xref: /openbmc/linux/fs/binfmt_elf.c (revision 78700c0a)
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/dax.h>
39 #include <asm/uaccess.h>
40 #include <asm/param.h>
41 #include <asm/page.h>
42 
43 #ifndef user_long_t
44 #define user_long_t long
45 #endif
46 #ifndef user_siginfo_t
47 #define user_siginfo_t siginfo_t
48 #endif
49 
50 static int load_elf_binary(struct linux_binprm *bprm);
51 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
52 				int, int, unsigned long);
53 
54 #ifdef CONFIG_USELIB
55 static int load_elf_library(struct file *);
56 #else
57 #define load_elf_library NULL
58 #endif
59 
60 /*
61  * If we don't support core dumping, then supply a NULL so we
62  * don't even try.
63  */
64 #ifdef CONFIG_ELF_CORE
65 static int elf_core_dump(struct coredump_params *cprm);
66 #else
67 #define elf_core_dump	NULL
68 #endif
69 
70 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
71 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
72 #else
73 #define ELF_MIN_ALIGN	PAGE_SIZE
74 #endif
75 
76 #ifndef ELF_CORE_EFLAGS
77 #define ELF_CORE_EFLAGS	0
78 #endif
79 
80 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
81 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
82 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
83 
84 static struct linux_binfmt elf_format = {
85 	.module		= THIS_MODULE,
86 	.load_binary	= load_elf_binary,
87 	.load_shlib	= load_elf_library,
88 	.core_dump	= elf_core_dump,
89 	.min_coredump	= ELF_EXEC_PAGESIZE,
90 };
91 
92 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
93 
94 static int set_brk(unsigned long start, unsigned long end)
95 {
96 	start = ELF_PAGEALIGN(start);
97 	end = ELF_PAGEALIGN(end);
98 	if (end > start) {
99 		int error = vm_brk(start, end - start);
100 		if (error)
101 			return error;
102 	}
103 	current->mm->start_brk = current->mm->brk = end;
104 	return 0;
105 }
106 
107 /* We need to explicitly zero any fractional pages
108    after the data section (i.e. bss).  This would
109    contain the junk from the file that should not
110    be in memory
111  */
112 static int padzero(unsigned long elf_bss)
113 {
114 	unsigned long nbyte;
115 
116 	nbyte = ELF_PAGEOFFSET(elf_bss);
117 	if (nbyte) {
118 		nbyte = ELF_MIN_ALIGN - nbyte;
119 		if (clear_user((void __user *) elf_bss, nbyte))
120 			return -EFAULT;
121 	}
122 	return 0;
123 }
124 
125 /* Let's use some macros to make this stack manipulation a little clearer */
126 #ifdef CONFIG_STACK_GROWSUP
127 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
128 #define STACK_ROUND(sp, items) \
129 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
130 #define STACK_ALLOC(sp, len) ({ \
131 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
132 	old_sp; })
133 #else
134 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
135 #define STACK_ROUND(sp, items) \
136 	(((unsigned long) (sp - items)) &~ 15UL)
137 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
138 #endif
139 
140 #ifndef ELF_BASE_PLATFORM
141 /*
142  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
143  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
144  * will be copied to the user stack in the same manner as AT_PLATFORM.
145  */
146 #define ELF_BASE_PLATFORM NULL
147 #endif
148 
149 static int
150 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
151 		unsigned long load_addr, unsigned long interp_load_addr)
152 {
153 	unsigned long p = bprm->p;
154 	int argc = bprm->argc;
155 	int envc = bprm->envc;
156 	elf_addr_t __user *argv;
157 	elf_addr_t __user *envp;
158 	elf_addr_t __user *sp;
159 	elf_addr_t __user *u_platform;
160 	elf_addr_t __user *u_base_platform;
161 	elf_addr_t __user *u_rand_bytes;
162 	const char *k_platform = ELF_PLATFORM;
163 	const char *k_base_platform = ELF_BASE_PLATFORM;
164 	unsigned char k_rand_bytes[16];
165 	int items;
166 	elf_addr_t *elf_info;
167 	int ei_index = 0;
168 	const struct cred *cred = current_cred();
169 	struct vm_area_struct *vma;
170 
171 	/*
172 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
173 	 * evictions by the processes running on the same package. One
174 	 * thing we can do is to shuffle the initial stack for them.
175 	 */
176 
177 	p = arch_align_stack(p);
178 
179 	/*
180 	 * If this architecture has a platform capability string, copy it
181 	 * to userspace.  In some cases (Sparc), this info is impossible
182 	 * for userspace to get any other way, in others (i386) it is
183 	 * merely difficult.
184 	 */
185 	u_platform = NULL;
186 	if (k_platform) {
187 		size_t len = strlen(k_platform) + 1;
188 
189 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
190 		if (__copy_to_user(u_platform, k_platform, len))
191 			return -EFAULT;
192 	}
193 
194 	/*
195 	 * If this architecture has a "base" platform capability
196 	 * string, copy it to userspace.
197 	 */
198 	u_base_platform = NULL;
199 	if (k_base_platform) {
200 		size_t len = strlen(k_base_platform) + 1;
201 
202 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
203 		if (__copy_to_user(u_base_platform, k_base_platform, len))
204 			return -EFAULT;
205 	}
206 
207 	/*
208 	 * Generate 16 random bytes for userspace PRNG seeding.
209 	 */
210 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
211 	u_rand_bytes = (elf_addr_t __user *)
212 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
213 	if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
214 		return -EFAULT;
215 
216 	/* Create the ELF interpreter info */
217 	elf_info = (elf_addr_t *)current->mm->saved_auxv;
218 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
219 #define NEW_AUX_ENT(id, val) \
220 	do { \
221 		elf_info[ei_index++] = id; \
222 		elf_info[ei_index++] = val; \
223 	} while (0)
224 
225 #ifdef ARCH_DLINFO
226 	/*
227 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
228 	 * AUXV.
229 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
230 	 * ARCH_DLINFO changes
231 	 */
232 	ARCH_DLINFO;
233 #endif
234 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
235 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
236 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
237 	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
238 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
239 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
240 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
241 	NEW_AUX_ENT(AT_FLAGS, 0);
242 	NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
243 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
244 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
245 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
246 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
247  	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
248 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
249 #ifdef ELF_HWCAP2
250 	NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
251 #endif
252 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
253 	if (k_platform) {
254 		NEW_AUX_ENT(AT_PLATFORM,
255 			    (elf_addr_t)(unsigned long)u_platform);
256 	}
257 	if (k_base_platform) {
258 		NEW_AUX_ENT(AT_BASE_PLATFORM,
259 			    (elf_addr_t)(unsigned long)u_base_platform);
260 	}
261 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
262 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
263 	}
264 #undef NEW_AUX_ENT
265 	/* AT_NULL is zero; clear the rest too */
266 	memset(&elf_info[ei_index], 0,
267 	       sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
268 
269 	/* And advance past the AT_NULL entry.  */
270 	ei_index += 2;
271 
272 	sp = STACK_ADD(p, ei_index);
273 
274 	items = (argc + 1) + (envc + 1) + 1;
275 	bprm->p = STACK_ROUND(sp, items);
276 
277 	/* Point sp at the lowest address on the stack */
278 #ifdef CONFIG_STACK_GROWSUP
279 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
280 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
281 #else
282 	sp = (elf_addr_t __user *)bprm->p;
283 #endif
284 
285 
286 	/*
287 	 * Grow the stack manually; some architectures have a limit on how
288 	 * far ahead a user-space access may be in order to grow the stack.
289 	 */
290 	vma = find_extend_vma(current->mm, bprm->p);
291 	if (!vma)
292 		return -EFAULT;
293 
294 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
295 	if (__put_user(argc, sp++))
296 		return -EFAULT;
297 	argv = sp;
298 	envp = argv + argc + 1;
299 
300 	/* Populate argv and envp */
301 	p = current->mm->arg_end = current->mm->arg_start;
302 	while (argc-- > 0) {
303 		size_t len;
304 		if (__put_user((elf_addr_t)p, argv++))
305 			return -EFAULT;
306 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
307 		if (!len || len > MAX_ARG_STRLEN)
308 			return -EINVAL;
309 		p += len;
310 	}
311 	if (__put_user(0, argv))
312 		return -EFAULT;
313 	current->mm->arg_end = current->mm->env_start = p;
314 	while (envc-- > 0) {
315 		size_t len;
316 		if (__put_user((elf_addr_t)p, envp++))
317 			return -EFAULT;
318 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
319 		if (!len || len > MAX_ARG_STRLEN)
320 			return -EINVAL;
321 		p += len;
322 	}
323 	if (__put_user(0, envp))
324 		return -EFAULT;
325 	current->mm->env_end = p;
326 
327 	/* Put the elf_info on the stack in the right place.  */
328 	sp = (elf_addr_t __user *)envp + 1;
329 	if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
330 		return -EFAULT;
331 	return 0;
332 }
333 
334 #ifndef elf_map
335 
336 static unsigned long elf_map(struct file *filep, unsigned long addr,
337 		struct elf_phdr *eppnt, int prot, int type,
338 		unsigned long total_size)
339 {
340 	unsigned long map_addr;
341 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
342 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
343 	addr = ELF_PAGESTART(addr);
344 	size = ELF_PAGEALIGN(size);
345 
346 	/* mmap() will return -EINVAL if given a zero size, but a
347 	 * segment with zero filesize is perfectly valid */
348 	if (!size)
349 		return addr;
350 
351 	/*
352 	* total_size is the size of the ELF (interpreter) image.
