xref: /linux/fs/binfmt_elf.c (revision 2dbc0838bcf24ca59cabc3130cf3b1d6809cdcd4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/fs/binfmt_elf.c
4  *
5  * These are the functions used to load ELF format executables as used
6  * on SVr4 machines.  Information on the format may be found in the book
7  * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
8  * Tools".
9  *
10  * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
11  */
12 
13 #include <linux/module.h>
14 #include <linux/kernel.h>
15 #include <linux/fs.h>
16 #include <linux/mm.h>
17 #include <linux/mman.h>
18 #include <linux/errno.h>
19 #include <linux/signal.h>
20 #include <linux/binfmts.h>
21 #include <linux/string.h>
22 #include <linux/file.h>
23 #include <linux/slab.h>
24 #include <linux/personality.h>
25 #include <linux/elfcore.h>
26 #include <linux/init.h>
27 #include <linux/highuid.h>
28 #include <linux/compiler.h>
29 #include <linux/highmem.h>
30 #include <linux/pagemap.h>
31 #include <linux/vmalloc.h>
32 #include <linux/security.h>
33 #include <linux/random.h>
34 #include <linux/elf.h>
35 #include <linux/elf-randomize.h>
36 #include <linux/utsname.h>
37 #include <linux/coredump.h>
38 #include <linux/sched.h>
39 #include <linux/sched/coredump.h>
40 #include <linux/sched/task_stack.h>
41 #include <linux/sched/cputime.h>
42 #include <linux/cred.h>
43 #include <linux/dax.h>
44 #include <linux/uaccess.h>
45 #include <asm/param.h>
46 #include <asm/page.h>
47 
48 #ifndef user_long_t
49 #define user_long_t long
50 #endif
51 #ifndef user_siginfo_t
52 #define user_siginfo_t siginfo_t
53 #endif
54 
55 /* That's for binfmt_elf_fdpic to deal with */
56 #ifndef elf_check_fdpic
57 #define elf_check_fdpic(ex) false
58 #endif
59 
60 static int load_elf_binary(struct linux_binprm *bprm);
61 
62 #ifdef CONFIG_USELIB
63 static int load_elf_library(struct file *);
64 #else
65 #define load_elf_library NULL
66 #endif
67 
68 /*
69  * If we don't support core dumping, then supply a NULL so we
70  * don't even try.
71  */
72 #ifdef CONFIG_ELF_CORE
73 static int elf_core_dump(struct coredump_params *cprm);
74 #else
75 #define elf_core_dump	NULL
76 #endif
77 
78 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
79 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
80 #else
81 #define ELF_MIN_ALIGN	PAGE_SIZE
82 #endif
83 
84 #ifndef ELF_CORE_EFLAGS
85 #define ELF_CORE_EFLAGS	0
86 #endif
87 
88 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
89 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
90 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
91 
92 static struct linux_binfmt elf_format = {
93 	.module		= THIS_MODULE,
94 	.load_binary	= load_elf_binary,
95 	.load_shlib	= load_elf_library,
96 	.core_dump	= elf_core_dump,
97 	.min_coredump	= ELF_EXEC_PAGESIZE,
98 };
99 
100 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
101 
102 static int set_brk(unsigned long start, unsigned long end, int prot)
103 {
104 	start = ELF_PAGEALIGN(start);
105 	end = ELF_PAGEALIGN(end);
106 	if (end > start) {
107 		/*
108 		 * Map the last of the bss segment.
109 		 * If the header is requesting these pages to be
110 		 * executable, honour that (ppc32 needs this).
111 		 */
112 		int error = vm_brk_flags(start, end - start,
113 				prot & PROT_EXEC ? VM_EXEC : 0);
114 		if (error)
115 			return error;
116 	}
117 	current->mm->start_brk = current->mm->brk = end;
118 	return 0;
119 }
120 
121 /* We need to explicitly zero any fractional pages
122    after the data section (i.e. bss).  This would
123    contain the junk from the file that should not
124    be in memory
125  */
126 static int padzero(unsigned long elf_bss)
127 {
128 	unsigned long nbyte;
129 
130 	nbyte = ELF_PAGEOFFSET(elf_bss);
131 	if (nbyte) {
132 		nbyte = ELF_MIN_ALIGN - nbyte;
133 		if (clear_user((void __user *) elf_bss, nbyte))
134 			return -EFAULT;
135 	}
136 	return 0;
137 }
138 
139 /* Let's use some macros to make this stack manipulation a little clearer */
140 #ifdef CONFIG_STACK_GROWSUP
141 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
142 #define STACK_ROUND(sp, items) \
143 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
144 #define STACK_ALLOC(sp, len) ({ \
145 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
146 	old_sp; })
147 #else
148 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
149 #define STACK_ROUND(sp, items) \
150 	(((unsigned long) (sp - items)) &~ 15UL)
151 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
152 #endif
153 
154 #ifndef ELF_BASE_PLATFORM
155 /*
156  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
157  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
158  * will be copied to the user stack in the same manner as AT_PLATFORM.
159  */
160 #define ELF_BASE_PLATFORM NULL
161 #endif
162 
163 static int
164 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
165 		unsigned long load_addr, unsigned long interp_load_addr)
166 {
167 	unsigned long p = bprm->p;
168 	int argc = bprm->argc;
169 	int envc = bprm->envc;
170 	elf_addr_t __user *sp;
171 	elf_addr_t __user *u_platform;
172 	elf_addr_t __user *u_base_platform;
173 	elf_addr_t __user *u_rand_bytes;
174 	const char *k_platform = ELF_PLATFORM;
175 	const char *k_base_platform = ELF_BASE_PLATFORM;
176 	unsigned char k_rand_bytes[16];
177 	int items;
178 	elf_addr_t *elf_info;
179 	int ei_index = 0;
180 	const struct cred *cred = current_cred();
181 	struct vm_area_struct *vma;
182 
183 	/*
184 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
185 	 * evictions by the processes running on the same package. One
186 	 * thing we can do is to shuffle the initial stack for them.
187 	 */
188 
189 	p = arch_align_stack(p);
190 
191 	/*
192 	 * If this architecture has a platform capability string, copy it
193 	 * to userspace.  In some cases (Sparc), this info is impossible
194 	 * for userspace to get any other way, in others (i386) it is
195 	 * merely difficult.
196 	 */
197 	u_platform = NULL;
198 	if (k_platform) {
199 		size_t len = strlen(k_platform) + 1;
200 
201 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
202 		if (__copy_to_user(u_platform, k_platform, len))
203 			return -EFAULT;
204 	}
205 
206 	/*
207 	 * If this architecture has a "base" platform capability
208 	 * string, copy it to userspace.
209 	 */
210 	u_base_platform = NULL;
211 	if (k_base_platform) {
212 		size_t len = strlen(k_base_platform) + 1;
213 
214 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
215 		if (__copy_to_user(u_base_platform, k_base_platform, len))
216 			return -EFAULT;
217 	}
218 
219 	/*
220 	 * Generate 16 random bytes for userspace PRNG seeding.
221 	 */
222 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
223 	u_rand_bytes = (elf_addr_t __user *)
224 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
225 	if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
226 		return -EFAULT;
227 
228 	/* Create the ELF interpreter info */
229 	elf_info = (elf_addr_t *)current->mm->saved_auxv;
230 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
231 #define NEW_AUX_ENT(id, val) \
232 	do { \
233 		elf_info[ei_index++] = id; \
234 		elf_info[ei_index++] = val; \
235 	} while (0)
236 
237 #ifdef ARCH_DLINFO
238 	/*
239 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
240 	 * AUXV.
