			return -EINVAL;
328 		p += len;
329 	}
330 	if (__put_user(0, envp))
331 		return -EFAULT;
332 	current->mm->env_end = p;
333 
334 	/* Put the elf_info on the stack in the right place.  */
335 	sp = (elf_addr_t __user *)envp + 1;
336 	if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
337 		return -EFAULT;
338 	return 0;
339 }
340 
341 #ifndef elf_map
342 
343 static unsigned long elf_map(struct file *filep, unsigned long addr,
344 		struct elf_phdr *eppnt, int prot, int type,
345 		unsigned long total_size)
346 {
347 	unsigned long map_addr;
348 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
349 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
350 	addr = ELF_PAGESTART(addr);
351 	size = ELF_PAGEALIGN(size);
352 
353 	/* mmap() will return -EINVAL if given a zero size, but a
354 	 * segment with zero filesize is perfectly valid */
355 	if (!size)
356 		return addr;
357 
358 	/*
359 	* total_size is the size of the ELF (interpreter) image.
360 	* The _first_ mmap needs to know the full size, otherwise
361 	* randomization might put this image into an overlapping
362 	* position with the ELF binary image. (since size < total_size)
363 	* So we first map the 'big' image - and unmap the remainder at
364 	* the end. (which unmap is needed for ELF images with holes.)
365 	*/
366 	if (total_size) {
367 		total_size = ELF_PAGEALIGN(total_size);
368 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
369 		if (!BAD_ADDR(map_addr))
370 			vm_munmap(map_addr+size, total_size-size);
371 	} else
372 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
373 
374 	return(map_addr);
375 }
376 
377 #endif /* !elf_map */
378 
379 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
380 {
381 	int i, first_idx = -1, last_idx = -1;
382 
383 	for (i = 0; i < nr; i++) {
384 		if (cmds[i].p_type == PT_LOAD) {
385 			last_idx = i;
386 			if (first_idx == -1)
387 				first_idx = i;
388 		}
389 	}
390 	if (first_idx == -1)
391 		return 0;
392 
393 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
394 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
395 }
396 
397 /**
398  * load_elf_phdrs() - load ELF program headers
399  * @elf_ex:   ELF header of the binary whose program headers should be loaded
400  * @elf_file: the opened ELF binary file
401  *
402  * Loads ELF program headers from the binary file elf_file, which has the ELF
403  * header pointed to by elf_ex, into a newly allocated array. The caller is
404  * responsible for freeing the allocated data. Returns an ERR_PTR upon failure.
405  */
406 static struct elf_phdr *load_elf_phdrs(struct elfhdr *elf_ex,
407 				       struct file *elf_file)
408 {
409 	struct elf_phdr *elf_phdata = NULL;
410 	int retval, size, err = -1;
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 (elf_ex->e_phentsize != sizeof(struct elf_phdr))
417 		goto out;
418 
419 	/* Sanity check the number of program headers... */
420 	if (elf_ex->e_phnum < 1 ||
421 		elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
422 		goto out;
423 
424 	/* ...and their total size. */
425 	size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
426 	if (size > ELF_MIN_ALIGN)
427 		goto out;
428 
429 	elf_phdata = kmalloc(size, GFP_KERNEL);
430 	if (!elf_phdata)
431 		goto out;
432 
433 	/* Read in the program headers */
434 	retval = kernel_read(elf_file, elf_ex->e_phoff,
435 			     (char *)elf_phdata, size);
436 	if (retval != size) {
437 		err = (retval < 0) ? retval : -EIO;
438 		goto out;
439 	}
440 
441 	/* Success! */
442 	err = 0;
443 out:
444 	if (err) {
445 		kfree(elf_phdata);
446 		elf_phdata = NULL;
447 	}
448 	return elf_phdata;
449 }
450 
451 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
452 
453 /**
454  * struct arch_elf_state - arch-specific ELF loading state
455  *
456  * This structure is used to preserve architecture specific data during
457  * the loading of an ELF file, throughout the checking of architecture
458  * specific ELF headers & through to the point where the ELF load is
459  * known to be proceeding (ie. SET_PERSONALITY).
460  *
461  * This implementation is a dummy for architectures which require no
462  * specific state.
463  */
464 struct arch_elf_state {
465 };
466 
467 #define INIT_ARCH_ELF_STATE {}
468 
469 /**
470  * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
471  * @ehdr:	The main ELF header
472  * @phdr:	The program header to check
473  * @elf:	The open ELF file
474  * @is_interp:	True if the phdr is from the interpreter of the ELF being
475  *		loaded, else false.
476  * @state:	Architecture-specific state preserved throughout the process
477  *		of loading the ELF.
478  *
479  * Inspects the program header phdr to validate its correctness and/or
480  * suitability for the system. Called once per ELF program header in the
481  * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
482  * interpreter.
483  *
484  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
485  *         with that return code.
486  */
487 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
488 				   struct elf_phdr *phdr,
489 				   struct file *elf, bool is_interp,
490 				   struct arch_elf_state *state)
491 {
492 	/* Dummy implementation, always proceed */
493 	return 0;
494 }
495 
496 /**
497  * arch_check_elf() - check an ELF executable
498  * @ehdr:	The main ELF header
499  * @has_interp:	True if the ELF has an interpreter, else false.
500  * @interp_ehdr: The interpreter's ELF header
501  * @state:	Architecture-specific state preserved throughout the process
502  *		of loading the ELF.
503  *
504  * Provides a final opportunity for architecture code to reject the loading
505  * of the ELF & cause an exec syscall to return an error. This is called after
506  * all program headers to be checked by arch_elf_pt_proc have been.
507  *
508  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
509  *         with that return code.
510  */
511 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
512 				 struct elfhdr *interp_ehdr,
513 				 struct arch_elf_state *state)
514 {
515 	/* Dummy implementation, always proceed */
516 	return 0;
517 }
518 
519 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
520 
521 /* This is much more generalized than the library routine read function,
522    so we keep this separate.  Technically the library read function
523    is only provided so that we can read a.out libraries that have
524    an ELF header */
525 
526 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
527 		struct file *interpreter, unsigned long *interp_map_addr,
528 		unsigned long no_base, struct elf_phdr *interp_elf_phdata)
529 {
530 	struct elf_phdr *eppnt;
531 	unsigned long load_addr = 0;
532 	int load_addr_set = 0;
533 	unsigned long last_bss = 0, elf_bss = 0;
534 	int bss_prot = 0;
535 	unsigned long error = ~0UL;
536 	unsigned long total_size;
537 	int i;
538 
539 	/* First of all, some simple consistency checks */
540 	if (interp_elf_ex->e_type != ET_EXEC &&
541 	    interp_elf_ex->e_type != ET_DYN)
542 		goto out;
543 	if (!elf_check_arch(interp_elf_ex))
544 		goto out;
545 	if (!interpreter->f_op->mmap)
546 		goto out;
547 
548 	total_size = total_mapping_size(interp_elf_phdata,
549 					interp_elf_ex->e_phnum);
550 	if (!total_size) {
551 		error = -EINVAL;
552 		goto out;
553 	}
554 
555 	eppnt = interp_elf_phdata;
556 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
557 		if (eppnt->p_type == PT_LOAD) {
558 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
559 			int elf_prot = 0;
560 			unsigned long vaddr = 0;
561 			unsigned long k, map_addr;
562 
563 			if (eppnt->p_flags & PF_R)
564 		    		elf_prot = PROT_READ;
565 			if (eppnt->p_flags & PF_W)
566 				elf_prot |= PROT_WRITE;
567 			if (eppnt->p_flags & PF_X)
568 				elf_prot |= PROT_EXEC;
569 			vaddr = eppnt->p_vaddr;
570 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
571 				elf_type |= MAP_FIXED;
572 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
573 				load_addr = -vaddr;
574 
575 			map_addr = elf_map(interpreter, load_addr + vaddr,
576 					eppnt, elf_prot, elf_type, total_size);
577 			total_size = 0;
578 			if (!*interp_map_addr)
579 				*interp_map_addr = map_addr;
580 			error = map_addr;
581 			if (BAD_ADDR(map_addr))
582 				goto out;
583 
584 			if (!load_addr_set &&
585 			    interp_elf_ex->e_type == ET_DYN) {
586 				load_addr = map_addr - ELF_PAGESTART(vaddr);
587 				load_addr_set = 1;
588 			}
589 
590 			/*
591 			 * Check to see if the section's size will overflow the
592 			 * allowed task size. Note that p_filesz must always be
593 			 * <= p_memsize so it's only necessary to check p_memsz.