353 	* The _first_ mmap needs to know the full size, otherwise
354 	* randomization might put this image into an overlapping
355 	* position with the ELF binary image. (since size < total_size)
356 	* So we first map the 'big' image - and unmap the remainder at
357 	* the end. (which unmap is needed for ELF images with holes.)
358 	*/
359 	if (total_size) {
360 		total_size = ELF_PAGEALIGN(total_size);
361 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
362 		if (!BAD_ADDR(map_addr))
363 			vm_munmap(map_addr+size, total_size-size);
364 	} else
365 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
366 
367 	return(map_addr);
368 }
369 
370 #endif /* !elf_map */
371 
372 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
373 {
374 	int i, first_idx = -1, last_idx = -1;
375 
376 	for (i = 0; i < nr; i++) {
377 		if (cmds[i].p_type == PT_LOAD) {
378 			last_idx = i;
379 			if (first_idx == -1)
380 				first_idx = i;
381 		}
382 	}
383 	if (first_idx == -1)
384 		return 0;
385 
386 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
387 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
388 }
389 
390 /**
391  * load_elf_phdrs() - load ELF program headers
392  * @elf_ex:   ELF header of the binary whose program headers should be loaded
393  * @elf_file: the opened ELF binary file
394  *
395  * Loads ELF program headers from the binary file elf_file, which has the ELF
396  * header pointed to by elf_ex, into a newly allocated array. The caller is
397  * responsible for freeing the allocated data. Returns an ERR_PTR upon failure.
398  */
399 static struct elf_phdr *load_elf_phdrs(struct elfhdr *elf_ex,
400 				       struct file *elf_file)
401 {
402 	struct elf_phdr *elf_phdata = NULL;
403 	int retval, size, err = -1;
404 
405 	/*
406 	 * If the size of this structure has changed, then punt, since
407 	 * we will be doing the wrong thing.
408 	 */
409 	if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
410 		goto out;
411 
412 	/* Sanity check the number of program headers... */
413 	if (elf_ex->e_phnum < 1 ||
414 		elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
415 		goto out;
416 
417 	/* ...and their total size. */
418 	size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
419 	if (size > ELF_MIN_ALIGN)
420 		goto out;
421 
422 	elf_phdata = kmalloc(size, GFP_KERNEL);
423 	if (!elf_phdata)
424 		goto out;
425 
426 	/* Read in the program headers */
427 	retval = kernel_read(elf_file, elf_ex->e_phoff,
428 			     (char *)elf_phdata, size);
429 	if (retval != size) {
430 		err = (retval < 0) ? retval : -EIO;
431 		goto out;
432 	}
433 
434 	/* Success! */
435 	err = 0;
436 out:
437 	if (err) {
438 		kfree(elf_phdata);
439 		elf_phdata = NULL;
440 	}
441 	return elf_phdata;
442 }
443 
444 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
445 
446 /**
447  * struct arch_elf_state - arch-specific ELF loading state
448  *
449  * This structure is used to preserve architecture specific data during
450  * the loading of an ELF file, throughout the checking of architecture
451  * specific ELF headers & through to the point where the ELF load is
452  * known to be proceeding (ie. SET_PERSONALITY).
453  *
454  * This implementation is a dummy for architectures which require no
455  * specific state.
456  */
457 struct arch_elf_state {
458 };
459 
460 #define INIT_ARCH_ELF_STATE {}
461 
462 /**
463  * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
464  * @ehdr:	The main ELF header
465  * @phdr:	The program header to check
466  * @elf:	The open ELF file
467  * @is_interp:	True if the phdr is from the interpreter of the ELF being
468  *		loaded, else false.
469  * @state:	Architecture-specific state preserved throughout the process
470  *		of loading the ELF.
471  *
472  * Inspects the program header phdr to validate its correctness and/or
473  * suitability for the system. Called once per ELF program header in the
474  * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
475  * interpreter.
476  *
477  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
478  *         with that return code.
479  */
480 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
481 				   struct elf_phdr *phdr,
482 				   struct file *elf, bool is_interp,
483 				   struct arch_elf_state *state)
484 {
485 	/* Dummy implementation, always proceed */
486 	return 0;
487 }
488 
489 /**
490  * arch_check_elf() - check an ELF executable
491  * @ehdr:	The main ELF header
492  * @has_interp:	True if the ELF has an interpreter, else false.
493  * @interp_ehdr: The interpreter's ELF header
494  * @state:	Architecture-specific state preserved throughout the process
495  *		of loading the ELF.
496  *
497  * Provides a final opportunity for architecture code to reject the loading
498  * of the ELF & cause an exec syscall to return an error. This is called after
499  * all program headers to be checked by arch_elf_pt_proc have been.
500  *
501  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
502  *         with that return code.
503  */
504 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
505 				 struct elfhdr *interp_ehdr,
506 				 struct arch_elf_state *state)
507 {
508 	/* Dummy implementation, always proceed */
509 	return 0;
510 }
511 
512 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
513 
514 /* This is much more generalized than the library routine read function,
515    so we keep this separate.  Technically the library read function
516    is only provided so that we can read a.out libraries that have
517    an ELF header */
518 
519 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
520 		struct file *interpreter, unsigned long *interp_map_addr,
521 		unsigned long no_base, struct elf_phdr *interp_elf_phdata)
522 {
523 	struct elf_phdr *eppnt;
524 	unsigned long load_addr = 0;
525 	int load_addr_set = 0;
526 	unsigned long last_bss = 0, elf_bss = 0;
527 	unsigned long error = ~0UL;
528 	unsigned long total_size;
529 	int i;
530 
531 	/* First of all, some simple consistency checks */
532 	if (interp_elf_ex->e_type != ET_EXEC &&
533 	    interp_elf_ex->e_type != ET_DYN)
534 		goto out;
535 	if (!elf_check_arch(interp_elf_ex))
536 		goto out;
537 	if (!interpreter->f_op->mmap)
538 		goto out;
539 
540 	total_size = total_mapping_size(interp_elf_phdata,
541 					interp_elf_ex->e_phnum);
542 	if (!total_size) {
543 		error = -EINVAL;
544 		goto out;
545 	}
546 
547 	eppnt = interp_elf_phdata;
548 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
549 		if (eppnt->p_type == PT_LOAD) {
550 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
551 			int elf_prot = 0;
552 			unsigned long vaddr = 0;
553 			unsigned long k, map_addr;
554 
555 			if (eppnt->p_flags & PF_R)
556 		    		elf_prot = PROT_READ;
557 			if (eppnt->p_flags & PF_W)
558 				elf_prot |= PROT_WRITE;
559 			if (eppnt->p_flags & PF_X)
560 				elf_prot |= PROT_EXEC;
561 			vaddr = eppnt->p_vaddr;
562 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
563 				elf_type |= MAP_FIXED;
564 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
565 				load_addr = -vaddr;
566 
567 			map_addr = elf_map(interpreter, load_addr + vaddr,
568 					eppnt, elf_prot, elf_type, total_size);
569 			total_size = 0;
570 			if (!*interp_map_addr)
571 				*interp_map_addr = map_addr;
572 			error = map_addr;
573 			if (BAD_ADDR(map_addr))
574 				goto out;
575 
576 			if (!load_addr_set &&
577 			    interp_elf_ex->e_type == ET_DYN) {
578 				load_addr = map_addr - ELF_PAGESTART(vaddr);
579 				load_addr_set = 1;
580 			}
581 
582 			/*
583 			 * Check to see if the section's size will overflow the
584 			 * allowed task size. Note that p_filesz must always be
585 			 * <= p_memsize so it's only necessary to check p_memsz.
586 			 */
587 			k = load_addr + eppnt->p_vaddr;
588 			if (BAD_ADDR(k) ||
589 			    eppnt->p_filesz > eppnt->p_memsz ||
590 			    eppnt->p_memsz > TASK_SIZE ||
591 			    TASK_SIZE - eppnt->p_memsz < k) {
592 				error = -ENOMEM;
593 				goto out;
594 			}
595 
596 			/*
597 			 * Find the end of the file mapping for this phdr, and
598 			 * keep track of the largest address we see for this.
599 			 */
600 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
601 			if (k > elf_bss)
602 				elf_bss = k;
603 
604 			/*
605 			 * Do the same thing for the memory mapping - between
606 			 * elf_bss and last_bss is the bss section.
607 			 */
608 			k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
609 			if (k > last_bss)
610 				last_bss = k;
611 		}
612 	}
613 
614 	if (last_bss > elf_bss) {
615 		/*
616 		 * Now fill out the bss section.  First pad the last page up
617 		 * to the page boundary, and then perform a mmap to make sure
618 		 * that there are zero-mapped pages up to and including the
619 		 * last bss page.
620 		 */
621 		if (padzero(elf_bss)) {
622 			error = -EFAULT;
623 			goto out;
624 		}
625 
626 		/* What we have mapped so far */
627 		elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
628 
629 		/* Map the last of the bss segment */
630 		error = vm_brk(elf_bss, last_bss - elf_bss);
631 		if (error)
632 			goto out;
633 	}
634 
635 	error = load_addr;
636 out:
637 	return error;
638 }
639 
640 /*
641  * These are the functions used to load ELF style executables and shared
642  * libraries.  There is no binary dependent code anywhere else.