241 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
242 	 * ARCH_DLINFO changes
243 	 */
244 	ARCH_DLINFO;
245 #endif
246 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
247 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
248 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
249 	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
250 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
251 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
252 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
253 	NEW_AUX_ENT(AT_FLAGS, 0);
254 	NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
255 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
256 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
257 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
258 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
259 	NEW_AUX_ENT(AT_SECURE, bprm->secureexec);
260 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
261 #ifdef ELF_HWCAP2
262 	NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
263 #endif
264 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
265 	if (k_platform) {
266 		NEW_AUX_ENT(AT_PLATFORM,
267 			    (elf_addr_t)(unsigned long)u_platform);
268 	}
269 	if (k_base_platform) {
270 		NEW_AUX_ENT(AT_BASE_PLATFORM,
271 			    (elf_addr_t)(unsigned long)u_base_platform);
272 	}
273 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
274 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
275 	}
276 #undef NEW_AUX_ENT
277 	/* AT_NULL is zero; clear the rest too */
278 	memset(&elf_info[ei_index], 0,
279 	       sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
280 
281 	/* And advance past the AT_NULL entry.  */
282 	ei_index += 2;
283 
284 	sp = STACK_ADD(p, ei_index);
285 
286 	items = (argc + 1) + (envc + 1) + 1;
287 	bprm->p = STACK_ROUND(sp, items);
288 
289 	/* Point sp at the lowest address on the stack */
290 #ifdef CONFIG_STACK_GROWSUP
291 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
292 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
293 #else
294 	sp = (elf_addr_t __user *)bprm->p;
295 #endif
296 
297 
298 	/*
299 	 * Grow the stack manually; some architectures have a limit on how
300 	 * far ahead a user-space access may be in order to grow the stack.
301 	 */
302 	vma = find_extend_vma(current->mm, bprm->p);
303 	if (!vma)
304 		return -EFAULT;
305 
306 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
307 	if (__put_user(argc, sp++))
308 		return -EFAULT;
309 
310 	/* Populate list of argv pointers back to argv strings. */
311 	p = current->mm->arg_end = current->mm->arg_start;
312 	while (argc-- > 0) {
313 		size_t len;
314 		if (__put_user((elf_addr_t)p, sp++))
315 			return -EFAULT;
316 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
317 		if (!len || len > MAX_ARG_STRLEN)
318 			return -EINVAL;
319 		p += len;
320 	}
321 	if (__put_user(0, sp++))
322 		return -EFAULT;
323 	current->mm->arg_end = p;
324 
325 	/* Populate list of envp pointers back to envp strings. */
326 	current->mm->env_end = current->mm->env_start = p;
327 	while (envc-- > 0) {
328 		size_t len;
329 		if (__put_user((elf_addr_t)p, sp++))
330 			return -EFAULT;
331 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
332 		if (!len || len > MAX_ARG_STRLEN)
333 			return -EINVAL;
334 		p += len;
335 	}
336 	if (__put_user(0, sp++))
337 		return -EFAULT;
338 	current->mm->env_end = p;
339 
340 	/* Put the elf_info on the stack in the right place.  */
341 	if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
342 		return -EFAULT;
343 	return 0;
344 }
345 
346 #ifndef elf_map
347 
348 static unsigned long elf_map(struct file *filep, unsigned long addr,
349 		const struct elf_phdr *eppnt, int prot, int type,
350 		unsigned long total_size)
351 {
352 	unsigned long map_addr;
353 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
354 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
355 	addr = ELF_PAGESTART(addr);
356 	size = ELF_PAGEALIGN(size);
357 
358 	/* mmap() will return -EINVAL if given a zero size, but a
359 	 * segment with zero filesize is perfectly valid */
360 	if (!size)
361 		return addr;
362 
363 	/*
364 	* total_size is the size of the ELF (interpreter) image.
365 	* The _first_ mmap needs to know the full size, otherwise
366 	* randomization might put this image into an overlapping
367 	* position with the ELF binary image. (since size < total_size)
368 	* So we first map the 'big' image - and unmap the remainder at
369 	* the end. (which unmap is needed for ELF images with holes.)
370 	*/
371 	if (total_size) {
372 		total_size = ELF_PAGEALIGN(total_size);
373 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
374 		if (!BAD_ADDR(map_addr))
375 			vm_munmap(map_addr+size, total_size-size);
376 	} else
377 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
378 
379 	if ((type & MAP_FIXED_NOREPLACE) &&
380 	    PTR_ERR((void *)map_addr) == -EEXIST)
381 		pr_info("%d (%s): Uhuuh, elf segment at %px requested but the memory is mapped already\n",
382 			task_pid_nr(current), current->comm, (void *)addr);
383 
384 	return(map_addr);
385 }
386 
387 #endif /* !elf_map */
388 
389 static unsigned long total_mapping_size(const struct elf_phdr *cmds, int nr)
390 {
391 	int i, first_idx = -1, last_idx = -1;
392 
393 	for (i = 0; i < nr; i++) {
394 		if (cmds[i].p_type == PT_LOAD) {
395 			last_idx = i;
396 			if (first_idx == -1)
397 				first_idx = i;
398 		}
399 	}
400 	if (first_idx == -1)
401 		return 0;
402 
403 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
404 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
405 }
406 
407 /**
408  * load_elf_phdrs() - load ELF program headers
409  * @elf_ex:   ELF header of the binary whose program headers should be loaded
410  * @elf_file: the opened ELF binary file
411  *
412  * Loads ELF program headers from the binary file elf_file, which has the ELF
413  * header pointed to by elf_ex, into a newly allocated array. The caller is
414  * responsible for freeing the allocated data. Returns an ERR_PTR upon failure.
415  */
416 static struct elf_phdr *load_elf_phdrs(const struct elfhdr *elf_ex,
417 				       struct file *elf_file)
418 {
419 	struct elf_phdr *elf_phdata = NULL;
420 	int retval, err = -1;
421 	loff_t pos = elf_ex->e_phoff;
422 	unsigned int size;
423 
424 	/*
425 	 * If the size of this structure has changed, then punt, since
426 	 * we will be doing the wrong thing.
427 	 */
428 	if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
429 		goto out;
430 
431 	/* Sanity check the number of program headers... */
432 	/* ...and their total size. */
433 	size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
434 	if (size == 0 || size > 65536 || size > ELF_MIN_ALIGN)
435 		goto out;
436 
437 	elf_phdata = kmalloc(size, GFP_KERNEL);
438 	if (!elf_phdata)
439 		goto out;
440 
441 	/* Read in the program headers */
442 	retval = kernel_read(elf_file, elf_phdata, size, &pos);
443 	if (retval != size) {
444 		err = (retval < 0) ? retval : -EIO;
445 		goto out;
446 	}
447 
448 	/* Success! */
449 	err = 0;
450 out:
451 	if (err) {
452 		kfree(elf_phdata);
453 		elf_phdata = NULL;
454 	}
455 	return elf_phdata;
456 }
457 
458 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
459 
460 /**
461  * struct arch_elf_state - arch-specific ELF loading state
462  *
463  * This structure is used to preserve architecture specific data during
464  * the loading of an ELF file, throughout the checking of architecture
465  * specific ELF headers & through to the point where the ELF load is
466  * known to be proceeding (ie. SET_PERSONALITY).
467  *
468  * This implementation is a dummy for architectures which require no
469  * specific state.