594 			 */
595 			k = load_addr + eppnt->p_vaddr;
596 			if (BAD_ADDR(k) ||
597 			    eppnt->p_filesz > eppnt->p_memsz ||
598 			    eppnt->p_memsz > TASK_SIZE ||
599 			    TASK_SIZE - eppnt->p_memsz < k) {
600 				error = -ENOMEM;
601 				goto out;
602 			}
603 
604 			/*
605 			 * Find the end of the file mapping for this phdr, and
606 			 * keep track of the largest address we see for this.
607 			 */
608 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
609 			if (k > elf_bss)
610 				elf_bss = k;
611 
612 			/*
613 			 * Do the same thing for the memory mapping - between
614 			 * elf_bss and last_bss is the bss section.
615 			 */
616 			k = load_addr + eppnt->p_vaddr + eppnt->p_memsz;
617 			if (k > last_bss) {
618 				last_bss = k;
619 				bss_prot = elf_prot;
620 			}
621 		}
622 	}
623 
624 	/*
625 	 * Now fill out the bss section: first pad the last page from
626 	 * the file up to the page boundary, and zero it from elf_bss
627 	 * up to the end of the page.
628 	 */
629 	if (padzero(elf_bss)) {
630 		error = -EFAULT;
631 		goto out;
632 	}
633 	/*
634 	 * Next, align both the file and mem bss up to the page size,
635 	 * since this is where elf_bss was just zeroed up to, and where
636 	 * last_bss will end after the vm_brk_flags() below.
637 	 */
638 	elf_bss = ELF_PAGEALIGN(elf_bss);
639 	last_bss = ELF_PAGEALIGN(last_bss);
640 	/* Finally, if there is still more bss to allocate, do it. */
641 	if (last_bss > elf_bss) {
642 		error = vm_brk_flags(elf_bss, last_bss - elf_bss,
643 				bss_prot & PROT_EXEC ? VM_EXEC : 0);
644 		if (error)
645 			goto out;
646 	}
647 
648 	error = load_addr;
649 out:
650 	return error;
651 }
652 
653 /*
654  * These are the functions used to load ELF style executables and shared
655  * libraries.  There is no binary dependent code anywhere else.
656  */
657 
658 #ifndef STACK_RND_MASK
659 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))	/* 8MB of VA */
660 #endif
661 
662 static unsigned long randomize_stack_top(unsigned long stack_top)
663 {
664 	unsigned long random_variable = 0;
665 
666 	if ((current->flags & PF_RANDOMIZE) &&
667 		!(current->personality & ADDR_NO_RANDOMIZE)) {
668 		random_variable = get_random_long();
669 		random_variable &= STACK_RND_MASK;
670 		random_variable <<= PAGE_SHIFT;
671 	}
672 #ifdef CONFIG_STACK_GROWSUP
673 	return PAGE_ALIGN(stack_top) + random_variable;
674 #else
675 	return PAGE_ALIGN(stack_top) - random_variable;
676 #endif
677 }
678 
679 static int load_elf_binary(struct linux_binprm *bprm)
680 {
681 	struct file *interpreter = NULL; /* to shut gcc up */
682  	unsigned long load_addr = 0, load_bias = 0;
683 	int load_addr_set = 0;
684 	char * elf_interpreter = NULL;
685 	unsigned long error;
686 	struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
687 	unsigned long elf_bss, elf_brk;
688 	int bss_prot = 0;
689 	int retval, i;
690 	unsigned long elf_entry;
691 	unsigned long interp_load_addr = 0;
692 	unsigned long start_code, end_code, start_data, end_data;
693 	unsigned long reloc_func_desc __maybe_unused = 0;
694 	int executable_stack = EXSTACK_DEFAULT;
695 	struct pt_regs *regs = current_pt_regs();
696 	struct {
697 		struct elfhdr elf_ex;
698 		struct elfhdr interp_elf_ex;
699 	} *loc;
700 	struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
701 
702 	loc = kmalloc(sizeof(*loc), GFP_KERNEL);
703 	if (!loc) {
704 		retval = -ENOMEM;
705 		goto out_ret;
706 	}
707 
708 	/* Get the exec-header */
709 	loc->elf_ex = *((struct elfhdr *)bprm->buf);
710 
711 	retval = -ENOEXEC;
712 	/* First of all, some simple consistency checks */
713 	if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
714 		goto out;
715 
716 	if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
717 		goto out;
718 	if (!elf_check_arch(&loc->elf_ex))
719 		goto out;
720 	if (!bprm->file->f_op->mmap)
721 		goto out;
722 
723 	elf_phdata = load_elf_phdrs(&loc->elf_ex, bprm->file);
724 	if (!elf_phdata)
725 		goto out;
726 
727 	elf_ppnt = elf_phdata;
728 	elf_bss = 0;
729 	elf_brk = 0;
730 
731 	start_code = ~0UL;
732 	end_code = 0;
733 	start_data = 0;
734 	end_data = 0;
735 
736 	for (i = 0; i < loc->elf_ex.e_phnum; i++) {
737 		if (elf_ppnt->p_type == PT_INTERP) {
738 			/* This is the program interpreter used for
739 			 * shared libraries - for now assume that this
740 			 * is an a.out format binary
741 			 */
742 			retval = -ENOEXEC;
743 			if (elf_ppnt->p_filesz > PATH_MAX ||
744 			    elf_ppnt->p_filesz < 2)
745 				goto out_free_ph;
746 
747 			retval = -ENOMEM;
748 			elf_interpreter = kmalloc(elf_ppnt->p_filesz,
749 						  GFP_KERNEL);
750 			if (!elf_interpreter)
751 				goto out_free_ph;
752 
753 			retval = kernel_read(bprm->file, elf_ppnt->p_offset,
754 					     elf_interpreter,
755 					     elf_ppnt->p_filesz);
756 			if (retval != elf_ppnt->p_filesz) {
757 				if (retval >= 0)
758 					retval = -EIO;
759 				goto out_free_interp;
760 			}
761 			/* make sure path is NULL terminated */
762 			retval = -ENOEXEC;
763 			if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
764 				goto out_free_interp;
765 
766 			interpreter = open_exec(elf_interpreter);
767 			retval = PTR_ERR(interpreter);
768 			if (IS_ERR(interpreter))
769 				goto out_free_interp;
770 
771 			/*
772 			 * If the binary is not readable then enforce
773 			 * mm->dumpable = 0 regardless of the interpreter's
774 			 * permissions.
775 			 */
776 			would_dump(bprm, interpreter);
777 
778 			/* Get the exec headers */
779 			retval = kernel_read(interpreter, 0,
780 					     (void *)&loc->interp_elf_ex,
781 					     sizeof(loc->interp_elf_ex));
782 			if (retval != sizeof(loc->interp_elf_ex)) {
783 				if (retval >= 0)
784 					retval = -EIO;
785 				goto out_free_dentry;
786 			}
787 
788 			break;
789 		}
790 		elf_ppnt++;
791 	}
792 
793 	elf_ppnt = elf_phdata;
794 	for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
795 		switch (elf_ppnt->p_type) {
796 		case PT_GNU_STACK:
797 			if (elf_ppnt->p_flags & PF_X)
798 				executable_stack = EXSTACK_ENABLE_X;
799 			else
800 				executable_stack = EXSTACK_DISABLE_X;
801 			break;
802 
803 		case PT_LOPROC ... PT_HIPROC:
804 			retval = arch_elf_pt_proc(&loc->elf_ex, elf_ppnt,
805 						  bprm->file, false,
806 						  &arch_state);
807 			if (retval)
808 				goto out_free_dentry;
809 			break;
810 		}
811 
812 	/* Some simple consistency checks for the interpreter */
813 	if (elf_interpreter) {
814 		retval = -ELIBBAD;
815 		/* Not an ELF interpreter */
816 		if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
817 			goto out_free_dentry;
818 		/* Verify the interpreter has a valid arch */
819 		if (!elf_check_arch(&loc->interp_elf_ex))
820 			goto out_free_dentry;
821 
822 		/* Load the interpreter program headers */
823 		interp_elf_phdata = load_elf_phdrs(&loc->interp_elf_ex,
824 						   interpreter);
825 		if (!interp_elf_phdata)
826 			goto out_free_dentry;
827 
828 		/* Pass PT_LOPROC..PT_HIPROC headers to arch code */
829 		elf_ppnt = interp_elf_phdata;
830 		for (i = 0; i < loc->interp_elf_ex.e_phnum; i++, elf_ppnt++)
831 			switch (elf_ppnt->p_type) {
832 			case PT_LOPROC ... PT_HIPROC:
833 				retval = arch_elf_pt_proc(&loc->interp_elf_ex,
834 							  elf_ppnt, interpreter,
835 							  true, &arch_state);
836 				if (retval)
837 					goto out_free_dentry;
838 				break;
839 			}
840 	}
841 
842 	/*
843 	 * Allow arch code to reject the ELF at this point, whilst it's
844 	 * still possible to return an error to the code that invoked
845 	 * the exec syscall.