643  */
644 
645 #ifndef STACK_RND_MASK
646 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))	/* 8MB of VA */
647 #endif
648 
649 static unsigned long randomize_stack_top(unsigned long stack_top)
650 {
651 	unsigned long random_variable = 0;
652 
653 	if ((current->flags & PF_RANDOMIZE) &&
654 		!(current->personality & ADDR_NO_RANDOMIZE)) {
655 		random_variable = get_random_long();
656 		random_variable &= STACK_RND_MASK;
657 		random_variable <<= PAGE_SHIFT;
658 	}
659 #ifdef CONFIG_STACK_GROWSUP
660 	return PAGE_ALIGN(stack_top) + random_variable;
661 #else
662 	return PAGE_ALIGN(stack_top) - random_variable;
663 #endif
664 }
665 
666 static int load_elf_binary(struct linux_binprm *bprm)
667 {
668 	struct file *interpreter = NULL; /* to shut gcc up */
669  	unsigned long load_addr = 0, load_bias = 0;
670 	int load_addr_set = 0;
671 	char * elf_interpreter = NULL;
672 	unsigned long error;
673 	struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
674 	unsigned long elf_bss, elf_brk;
675 	int retval, i;
676 	unsigned long elf_entry;
677 	unsigned long interp_load_addr = 0;
678 	unsigned long start_code, end_code, start_data, end_data;
679 	unsigned long reloc_func_desc __maybe_unused = 0;
680 	int executable_stack = EXSTACK_DEFAULT;
681 	struct pt_regs *regs = current_pt_regs();
682 	struct {
683 		struct elfhdr elf_ex;
684 		struct elfhdr interp_elf_ex;
685 	} *loc;
686 	struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
687 
688 	loc = kmalloc(sizeof(*loc), GFP_KERNEL);
689 	if (!loc) {
690 		retval = -ENOMEM;
691 		goto out_ret;
692 	}
693 
694 	/* Get the exec-header */
695 	loc->elf_ex = *((struct elfhdr *)bprm->buf);
696 
697 	retval = -ENOEXEC;
698 	/* First of all, some simple consistency checks */
699 	if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
700 		goto out;
701 
702 	if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
703 		goto out;
704 	if (!elf_check_arch(&loc->elf_ex))
705 		goto out;
706 	if (!bprm->file->f_op->mmap)
707 		goto out;
708 
709 	elf_phdata = load_elf_phdrs(&loc->elf_ex, bprm->file);
710 	if (!elf_phdata)
711 		goto out;
712 
713 	elf_ppnt = elf_phdata;
714 	elf_bss = 0;
715 	elf_brk = 0;
716 
717 	start_code = ~0UL;
718 	end_code = 0;
719 	start_data = 0;
720 	end_data = 0;
721 
722 	for (i = 0; i < loc->elf_ex.e_phnum; i++) {
723 		if (elf_ppnt->p_type == PT_INTERP) {
724 			/* This is the program interpreter used for
725 			 * shared libraries - for now assume that this
726 			 * is an a.out format binary
727 			 */
728 			retval = -ENOEXEC;
729 			if (elf_ppnt->p_filesz > PATH_MAX ||
730 			    elf_ppnt->p_filesz < 2)
731 				goto out_free_ph;
732 
733 			retval = -ENOMEM;
734 			elf_interpreter = kmalloc(elf_ppnt->p_filesz,
735 						  GFP_KERNEL);
736 			if (!elf_interpreter)
737 				goto out_free_ph;
738 
739 			retval = kernel_read(bprm->file, elf_ppnt->p_offset,
740 					     elf_interpreter,
741 					     elf_ppnt->p_filesz);
742 			if (retval != elf_ppnt->p_filesz) {
743 				if (retval >= 0)
744 					retval = -EIO;
745 				goto out_free_interp;
746 			}
747 			/* make sure path is NULL terminated */
748 			retval = -ENOEXEC;
749 			if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
750 				goto out_free_interp;
751 
752 			interpreter = open_exec(elf_interpreter);
753 			retval = PTR_ERR(interpreter);
754 			if (IS_ERR(interpreter))
755 				goto out_free_interp;
756 
757 			/*
758 			 * If the binary is not readable then enforce
759 			 * mm->dumpable = 0 regardless of the interpreter's
760 			 * permissions.
761 			 */
762 			would_dump(bprm, interpreter);
763 
764 			/* Get the exec headers */
765 			retval = kernel_read(interpreter, 0,
766 					     (void *)&loc->interp_elf_ex,
767 					     sizeof(loc->interp_elf_ex));
768 			if (retval != sizeof(loc->interp_elf_ex)) {
769 				if (retval >= 0)
770 					retval = -EIO;
771 				goto out_free_dentry;
772 			}
773 
774 			break;
775 		}
776 		elf_ppnt++;
777 	}
778 
779 	elf_ppnt = elf_phdata;
780 	for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
781 		switch (elf_ppnt->p_type) {
782 		case PT_GNU_STACK:
783 			if (elf_ppnt->p_flags & PF_X)
784 				executable_stack = EXSTACK_ENABLE_X;
785 			else
786 				executable_stack = EXSTACK_DISABLE_X;
787 			break;
788 
789 		case PT_LOPROC ... PT_HIPROC:
790 			retval = arch_elf_pt_proc(&loc->elf_ex, elf_ppnt,
791 						  bprm->file, false,
792 						  &arch_state);
793 			if (retval)
794 				goto out_free_dentry;
795 			break;
796 		}
797 
798 	/* Some simple consistency checks for the interpreter */
799 	if (elf_interpreter) {
800 		retval = -ELIBBAD;
801 		/* Not an ELF interpreter */
802 		if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
803 			goto out_free_dentry;
804 		/* Verify the interpreter has a valid arch */
805 		if (!elf_check_arch(&loc->interp_elf_ex))
806 			goto out_free_dentry;
807 
808 		/* Load the interpreter program headers */
809 		interp_elf_phdata = load_elf_phdrs(&loc->interp_elf_ex,
810 						   interpreter);
811 		if (!interp_elf_phdata)
812 			goto out_free_dentry;
813 
814 		/* Pass PT_LOPROC..PT_HIPROC headers to arch code */
815 		elf_ppnt = interp_elf_phdata;
816 		for (i = 0; i < loc->interp_elf_ex.e_phnum; i++, elf_ppnt++)
817 			switch (elf_ppnt->p_type) {
818 			case PT_LOPROC ... PT_HIPROC:
819 				retval = arch_elf_pt_proc(&loc->interp_elf_ex,
820 							  elf_ppnt, interpreter,
821 							  true, &arch_state);
822 				if (retval)
823 					goto out_free_dentry;
824 				break;
825 			}
826 	}
827 
828 	/*
829 	 * Allow arch code to reject the ELF at this point, whilst it's
830 	 * still possible to return an error to the code that invoked
831 	 * the exec syscall.
832 	 */
833 	retval = arch_check_elf(&loc->elf_ex,
834 				!!interpreter, &loc->interp_elf_ex,
835 				&arch_state);
836 	if (retval)
837 		goto out_free_dentry;
838 
839 	/* Flush all traces of the currently running executable */
840 	retval = flush_old_exec(bprm);
841 	if (retval)
842 		goto out_free_dentry;
843 
844 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
845 	   may depend on the personality.  */
846 	SET_PERSONALITY2(loc->elf_ex, &arch_state);
847 	if (elf_read_implies_exec(loc->elf_ex, executable_stack))
848 		current->personality |= READ_IMPLIES_EXEC;
849 
850 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
851 		current->flags |= PF_RANDOMIZE;
852 
853 	setup_new_exec(bprm);
854 
855 	/* Do this so that we can load the interpreter, if need be.  We will
856 	   change some of these later */
857 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
858 				 executable_stack);
859 	if (retval < 0)
860 		goto out_free_dentry;
861 
862 	current->mm->start_stack = bprm->p;
863 
864 	/* Now we do a little grungy work by mmapping the ELF image into
865 	   the correct location in memory. */
866 	for(i = 0, elf_ppnt = elf_phdata;
867 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
868 		int elf_prot = 0, elf_flags;
869 		unsigned long k, vaddr;
870 		unsigned long total_size = 0;
871 
872 		if (elf_ppnt->p_type != PT_LOAD)
873 			continue;
874 
875 		if (unlikely (elf_brk > elf_bss)) {
876 			unsigned long nbyte;
877 
878 			/* There was a PT_LOAD segment with p_memsz > p_filesz
879 			   before this one. Map anonymous pages, if needed,
880 			   and clear the area.  */
881 			retval = set_brk(elf_bss + load_bias,
882 					 elf_brk + load_bias);
883 			if (retval)
884 				goto out_free_dentry;
885 			nbyte = ELF_PAGEOFFSET(elf_bss);
886 			if (nbyte) {
887 				nbyte = ELF_MIN_ALIGN - nbyte;
888 				if (nbyte > elf_brk - elf_bss)
889 					nbyte = elf_brk - elf_bss;
890 				if (clear_user((void __user *)elf_bss +
891 							load_bias, nbyte)) {
892 					/*
893 					 * This bss-zeroing can fail if the ELF
894 					 * file specifies odd protections. So
895 					 * we don't check the return value
896 					 */
897 				}
898 			}
899 		}
900 
901 		if (elf_ppnt->p_flags & PF_R)
902 			elf_prot |= PROT_READ;
903 		if (elf_ppnt->p_flags & PF_W)
904 			elf_prot |= PROT_WRITE;
905 		if (elf_ppnt->p_flags & PF_X)
906 			elf_prot |= PROT_EXEC;
907 
908 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
909 
910 		vaddr = elf_ppnt->p_vaddr;
911 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
912 			elf_flags |= MAP_FIXED;
913 		} else if (loc->elf_ex.e_type == ET_DYN) {
914 			/* Try and get dynamic programs out of the way of the
915 			 * default mmap base, as well as whatever program they
916 			 * might try to exec.  This is because the brk will
917 			 * follow the loader, and is not movable.  */
918 			load_bias = ELF_ET_DYN_BASE - vaddr;
919 			if (current->flags & PF_RANDOMIZE)
920 				load_bias += arch_mmap_rnd();
921 			load_bias = ELF_PAGESTART(load_bias);
922 			total_size = total_mapping_size(elf_phdata,
923 							loc->elf_ex.e_phnum);
924 			if (!total_size) {
925 				retval = -EINVAL;
926 				goto out_free_dentry;
927 			}
928 		}
929 
930 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
931 				elf_prot, elf_flags, total_size);
932 		if (BAD_ADDR(error)) {
933 			retval = IS_ERR((void *)error) ?