470  */
471 struct arch_elf_state {
472 };
473 
474 #define INIT_ARCH_ELF_STATE {}
475 
476 /**
477  * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
478  * @ehdr:	The main ELF header
479  * @phdr:	The program header to check
480  * @elf:	The open ELF file
481  * @is_interp:	True if the phdr is from the interpreter of the ELF being
482  *		loaded, else false.
483  * @state:	Architecture-specific state preserved throughout the process
484  *		of loading the ELF.
485  *
486  * Inspects the program header phdr to validate its correctness and/or
487  * suitability for the system. Called once per ELF program header in the
488  * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
489  * interpreter.
490  *
491  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
492  *         with that return code.
493  */
494 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
495 				   struct elf_phdr *phdr,
496 				   struct file *elf, bool is_interp,
497 				   struct arch_elf_state *state)
498 {
499 	/* Dummy implementation, always proceed */
500 	return 0;
501 }
502 
503 /**
504  * arch_check_elf() - check an ELF executable
505  * @ehdr:	The main ELF header
506  * @has_interp:	True if the ELF has an interpreter, else false.
507  * @interp_ehdr: The interpreter's ELF header
508  * @state:	Architecture-specific state preserved throughout the process
509  *		of loading the ELF.
510  *
511  * Provides a final opportunity for architecture code to reject the loading
512  * of the ELF & cause an exec syscall to return an error. This is called after
513  * all program headers to be checked by arch_elf_pt_proc have been.
514  *
515  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
516  *         with that return code.
517  */
518 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
519 				 struct elfhdr *interp_ehdr,
520 				 struct arch_elf_state *state)
521 {
522 	/* Dummy implementation, always proceed */
523 	return 0;
524 }
525 
526 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
527 
528 static inline int make_prot(u32 p_flags)
529 {
530 	int prot = 0;
531 
532 	if (p_flags & PF_R)
533 		prot |= PROT_READ;
534 	if (p_flags & PF_W)
535 		prot |= PROT_WRITE;
536 	if (p_flags & PF_X)
537 		prot |= PROT_EXEC;
538 	return prot;
539 }
540 
541 /* This is much more generalized than the library routine read function,
542    so we keep this separate.  Technically the library read function
543    is only provided so that we can read a.out libraries that have
544    an ELF header */
545 
546 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
547 		struct file *interpreter, unsigned long *interp_map_addr,
548 		unsigned long no_base, struct elf_phdr *interp_elf_phdata)
549 {
550 	struct elf_phdr *eppnt;
551 	unsigned long load_addr = 0;
552 	int load_addr_set = 0;
553 	unsigned long last_bss = 0, elf_bss = 0;
554 	int bss_prot = 0;
555 	unsigned long error = ~0UL;
556 	unsigned long total_size;
557 	int i;
558 
559 	/* First of all, some simple consistency checks */
560 	if (interp_elf_ex->e_type != ET_EXEC &&
561 	    interp_elf_ex->e_type != ET_DYN)
562 		goto out;
563 	if (!elf_check_arch(interp_elf_ex) ||
564 	    elf_check_fdpic(interp_elf_ex))
565 		goto out;
566 	if (!interpreter->f_op->mmap)
567 		goto out;
568 
569 	total_size = total_mapping_size(interp_elf_phdata,
570 					interp_elf_ex->e_phnum);
571 	if (!total_size) {
572 		error = -EINVAL;
573 		goto out;
574 	}
575 
576 	eppnt = interp_elf_phdata;
577 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
578 		if (eppnt->p_type == PT_LOAD) {
579 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
580 			int elf_prot = make_prot(eppnt->p_flags);
581 			unsigned long vaddr = 0;
582 			unsigned long k, map_addr;
583 
584 			vaddr = eppnt->p_vaddr;
585 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
586 				elf_type |= MAP_FIXED_NOREPLACE;
587 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
588 				load_addr = -vaddr;
589 
590 			map_addr = elf_map(interpreter, load_addr + vaddr,
591 					eppnt, elf_prot, elf_type, total_size);
592 			total_size = 0;
593 			if (!*interp_map_addr)
594 				*interp_map_addr = map_addr;
595 			error = map_addr;
596 			if (BAD_ADDR(map_addr))
597 				goto out;
598 
599 			if (!load_addr_set &&
600 			    interp_elf_ex->e_type == ET_DYN) {
601 				load_addr = map_addr - ELF_PAGESTART(vaddr);
602 				load_addr_set = 1;
603 			}
604 
605 			/*
606 			 * Check to see if the section's size will overflow the
607 			 * allowed task size. Note that p_filesz must always be
608 			 * <= p_memsize so it's only necessary to check p_memsz.
609 			 */
610 			k = load_addr + eppnt->p_vaddr;
611 			if (BAD_ADDR(k) ||
612 			    eppnt->p_filesz > eppnt->p_memsz ||
613 			    eppnt->p_memsz > TASK_SIZE ||
614 			    TASK_SIZE - eppnt->p_memsz < k) {
615 				error = -ENOMEM;
616 				goto out;
617 			}
618 
619 			/*
620 			 * Find the end of the file mapping for this phdr, and
621 			 * keep track of the largest address we see for this.
622 			 */
623 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
624 			if (k > elf_bss)
625 				elf_bss = k;
626 
627 			/*
628 			 * Do the same thing for the memory mapping - between
629 			 * elf_bss and last_bss is the bss section.
630 			 */
631 			k = load_addr + eppnt->p_vaddr + eppnt->p_memsz;
632 			if (k > last_bss) {
633 				last_bss = k;
634 				bss_prot = elf_prot;
635 			}
636 		}
637 	}
638 
639 	/*
640 	 * Now fill out the bss section: first pad the last page from
641 	 * the file up to the page boundary, and zero it from elf_bss
642 	 * up to the end of the page.
643 	 */
644 	if (padzero(elf_bss)) {
645 		error = -EFAULT;
646 		goto out;
647 	}
648 	/*
649 	 * Next, align both the file and mem bss up to the page size,
650 	 * since this is where elf_bss was just zeroed up to, and where
651 	 * last_bss will end after the vm_brk_flags() below.
652 	 */
653 	elf_bss = ELF_PAGEALIGN(elf_bss);
654 	last_bss = ELF_PAGEALIGN(last_bss);
655 	/* Finally, if there is still more bss to allocate, do it. */
656 	if (last_bss > elf_bss) {
657 		error = vm_brk_flags(elf_bss, last_bss - elf_bss,
658 				bss_prot & PROT_EXEC ? VM_EXEC : 0);
659 		if (error)
660 			goto out;
661 	}
662 
663 	error = load_addr;
664 out:
665 	return error;
666 }
667 
668 /*
669  * These are the functions used to load ELF style executables and shared
670  * libraries.  There is no binary dependent code anywhere else.