846 	 */
847 	retval = arch_check_elf(&loc->elf_ex,
848 				!!interpreter, &loc->interp_elf_ex,
849 				&arch_state);
850 	if (retval)
851 		goto out_free_dentry;
852 
853 	/* Flush all traces of the currently running executable */
854 	retval = flush_old_exec(bprm);
855 	if (retval)
856 		goto out_free_dentry;
857 
858 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
859 	   may depend on the personality.  */
860 	SET_PERSONALITY2(loc->elf_ex, &arch_state);
861 	if (elf_read_implies_exec(loc->elf_ex, executable_stack))
862 		current->personality |= READ_IMPLIES_EXEC;
863 
864 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
865 		current->flags |= PF_RANDOMIZE;
866 
867 	setup_new_exec(bprm);
868 	install_exec_creds(bprm);
869 
870 	/* Do this so that we can load the interpreter, if need be.  We will
871 	   change some of these later */
872 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
873 				 executable_stack);
874 	if (retval < 0)
875 		goto out_free_dentry;
876 
877 	current->mm->start_stack = bprm->p;
878 
879 	/* Now we do a little grungy work by mmapping the ELF image into
880 	   the correct location in memory. */
881 	for(i = 0, elf_ppnt = elf_phdata;
882 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
883 		int elf_prot = 0, elf_flags;
884 		unsigned long k, vaddr;
885 		unsigned long total_size = 0;
886 
887 		if (elf_ppnt->p_type != PT_LOAD)
888 			continue;
889 
890 		if (unlikely (elf_brk > elf_bss)) {
891 			unsigned long nbyte;
892 
893 			/* There was a PT_LOAD segment with p_memsz > p_filesz
894 			   before this one. Map anonymous pages, if needed,
895 			   and clear the area.  */
896 			retval = set_brk(elf_bss + load_bias,
897 					 elf_brk + load_bias,
898 					 bss_prot);
899 			if (retval)
900 				goto out_free_dentry;
901 			nbyte = ELF_PAGEOFFSET(elf_bss);
902 			if (nbyte) {
903 				nbyte = ELF_MIN_ALIGN - nbyte;
904 				if (nbyte > elf_brk - elf_bss)
905 					nbyte = elf_brk - elf_bss;
906 				if (clear_user((void __user *)elf_bss +
907 							load_bias, nbyte)) {
908 					/*
909 					 * This bss-zeroing can fail if the ELF
910 					 * file specifies odd protections. So
911 					 * we don't check the return value
912 					 */
913 				}
914 			}
915 		}
916 
917 		if (elf_ppnt->p_flags & PF_R)
918 			elf_prot |= PROT_READ;
919 		if (elf_ppnt->p_flags & PF_W)
920 			elf_prot |= PROT_WRITE;
921 		if (elf_ppnt->p_flags & PF_X)
922 			elf_prot |= PROT_EXEC;
923 
924 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
925 
926 		vaddr = elf_ppnt->p_vaddr;
927 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
928 			elf_flags |= MAP_FIXED;
929 		} else if (loc->elf_ex.e_type == ET_DYN) {
930 			/* Try and get dynamic programs out of the way of the
931 			 * default mmap base, as well as whatever program they
932 			 * might try to exec.  This is because the brk will
933 			 * follow the loader, and is not movable.  */
934 			load_bias = ELF_ET_DYN_BASE - vaddr;
935 			if (current->flags & PF_RANDOMIZE)
936 				load_bias += arch_mmap_rnd();
937 			load_bias = ELF_PAGESTART(load_bias);
938 			total_size = total_mapping_size(elf_phdata,
939 							loc->elf_ex.e_phnum);
940 			if (!total_size) {
941 				retval = -EINVAL;
942 				goto out_free_dentry;
943 			}
944 		}
945 
946 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
947 				elf_prot, elf_flags, total_size);
948 		if (BAD_ADDR(error)) {
949 			retval = IS_ERR((void *)error) ?
950 				PTR_ERR((void*)error) : -EINVAL;
951 			goto out_free_dentry;
952 		}
953 
954 		if (!load_addr_set) {
955 			load_addr_set = 1;
956 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
957 			if (loc->elf_ex.e_type == ET_DYN) {
958 				load_bias += error -
959 				             ELF_PAGESTART(load_bias + vaddr);
960 				load_addr += load_bias;
961 				reloc_func_desc = load_bias;
962 			}
963 		}
964 		k = elf_ppnt->p_vaddr;
965 		if (k < start_code)
966 			start_code = k;
967 		if (start_data < k)
968 			start_data = k;
969 
970 		/*
971 		 * Check to see if the section's size will overflow the
972 		 * allowed task size. Note that p_filesz must always be
973 		 * <= p_memsz so it is only necessary to check p_memsz.
974 		 */
975 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
976 		    elf_ppnt->p_memsz > TASK_SIZE ||
977 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
978 			/* set_brk can never work. Avoid overflows. */
979 			retval = -EINVAL;
980 			goto out_free_dentry;
981 		}
982 
983 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
984 
985 		if (k > elf_bss)
986 			elf_bss = k;
987 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
988 			end_code = k;
989 		if (end_data < k)
990 			end_data = k;
991 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
992 		if (k > elf_brk) {
993 			bss_prot = elf_prot;
994 			elf_brk = k;
995 		}
996 	}
997 
998 	loc->elf_ex.e_entry += load_bias;
999 	elf_bss += load_bias;
1000 	elf_brk += load_bias;
1001 	start_code += load_bias;
1002 	end_code += load_bias;
1003 	start_data += load_bias;
1004 	end_data += load_bias;
1005 
1006 	/* Calling set_brk effectively mmaps the pages that we need
1007 	 * for the bss and break sections.  We must do this before
1008 	 * mapping in the interpreter, to make sure it doesn't wind
1009 	 * up getting placed where the bss needs to go.
1010 	 */
1011 	retval = set_brk(elf_bss, elf_brk, bss_prot);
1012 	if (retval)
1013 		goto out_free_dentry;
1014 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
1015 		retval = -EFAULT; /* Nobody gets to see this, but.. */
1016 		goto out_free_dentry;
1017 	}
1018 
1019 	if (elf_interpreter) {
1020 		unsigned long interp_map_addr = 0;
1021 
1022 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
1023 					    interpreter,
1024 					    &interp_map_addr,
1025 					    load_bias, interp_elf_phdata);
1026 		if (!IS_ERR((void *)elf_entry)) {
1027 			/*
1028 			 * load_elf_interp() returns relocation
1029 			 * adjustment
1030 			 */
1031 			interp_load_addr = elf_entry;
1032 			elf_entry += loc->interp_elf_ex.e_entry;
1033 		}
1034 		if (BAD_ADDR(elf_entry)) {
1035 			retval = IS_ERR((void *)elf_entry) ?
1036 					(int)elf_entry : -EINVAL;
1037 			goto out_free_dentry;
1038 		}
1039 		reloc_func_desc = interp_load_addr;
1040 
1041 		allow_write_access(interpreter);
1042 		fput(interpreter);
1043 		kfree(elf_interpreter);
1044 	} else {
1045 		elf_entry = loc->elf_ex.e_entry;
1046 		if (BAD_ADDR(elf_entry)) {
1047 			retval = -EINVAL;
1048 			goto out_free_dentry;
1049 		}
1050 	}
1051 
1052 	kfree(interp_elf_phdata);
1053 	kfree(elf_phdata);
1054 
1055 	set_binfmt(&elf_format);
1056 
1057 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1058 	retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
1059 	if (retval < 0)
1060 		goto out;
1061 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1062 
1063 	retval = create_elf_tables(bprm, &loc->elf_ex,
1064 			  load_addr, interp_load_addr);
1065 	if (retval < 0)
1066 		goto out;
1067 	/* N.B. passed_fileno might not be initialized? */
1068 	current->mm->end_code = end_code;
1069 	current->mm->start_code = start_code;
1070 	current->mm->start_data = start_data;
1071 	current->mm->end_data = end_data;
1072 	current->mm->start_stack = bprm->p;
1073 
1074 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1075 		current->mm->brk = current->mm->start_brk =
1076 			arch_randomize_brk(current->mm);
1077 #ifdef compat_brk_randomized
1078 		current->brk_randomized = 1;
1079 #endif
1080 	}
1081 
1082 	if (current->personality & MMAP_PAGE_ZERO) {
1083 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
1084 		   and some applications "depend" upon this behavior.