934 				PTR_ERR((void*)error) : -EINVAL;
935 			goto out_free_dentry;
936 		}
937 
938 		if (!load_addr_set) {
939 			load_addr_set = 1;
940 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
941 			if (loc->elf_ex.e_type == ET_DYN) {
942 				load_bias += error -
943 				             ELF_PAGESTART(load_bias + vaddr);
944 				load_addr += load_bias;
945 				reloc_func_desc = load_bias;
946 			}
947 		}
948 		k = elf_ppnt->p_vaddr;
949 		if (k < start_code)
950 			start_code = k;
951 		if (start_data < k)
952 			start_data = k;
953 
954 		/*
955 		 * Check to see if the section's size will overflow the
956 		 * allowed task size. Note that p_filesz must always be
957 		 * <= p_memsz so it is only necessary to check p_memsz.
958 		 */
959 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
960 		    elf_ppnt->p_memsz > TASK_SIZE ||
961 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
962 			/* set_brk can never work. Avoid overflows. */
963 			retval = -EINVAL;
964 			goto out_free_dentry;
965 		}
966 
967 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
968 
969 		if (k > elf_bss)
970 			elf_bss = k;
971 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
972 			end_code = k;
973 		if (end_data < k)
974 			end_data = k;
975 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
976 		if (k > elf_brk)
977 			elf_brk = k;
978 	}
979 
980 	loc->elf_ex.e_entry += load_bias;
981 	elf_bss += load_bias;
982 	elf_brk += load_bias;
983 	start_code += load_bias;
984 	end_code += load_bias;
985 	start_data += load_bias;
986 	end_data += load_bias;
987 
988 	/* Calling set_brk effectively mmaps the pages that we need
989 	 * for the bss and break sections.  We must do this before
990 	 * mapping in the interpreter, to make sure it doesn't wind
991 	 * up getting placed where the bss needs to go.
992 	 */
993 	retval = set_brk(elf_bss, elf_brk);
994 	if (retval)
995 		goto out_free_dentry;
996 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
997 		retval = -EFAULT; /* Nobody gets to see this, but.. */
998 		goto out_free_dentry;
999 	}
1000 
1001 	if (elf_interpreter) {
1002 		unsigned long interp_map_addr = 0;
1003 
1004 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
1005 					    interpreter,
1006 					    &interp_map_addr,
1007 					    load_bias, interp_elf_phdata);
1008 		if (!IS_ERR((void *)elf_entry)) {
1009 			/*
1010 			 * load_elf_interp() returns relocation
1011 			 * adjustment
1012 			 */
1013 			interp_load_addr = elf_entry;
1014 			elf_entry += loc->interp_elf_ex.e_entry;
1015 		}
1016 		if (BAD_ADDR(elf_entry)) {
1017 			retval = IS_ERR((void *)elf_entry) ?
1018 					(int)elf_entry : -EINVAL;
1019 			goto out_free_dentry;
1020 		}
1021 		reloc_func_desc = interp_load_addr;
1022 
1023 		allow_write_access(interpreter);
1024 		fput(interpreter);
1025 		kfree(elf_interpreter);
1026 	} else {
1027 		elf_entry = loc->elf_ex.e_entry;
1028 		if (BAD_ADDR(elf_entry)) {
1029 			retval = -EINVAL;
1030 			goto out_free_dentry;
1031 		}
1032 	}
1033 
1034 	kfree(interp_elf_phdata);
1035 	kfree(elf_phdata);
1036 
1037 	set_binfmt(&elf_format);
1038 
1039 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1040 	retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
1041 	if (retval < 0)
1042 		goto out;
1043 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1044 
1045 	install_exec_creds(bprm);
1046 	retval = create_elf_tables(bprm, &loc->elf_ex,
1047 			  load_addr, interp_load_addr);
1048 	if (retval < 0)
1049 		goto out;
1050 	/* N.B. passed_fileno might not be initialized? */
1051 	current->mm->end_code = end_code;
1052 	current->mm->start_code = start_code;
1053 	current->mm->start_data = start_data;
1054 	current->mm->end_data = end_data;
1055 	current->mm->start_stack = bprm->p;
1056 
1057 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1058 		current->mm->brk = current->mm->start_brk =
1059 			arch_randomize_brk(current->mm);
1060 #ifdef compat_brk_randomized
1061 		current->brk_randomized = 1;
1062 #endif
1063 	}
1064 
1065 	if (current->personality & MMAP_PAGE_ZERO) {
1066 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
1067 		   and some applications "depend" upon this behavior.
1068 		   Since we do not have the power to recompile these, we
1069 		   emulate the SVr4 behavior. Sigh. */
1070 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1071 				MAP_FIXED | MAP_PRIVATE, 0);
1072 	}
1073 
1074 #ifdef ELF_PLAT_INIT
1075 	/*
1076 	 * The ABI may specify that certain registers be set up in special
1077 	 * ways (on i386 %edx is the address of a DT_FINI function, for
1078 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
1079 	 * that the e_entry field is the address of the function descriptor
1080 	 * for the startup routine, rather than the address of the startup
1081 	 * routine itself.  This macro performs whatever initialization to
1082 	 * the regs structure is required as well as any relocations to the
1083 	 * function descriptor entries when executing dynamically links apps.
1084 	 */
1085 	ELF_PLAT_INIT(regs, reloc_func_desc);
1086 #endif
1087 
1088 	start_thread(regs, elf_entry, bprm->p);
1089 	retval = 0;
1090 out:
1091 	kfree(loc);
1092 out_ret:
1093 	return retval;
1094 
1095 	/* error cleanup */
1096 out_free_dentry:
1097 	kfree(interp_elf_phdata);
1098 	allow_write_access(interpreter);
1099 	if (interpreter)
1100 		fput(interpreter);
1101 out_free_interp:
1102 	kfree(elf_interpreter);
1103 out_free_ph:
1104 	kfree(elf_phdata);
1105 	goto out;
1106 }
1107 
1108 #ifdef CONFIG_USELIB
1109 /* This is really simpleminded and specialized - we are loading an
1110    a.out library that is given an ELF header. */
1111 static int load_elf_library(struct file *file)
1112 {
1113 	struct elf_phdr *elf_phdata;
1114 	struct elf_phdr *eppnt;
1115 	unsigned long elf_bss, bss, len;
1116 	int retval, error, i, j;
1117 	struct elfhdr elf_ex;
1118 
1119 	error = -ENOEXEC;
1120 	retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1121 	if (retval != sizeof(elf_ex))
1122 		goto out;
1123 
1124 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1125 		goto out;
1126 
1127 	/* First of all, some simple consistency checks */
1128 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1129 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1130 		goto out;
1131 
1132 	/* Now read in all of the header information */
1133 
1134 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1135 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1136 
1137 	error = -ENOMEM;
1138 	elf_phdata = kmalloc(j, GFP_KERNEL);
1139 	if (!elf_phdata)
1140 		goto out;
1141 
1142 	eppnt = elf_phdata;
1143 	error = -ENOEXEC;
1144 	retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1145 	if (retval != j)
1146 		goto out_free_ph;
1147 
1148 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1149 		if ((eppnt + i)->p_type == PT_LOAD)
1150 			j++;
1151 	if (j != 1)
1152 		goto out_free_ph;
1153 
1154 	while (eppnt->p_type != PT_LOAD)
1155 		eppnt++;
1156 
1157 	/* Now use mmap to map the library into memory. */
1158 	error = vm_mmap(file,
1159 			ELF_PAGESTART(eppnt->p_vaddr),
1160 			(eppnt->p_filesz +
1161 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1162 			PROT_READ | PROT_WRITE | PROT_EXEC,
1163 			MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1164 			(eppnt->p_offset -
1165 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1166 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1167 		goto out_free_ph;
1168 
1169 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1170 	if (padzero(elf_bss)) {
1171 		error = -EFAULT;
1172 		goto out_free_ph;
1173 	}
1174 
1175 	len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1176 			    ELF_MIN_ALIGN - 1);
1177 	bss = eppnt->p_memsz + eppnt->p_vaddr;
1178 	if (bss > len) {
1179 		error = vm_brk(len, bss - len);
1180 		if (error)
1181 			goto out_free_ph;
1182 	}
1183 	error = 0;
1184 
1185 out_free_ph:
1186 	kfree(elf_phdata);
1187 out:
1188 	return error;
1189 }
1190 #endif /* #ifdef CONFIG_USELIB */
1191 
1192 #ifdef CONFIG_ELF_CORE
1193 /*
1194  * ELF core dumper
1195  *
1196  * Modelled on fs/exec.c:aout_core_dump()
1197  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1198  */
1199 
1200 /*
1201  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1202  * that are useful for post-mortem analysis are included in every core dump.