671  */
672 
673 #ifndef STACK_RND_MASK
674 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))	/* 8MB of VA */
675 #endif
676 
677 static unsigned long randomize_stack_top(unsigned long stack_top)
678 {
679 	unsigned long random_variable = 0;
680 
681 	if (current->flags & PF_RANDOMIZE) {
682 		random_variable = get_random_long();
683 		random_variable &= STACK_RND_MASK;
684 		random_variable <<= PAGE_SHIFT;
685 	}
686 #ifdef CONFIG_STACK_GROWSUP
687 	return PAGE_ALIGN(stack_top) + random_variable;
688 #else
689 	return PAGE_ALIGN(stack_top) - random_variable;
690 #endif
691 }
692 
693 static int load_elf_binary(struct linux_binprm *bprm)
694 {
695 	struct file *interpreter = NULL; /* to shut gcc up */
696  	unsigned long load_addr = 0, load_bias = 0;
697 	int load_addr_set = 0;
698 	unsigned long error;
699 	struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
700 	unsigned long elf_bss, elf_brk;
701 	int bss_prot = 0;
702 	int retval, i;
703 	unsigned long elf_entry;
704 	unsigned long interp_load_addr = 0;
705 	unsigned long start_code, end_code, start_data, end_data;
706 	unsigned long reloc_func_desc __maybe_unused = 0;
707 	int executable_stack = EXSTACK_DEFAULT;
708 	struct {
709 		struct elfhdr elf_ex;
710 		struct elfhdr interp_elf_ex;
711 	} *loc;
712 	struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
713 	struct pt_regs *regs;
714 
715 	loc = kmalloc(sizeof(*loc), GFP_KERNEL);
716 	if (!loc) {
717 		retval = -ENOMEM;
718 		goto out_ret;
719 	}
720 
721 	/* Get the exec-header */
722 	loc->elf_ex = *((struct elfhdr *)bprm->buf);
723 
724 	retval = -ENOEXEC;
725 	/* First of all, some simple consistency checks */
726 	if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
727 		goto out;
728 
729 	if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
730 		goto out;
731 	if (!elf_check_arch(&loc->elf_ex))
732 		goto out;
733 	if (elf_check_fdpic(&loc->elf_ex))
734 		goto out;
735 	if (!bprm->file->f_op->mmap)
736 		goto out;
737 
738 	elf_phdata = load_elf_phdrs(&loc->elf_ex, bprm->file);
739 	if (!elf_phdata)
740 		goto out;
741 
742 	elf_ppnt = elf_phdata;
743 	for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
744 		char *elf_interpreter;
745 		loff_t pos;
746 
747 		if (elf_ppnt->p_type != PT_INTERP)
748 			continue;
749 
750 		/*
751 		 * This is the program interpreter used for shared libraries -
752 		 * for now assume that this is an a.out format binary.
753 		 */
754 		retval = -ENOEXEC;
755 		if (elf_ppnt->p_filesz > PATH_MAX || elf_ppnt->p_filesz < 2)
756 			goto out_free_ph;
757 
758 		retval = -ENOMEM;
759 		elf_interpreter = kmalloc(elf_ppnt->p_filesz, GFP_KERNEL);
760 		if (!elf_interpreter)
761 			goto out_free_ph;
762 
763 		pos = elf_ppnt->p_offset;
764 		retval = kernel_read(bprm->file, elf_interpreter,
765 				     elf_ppnt->p_filesz, &pos);
766 		if (retval != elf_ppnt->p_filesz) {
767 			if (retval >= 0)
768 				retval = -EIO;
769 			goto out_free_interp;
770 		}
771 		/* make sure path is NULL terminated */
772 		retval = -ENOEXEC;
773 		if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
774 			goto out_free_interp;
775 
776 		interpreter = open_exec(elf_interpreter);
777 		kfree(elf_interpreter);
778 		retval = PTR_ERR(interpreter);
779 		if (IS_ERR(interpreter))
780 			goto out_free_ph;
781 
782 		/*
783 		 * If the binary is not readable then enforce mm->dumpable = 0
784 		 * regardless of the interpreter's permissions.
785 		 */
786 		would_dump(bprm, interpreter);
787 
788 		/* Get the exec headers */
789 		pos = 0;
790 		retval = kernel_read(interpreter, &loc->interp_elf_ex,
791 				     sizeof(loc->interp_elf_ex), &pos);
792 		if (retval != sizeof(loc->interp_elf_ex)) {
793 			if (retval >= 0)
794 				retval = -EIO;
795 			goto out_free_dentry;
796 		}
797 
798 		break;
799 
800 out_free_interp:
801 		kfree(elf_interpreter);
802 		goto out_free_ph;
803 	}
804 
805 	elf_ppnt = elf_phdata;
806 	for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
807 		switch (elf_ppnt->p_type) {
808 		case PT_GNU_STACK:
809 			if (elf_ppnt->p_flags & PF_X)
810 				executable_stack = EXSTACK_ENABLE_X;
811 			else
812 				executable_stack = EXSTACK_DISABLE_X;
813 			break;
814 
815 		case PT_LOPROC ... PT_HIPROC:
816 			retval = arch_elf_pt_proc(&loc->elf_ex, elf_ppnt,
817 						  bprm->file, false,
818 						  &arch_state);
819 			if (retval)
820 				goto out_free_dentry;
821 			break;
822 		}
823 
824 	/* Some simple consistency checks for the interpreter */
825 	if (interpreter) {
826 		retval = -ELIBBAD;
827 		/* Not an ELF interpreter */
828 		if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
829 			goto out_free_dentry;
830 		/* Verify the interpreter has a valid arch */
831 		if (!elf_check_arch(&loc->interp_elf_ex) ||
832 		    elf_check_fdpic(&loc->interp_elf_ex))
833 			goto out_free_dentry;
834 
835 		/* Load the interpreter program headers */
836 		interp_elf_phdata = load_elf_phdrs(&loc->interp_elf_ex,
837 						   interpreter);
838 		if (!interp_elf_phdata)
839 			goto out_free_dentry;
840 
841 		/* Pass PT_LOPROC..PT_HIPROC headers to arch code */
842 		elf_ppnt = interp_elf_phdata;
843 		for (i = 0; i < loc->interp_elf_ex.e_phnum; i++, elf_ppnt++)
844 			switch (elf_ppnt->p_type) {
845 			case PT_LOPROC ... PT_HIPROC:
846 				retval = arch_elf_pt_proc(&loc->interp_elf_ex,
847 							  elf_ppnt, interpreter,
848 							  true, &arch_state);
849 				if (retval)
850 					goto out_free_dentry;
851 				break;
852 			}
853 	}
854 
855 	/*
856 	 * Allow arch code to reject the ELF at this point, whilst it's
857 	 * still possible to return an error to the code that invoked
858 	 * the exec syscall.
859 	 */
860 	retval = arch_check_elf(&loc->elf_ex,
861 				!!interpreter, &loc->interp_elf_ex,
862 				&arch_state);
863 	if (retval)
864 		goto out_free_dentry;
865 
866 	/* Flush all traces of the currently running executable */
867 	retval = flush_old_exec(bprm);
868 	if (retval)
869 		goto out_free_dentry;
870 
871 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
872 	   may depend on the personality.  */
873 	SET_PERSONALITY2(loc->elf_ex, &arch_state);
874 	if (elf_read_implies_exec(loc->elf_ex, executable_stack))
875 		current->personality |= READ_IMPLIES_EXEC;
876 
877 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
878 		current->flags |= PF_RANDOMIZE;
879 
880 	setup_new_exec(bprm);
881 	install_exec_creds(bprm);
882 
883 	/* Do this so that we can load the interpreter, if need be.  We will
884 	   change some of these later */
885 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
886 				 executable_stack);
887 	if (retval < 0)
888 		goto out_free_dentry;
889 
890 	elf_bss = 0;
891 	elf_brk = 0;
892 
893 	start_code = ~0UL;
894 	end_code = 0;
895 	start_data = 0;
896 	end_data = 0;
897 
898 	/* Now we do a little grungy work by mmapping the ELF image into
899 	   the correct location in memory. */
900 	for(i = 0, elf_ppnt = elf_phdata;
901 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
902 		int elf_prot, elf_flags, elf_fixed = MAP_FIXED_NOREPLACE;
903 		unsigned long k, vaddr;
904 		unsigned long total_size = 0;
905 
906 		if (elf_ppnt->p_type != PT_LOAD)
907 			continue;
908 
909 		if (unlikely (elf_brk > elf_bss)) {
910 			unsigned long nbyte;
911 
912 			/* There was a PT_LOAD segment with p_memsz > p_filesz
913 			   before this one. Map anonymous pages, if needed,
914 			   and clear the area.  */
915 			retval = set_brk(elf_bss + load_bias,
916 					 elf_brk + load_bias,
917 					 bss_prot);
918 			if (retval)
919 				goto out_free_dentry;
920 			nbyte = ELF_PAGEOFFSET(elf_bss);
921 			if (nbyte) {
922 				nbyte = ELF_MIN_ALIGN - nbyte;
923 				if (nbyte > elf_brk - elf_bss)
924 					nbyte = elf_brk - elf_bss;
925 				if (clear_user((void __user *)elf_bss +
926 							load_bias, nbyte)) {
927 					/*
928 					 * This bss-zeroing can fail if the ELF
929 					 * file specifies odd protections. So
930 					 * we don't check the return value
931 					 */
932 				}
933 			}
934 
935 			/*
936 			 * Some binaries have overlapping elf segments and then
937 			 * we have to forcefully map over an existing mapping
938 			 * e.g. over this newly established brk mapping.