1085 		   Since we do not have the power to recompile these, we
1086 		   emulate the SVr4 behavior. Sigh. */
1087 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1088 				MAP_FIXED | MAP_PRIVATE, 0);
1089 	}
1090 
1091 #ifdef ELF_PLAT_INIT
1092 	/*
1093 	 * The ABI may specify that certain registers be set up in special
1094 	 * ways (on i386 %edx is the address of a DT_FINI function, for
1095 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
1096 	 * that the e_entry field is the address of the function descriptor
1097 	 * for the startup routine, rather than the address of the startup
1098 	 * routine itself.  This macro performs whatever initialization to
1099 	 * the regs structure is required as well as any relocations to the
1100 	 * function descriptor entries when executing dynamically links apps.
1101 	 */
1102 	ELF_PLAT_INIT(regs, reloc_func_desc);
1103 #endif
1104 
1105 	start_thread(regs, elf_entry, bprm->p);
1106 	retval = 0;
1107 out:
1108 	kfree(loc);
1109 out_ret:
1110 	return retval;
1111 
1112 	/* error cleanup */
1113 out_free_dentry:
1114 	kfree(interp_elf_phdata);
1115 	allow_write_access(interpreter);
1116 	if (interpreter)
1117 		fput(interpreter);
1118 out_free_interp:
1119 	kfree(elf_interpreter);
1120 out_free_ph:
1121 	kfree(elf_phdata);
1122 	goto out;
1123 }
1124 
1125 #ifdef CONFIG_USELIB
1126 /* This is really simpleminded and specialized - we are loading an
1127    a.out library that is given an ELF header. */
1128 static int load_elf_library(struct file *file)
1129 {
1130 	struct elf_phdr *elf_phdata;
1131 	struct elf_phdr *eppnt;
1132 	unsigned long elf_bss, bss, len;
1133 	int retval, error, i, j;
1134 	struct elfhdr elf_ex;
1135 
1136 	error = -ENOEXEC;
1137 	retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1138 	if (retval != sizeof(elf_ex))
1139 		goto out;
1140 
1141 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1142 		goto out;
1143 
1144 	/* First of all, some simple consistency checks */
1145 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1146 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1147 		goto out;
1148 
1149 	/* Now read in all of the header information */
1150 
1151 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1152 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1153 
1154 	error = -ENOMEM;
1155 	elf_phdata = kmalloc(j, GFP_KERNEL);
1156 	if (!elf_phdata)
1157 		goto out;
1158 
1159 	eppnt = elf_phdata;
1160 	error = -ENOEXEC;
1161 	retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1162 	if (retval != j)
1163 		goto out_free_ph;
1164 
1165 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1166 		if ((eppnt + i)->p_type == PT_LOAD)
1167 			j++;
1168 	if (j != 1)
1169 		goto out_free_ph;
1170 
1171 	while (eppnt->p_type != PT_LOAD)
1172 		eppnt++;
1173 
1174 	/* Now use mmap to map the library into memory. */
1175 	error = vm_mmap(file,
1176 			ELF_PAGESTART(eppnt->p_vaddr),
1177 			(eppnt->p_filesz +
1178 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1179 			PROT_READ | PROT_WRITE | PROT_EXEC,
1180 			MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1181 			(eppnt->p_offset -
1182 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1183 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1184 		goto out_free_ph;
1185 
1186 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1187 	if (padzero(elf_bss)) {
1188 		error = -EFAULT;
1189 		goto out_free_ph;
1190 	}
1191 
1192 	len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1193 			    ELF_MIN_ALIGN - 1);
1194 	bss = eppnt->p_memsz + eppnt->p_vaddr;
1195 	if (bss > len) {
1196 		error = vm_brk(len, bss - len);
1197 		if (error)
1198 			goto out_free_ph;
1199 	}
1200 	error = 0;
1201 
1202 out_free_ph:
1203 	kfree(elf_phdata);
1204 out:
1205 	return error;
1206 }
1207 #endif /* #ifdef CONFIG_USELIB */
1208 
1209 #ifdef CONFIG_ELF_CORE
1210 /*
1211  * ELF core dumper
1212  *
1213  * Modelled on fs/exec.c:aout_core_dump()
1214  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1215  */
1216 
1217 /*
1218  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1219  * that are useful for post-mortem analysis are included in every core dump.
1220  * In that way we ensure that the core dump is fully interpretable later
1221  * without matching up the same kernel and hardware config to see what PC values
1222  * meant. These special mappings include - vDSO, vsyscall, and other
1223  * architecture specific mappings
1224  */
1225 static bool always_dump_vma(struct vm_area_struct *vma)
1226 {
1227 	/* Any vsyscall mappings? */
1228 	if (vma == get_gate_vma(vma->vm_mm))
1229 		return true;
1230 
1231 	/*
1232 	 * Assume that all vmas with a .name op should always be dumped.
1233 	 * If this changes, a new vm_ops field can easily be added.
1234 	 */
1235 	if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1236 		return true;
1237 
1238 	/*
1239 	 * arch_vma_name() returns non-NULL for special architecture mappings,
1240 	 * such as vDSO sections.
1241 	 */
1242 	if (arch_vma_name(vma))
1243 		return true;
1244 
1245 	return false;
1246 }
1247 
1248 /*
1249  * Decide what to dump of a segment, part, all or none.
1250  */
1251 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1252 				   unsigned long mm_flags)
1253 {
1254 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1255 
1256 	/* always dump the vdso and vsyscall sections */
1257 	if (always_dump_vma(vma))
1258 		goto whole;
1259 
1260 	if (vma->vm_flags & VM_DONTDUMP)
1261 		return 0;
1262 
1263 	/* support for DAX */
1264 	if (vma_is_dax(vma)) {
1265 		if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1266 			goto whole;
1267 		if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1268 			goto whole;
1269 		return 0;
1270 	}
1271 
1272 	/* Hugetlb memory check */
1273 	if (vma->vm_flags & VM_HUGETLB) {
1274 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1275 			goto whole;
1276 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1277 			goto whole;
1278 		return 0;
1279 	}
1280 
1281 	/* Do not dump I/O mapped devices or special mappings */
1282 	if (vma->vm_flags & VM_IO)
1283 		return 0;
1284 
1285 	/* By default, dump shared memory if mapped from an anonymous file. */
1286 	if (vma->vm_flags & VM_SHARED) {
1287 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1288 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1289 			goto whole;
1290 		return 0;
1291 	}
1292 
1293 	/* Dump segments that have been written to.  */
1294 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1295 		goto whole;
1296 	if (vma->vm_file == NULL)
1297 		return 0;
1298 
1299 	if (FILTER(MAPPED_PRIVATE))
1300 		goto whole;
1301 
1302 	/*
1303 	 * If this looks like the beginning of a DSO or executable mapping,
1304 	 * check for an ELF header.  If we find one, dump the first page to
1305 	 * aid in determining what was mapped here.
1306 	 */
1307 	if (FILTER(ELF_HEADERS) &&
1308 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1309 		u32 __user *header = (u32 __user *) vma->vm_start;
1310 		u32 word;
1311 		mm_segment_t fs = get_fs();
1312 		/*
1313 		 * Doing it this way gets the constant folded by GCC.
1314 		 */
1315 		union {
1316 			u32 cmp;
1317 			char elfmag[SELFMAG];
1318 		} magic;
1319 		BUILD_BUG_ON(SELFMAG != sizeof word);
1320 		magic.elfmag[EI_MAG0] = ELFMAG0;
1321 		magic.elfmag[EI_MAG1] = ELFMAG1;
1322 		magic.elfmag[EI_MAG2] = ELFMAG2;
1323 		magic.elfmag[EI_MAG3] = ELFMAG3;
1324 		/*
1325 		 * Switch to the user "segment" for get_user(),
1326 		 * then put back what elf_core_dump() had in place.