1203  * In that way we ensure that the core dump is fully interpretable later
1204  * without matching up the same kernel and hardware config to see what PC values
1205  * meant. These special mappings include - vDSO, vsyscall, and other
1206  * architecture specific mappings
1207  */
1208 static bool always_dump_vma(struct vm_area_struct *vma)
1209 {
1210 	/* Any vsyscall mappings? */
1211 	if (vma == get_gate_vma(vma->vm_mm))
1212 		return true;
1213 
1214 	/*
1215 	 * Assume that all vmas with a .name op should always be dumped.
1216 	 * If this changes, a new vm_ops field can easily be added.
1217 	 */
1218 	if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1219 		return true;
1220 
1221 	/*
1222 	 * arch_vma_name() returns non-NULL for special architecture mappings,
1223 	 * such as vDSO sections.
1224 	 */
1225 	if (arch_vma_name(vma))
1226 		return true;
1227 
1228 	return false;
1229 }
1230 
1231 /*
1232  * Decide what to dump of a segment, part, all or none.
1233  */
1234 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1235 				   unsigned long mm_flags)
1236 {
1237 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1238 
1239 	/* always dump the vdso and vsyscall sections */
1240 	if (always_dump_vma(vma))
1241 		goto whole;
1242 
1243 	if (vma->vm_flags & VM_DONTDUMP)
1244 		return 0;
1245 
1246 	/* support for DAX */
1247 	if (vma_is_dax(vma)) {
1248 		if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1249 			goto whole;
1250 		if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1251 			goto whole;
1252 		return 0;
1253 	}
1254 
1255 	/* Hugetlb memory check */
1256 	if (vma->vm_flags & VM_HUGETLB) {
1257 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1258 			goto whole;
1259 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1260 			goto whole;
1261 		return 0;
1262 	}
1263 
1264 	/* Do not dump I/O mapped devices or special mappings */
1265 	if (vma->vm_flags & VM_IO)
1266 		return 0;
1267 
1268 	/* By default, dump shared memory if mapped from an anonymous file. */
1269 	if (vma->vm_flags & VM_SHARED) {
1270 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1271 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1272 			goto whole;
1273 		return 0;
1274 	}
1275 
1276 	/* Dump segments that have been written to.  */
1277 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1278 		goto whole;
1279 	if (vma->vm_file == NULL)
1280 		return 0;
1281 
1282 	if (FILTER(MAPPED_PRIVATE))
1283 		goto whole;
1284 
1285 	/*
1286 	 * If this looks like the beginning of a DSO or executable mapping,
1287 	 * check for an ELF header.  If we find one, dump the first page to
1288 	 * aid in determining what was mapped here.
1289 	 */
1290 	if (FILTER(ELF_HEADERS) &&
1291 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1292 		u32 __user *header = (u32 __user *) vma->vm_start;
1293 		u32 word;
1294 		mm_segment_t fs = get_fs();
1295 		/*
1296 		 * Doing it this way gets the constant folded by GCC.
1297 		 */
1298 		union {
1299 			u32 cmp;
1300 			char elfmag[SELFMAG];
1301 		} magic;
1302 		BUILD_BUG_ON(SELFMAG != sizeof word);
1303 		magic.elfmag[EI_MAG0] = ELFMAG0;
1304 		magic.elfmag[EI_MAG1] = ELFMAG1;
1305 		magic.elfmag[EI_MAG2] = ELFMAG2;
1306 		magic.elfmag[EI_MAG3] = ELFMAG3;
1307 		/*
1308 		 * Switch to the user "segment" for get_user(),
1309 		 * then put back what elf_core_dump() had in place.
1310 		 */
1311 		set_fs(USER_DS);
1312 		if (unlikely(get_user(word, header)))
1313 			word = 0;
1314 		set_fs(fs);
1315 		if (word == magic.cmp)
1316 			return PAGE_SIZE;
1317 	}
1318 
1319 #undef	FILTER
1320 
1321 	return 0;
1322 
1323 whole:
1324 	return vma->vm_end - vma->vm_start;
1325 }
1326 
1327 /* An ELF note in memory */
1328 struct memelfnote
1329 {
1330 	const char *name;
1331 	int type;
1332 	unsigned int datasz;
1333 	void *data;
1334 };
1335 
1336 static int notesize(struct memelfnote *en)
1337 {
1338 	int sz;
1339 
1340 	sz = sizeof(struct elf_note);
1341 	sz += roundup(strlen(en->name) + 1, 4);
1342 	sz += roundup(en->datasz, 4);
1343 
1344 	return sz;
1345 }
1346 
1347 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1348 {
1349 	struct elf_note en;
1350 	en.n_namesz = strlen(men->name) + 1;
1351 	en.n_descsz = men->datasz;
1352 	en.n_type = men->type;
1353 
1354 	return dump_emit(cprm, &en, sizeof(en)) &&
1355 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1356 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1357 }
1358 
1359 static void fill_elf_header(struct elfhdr *elf, int segs,
1360 			    u16 machine, u32 flags)
1361 {
1362 	memset(elf, 0, sizeof(*elf));
1363 
1364 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1365 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1366 	elf->e_ident[EI_DATA] = ELF_DATA;
1367 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1368 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1369 
1370 	elf->e_type = ET_CORE;
1371 	elf->e_machine = machine;
1372 	elf->e_version = EV_CURRENT;
1373 	elf->e_phoff = sizeof(struct elfhdr);
1374 	elf->e_flags = flags;
1375 	elf->e_ehsize = sizeof(struct elfhdr);
1376 	elf->e_phentsize = sizeof(struct elf_phdr);
1377 	elf->e_phnum = segs;
1378 
1379 	return;
1380 }
1381 
1382 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1383 {
1384 	phdr->p_type = PT_NOTE;
1385 	phdr->p_offset = offset;
1386 	phdr->p_vaddr = 0;
1387 	phdr->p_paddr = 0;
1388 	phdr->p_filesz = sz;
1389 	phdr->p_memsz = 0;
1390 	phdr->p_flags = 0;
1391 	phdr->p_align = 0;
1392 	return;
1393 }
1394 
1395 static void fill_note(struct memelfnote *note, const char *name, int type,
1396 		unsigned int sz, void *data)
1397 {
1398 	note->name = name;
1399 	note->type = type;
1400 	note->datasz = sz;
1401 	note->data = data;
1402 	return;
1403 }
1404 
1405 /*
1406  * fill up all the fields in prstatus from the given task struct, except
1407  * registers which need to be filled up separately.
1408  */
1409 static void fill_prstatus(struct elf_prstatus *prstatus,
1410 		struct task_struct *p, long signr)
1411 {
1412 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1413 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1414 	prstatus->pr_sighold = p->blocked.sig[0];
1415 	rcu_read_lock();
1416 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1417 	rcu_read_unlock();
1418 	prstatus->pr_pid = task_pid_vnr(p);
1419 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1420 	prstatus->pr_sid = task_session_vnr(p);
1421 	if (thread_group_leader(p)) {
1422 		struct task_cputime cputime;
1423 
1424 		/*
1425 		 * This is the record for the group leader.  It shows the
1426 		 * group-wide total, not its individual thread total.