939 			 */
940 			elf_fixed = MAP_FIXED;
941 		}
942 
943 		elf_prot = make_prot(elf_ppnt->p_flags);
944 
945 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
946 
947 		vaddr = elf_ppnt->p_vaddr;
948 		/*
949 		 * If we are loading ET_EXEC or we have already performed
950 		 * the ET_DYN load_addr calculations, proceed normally.
951 		 */
952 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
953 			elf_flags |= elf_fixed;
954 		} else if (loc->elf_ex.e_type == ET_DYN) {
955 			/*
956 			 * This logic is run once for the first LOAD Program
957 			 * Header for ET_DYN binaries to calculate the
958 			 * randomization (load_bias) for all the LOAD
959 			 * Program Headers, and to calculate the entire
960 			 * size of the ELF mapping (total_size). (Note that
961 			 * load_addr_set is set to true later once the
962 			 * initial mapping is performed.)
963 			 *
964 			 * There are effectively two types of ET_DYN
965 			 * binaries: programs (i.e. PIE: ET_DYN with INTERP)
966 			 * and loaders (ET_DYN without INTERP, since they
967 			 * _are_ the ELF interpreter). The loaders must
968 			 * be loaded away from programs since the program
969 			 * may otherwise collide with the loader (especially
970 			 * for ET_EXEC which does not have a randomized
971 			 * position). For example to handle invocations of
972 			 * "./ld.so someprog" to test out a new version of
973 			 * the loader, the subsequent program that the
974 			 * loader loads must avoid the loader itself, so
975 			 * they cannot share the same load range. Sufficient
976 			 * room for the brk must be allocated with the
977 			 * loader as well, since brk must be available with
978 			 * the loader.
979 			 *
980 			 * Therefore, programs are loaded offset from
981 			 * ELF_ET_DYN_BASE and loaders are loaded into the
982 			 * independently randomized mmap region (0 load_bias
983 			 * without MAP_FIXED).
984 			 */
985 			if (interpreter) {
986 				load_bias = ELF_ET_DYN_BASE;
987 				if (current->flags & PF_RANDOMIZE)
988 					load_bias += arch_mmap_rnd();
989 				elf_flags |= elf_fixed;
990 			} else
991 				load_bias = 0;
992 
993 			/*
994 			 * Since load_bias is used for all subsequent loading
995 			 * calculations, we must lower it by the first vaddr
996 			 * so that the remaining calculations based on the
997 			 * ELF vaddrs will be correctly offset. The result
998 			 * is then page aligned.
999 			 */
1000 			load_bias = ELF_PAGESTART(load_bias - vaddr);
1001 
1002 			total_size = total_mapping_size(elf_phdata,
1003 							loc->elf_ex.e_phnum);
1004 			if (!total_size) {
1005 				retval = -EINVAL;
1006 				goto out_free_dentry;
1007 			}
1008 		}
1009 
1010 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
1011 				elf_prot, elf_flags, total_size);
1012 		if (BAD_ADDR(error)) {
1013 			retval = IS_ERR((void *)error) ?
1014 				PTR_ERR((void*)error) : -EINVAL;
1015 			goto out_free_dentry;
1016 		}
1017 
1018 		if (!load_addr_set) {
1019 			load_addr_set = 1;
1020 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
1021 			if (loc->elf_ex.e_type == ET_DYN) {
1022 				load_bias += error -
1023 				             ELF_PAGESTART(load_bias + vaddr);
1024 				load_addr += load_bias;
1025 				reloc_func_desc = load_bias;
1026 			}
1027 		}
1028 		k = elf_ppnt->p_vaddr;
1029 		if (k < start_code)
1030 			start_code = k;
1031 		if (start_data < k)
1032 			start_data = k;
1033 
1034 		/*
1035 		 * Check to see if the section's size will overflow the
1036 		 * allowed task size. Note that p_filesz must always be
1037 		 * <= p_memsz so it is only necessary to check p_memsz.
1038 		 */
1039 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
1040 		    elf_ppnt->p_memsz > TASK_SIZE ||
1041 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
1042 			/* set_brk can never work. Avoid overflows. */
1043 			retval = -EINVAL;
1044 			goto out_free_dentry;
1045 		}
1046 
1047 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
1048 
1049 		if (k > elf_bss)
1050 			elf_bss = k;
1051 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
1052 			end_code = k;
1053 		if (end_data < k)
1054 			end_data = k;
1055 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
1056 		if (k > elf_brk) {
1057 			bss_prot = elf_prot;
1058 			elf_brk = k;
1059 		}
1060 	}
1061 
1062 	loc->elf_ex.e_entry += load_bias;
1063 	elf_bss += load_bias;
1064 	elf_brk += load_bias;
1065 	start_code += load_bias;
1066 	end_code += load_bias;
1067 	start_data += load_bias;
1068 	end_data += load_bias;
1069 
1070 	/* Calling set_brk effectively mmaps the pages that we need
1071 	 * for the bss and break sections.  We must do this before
1072 	 * mapping in the interpreter, to make sure it doesn't wind
1073 	 * up getting placed where the bss needs to go.
1074 	 */
1075 	retval = set_brk(elf_bss, elf_brk, bss_prot);
1076 	if (retval)
1077 		goto out_free_dentry;
1078 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
1079 		retval = -EFAULT; /* Nobody gets to see this, but.. */
1080 		goto out_free_dentry;
1081 	}
1082 
1083 	if (interpreter) {
1084 		unsigned long interp_map_addr = 0;
1085 
1086 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
1087 					    interpreter,
1088 					    &interp_map_addr,
1089 					    load_bias, interp_elf_phdata);
1090 		if (!IS_ERR((void *)elf_entry)) {
1091 			/*
1092 			 * load_elf_interp() returns relocation
1093 			 * adjustment
1094 			 */
1095 			interp_load_addr = elf_entry;
1096 			elf_entry += loc->interp_elf_ex.e_entry;
1097 		}
1098 		if (BAD_ADDR(elf_entry)) {
1099 			retval = IS_ERR((void *)elf_entry) ?