1327 		 */
1328 		set_fs(USER_DS);
1329 		if (unlikely(get_user(word, header)))
1330 			word = 0;
1331 		set_fs(fs);
1332 		if (word == magic.cmp)
1333 			return PAGE_SIZE;
1334 	}
1335 
1336 #undef	FILTER
1337 
1338 	return 0;
1339 
1340 whole:
1341 	return vma->vm_end - vma->vm_start;
1342 }
1343 
1344 /* An ELF note in memory */
1345 struct memelfnote
1346 {
1347 	const char *name;
1348 	int type;
1349 	unsigned int datasz;
1350 	void *data;
1351 };
1352 
1353 static int notesize(struct memelfnote *en)
1354 {
1355 	int sz;
1356 
1357 	sz = sizeof(struct elf_note);
1358 	sz += roundup(strlen(en->name) + 1, 4);
1359 	sz += roundup(en->datasz, 4);
1360 
1361 	return sz;
1362 }
1363 
1364 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1365 {
1366 	struct elf_note en;
1367 	en.n_namesz = strlen(men->name) + 1;
1368 	en.n_descsz = men->datasz;
1369 	en.n_type = men->type;
1370 
1371 	return dump_emit(cprm, &en, sizeof(en)) &&
1372 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1373 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1374 }
1375 
1376 static void fill_elf_header(struct elfhdr *elf, int segs,
1377 			    u16 machine, u32 flags)
1378 {
1379 	memset(elf, 0, sizeof(*elf));
1380 
1381 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1382 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1383 	elf->e_ident[EI_DATA] = ELF_DATA;
1384 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1385 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1386 
1387 	elf->e_type = ET_CORE;
1388 	elf->e_machine = machine;
1389 	elf->e_version = EV_CURRENT;
1390 	elf->e_phoff = sizeof(struct elfhdr);
1391 	elf->e_flags = flags;
1392 	elf->e_ehsize = sizeof(struct elfhdr);
1393 	elf->e_phentsize = sizeof(struct elf_phdr);
1394 	elf->e_phnum = segs;
1395 
1396 	return;
1397 }
1398 
1399 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1400 {
1401 	phdr->p_type = PT_NOTE;
1402 	phdr->p_offset = offset;
1403 	phdr->p_vaddr = 0;
1404 	phdr->p_paddr = 0;
1405 	phdr->p_filesz = sz;
1406 	phdr->p_memsz = 0;
1407 	phdr->p_flags = 0;
1408 	phdr->p_align = 0;
1409 	return;
1410 }
1411 
1412 static void fill_note(struct memelfnote *note, const char *name, int type,
1413 		unsigned int sz, void *data)
1414 {
1415 	note->name = name;
1416 	note->type = type;
1417 	note->datasz = sz;
1418 	note->data = data;
1419 	return;
1420 }
1421 
1422 /*
1423  * fill up all the fields in prstatus from the given task struct, except
1424  * registers which need to be filled up separately.
1425  */
1426 static void fill_prstatus(struct elf_prstatus *prstatus,
1427 		struct task_struct *p, long signr)
1428 {
1429 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1430 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1431 	prstatus->pr_sighold = p->blocked.sig[0];
1432 	rcu_read_lock();
1433 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1434 	rcu_read_unlock();
1435 	prstatus->pr_pid = task_pid_vnr(p);
1436 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1437 	prstatus->pr_sid = task_session_vnr(p);
1438 	if (thread_group_leader(p)) {
1439 		struct task_cputime cputime;
1440 
1441 		/*
1442 		 * This is the record for the group leader.  It shows the
1443 		 * group-wide total, not its individual thread total.
1444 		 */
1445 		thread_group_cputime(p, &cputime);
1446 		prstatus->pr_utime = ns_to_timeval(cputime.utime);
1447 		prstatus->pr_stime = ns_to_timeval(cputime.stime);
1448 	} else {
1449 		u64 utime, stime;
1450 
1451 		task_cputime(p, &utime, &stime);
1452 		prstatus->pr_utime = ns_to_timeval(utime);
1453 		prstatus->pr_stime = ns_to_timeval(stime);
1454 	}
1455 
1456 	prstatus->pr_cutime = ns_to_timeval(p->signal->cutime);
1457 	prstatus->pr_cstime = ns_to_timeval(p->signal->cstime);
1458 }
1459 
1460 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1461 		       struct mm_struct *mm)
1462 {
1463 	const struct cred *cred;
1464 	unsigned int i, len;
1465 
1466 	/* first copy the parameters from user space */
1467 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1468 
1469 	len = mm->arg_end - mm->arg_start;
1470 	if (len >= ELF_PRARGSZ)
1471 		len = ELF_PRARGSZ-1;
1472 	if (copy_from_user(&psinfo->pr_psargs,
1473 		           (const char __user *)mm->arg_start, len))
1474 		return -EFAULT;
1475 	for(i = 0; i < len; i++)
1476 		if (psinfo->pr_psargs[i] == 0)
1477 			psinfo->pr_psargs[i] = ' ';
1478 	psinfo->pr_psargs[len] = 0;
1479 
1480 	rcu_read_lock();
1481 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1482 	rcu_read_unlock();
1483 	psinfo->pr_pid = task_pid_vnr(p);
1484 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1485 	psinfo->pr_sid = task_session_vnr(p);
1486 
1487 	i = p->state ? ffz(~p->state) + 1 : 0;
1488 	psinfo->pr_state = i;
1489 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1490 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1491 	psinfo->pr_nice = task_nice(p);
1492 	psinfo->pr_flag = p->flags;
1493 	rcu_read_lock();
1494 	cred = __task_cred(p);
1495 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1496 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1497 	rcu_read_unlock();
1498 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1499 
1500 	return 0;
1501 }
1502 
1503 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1504 {
1505 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1506 	int i = 0;
1507 	do
1508 		i += 2;
1509 	while (auxv[i - 2] != AT_NULL);
1510 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1511 }
1512 
1513 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1514 		const siginfo_t *siginfo)
1515 {
1516 	mm_segment_t old_fs = get_fs();
1517 	set_fs(KERNEL_DS);
1518 	copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1519 	set_fs(old_fs);
1520 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1521 }
1522 
1523 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1524 /*
1525  * Format of NT_FILE note:
1526  *
1527  * long count     -- how many files are mapped
1528  * long page_size -- units for file_ofs
1529  * array of [COUNT] elements of
1530  *   long start
1531  *   long end
1532  *   long file_ofs
1533  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1534  */
1535 static int fill_files_note(struct memelfnote *note)
1536 {
1537 	struct vm_area_struct *vma;
1538 	unsigned count, size, names_ofs, remaining, n;
1539 	user_long_t *data;
1540 	user_long_t *start_end_ofs;
1541 	char *name_base, *name_curpos;
1542 
1543 	/* *Estimated* file count and total data size needed */
1544 	count = current->mm->map_count;
1545 	size = count * 64;
1546 
1547 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1548  alloc:
1549 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1550 		return -EINVAL;
1551 	size = round_up(size, PAGE_SIZE);
1552 	data = vmalloc(size);
1553 	if (!data)
1554 		return -ENOMEM;
1555 
1556 	start_end_ofs = data + 2;
1557 	name_base = name_curpos = ((char *)data) + names_ofs;
1558 	remaining = size - names_ofs;
1559 	count = 0;
1560 	for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1561 		struct file *file;
1562 		const char *filename;
1563 
1564 		file = vma->vm_file;
1565 		if (!file)
1566 			continue;
1567 		filename = file_path(file, name_curpos, remaining);
1568 		if (IS_ERR(filename)) {
1569 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1570 				vfree(data);
1571 				size = size * 5 / 4;
1572 				goto alloc;
1573 			}
1574 			continue;
1575 		}
1576 
1577 		/* file_path() fills at the end, move name down */
1578 		/* n = strlen(filename) + 1: */
1579 		n = (name_curpos + remaining) - filename;
1580 		remaining = filename - name_curpos;
1581 		memmove(name_curpos, filename, n);
1582 		name_curpos += n;
1583 
1584 		*start_end_ofs++ = vma->vm_start;
1585 		*start_end_ofs++ = vma->vm_end;
1586 		*start_end_ofs++ = vma->vm_pgoff;
1587 		count++;
1588 	}
1589 
1590 	/* Now we know exact count of files, can store it */
1591 	data[0] = count;
1592 	data[1] = PAGE_SIZE;
1593 	/*
1594 	 * Count usually is less than current->mm->map_count,
1595 	 * we need to move filenames down.