1427 		 */
1428 		thread_group_cputime(p, &cputime);
1429 		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1430 		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1431 	} else {
1432 		cputime_t utime, stime;
1433 
1434 		task_cputime(p, &utime, &stime);
1435 		cputime_to_timeval(utime, &prstatus->pr_utime);
1436 		cputime_to_timeval(stime, &prstatus->pr_stime);
1437 	}
1438 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1439 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1440 }
1441 
1442 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1443 		       struct mm_struct *mm)
1444 {
1445 	const struct cred *cred;
1446 	unsigned int i, len;
1447 
1448 	/* first copy the parameters from user space */
1449 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1450 
1451 	len = mm->arg_end - mm->arg_start;
1452 	if (len >= ELF_PRARGSZ)
1453 		len = ELF_PRARGSZ-1;
1454 	if (copy_from_user(&psinfo->pr_psargs,
1455 		           (const char __user *)mm->arg_start, len))
1456 		return -EFAULT;
1457 	for(i = 0; i < len; i++)
1458 		if (psinfo->pr_psargs[i] == 0)
1459 			psinfo->pr_psargs[i] = ' ';
1460 	psinfo->pr_psargs[len] = 0;
1461 
1462 	rcu_read_lock();
1463 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1464 	rcu_read_unlock();
1465 	psinfo->pr_pid = task_pid_vnr(p);
1466 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1467 	psinfo->pr_sid = task_session_vnr(p);
1468 
1469 	i = p->state ? ffz(~p->state) + 1 : 0;
1470 	psinfo->pr_state = i;
1471 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1472 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1473 	psinfo->pr_nice = task_nice(p);
1474 	psinfo->pr_flag = p->flags;
1475 	rcu_read_lock();
1476 	cred = __task_cred(p);
1477 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1478 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1479 	rcu_read_unlock();
1480 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1481 
1482 	return 0;
1483 }
1484 
1485 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1486 {
1487 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1488 	int i = 0;
1489 	do
1490 		i += 2;
1491 	while (auxv[i - 2] != AT_NULL);
1492 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1493 }
1494 
1495 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1496 		const siginfo_t *siginfo)
1497 {
1498 	mm_segment_t old_fs = get_fs();
1499 	set_fs(KERNEL_DS);
1500 	copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1501 	set_fs(old_fs);
1502 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1503 }
1504 
1505 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1506 /*
1507  * Format of NT_FILE note:
1508  *
1509  * long count     -- how many files are mapped
1510  * long page_size -- units for file_ofs
1511  * array of [COUNT] elements of
1512  *   long start
1513  *   long end
1514  *   long file_ofs
1515  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1516  */
1517 static int fill_files_note(struct memelfnote *note)
1518 {
1519 	struct vm_area_struct *vma;
1520 	unsigned count, size, names_ofs, remaining, n;
1521 	user_long_t *data;
1522 	user_long_t *start_end_ofs;
1523 	char *name_base, *name_curpos;
1524 
1525 	/* *Estimated* file count and total data size needed */
1526 	count = current->mm->map_count;
1527 	size = count * 64;
1528 
1529 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1530  alloc:
1531 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1532 		return -EINVAL;
1533 	size = round_up(size, PAGE_SIZE);
1534 	data = vmalloc(size);
1535 	if (!data)
1536 		return -ENOMEM;
1537 
1538 	start_end_ofs = data + 2;
1539 	name_base = name_curpos = ((char *)data) + names_ofs;
1540 	remaining = size - names_ofs;
1541 	count = 0;
1542 	for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1543 		struct file *file;
1544 		const char *filename;
1545 
1546 		file = vma->vm_file;
1547 		if (!file)
1548 			continue;
1549 		filename = file_path(file, name_curpos, remaining);
1550 		if (IS_ERR(filename)) {
1551 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1552 				vfree(data);
1553 				size = size * 5 / 4;
1554 				goto alloc;
1555 			}
1556 			continue;
1557 		}
1558 
1559 		/* file_path() fills at the end, move name down */
1560 		/* n = strlen(filename) + 1: */
1561 		n = (name_curpos + remaining) - filename;
1562 		remaining = filename - name_curpos;
1563 		memmove(name_curpos, filename, n);
1564 		name_curpos += n;
1565 
1566 		*start_end_ofs++ = vma->vm_start;
1567 		*start_end_ofs++ = vma->vm_end;
1568 		*start_end_ofs++ = vma->vm_pgoff;
1569 		count++;
1570 	}
1571 
1572 	/* Now we know exact count of files, can store it */
1573 	data[0] = count;
1574 	data[1] = PAGE_SIZE;
1575 	/*
1576 	 * Count usually is less than current->mm->map_count,
1577 	 * we need to move filenames down.
1578 	 */
1579 	n = current->mm->map_count - count;
1580 	if (n != 0) {
1581 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1582 		memmove(name_base - shift_bytes, name_base,
1583 			name_curpos - name_base);
1584 		name_curpos -= shift_bytes;
1585 	}
1586 
1587 	size = name_curpos - (char *)data;
1588 	fill_note(note, "CORE", NT_FILE, size, data);
1589 	return 0;
1590 }
1591 
1592 #ifdef CORE_DUMP_USE_REGSET
1593 #include <linux/regset.h>
1594 
1595 struct elf_thread_core_info {
1596 	struct elf_thread_core_info *next;
1597 	struct task_struct *task;
1598 	struct elf_prstatus prstatus;
1599 	struct memelfnote notes[0];
1600 };
1601 
1602 struct elf_note_info {
1603 	struct elf_thread_core_info *thread;
1604 	struct memelfnote psinfo;
1605 	struct memelfnote signote;
1606 	struct memelfnote auxv;
1607 	struct memelfnote files;
1608 	user_siginfo_t csigdata;
1609 	size_t size;
1610 	int thread_notes;
1611 };
1612 
1613 /*
1614  * When a regset has a writeback hook, we call it on each thread before
1615  * dumping user memory.  On register window machines, this makes sure the
1616  * user memory backing the register data is up to date before we read it.
1617  */
1618 static void do_thread_regset_writeback(struct task_struct *task,
1619 				       const struct user_regset *regset)
1620 {
1621 	if (regset->writeback)
1622 		regset->writeback(task, regset, 1);
1623 }
1624 
1625 #ifndef PR_REG_SIZE
1626 #define PR_REG_SIZE(S) sizeof(S)
1627 #endif
1628 
1629 #ifndef PRSTATUS_SIZE
1630 #define PRSTATUS_SIZE(S) sizeof(S)
1631 #endif
1632 
1633 #ifndef PR_REG_PTR
1634 #define PR_REG_PTR(S) (&((S)->pr_reg))
1635 #endif
1636 
1637 #ifndef SET_PR_FPVALID
1638 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1639 #endif
1640 
1641 static int fill_thread_core_info(struct elf_thread_core_info *t,
1642 				 const struct user_regset_view *view,
1643 				 long signr, size_t *total)
1644 {
1645 	unsigned int i;
1646 
1647 	/*
1648 	 * NT_PRSTATUS is the one special case, because the regset data
1649 	 * goes into the pr_reg field inside the note contents, rather
1650 	 * than being the whole note contents.  We fill the reset in here.
1651 	 * We assume that regset 0 is NT_PRSTATUS.
1652 	 */
1653 	fill_prstatus(&t->prstatus, t->task, signr);
1654 	(void) view->regsets[0].get(t->task, &view->regsets[0],
1655 				    0, PR_REG_SIZE(t->prstatus.pr_reg),
1656 				    PR_REG_PTR(&t->prstatus), NULL);
1657 
1658 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1659 		  PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1660 	*total += notesize(&t->notes[0]);
1661 
1662 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1663 
1664 	/*
1665 	 * Each other regset might generate a note too.  For each regset
1666 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1667 	 * all zero and we'll know to skip writing it later.
1668 	 */
1669 	for (i = 1; i < view->n; ++i) {
1670 		const struct user_regset *regset = &view->regsets[i];
1671 		do_thread_regset_writeback(t->task, regset);
1672 		if (regset->core_note_type && regset->get &&
1673 		    (!regset->active || regset->active(t->task, regset))) {
1674 			int ret;
1675 			size_t size = regset->n * regset->size;
1676 			void *data = kmalloc(size, GFP_KERNEL);
1677 			if (unlikely(!data))
1678 				return 0;
1679 			ret = regset->get(t->task, regset,
1680 					  0, size, data, NULL);
1681 			if (unlikely(ret))
1682 				kfree(data);
1683 			else {
1684 				if (regset->core_note_type != NT_PRFPREG)
1685 					fill_note(&t->notes[i], "LINUX",
1686 						  regset->core_note_type,
1687 						  size, data);
1688 				else {
1689 					SET_PR_FPVALID(&t->prstatus, 1);
1690 					fill_note(&t->notes[i], "CORE",
1691 						  NT_PRFPREG, size, data);
1692 				}
1693 				*total += notesize(&t->notes[i]);
1694 			}
1695 		}
1696 	}
1697 
1698 	return 1;
1699 }
1700 
1701 static int fill_note_info(struct elfhdr *elf, int phdrs,
1702 			  struct elf_note_info *info,
1703 			  const siginfo_t *siginfo, struct pt_regs *regs)
1704 {
1705 	struct task_struct *dump_task = current;
1706 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1707 	struct elf_thread_core_info *t;
1708 	struct elf_prpsinfo *psinfo;
1709 	struct core_thread *ct;
1710 	unsigned int i;
1711 
1712 	info->size = 0;
1713 	info->thread = NULL;
1714 
1715 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1716 	if (psinfo == NULL) {
1717 		info->psinfo.data = NULL; /* So we don't free this wrongly */
1718 		return 0;
1719 	}
1720 
1721 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1722 
1723 	/*
1724 	 * Figure out how many notes we're going to need for each thread.
1725 	 */
1726 	info->thread_notes = 0;
1727 	for (i = 0; i < view->n; ++i)
1728 		if (view->regsets[i].core_note_type != 0)
1729 			++info->thread_notes;
1730 
1731 	/*
1732 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1733 	 * since it is our one special case.
1734 	 */
1735 	if (unlikely(info->thread_notes == 0) ||
1736 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1737 		WARN_ON(1);
1738 		return 0;
1739 	}
1740 
1741 	/*
1742 	 * Initialize the ELF file header.
1743 	 */
1744 	fill_elf_header(elf, phdrs,
1745 			view->e_machine, view->e_flags);
1746 
1747 	/*
1748 	 * Allocate a structure for each thread.