1100 					(int)elf_entry : -EINVAL;
1101 			goto out_free_dentry;
1102 		}
1103 		reloc_func_desc = interp_load_addr;
1104 
1105 		allow_write_access(interpreter);
1106 		fput(interpreter);
1107 	} else {
1108 		elf_entry = loc->elf_ex.e_entry;
1109 		if (BAD_ADDR(elf_entry)) {
1110 			retval = -EINVAL;
1111 			goto out_free_dentry;
1112 		}
1113 	}
1114 
1115 	kfree(interp_elf_phdata);
1116 	kfree(elf_phdata);
1117 
1118 	set_binfmt(&elf_format);
1119 
1120 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1121 	retval = arch_setup_additional_pages(bprm, !!interpreter);
1122 	if (retval < 0)
1123 		goto out;
1124 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1125 
1126 	retval = create_elf_tables(bprm, &loc->elf_ex,
1127 			  load_addr, interp_load_addr);
1128 	if (retval < 0)
1129 		goto out;
1130 	/* N.B. passed_fileno might not be initialized? */
1131 	current->mm->end_code = end_code;
1132 	current->mm->start_code = start_code;
1133 	current->mm->start_data = start_data;
1134 	current->mm->end_data = end_data;
1135 	current->mm->start_stack = bprm->p;
1136 
1137 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1138 		/*
1139 		 * For architectures with ELF randomization, when executing
1140 		 * a loader directly (i.e. no interpreter listed in ELF
1141 		 * headers), move the brk area out of the mmap region
1142 		 * (since it grows up, and may collide early with the stack
1143 		 * growing down), and into the unused ELF_ET_DYN_BASE region.
1144 		 */
1145 		if (IS_ENABLED(CONFIG_ARCH_HAS_ELF_RANDOMIZE) && !interpreter)
1146 			current->mm->brk = current->mm->start_brk =
1147 				ELF_ET_DYN_BASE;
1148 
1149 		current->mm->brk = current->mm->start_brk =
1150 			arch_randomize_brk(current->mm);
1151 #ifdef compat_brk_randomized
1152 		current->brk_randomized = 1;
1153 #endif
1154 	}
1155 
1156 	if (current->personality & MMAP_PAGE_ZERO) {
1157 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
1158 		   and some applications "depend" upon this behavior.
1159 		   Since we do not have the power to recompile these, we
1160 		   emulate the SVr4 behavior. Sigh. */
1161 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1162 				MAP_FIXED | MAP_PRIVATE, 0);
1163 	}
1164 
1165 	regs = current_pt_regs();
1166 #ifdef ELF_PLAT_INIT
1167 	/*
1168 	 * The ABI may specify that certain registers be set up in special
1169 	 * ways (on i386 %edx is the address of a DT_FINI function, for
1170 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
1171 	 * that the e_entry field is the address of the function descriptor
1172 	 * for the startup routine, rather than the address of the startup
1173 	 * routine itself.  This macro performs whatever initialization to
1174 	 * the regs structure is required as well as any relocations to the
1175 	 * function descriptor entries when executing dynamically links apps.
1176 	 */
1177 	ELF_PLAT_INIT(regs, reloc_func_desc);
1178 #endif
1179 
1180 	finalize_exec(bprm);
1181 	start_thread(regs, elf_entry, bprm->p);
1182 	retval = 0;
1183 out:
1184 	kfree(loc);
1185 out_ret:
1186 	return retval;
1187 
1188 	/* error cleanup */
1189 out_free_dentry:
1190 	kfree(interp_elf_phdata);
1191 	allow_write_access(interpreter);
1192 	if (interpreter)
1193 		fput(interpreter);
1194 out_free_ph:
1195 	kfree(elf_phdata);
1196 	goto out;
1197 }
1198 
1199 #ifdef CONFIG_USELIB
1200 /* This is really simpleminded and specialized - we are loading an
1201    a.out library that is given an ELF header. */
1202 static int load_elf_library(struct file *file)
1203 {
1204 	struct elf_phdr *elf_phdata;
1205 	struct elf_phdr *eppnt;
1206 	unsigned long elf_bss, bss, len;
1207 	int retval, error, i, j;
1208 	struct elfhdr elf_ex;
1209 	loff_t pos = 0;
1210 
1211 	error = -ENOEXEC;
1212 	retval = kernel_read(file, &elf_ex, sizeof(elf_ex), &pos);
1213 	if (retval != sizeof(elf_ex))
1214 		goto out;
1215 
1216 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1217 		goto out;
1218 
1219 	/* First of all, some simple consistency checks */
1220 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1221 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1222 		goto out;
1223 	if (elf_check_fdpic(&elf_ex))
1224 		goto out;
1225 
1226 	/* Now read in all of the header information */
1227 
1228 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1229 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1230 
1231 	error = -ENOMEM;
1232 	elf_phdata = kmalloc(j, GFP_KERNEL);
1233 	if (!elf_phdata)
1234 		goto out;
1235 
1236 	eppnt = elf_phdata;
1237 	error = -ENOEXEC;
1238 	pos =  elf_ex.e_phoff;
1239 	retval = kernel_read(file, eppnt, j, &pos);
1240 	if (retval != j)
1241 		goto out_free_ph;
1242 
1243 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1244 		if ((eppnt + i)->p_type == PT_LOAD)
1245 			j++;
1246 	if (j != 1)
1247 		goto out_free_ph;
1248 
1249 	while (eppnt->p_type != PT_LOAD)
1250 		eppnt++;
1251 
1252 	/* Now use mmap to map the library into memory. */
1253 	error = vm_mmap(file,
1254 			ELF_PAGESTART(eppnt->p_vaddr),
1255 			(eppnt->p_filesz +
1256 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1257 			PROT_READ | PROT_WRITE | PROT_EXEC,
1258 			MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_DENYWRITE,
1259 			(eppnt->p_offset -
1260 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1261 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1262 		goto out_free_ph;
1263 
1264 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1265 	if (padzero(elf_bss)) {
1266 		error = -EFAULT;
1267 		goto out_free_ph;
1268 	}
1269 
1270 	len = ELF_PAGEALIGN(eppnt->p_filesz + eppnt->p_vaddr);
1271 	bss = ELF_PAGEALIGN(eppnt->p_memsz + eppnt->p_vaddr);
1272 	if (bss > len) {
1273 		error = vm_brk(len, bss - len);
1274 		if (error)
1275 			goto out_free_ph;
1276 	}
1277 	error = 0;
1278 
1279 out_free_ph:
1280 	kfree(elf_phdata);
1281 out:
1282 	return error;
1283 }
1284 #endif /* #ifdef CONFIG_USELIB */
1285 
1286 #ifdef CONFIG_ELF_CORE
1287 /*
1288  * ELF core dumper
1289  *
1290  * Modelled on fs/exec.c:aout_core_dump()
1291  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1292  */
1293 
1294 /*
1295  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1296  * that are useful for post-mortem analysis are included in every core dump.
1297  * In that way we ensure that the core dump is fully interpretable later
1298  * without matching up the same kernel and hardware config to see what PC values
1299  * meant. These special mappings include - vDSO, vsyscall, and other
1300  * architecture specific mappings
1301  */
1302 static bool always_dump_vma(struct vm_area_struct *vma)
1303 {
1304 	/* Any vsyscall mappings? */
1305 	if (vma == get_gate_vma(vma->vm_mm))
1306 		return true;
1307 
1308 	/*
1309 	 * Assume that all vmas with a .name op should always be dumped.
1310 	 * If this changes, a new vm_ops field can easily be added.
1311 	 */
1312 	if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1313 		return true;
1314 
1315 	/*
1316 	 * arch_vma_name() returns non-NULL for special architecture mappings,
1317 	 * such as vDSO sections.