1596 	 */
1597 	n = current->mm->map_count - count;
1598 	if (n != 0) {
1599 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1600 		memmove(name_base - shift_bytes, name_base,
1601 			name_curpos - name_base);
1602 		name_curpos -= shift_bytes;
1603 	}
1604 
1605 	size = name_curpos - (char *)data;
1606 	fill_note(note, "CORE", NT_FILE, size, data);
1607 	return 0;
1608 }
1609 
1610 #ifdef CORE_DUMP_USE_REGSET
1611 #include <linux/regset.h>
1612 
1613 struct elf_thread_core_info {
1614 	struct elf_thread_core_info *next;
1615 	struct task_struct *task;
1616 	struct elf_prstatus prstatus;
1617 	struct memelfnote notes[0];
1618 };
1619 
1620 struct elf_note_info {
1621 	struct elf_thread_core_info *thread;
1622 	struct memelfnote psinfo;
1623 	struct memelfnote signote;
1624 	struct memelfnote auxv;
1625 	struct memelfnote files;
1626 	user_siginfo_t csigdata;
1627 	size_t size;
1628 	int thread_notes;
1629 };
1630 
1631 /*
1632  * When a regset has a writeback hook, we call it on each thread before
1633  * dumping user memory.  On register window machines, this makes sure the
1634  * user memory backing the register data is up to date before we read it.
1635  */
1636 static void do_thread_regset_writeback(struct task_struct *task,
1637 				       const struct user_regset *regset)
1638 {
1639 	if (regset->writeback)
1640 		regset->writeback(task, regset, 1);
1641 }
1642 
1643 #ifndef PRSTATUS_SIZE
1644 #define PRSTATUS_SIZE(S, R) sizeof(S)
1645 #endif
1646 
1647 #ifndef SET_PR_FPVALID
1648 #define SET_PR_FPVALID(S, V, R) ((S)->pr_fpvalid = (V))
1649 #endif
1650 
1651 static int fill_thread_core_info(struct elf_thread_core_info *t,
1652 				 const struct user_regset_view *view,
1653 				 long signr, size_t *total)
1654 {
1655 	unsigned int i;
1656 	unsigned int regset_size = view->regsets[0].n * view->regsets[0].size;
1657 
1658 	/*
1659 	 * NT_PRSTATUS is the one special case, because the regset data
1660 	 * goes into the pr_reg field inside the note contents, rather
1661 	 * than being the whole note contents.  We fill the reset in here.
1662 	 * We assume that regset 0 is NT_PRSTATUS.
1663 	 */
1664 	fill_prstatus(&t->prstatus, t->task, signr);
1665 	(void) view->regsets[0].get(t->task, &view->regsets[0], 0, regset_size,
1666 				    &t->prstatus.pr_reg, NULL);
1667 
1668 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1669 		  PRSTATUS_SIZE(t->prstatus, regset_size), &t->prstatus);
1670 	*total += notesize(&t->notes[0]);
1671 
1672 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1673 
1674 	/*
1675 	 * Each other regset might generate a note too.  For each regset
1676 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1677 	 * all zero and we'll know to skip writing it later.
1678 	 */
1679 	for (i = 1; i < view->n; ++i) {
1680 		const struct user_regset *regset = &view->regsets[i];
1681 		do_thread_regset_writeback(t->task, regset);
1682 		if (regset->core_note_type && regset->get &&
1683 		    (!regset->active || regset->active(t->task, regset))) {
1684 			int ret;
1685 			size_t size = regset->n * regset->size;
1686 			void *data = kmalloc(size, GFP_KERNEL);
1687 			if (unlikely(!data))
1688 				return 0;
1689 			ret = regset->get(t->task, regset,
1690 					  0, size, data, NULL);
1691 			if (unlikely(ret))
1692 				kfree(data);
1693 			else {
1694 				if (regset->core_note_type != NT_PRFPREG)
1695 					fill_note(&t->notes[i], "LINUX",
1696 						  regset->core_note_type,
1697 						  size, data);
1698 				else {
1699 					SET_PR_FPVALID(&t->prstatus,
1700 							1, regset_size);
1701 					fill_note(&t->notes[i], "CORE",
1702 						  NT_PRFPREG, size, data);
1703 				}
1704 				*total += notesize(&t->notes[i]);
1705 			}
1706 		}
1707 	}
1708 
1709 	return 1;
1710 }
1711 
1712 static int fill_note_info(struct elfhdr *elf, int phdrs,
1713 			  struct elf_note_info *info,
1714 			  const siginfo_t *siginfo, struct pt_regs *regs)
1715 {
1716 	struct task_struct *dump_task = current;
1717 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1718 	struct elf_thread_core_info *t;
1719 	struct elf_prpsinfo *psinfo;
1720 	struct core_thread *ct;
1721 	unsigned int i;
1722 
1723 	info->size = 0;
1724 	info->thread = NULL;
1725 
1726 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1727 	if (psinfo == NULL) {
1728 		info->psinfo.data = NULL; /* So we don't free this wrongly */
1729 		return 0;
1730 	}
1731 
1732 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1733 
1734 	/*
1735 	 * Figure out how many notes we're going to need for each thread.
1736 	 */
1737 	info->thread_notes = 0;
1738 	for (i = 0; i < view->n; ++i)
1739 		if (view->regsets[i].core_note_type != 0)
1740 			++info->thread_notes;
1741 
1742 	/*
1743 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1744 	 * since it is our one special case.
1745 	 */
1746 	if (unlikely(info->thread_notes == 0) ||
1747 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1748 		WARN_ON(1);
1749 		return 0;
1750 	}
1751 
1752 	/*
1753 	 * Initialize the ELF file header.
1754 	 */
1755 	fill_elf_header(elf, phdrs,
1756 			view->e_machine, view->e_flags);
1757 
1758 	/*
1759 	 * Allocate a structure for each thread.
1760 	 */
1761 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1762 		t = kzalloc(offsetof(struct elf_thread_core_info,
1763 				     notes[info->thread_notes]),
1764 			    GFP_KERNEL);
1765 		if (unlikely(!t))
1766 			return 0;
1767 
1768 		t->task = ct->task;
1769 		if (ct->task == dump_task || !info->thread) {
1770 			t->next = info->thread;
1771 			info->thread = t;
1772 		} else {
1773 			/*
1774 			 * Make sure to keep the original task at
1775 			 * the head of the list.
1776 			 */
1777 			t->next = info->thread->next;
1778 			info->thread->next = t;
1779 		}
1780 	}
1781 
1782 	/*
1783 	 * Now fill in each thread's information.
1784 	 */
1785 	for (t = info->thread; t != NULL; t = t->next)
1786 		if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1787 			return 0;
1788 
1789 	/*
1790 	 * Fill in the two process-wide notes.
1791 	 */
1792 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1793 	info->size += notesize(&info->psinfo);
1794 
1795 	fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1796 	info->size += notesize(&info->signote);
1797 
1798 	fill_auxv_note(&info->auxv, current->mm);
1799 	info->size += notesize(&info->auxv);
1800 
1801 	if (fill_files_note(&info->files) == 0)
1802 		info->size += notesize(&info->files);
1803 
1804 	return 1;
1805 }
1806 
1807 static size_t get_note_info_size(struct elf_note_info *info)
1808 {
1809 	return info->size;
1810 }
1811 
1812 /*
1813  * Write all the notes for each thread.  When writing the first thread, the
1814  * process-wide notes are interleaved after the first thread-specific note.