1749 	 */
1750 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1751 		t = kzalloc(offsetof(struct elf_thread_core_info,
1752 				     notes[info->thread_notes]),
1753 			    GFP_KERNEL);
1754 		if (unlikely(!t))
1755 			return 0;
1756 
1757 		t->task = ct->task;
1758 		if (ct->task == dump_task || !info->thread) {
1759 			t->next = info->thread;
1760 			info->thread = t;
1761 		} else {
1762 			/*
1763 			 * Make sure to keep the original task at
1764 			 * the head of the list.
1765 			 */
1766 			t->next = info->thread->next;
1767 			info->thread->next = t;
1768 		}
1769 	}
1770 
1771 	/*
1772 	 * Now fill in each thread's information.
1773 	 */
1774 	for (t = info->thread; t != NULL; t = t->next)
1775 		if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1776 			return 0;
1777 
1778 	/*
1779 	 * Fill in the two process-wide notes.
1780 	 */
1781 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1782 	info->size += notesize(&info->psinfo);
1783 
1784 	fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1785 	info->size += notesize(&info->signote);
1786 
1787 	fill_auxv_note(&info->auxv, current->mm);
1788 	info->size += notesize(&info->auxv);
1789 
1790 	if (fill_files_note(&info->files) == 0)
1791 		info->size += notesize(&info->files);
1792 
1793 	return 1;
1794 }
1795 
1796 static size_t get_note_info_size(struct elf_note_info *info)
1797 {
1798 	return info->size;
1799 }
1800 
1801 /*
1802  * Write all the notes for each thread.  When writing the first thread, the
1803  * process-wide notes are interleaved after the first thread-specific note.
1804  */
1805 static int write_note_info(struct elf_note_info *info,
1806 			   struct coredump_params *cprm)
1807 {
1808 	bool first = true;
1809 	struct elf_thread_core_info *t = info->thread;
1810 
1811 	do {
1812 		int i;
1813 
1814 		if (!writenote(&t->notes[0], cprm))
1815 			return 0;
1816 
1817 		if (first && !writenote(&info->psinfo, cprm))
1818 			return 0;
1819 		if (first && !writenote(&info->signote, cprm))
1820 			return 0;
1821 		if (first && !writenote(&info->auxv, cprm))
1822 			return 0;
1823 		if (first && info->files.data &&
1824 				!writenote(&info->files, cprm))
1825 			return 0;
1826 
1827 		for (i = 1; i < info->thread_notes; ++i)
1828 			if (t->notes[i].data &&
1829 			    !writenote(&t->notes[i], cprm))
1830 				return 0;
1831 
1832 		first = false;
1833 		t = t->next;
1834 	} while (t);
1835 
1836 	return 1;
1837 }
1838 
1839 static void free_note_info(struct elf_note_info *info)
1840 {
1841 	struct elf_thread_core_info *threads = info->thread;
1842 	while (threads) {
1843 		unsigned int i;
1844 		struct elf_thread_core_info *t = threads;
1845 		threads = t->next;
1846 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1847 		for (i = 1; i < info->thread_notes; ++i)
1848 			kfree(t->notes[i].data);
1849 		kfree(t);
1850 	}
1851 	kfree(info->psinfo.data);
1852 	vfree(info->files.data);
1853 }
1854 
1855 #else
1856 
1857 /* Here is the structure in which status of each thread is captured. */
1858 struct elf_thread_status
1859 {
1860 	struct list_head list;
1861 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1862 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1863 	struct task_struct *thread;
1864 #ifdef ELF_CORE_COPY_XFPREGS
1865 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1866 #endif
1867 	struct memelfnote notes[3];
1868 	int num_notes;
1869 };
1870 
1871 /*
1872  * In order to add the specific thread information for the elf file format,
1873  * we need to keep a linked list of every threads pr_status and then create
1874  * a single section for them in the final core file.
1875  */
1876 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1877 {
1878 	int sz = 0;
1879 	struct task_struct *p = t->thread;
1880 	t->num_notes = 0;
1881 
1882 	fill_prstatus(&t->prstatus, p, signr);
1883 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1884 
1885 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1886 		  &(t->prstatus));
1887 	t->num_notes++;
1888 	sz += notesize(&t->notes[0]);
1889 
1890 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1891 								&t->fpu))) {
1892 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1893 			  &(t->fpu));
1894 		t->num_notes++;
1895 		sz += notesize(&t->notes[1]);
1896 	}
1897 
1898 #ifdef ELF_CORE_COPY_XFPREGS
1899 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1900 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1901 			  sizeof(t->xfpu), &t->xfpu);
1902 		t->num_notes++;
1903 		sz += notesize(&t->notes[2]);
1904 	}
1905 #endif
1906 	return sz;
1907 }
1908 
1909 struct elf_note_info {
1910 	struct memelfnote *notes;
1911 	struct memelfnote *notes_files;
1912 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1913 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1914 	struct list_head thread_list;
1915 	elf_fpregset_t *fpu;
1916 #ifdef ELF_CORE_COPY_XFPREGS
1917 	elf_fpxregset_t *xfpu;
1918 #endif
1919 	user_siginfo_t csigdata;
1920 	int thread_status_size;
1921 	int numnote;
1922 };
1923 
1924 static int elf_note_info_init(struct elf_note_info *info)
1925 {
1926 	memset(info, 0, sizeof(*info));
1927 	INIT_LIST_HEAD(&info->thread_list);
1928 
1929 	/* Allocate space for ELF notes */
1930 	info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1931 	if (!info->notes)
1932 		return 0;
1933 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1934 	if (!info->psinfo)
1935 		return 0;
1936 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1937 	if (!info->prstatus)
1938 		return 0;
1939 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1940 	if (!info->fpu)
1941 		return 0;
1942 #ifdef ELF_CORE_COPY_XFPREGS
1943 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1944 	if (!info->xfpu)
1945 		return 0;
1946 #endif
1947 	return 1;
1948 }
1949 
1950 static int fill_note_info(struct elfhdr *elf, int phdrs,
1951 			  struct elf_note_info *info,
1952 			  const siginfo_t *siginfo, struct pt_regs *regs)
1953 {
1954 	struct list_head *t;
1955 	struct core_thread *ct;
1956 	struct elf_thread_status *ets;
1957 
1958 	if (!elf_note_info_init(info))
1959 		return 0;
1960 
1961 	for (ct = current->mm->core_state->dumper.next;
1962 					ct; ct = ct->next) {
1963 		ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1964 		if (!ets)
1965 			return 0;
1966 
1967 		ets->thread = ct->task;
1968 		list_add(&ets->list, &info->thread_list);
1969 	}
1970 
1971 	list_for_each(t, &info->thread_list) {
1972 		int sz;
1973 
1974 		ets = list_entry(t, struct elf_thread_status, list);
1975 		sz = elf_dump_thread_status(siginfo->si_signo, ets);
1976 		info->thread_status_size += sz;
1977 	}
1978 	/* now collect the dump for the current */
1979 	memset(info->prstatus, 0, sizeof(*info->prstatus));
1980 	fill_prstatus(info->prstatus, current, siginfo->si_signo);
1981 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1982 
1983 	/* Set up header */
1984 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1985 
1986 	/*
1987 	 * Set up the notes in similar form to SVR4 core dumps made
1988 	 * with info from their /proc.
1989 	 */
1990 
1991 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1992 		  sizeof(*info->prstatus), info->prstatus);
1993 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
1994 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1995 		  sizeof(*info->psinfo), info->psinfo);
1996 
1997 	fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1998 	fill_auxv_note(info->notes + 3, current->mm);
1999 	info->numnote = 4;
2000 
2001 	if (fill_files_note(info->notes + info->numnote) == 0) {
2002 		info->notes_files = info->notes + info->numnote;
2003 		info->numnote++;
2004 	}
2005 
2006 	/* Try to dump the FPU. */
2007 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2008 							       info->fpu);
2009 	if (info->prstatus->pr_fpvalid)
2010 		fill_note(info->notes + info->numnote++,
2011 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2012 #ifdef ELF_CORE_COPY_XFPREGS
2013 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
2014 		fill_note(info->notes + info->numnote++,
2015 			  "LINUX", ELF_CORE_XFPREG_TYPE,
2016 			  sizeof(*info->xfpu), info->xfpu);
2017 #endif
2018 
2019 	return 1;
2020 }
2021 
2022 static size_t get_note_info_size(struct elf_note_info *info)
2023 {
2024 	int sz = 0;
2025 	int i;
2026 
2027 	for (i = 0; i < info->numnote; i++)
2028 		sz += notesize(info->notes + i);
2029 
2030 	sz += info->thread_status_size;
2031 
2032 	return sz;
2033 }
2034 
2035 static int write_note_info(struct elf_note_info *info,
2036 			   struct coredump_params *cprm)
2037 {
2038 	int i;
2039 	struct list_head *t;
2040 
2041 	for (i = 0; i < info->numnote; i++)
2042 		if (!writenote(info->notes + i, cprm))
2043 			return 0;
2044 
2045 	/* write out the thread status notes section */
2046 	list_for_each(t, &info->thread_list) {
2047 		struct elf_thread_status *tmp =
2048 				list_entry(t, struct elf_thread_status, list);
2049 
2050 		for (i = 0; i < tmp->num_notes; i++)
2051 			if (!writenote(&tmp->notes[i], cprm))
2052 				return 0;
2053 	}
2054 
2055 	return 1;
2056 }
2057 
2058 static void free_note_info(struct elf_note_info *info)
2059 {
2060 	while (!list_empty(&info->thread_list)) {
2061 		struct list_head *tmp = info->thread_list.next;
2062 		list_del(tmp);
2063 		kfree(list_entry(tmp, struct elf_thread_status, list));
2064 	}
2065 
2066 	/* Free data possibly allocated by fill_files_note(): */
2067 	if (info->notes_files)
2068 		vfree(info->notes_files->data);
2069 
2070 	kfree(info->prstatus);
2071 	kfree(info->psinfo);
2072 	kfree(info->notes);
2073 	kfree(info->fpu);
2074 #ifdef ELF_CORE_COPY_XFPREGS
2075 	kfree(info->xfpu);
2076 #endif
2077 }
2078 
2079 #endif
2080 
2081 static struct vm_area_struct *first_vma(struct task_struct *tsk,
2082 					struct vm_area_struct *gate_vma)
2083 {
2084 	struct vm_area_struct *ret = tsk->mm->mmap;
2085 
2086 	if (ret)
2087 		return ret;
2088 	return gate_vma;
2089 }
2090 /*
2091  * Helper function for iterating across a vma list.  It ensures that the caller
2092  * will visit `gate_vma' prior to terminating the search.