1318 	 */
1319 	if (arch_vma_name(vma))
1320 		return true;
1321 
1322 	return false;
1323 }
1324 
1325 /*
1326  * Decide what to dump of a segment, part, all or none.
1327  */
1328 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1329 				   unsigned long mm_flags)
1330 {
1331 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1332 
1333 	/* always dump the vdso and vsyscall sections */
1334 	if (always_dump_vma(vma))
1335 		goto whole;
1336 
1337 	if (vma->vm_flags & VM_DONTDUMP)
1338 		return 0;
1339 
1340 	/* support for DAX */
1341 	if (vma_is_dax(vma)) {
1342 		if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1343 			goto whole;
1344 		if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1345 			goto whole;
1346 		return 0;
1347 	}
1348 
1349 	/* Hugetlb memory check */
1350 	if (vma->vm_flags & VM_HUGETLB) {
1351 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1352 			goto whole;
1353 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1354 			goto whole;
1355 		return 0;
1356 	}
1357 
1358 	/* Do not dump I/O mapped devices or special mappings */
1359 	if (vma->vm_flags & VM_IO)
1360 		return 0;
1361 
1362 	/* By default, dump shared memory if mapped from an anonymous file. */
1363 	if (vma->vm_flags & VM_SHARED) {
1364 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1365 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1366 			goto whole;
1367 		return 0;
1368 	}
1369 
1370 	/* Dump segments that have been written to.  */
1371 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1372 		goto whole;
1373 	if (vma->vm_file == NULL)
1374 		return 0;
1375 
1376 	if (FILTER(MAPPED_PRIVATE))
1377 		goto whole;
1378 
1379 	/*
1380 	 * If this looks like the beginning of a DSO or executable mapping,
1381 	 * check for an ELF header.  If we find one, dump the first page to
1382 	 * aid in determining what was mapped here.
1383 	 */
1384 	if (FILTER(ELF_HEADERS) &&
1385 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1386 		u32 __user *header = (u32 __user *) vma->vm_start;
1387 		u32 word;
1388 		mm_segment_t fs = get_fs();
1389 		/*
1390 		 * Doing it this way gets the constant folded by GCC.
1391 		 */
1392 		union {
1393 			u32 cmp;
1394 			char elfmag[SELFMAG];
1395 		} magic;
1396 		BUILD_BUG_ON(SELFMAG != sizeof word);
1397 		magic.elfmag[EI_MAG0] = ELFMAG0;
1398 		magic.elfmag[EI_MAG1] = ELFMAG1;
1399 		magic.elfmag[EI_MAG2] = ELFMAG2;
1400 		magic.elfmag[EI_MAG3] = ELFMAG3;
1401 		/*
1402 		 * Switch to the user "segment" for get_user(),
1403 		 * then put back what elf_core_dump() had in place.
1404 		 */
1405 		set_fs(USER_DS);
1406 		if (unlikely(get_user(word, header)))
1407 			word = 0;
1408 		set_fs(fs);
1409 		if (word == magic.cmp)
1410 			return PAGE_SIZE;
1411 	}
1412 
1413 #undef	FILTER
1414 
1415 	return 0;
1416 
1417 whole:
1418 	return vma->vm_end - vma->vm_start;
1419 }
1420 
1421 /* An ELF note in memory */
1422 struct memelfnote
1423 {
1424 	const char *name;
1425 	int type;
1426 	unsigned int datasz;
1427 	void *data;
1428 };
1429 
1430 static int notesize(struct memelfnote *en)
1431 {
1432 	int sz;
1433 
1434 	sz = sizeof(struct elf_note);
1435 	sz += roundup(strlen(en->name) + 1, 4);
1436 	sz += roundup(en->datasz, 4);
1437 
1438 	return sz;
1439 }
1440 
1441 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1442 {
1443 	struct elf_note en;
1444 	en.n_namesz = strlen(men->name) + 1;
1445 	en.n_descsz = men->datasz;
1446 	en.n_type = men->type;
1447 
1448 	return dump_emit(cprm, &en, sizeof(en)) &&
1449 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1450 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1451 }
1452 
1453 static void fill_elf_header(struct elfhdr *elf, int segs,
1454 			    u16 machine, u32 flags)
1455 {
1456 	memset(elf, 0, sizeof(*elf));
1457 
1458 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1459 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1460 	elf->e_ident[EI_DATA] = ELF_DATA;
1461 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1462 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1463 
1464 	elf->e_type = ET_CORE;
1465 	elf->e_machine = machine;
1466 	elf->e_version = EV_CURRENT;
1467 	elf->e_phoff = sizeof(struct elfhdr);
1468 	elf->e_flags = flags;
1469 	elf->e_ehsize = sizeof(struct elfhdr);
1470 	elf->e_phentsize = sizeof(struct elf_phdr);
1471 	elf->e_phnum = segs;
1472 }
1473 
1474 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1475 {
1476 	phdr->p_type = PT_NOTE;
1477 	phdr->p_offset = offset;
1478 	phdr->p_vaddr = 0;
1479 	phdr->p_paddr = 0;
1480 	phdr->p_filesz = sz;
1481 	phdr->p_memsz = 0;
1482 	phdr->p_flags = 0;
1483 	phdr->p_align = 0;
1484 }
1485 
1486 static void fill_note(struct memelfnote *note, const char *name, int type,
1487 		unsigned int sz, void *data)
1488 {
1489 	note->name = name;
1490 	note->type = type;
1491 	note->datasz = sz;
1492 	note->data = data;
1493 }
1494 
1495 /*
1496  * fill up all the fields in prstatus from the given task struct, except
1497  * registers which need to be filled up separately.
1498  */
1499 static void fill_prstatus(struct elf_prstatus *prstatus,
1500 		struct task_struct *p, long signr)
1501 {
1502 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1503 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1504 	prstatus->pr_sighold = p->blocked.sig[0];
1505 	rcu_read_lock();
1506 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1507 	rcu_read_unlock();
1508 	prstatus->pr_pid = task_pid_vnr(p);
1509 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1510 	prstatus->pr_sid = task_session_vnr(p);
1511 	if (thread_group_leader(p)) {
1512 		struct task_cputime cputime;
1513 
1514 		/*
1515 		 * This is the record for the group leader.  It shows the
1516 		 * group-wide total, not its individual thread total.