1815  */
1816 static int write_note_info(struct elf_note_info *info,
1817 			   struct coredump_params *cprm)
1818 {
1819 	bool first = true;
1820 	struct elf_thread_core_info *t = info->thread;
1821 
1822 	do {
1823 		int i;
1824 
1825 		if (!writenote(&t->notes[0], cprm))
1826 			return 0;
1827 
1828 		if (first && !writenote(&info->psinfo, cprm))
1829 			return 0;
1830 		if (first && !writenote(&info->signote, cprm))
1831 			return 0;
1832 		if (first && !writenote(&info->auxv, cprm))
1833 			return 0;
1834 		if (first && info->files.data &&
1835 				!writenote(&info->files, cprm))
1836 			return 0;
1837 
1838 		for (i = 1; i < info->thread_notes; ++i)
1839 			if (t->notes[i].data &&
1840 			    !writenote(&t->notes[i], cprm))
1841 				return 0;
1842 
1843 		first = false;
1844 		t = t->next;
1845 	} while (t);
1846 
1847 	return 1;
1848 }
1849 
1850 static void free_note_info(struct elf_note_info *info)
1851 {
1852 	struct elf_thread_core_info *threads = info->thread;
1853 	while (threads) {
1854 		unsigned int i;
1855 		struct elf_thread_core_info *t = threads;
1856 		threads = t->next;
1857 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1858 		for (i = 1; i < info->thread_notes; ++i)
1859 			kfree(t->notes[i].data);
1860 		kfree(t);
1861 	}
1862 	kfree(info->psinfo.data);
1863 	vfree(info->files.data);
1864 }
1865 
1866 #else
1867 
1868 /* Here is the structure in which status of each thread is captured. */
1869 struct elf_thread_status
1870 {
1871 	struct list_head list;
1872 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1873 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1874 	struct task_struct *thread;
1875 #ifdef ELF_CORE_COPY_XFPREGS
1876 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1877 #endif
1878 	struct memelfnote notes[3];
1879 	int num_notes;
1880 };
1881 
1882 /*
1883  * In order to add the specific thread information for the elf file format,
1884  * we need to keep a linked list of every threads pr_status and then create
1885  * a single section for them in the final core file.
1886  */
1887 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1888 {
1889 	int sz = 0;
1890 	struct task_struct *p = t->thread;
1891 	t->num_notes = 0;
1892 
1893 	fill_prstatus(&t->prstatus, p, signr);
1894 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1895 
1896 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1897 		  &(t->prstatus));
1898 	t->num_notes++;
1899 	sz += notesize(&t->notes[0]);
1900 
1901 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1902 								&t->fpu))) {
1903 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1904 			  &(t->fpu));
1905 		t->num_notes++;
1906 		sz += notesize(&t->notes[1]);
1907 	}
1908 
1909 #ifdef ELF_CORE_COPY_XFPREGS
1910 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1911 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1912 			  sizeof(t->xfpu), &t->xfpu);
1913 		t->num_notes++;
1914 		sz += notesize(&t->notes[2]);
1915 	}
1916 #endif
1917 	return sz;
1918 }
1919 
1920 struct elf_note_info {
1921 	struct memelfnote *notes;
1922 	struct memelfnote *notes_files;
1923 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1924 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1925 	struct list_head thread_list;
1926 	elf_fpregset_t *fpu;
1927 #ifdef ELF_CORE_COPY_XFPREGS
1928 	elf_fpxregset_t *xfpu;
1929 #endif
1930 	user_siginfo_t csigdata;
1931 	int thread_status_size;
1932 	int numnote;
1933 };
1934 
1935 static int elf_note_info_init(struct elf_note_info *info)
1936 {
1937 	memset(info, 0, sizeof(*info));
1938 	INIT_LIST_HEAD(&info->thread_list);
1939 
1940 	/* Allocate space for ELF notes */
1941 	info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1942 	if (!info->notes)
1943 		return 0;
1944 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1945 	if (!info->psinfo)
1946 		return 0;
1947 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1948 	if (!info->prstatus)
1949 		return 0;
1950 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1951 	if (!info->fpu)
1952 		return 0;
1953 #ifdef ELF_CORE_COPY_XFPREGS
1954 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1955 	if (!info->xfpu)
1956 		return 0;
1957 #endif
1958 	return 1;
1959 }
1960 
1961 static int fill_note_info(struct elfhdr *elf, int phdrs,
1962 			  struct elf_note_info *info,
1963 			  const siginfo_t *siginfo, struct pt_regs *regs)
1964 {
1965 	struct list_head *t;
1966 	struct core_thread *ct;
1967 	struct elf_thread_status *ets;
1968 
1969 	if (!elf_note_info_init(info))
1970 		return 0;
1971 
1972 	for (ct = current->mm->core_state->dumper.next;
1973 					ct; ct = ct->next) {
1974 		ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1975 		if (!ets)
1976 			return 0;
1977 
1978 		ets->thread = ct->task;
1979 		list_add(&ets->list, &info->thread_list);
1980 	}
1981 
1982 	list_for_each(t, &info->thread_list) {
1983 		int sz;
1984 
1985 		ets = list_entry(t, struct elf_thread_status, list);
1986 		sz = elf_dump_thread_status(siginfo->si_signo, ets);
1987 		info->thread_status_size += sz;
1988 	}
1989 	/* now collect the dump for the current */
1990 	memset(info->prstatus, 0, sizeof(*info->prstatus));
1991 	fill_prstatus(info->prstatus, current, siginfo->si_signo);
1992 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1993 
1994 	/* Set up header */
1995 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1996 
1997 	/*
1998 	 * Set up the notes in similar form to SVR4 core dumps made
1999 	 * with info from their /proc.
2000 	 */
2001 
2002 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
2003 		  sizeof(*info->prstatus), info->prstatus);
2004 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
2005 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
2006 		  sizeof(*info->psinfo), info->psinfo);
2007 
2008 	fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
2009 	fill_auxv_note(info->notes + 3, current->mm);
2010 	info->numnote = 4;
2011 
2012 	if (fill_files_note(info->notes + info->numnote) == 0) {
2013 		info->notes_files = info->notes + info->numnote;
2014 		info->numnote++;
2015 	}
2016 
2017 	/* Try to dump the FPU. */
2018 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2019 							       info->fpu);
2020 	if (info->prstatus->pr_fpvalid)
2021 		fill_note(info->notes + info->numnote++,
2022 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2023 #ifdef ELF_CORE_COPY_XFPREGS
2024 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
2025 		fill_note(info->notes + info->numnote++,
2026 			  "LINUX", ELF_CORE_XFPREG_TYPE,
2027 			  sizeof(*info->xfpu), info->xfpu);
2028 #endif
2029 
2030 	return 1;
2031 }
2032 
2033 static size_t get_note_info_size(struct elf_note_info *info)
2034 {
2035 	int sz = 0;
2036 	int i;
2037 
2038 	for (i = 0; i < info->numnote; i++)
2039 		sz += notesize(info->notes + i);
2040 
2041 	sz += info->thread_status_size;
2042 
2043 	return sz;
2044 }
2045 
2046 static int write_note_info(struct elf_note_info *info,
2047 			   struct coredump_params *cprm)
2048 {
2049 	int i;
2050 	struct list_head *t;
2051 
2052 	for (i = 0; i < info->numnote; i++)
2053 		if (!writenote(info->notes + i, cprm))
2054 			return 0;
2055 
2056 	/* write out the thread status notes section */
2057 	list_for_each(t, &info->thread_list) {
2058 		struct elf_thread_status *tmp =
2059 				list_entry(t, struct elf_thread_status, list);
2060 
2061 		for (i = 0; i < tmp->num_notes; i++)
2062 			if (!writenote(&tmp->notes[i], cprm))
2063 				return 0;
2064 	}
2065 
2066 	return 1;
2067 }
2068 
2069 static void free_note_info(struct elf_note_info *info)
2070 {
2071 	while (!list_empty(&info->thread_list)) {
2072 		struct list_head *tmp = info->thread_list.next;
2073 		list_del(tmp);
2074 		kfree(list_entry(tmp, struct elf_thread_status, list));
2075 	}
2076 
2077 	/* Free data possibly allocated by fill_files_note(): */
2078 	if (info->notes_files)
2079 		vfree(info->notes_files->data);
2080 
2081 	kfree(info->prstatus);
2082 	kfree(info->psinfo);
2083 	kfree(info->notes);
2084 	kfree(info->fpu);
2085 #ifdef ELF_CORE_COPY_XFPREGS
2086 	kfree(info->xfpu);
2087 #endif
2088 }
2089 
2090 #endif
2091 
2092 static struct vm_area_struct *first_vma(struct task_struct *tsk,
2093 					struct vm_area_struct *gate_vma)
2094 {
2095 	struct vm_area_struct *ret = tsk->mm->mmap;
2096 
2097 	if (ret)
2098 		return ret;
2099 	return gate_vma;
2100 }
2101 /*
2102  * Helper function for iterating across a vma list.  It ensures that the caller
2103  * will visit `gate_vma' prior to terminating the search.