2093  */
2094 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2095 					struct vm_area_struct *gate_vma)
2096 {
2097 	struct vm_area_struct *ret;
2098 
2099 	ret = this_vma->vm_next;
2100 	if (ret)
2101 		return ret;
2102 	if (this_vma == gate_vma)
2103 		return NULL;
2104 	return gate_vma;
2105 }
2106 
2107 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2108 			     elf_addr_t e_shoff, int segs)
2109 {
2110 	elf->e_shoff = e_shoff;
2111 	elf->e_shentsize = sizeof(*shdr4extnum);
2112 	elf->e_shnum = 1;
2113 	elf->e_shstrndx = SHN_UNDEF;
2114 
2115 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2116 
2117 	shdr4extnum->sh_type = SHT_NULL;
2118 	shdr4extnum->sh_size = elf->e_shnum;
2119 	shdr4extnum->sh_link = elf->e_shstrndx;
2120 	shdr4extnum->sh_info = segs;
2121 }
2122 
2123 /*
2124  * Actual dumper
2125  *
2126  * This is a two-pass process; first we find the offsets of the bits,
2127  * and then they are actually written out.  If we run out of core limit
2128  * we just truncate.
2129  */
2130 static int elf_core_dump(struct coredump_params *cprm)
2131 {
2132 	int has_dumped = 0;
2133 	mm_segment_t fs;
2134 	int segs, i;
2135 	size_t vma_data_size = 0;
2136 	struct vm_area_struct *vma, *gate_vma;
2137 	struct elfhdr *elf = NULL;
2138 	loff_t offset = 0, dataoff;
2139 	struct elf_note_info info = { };
2140 	struct elf_phdr *phdr4note = NULL;
2141 	struct elf_shdr *shdr4extnum = NULL;
2142 	Elf_Half e_phnum;
2143 	elf_addr_t e_shoff;
2144 	elf_addr_t *vma_filesz = NULL;
2145 
2146 	/*
2147 	 * We no longer stop all VM operations.
2148 	 *
2149 	 * This is because those proceses that could possibly change map_count
2150 	 * or the mmap / vma pages are now blocked in do_exit on current
2151 	 * finishing this core dump.
2152 	 *
2153 	 * Only ptrace can touch these memory addresses, but it doesn't change
2154 	 * the map_count or the pages allocated. So no possibility of crashing
2155 	 * exists while dumping the mm->vm_next areas to the core file.
2156 	 */
2157 
2158 	/* alloc memory for large data structures: too large to be on stack */
2159 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2160 	if (!elf)
2161 		goto out;
2162 	/*
2163 	 * The number of segs are recored into ELF header as 16bit value.
2164 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2165 	 */
2166 	segs = current->mm->map_count;
2167 	segs += elf_core_extra_phdrs();
2168 
2169 	gate_vma = get_gate_vma(current->mm);
2170 	if (gate_vma != NULL)
2171 		segs++;
2172 
2173 	/* for notes section */
2174 	segs++;
2175 
2176 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2177 	 * this, kernel supports extended numbering. Have a look at
2178 	 * include/linux/elf.h for further information. */
2179 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2180 
2181 	/*
2182 	 * Collect all the non-memory information about the process for the
2183 	 * notes.  This also sets up the file header.
2184 	 */
2185 	if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2186 		goto cleanup;
2187 
2188 	has_dumped = 1;
2189 
2190 	fs = get_fs();
2191 	set_fs(KERNEL_DS);
2192 
2193 	offset += sizeof(*elf);				/* Elf header */
2194 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2195 
2196 	/* Write notes phdr entry */
2197 	{
2198 		size_t sz = get_note_info_size(&info);
2199 
2200 		sz += elf_coredump_extra_notes_size();
2201 
2202 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2203 		if (!phdr4note)
2204 			goto end_coredump;
2205 
2206 		fill_elf_note_phdr(phdr4note, sz, offset);
2207 		offset += sz;
2208 	}
2209 
2210 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2211 
2212 	vma_filesz = kmalloc_array(segs - 1, sizeof(*vma_filesz), GFP_KERNEL);
2213 	if (!vma_filesz)
2214 		goto end_coredump;
2215 
2216 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2217 			vma = next_vma(vma, gate_vma)) {
2218 		unsigned long dump_size;
2219 
2220 		dump_size = vma_dump_size(vma, cprm->mm_flags);
2221 		vma_filesz[i++] = dump_size;
2222 		vma_data_size += dump_size;
2223 	}
2224 
2225 	offset += vma_data_size;
2226 	offset += elf_core_extra_data_size();
2227 	e_shoff = offset;
2228 
2229 	if (e_phnum == PN_XNUM) {
2230 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2231 		if (!shdr4extnum)
2232 			goto end_coredump;
2233 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2234 	}
2235 
2236 	offset = dataoff;
2237 
2238 	if (!dump_emit(cprm, elf, sizeof(*elf)))
2239 		goto end_coredump;
2240 
2241 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2242 		goto end_coredump;
2243 
2244 	/* Write program headers for segments dump */
2245 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2246 			vma = next_vma(vma, gate_vma)) {
2247 		struct elf_phdr phdr;
2248 
2249 		phdr.p_type = PT_LOAD;
2250 		phdr.p_offset = offset;
2251 		phdr.p_vaddr = vma->vm_start;
2252 		phdr.p_paddr = 0;
2253 		phdr.p_filesz = vma_filesz[i++];
2254 		phdr.p_memsz = vma->vm_end - vma->vm_start;
2255 		offset += phdr.p_filesz;
2256 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2257 		if (vma->vm_flags & VM_WRITE)
2258 			phdr.p_flags |= PF_W;
2259 		if (vma->vm_flags & VM_EXEC)
2260 			phdr.p_flags |= PF_X;
2261 		phdr.p_align = ELF_EXEC_PAGESIZE;
2262 
2263 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2264 			goto end_coredump;
2265 	}
2266 
2267 	if (!elf_core_write_extra_phdrs(cprm, offset))
2268 		goto end_coredump;
2269 
2270  	/* write out the notes section */
2271 	if (!write_note_info(&info, cprm))
2272 		goto end_coredump;
2273 
2274 	if (elf_coredump_extra_notes_write(cprm))
2275 		goto end_coredump;
2276 
2277 	/* Align to page */
2278 	if (!dump_skip(cprm, dataoff - cprm->file->f_pos))
2279 		goto end_coredump;
2280 
2281 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2282 			vma = next_vma(vma, gate_vma)) {
2283 		unsigned long addr;
2284 		unsigned long end;
2285 
2286 		end = vma->vm_start + vma_filesz[i++];
2287 
2288 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2289 			struct page *page;
2290 			int stop;
2291 
2292 			page = get_dump_page(addr);
2293 			if (page) {
2294 				void *kaddr = kmap(page);
2295 				stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2296 				kunmap(page);
2297 				put_page(page);
2298 			} else
2299 				stop = !dump_skip(cprm, PAGE_SIZE);
2300 			if (stop)
2301 				goto end_coredump;
2302 		}
2303 	}
2304 
2305 	if (!elf_core_write_extra_data(cprm))
2306 		goto end_coredump;
2307 
2308 	if (e_phnum == PN_XNUM) {
2309 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2310 			goto end_coredump;
2311 	}
2312 
2313 end_coredump:
2314 	set_fs(fs);
2315 
2316 cleanup:
2317 	free_note_info(&info);
2318 	kfree(shdr4extnum);
2319 	kfree(vma_filesz);
2320 	kfree(phdr4note);
2321 	kfree(elf);
2322 out:
2323 	return has_dumped;
2324 }
2325 
2326 #endif		/* CONFIG_ELF_CORE */
2327 
2328 static int __init init_elf_binfmt(void)
2329 {
2330 	register_binfmt(&elf_format);
2331 	return 0;
2332 }
2333 
2334 static void __exit exit_elf_binfmt(void)
2335 {
2336 	/* Remove the COFF and ELF loaders. */
2337 	unregister_binfmt(&elf_format);
2338 }
2339 
2340 core_initcall(init_elf_binfmt);
2341 module_exit(exit_elf_binfmt);
2342 MODULE_LICENSE("GPL");
2343