2104  */
2105 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2106 					struct vm_area_struct *gate_vma)
2107 {
2108 	struct vm_area_struct *ret;
2109 
2110 	ret = this_vma->vm_next;
2111 	if (ret)
2112 		return ret;
2113 	if (this_vma == gate_vma)
2114 		return NULL;
2115 	return gate_vma;
2116 }
2117 
2118 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2119 			     elf_addr_t e_shoff, int segs)
2120 {
2121 	elf->e_shoff = e_shoff;
2122 	elf->e_shentsize = sizeof(*shdr4extnum);
2123 	elf->e_shnum = 1;
2124 	elf->e_shstrndx = SHN_UNDEF;
2125 
2126 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2127 
2128 	shdr4extnum->sh_type = SHT_NULL;
2129 	shdr4extnum->sh_size = elf->e_shnum;
2130 	shdr4extnum->sh_link = elf->e_shstrndx;
2131 	shdr4extnum->sh_info = segs;
2132 }
2133 
2134 /*
2135  * Actual dumper
2136  *
2137  * This is a two-pass process; first we find the offsets of the bits,
2138  * and then they are actually written out.  If we run out of core limit
2139  * we just truncate.
2140  */
2141 static int elf_core_dump(struct coredump_params *cprm)
2142 {
2143 	int has_dumped = 0;
2144 	mm_segment_t fs;
2145 	int segs, i;
2146 	size_t vma_data_size = 0;
2147 	struct vm_area_struct *vma, *gate_vma;
2148 	struct elfhdr *elf = NULL;
2149 	loff_t offset = 0, dataoff;
2150 	struct elf_note_info info = { };
2151 	struct elf_phdr *phdr4note = NULL;
2152 	struct elf_shdr *shdr4extnum = NULL;
2153 	Elf_Half e_phnum;
2154 	elf_addr_t e_shoff;
2155 	elf_addr_t *vma_filesz = NULL;
2156 
2157 	/*
2158 	 * We no longer stop all VM operations.
2159 	 *
2160 	 * This is because those proceses that could possibly change map_count
2161 	 * or the mmap / vma pages are now blocked in do_exit on current
2162 	 * finishing this core dump.
2163 	 *
2164 	 * Only ptrace can touch these memory addresses, but it doesn't change
2165 	 * the map_count or the pages allocated. So no possibility of crashing
2166 	 * exists while dumping the mm->vm_next areas to the core file.
2167 	 */
2168 
2169 	/* alloc memory for large data structures: too large to be on stack */
2170 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2171 	if (!elf)
2172 		goto out;
2173 	/*
2174 	 * The number of segs are recored into ELF header as 16bit value.
2175 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2176 	 */
2177 	segs = current->mm->map_count;
2178 	segs += elf_core_extra_phdrs();
2179 
2180 	gate_vma = get_gate_vma(current->mm);
2181 	if (gate_vma != NULL)
2182 		segs++;
2183 
2184 	/* for notes section */
2185 	segs++;
2186 
2187 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2188 	 * this, kernel supports extended numbering. Have a look at
2189 	 * include/linux/elf.h for further information. */
2190 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2191 
2192 	/*
2193 	 * Collect all the non-memory information about the process for the
2194 	 * notes.  This also sets up the file header.
2195 	 */
2196 	if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2197 		goto cleanup;
2198 
2199 	has_dumped = 1;
2200 
2201 	fs = get_fs();
2202 	set_fs(KERNEL_DS);
2203 
2204 	offset += sizeof(*elf);				/* Elf header */
2205 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2206 
2207 	/* Write notes phdr entry */
2208 	{
2209 		size_t sz = get_note_info_size(&info);
2210 
2211 		sz += elf_coredump_extra_notes_size();
2212 
2213 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2214 		if (!phdr4note)
2215 			goto end_coredump;
2216 
2217 		fill_elf_note_phdr(phdr4note, sz, offset);
2218 		offset += sz;
2219 	}
2220 
2221 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2222 
2223 	if (segs - 1 > ULONG_MAX / sizeof(*vma_filesz))
2224 		goto end_coredump;
2225 	vma_filesz = vmalloc((segs - 1) * sizeof(*vma_filesz));
2226 	if (!vma_filesz)
2227 		goto end_coredump;
2228 
2229 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2230 			vma = next_vma(vma, gate_vma)) {
2231 		unsigned long dump_size;
2232 
2233 		dump_size = vma_dump_size(vma, cprm->mm_flags);
2234 		vma_filesz[i++] = dump_size;
2235 		vma_data_size += dump_size;
2236 	}
2237 
2238 	offset += vma_data_size;
2239 	offset += elf_core_extra_data_size();
2240 	e_shoff = offset;
2241 
2242 	if (e_phnum == PN_XNUM) {
2243 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2244 		if (!shdr4extnum)
2245 			goto end_coredump;
2246 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2247 	}
2248 
2249 	offset = dataoff;
2250 
2251 	if (!dump_emit(cprm, elf, sizeof(*elf)))
2252 		goto end_coredump;
2253 
2254 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2255 		goto end_coredump;
2256 
2257 	/* Write program headers for segments dump */
2258 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2259 			vma = next_vma(vma, gate_vma)) {
2260 		struct elf_phdr phdr;
2261 
2262 		phdr.p_type = PT_LOAD;
2263 		phdr.p_offset = offset;
2264 		phdr.p_vaddr = vma->vm_start;
2265 		phdr.p_paddr = 0;
2266 		phdr.p_filesz = vma_filesz[i++];
2267 		phdr.p_memsz = vma->vm_end - vma->vm_start;
2268 		offset += phdr.p_filesz;
2269 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2270 		if (vma->vm_flags & VM_WRITE)
2271 			phdr.p_flags |= PF_W;
2272 		if (vma->vm_flags & VM_EXEC)
2273 			phdr.p_flags |= PF_X;
2274 		phdr.p_align = ELF_EXEC_PAGESIZE;
2275 
2276 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2277 			goto end_coredump;
2278 	}
2279 
2280 	if (!elf_core_write_extra_phdrs(cprm, offset))
2281 		goto end_coredump;
2282 
2283  	/* write out the notes section */
2284 	if (!write_note_info(&info, cprm))
2285 		goto end_coredump;
2286 
2287 	if (elf_coredump_extra_notes_write(cprm))
2288 		goto end_coredump;
2289 
2290 	/* Align to page */
2291 	if (!dump_skip(cprm, dataoff - cprm->pos))
2292 		goto end_coredump;
2293 
2294 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2295 			vma = next_vma(vma, gate_vma)) {
2296 		unsigned long addr;
2297 		unsigned long end;
2298 
2299 		end = vma->vm_start + vma_filesz[i++];
2300 
2301 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2302 			struct page *page;
2303 			int stop;
2304 
2305 			page = get_dump_page(addr);
2306 			if (page) {
2307 				void *kaddr = kmap(page);
2308 				stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2309 				kunmap(page);
2310 				put_page(page);
2311 			} else
2312 				stop = !dump_skip(cprm, PAGE_SIZE);
2313 			if (stop)
2314 				goto end_coredump;
2315 		}
2316 	}
2317 	dump_truncate(cprm);
2318 
2319 	if (!elf_core_write_extra_data(cprm))
2320 		goto end_coredump;
2321 
2322 	if (e_phnum == PN_XNUM) {
2323 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2324 			goto end_coredump;
2325 	}
2326 
2327 end_coredump:
2328 	set_fs(fs);
2329 
2330 cleanup:
2331 	free_note_info(&info);
2332 	kfree(shdr4extnum);
2333 	vfree(vma_filesz);
2334 	kfree(phdr4note);
2335 	kfree(elf);
2336 out:
2337 	return has_dumped;
2338 }
2339 
2340 #endif		/* CONFIG_ELF_CORE */
2341 
2342 static int __init init_elf_binfmt(void)
2343 {
2344 	register_binfmt(&elf_format);
2345 	return 0;
2346 }
2347 
2348 static void __exit exit_elf_binfmt(void)
2349 {
2350 	/* Remove the COFF and ELF loaders. */
2351 	unregister_binfmt(&elf_format);
2352 }
2353 
2354 core_initcall(init_elf_binfmt);
2355 module_exit(exit_elf_binfmt);
2356 MODULE_LICENSE("GPL");
2357