1517 		 */
1518 		thread_group_cputime(p, &cputime);
1519 		prstatus->pr_utime = ns_to_timeval(cputime.utime);
1520 		prstatus->pr_stime = ns_to_timeval(cputime.stime);
1521 	} else {
1522 		u64 utime, stime;
1523 
1524 		task_cputime(p, &utime, &stime);
1525 		prstatus->pr_utime = ns_to_timeval(utime);
1526 		prstatus->pr_stime = ns_to_timeval(stime);
1527 	}
1528 
1529 	prstatus->pr_cutime = ns_to_timeval(p->signal->cutime);
1530 	prstatus->pr_cstime = ns_to_timeval(p->signal->cstime);
1531 }
1532 
1533 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1534 		       struct mm_struct *mm)
1535 {
1536 	const struct cred *cred;
1537 	unsigned int i, len;
1538 
1539 	/* first copy the parameters from user space */
1540 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1541 
1542 	len = mm->arg_end - mm->arg_start;
1543 	if (len >= ELF_PRARGSZ)
1544 		len = ELF_PRARGSZ-1;
1545 	if (copy_from_user(&psinfo->pr_psargs,
1546 		           (const char __user *)mm->arg_start, len))
1547 		return -EFAULT;
1548 	for(i = 0; i < len; i++)
1549 		if (psinfo->pr_psargs[i] == 0)
1550 			psinfo->pr_psargs[i] = ' ';
1551 	psinfo->pr_psargs[len] = 0;
1552 
1553 	rcu_read_lock();
1554 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1555 	rcu_read_unlock();
1556 	psinfo->pr_pid = task_pid_vnr(p);
1557 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1558 	psinfo->pr_sid = task_session_vnr(p);
1559 
1560 	i = p->state ? ffz(~p->state) + 1 : 0;
1561 	psinfo->pr_state = i;
1562 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1563 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1564 	psinfo->pr_nice = task_nice(p);
1565 	psinfo->pr_flag = p->flags;
1566 	rcu_read_lock();
1567 	cred = __task_cred(p);
1568 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1569 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1570 	rcu_read_unlock();
1571 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1572 
1573 	return 0;
1574 }
1575 
1576 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1577 {
1578 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1579 	int i = 0;
1580 	do
1581 		i += 2;
1582 	while (auxv[i - 2] != AT_NULL);
1583 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1584 }
1585 
1586 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1587 		const kernel_siginfo_t *siginfo)
1588 {
1589 	mm_segment_t old_fs = get_fs();
1590 	set_fs(KERNEL_DS);
1591 	copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1592 	set_fs(old_fs);
1593 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1594 }
1595 
1596 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1597 /*
1598  * Format of NT_FILE note:
1599  *
1600  * long count     -- how many files are mapped
1601  * long page_size -- units for file_ofs
1602  * array of [COUNT] elements of
1603  *   long start
1604  *   long end
1605  *   long file_ofs
1606  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1607  */
1608 static int fill_files_note(struct memelfnote *note)
1609 {
1610 	struct vm_area_struct *vma;
1611 	unsigned count, size, names_ofs, remaining, n;
1612 	user_long_t *data;
1613 	user_long_t *start_end_ofs;
1614 	char *name_base, *name_curpos;
1615 
1616 	/* *Estimated* file count and total data size needed */
1617 	count = current->mm->map_count;
1618 	if (count > UINT_MAX / 64)
1619 		return -EINVAL;
1620 	size = count * 64;
1621 
1622 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1623  alloc:
1624 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1625 		return -EINVAL;
1626 	size = round_up(size, PAGE_SIZE);
1627 	data = kvmalloc(size, GFP_KERNEL);
1628 	if (ZERO_OR_NULL_PTR(data))
1629 		return -ENOMEM;
1630 
1631 	start_end_ofs = data + 2;
1632 	name_base = name_curpos = ((char *)data) + names_ofs;
1633 	remaining = size - names_ofs;
1634 	count = 0;
1635 	for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1636 		struct file *file;
1637 		const char *filename;
1638 
1639 		file = vma->vm_file;
1640 		if (!file)
1641 			continue;
1642 		filename = file_path(file, name_curpos, remaining);
1643 		if (IS_ERR(filename)) {
1644 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1645 				kvfree(data);
1646 				size = size * 5 / 4;
1647 				goto alloc;
1648 			}
1649 			continue;
1650 		}
1651 
1652 		/* file_path() fills at the end, move name down */
1653 		/* n = strlen(filename) + 1: */
1654 		n = (name_curpos + remaining) - filename;
1655 		remaining = filename - name_curpos;
1656 		memmove(name_curpos, filename, n);
1657 		name_curpos += n;
1658 
1659 		*start_end_ofs++ = vma->vm_start;
1660 		*start_end_ofs++ = vma->vm_end;
1661 		*start_end_ofs++ = vma->vm_pgoff;
1662 		count++;
1663 	}
1664 
1665 	/* Now we know exact count of files, can store it */
1666 	data[0] = count;
1667 	data[1] = PAGE_SIZE;
1668 	/*
1669 	 * Count usually is less than current->mm->map_count,
1670 	 * we need to move filenames down.
1671 	 */
1672 	n = current->mm->map_count - count;
1673 	if (n != 0) {
1674 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1675 		memmove(name_base - shift_bytes, name_base,
1676 			name_curpos - name_base);
1677 		name_curpos -= shift_bytes;
1678 	}
1679 
1680 	size = name_curpos - (char *)data;
1681 	fill_note(note, "CORE", NT_FILE, size, data);
1682 	return 0;
1683 }
1684 
1685 #ifdef CORE_DUMP_USE_REGSET
1686 #include <linux/regset.h>
1687 
1688 struct elf_thread_core_info {
1689 	struct elf_thread_core_info *next;
1690 	struct task_struct *task;
1691 	struct elf_prstatus prstatus;
1692 	struct memelfnote notes[0];
1693 };
1694 
1695 struct elf_note_info {
1696 	struct elf_thread_core_info *thread;
1697 	struct memelfnote psinfo;
1698 	struct memelfnote signote;
1699 	struct memelfnote auxv;
1700 	struct memelfnote files;
1701 	user_siginfo_t csigdata;
1702 	size_t size;
1703 	int thread_notes;
1704 };
1705 
1706 /*
1707  * When a regset has a writeback hook, we call it on each thread before
1708  * dumping user memory.  On register window machines, this makes sure the
1709  * user memory backing the register data is up to date before we read it.
1710  */
1711 static void do_thread_regset_writeback(struct task_struct *task,
1712 				       const struct user_regset *regset)
1713 {
1714 	if (regset->writeback)
1715 		regset->writeback(task, regset, 1);
1716 }
1717 
1718 #ifndef PRSTATUS_SIZE
1719 #define PRSTATUS_SIZE(S, R) sizeof(S)
1720 #endif
1721 
1722 #ifndef SET_PR_FPVALID
1723 #define SET_PR_FPVALID(S, V, R) ((S)->pr_fpvalid = (V))
1724 #endif
1725 
1726 static int fill_thread_core_info(struct elf_thread_core_info *t,
1727 				 const struct user_regset_view *view,
1728 				 long signr, size_t *total)
1729 {
1730 	unsigned int i;
1731 	unsigned int regset0_size = regset_size(t->task, &view->regsets[0]);
1732 
1733 	/*
1734 	 * NT_PRSTATUS is the one special case, because the regset data
1735 	 * goes into the pr_reg field inside the note contents, rather
1736 	 * than being the whole note contents.  We fill the reset in here.
1737 	 * We assume that regset 0 is NT_PRSTATUS.
1738 	 */
1739 	fill_prstatus(&t->prstatus, t->task, signr);
1740 	(void) view->regsets[0].get(t->task, &view->regsets[0], 0, regset0_size,
1741 				    &t->prstatus.pr_reg, NULL);
1742 
1743 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1744 		  PRSTATUS_SIZE(t->prstatus, regset0_size), &t->prstatus);
1745 	*total += notesize(&t->notes[0]);
1746 
1747 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1748 
1749 	/*
1750 	 * Each other regset might generate a note too.  For each regset
1751 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1752 	 * all zero and we'll know to skip writing it later.
1753 	 */
1754 	for (i = 1; i < view->n; ++i) {
1755 		const struct user_regset *regset = &view->regsets[i];
1756 		do_thread_regset_writeback(t->task, regset);
1757 		if (regset->core_note_type && regset->get &&
1758 		    (!regset->active || regset->active(t->task, regset) > 0)) {
1759 			int ret;
1760 			size_t size = regset_size(t->task, regset);
1761 			void *data = kmalloc(size, GFP_KERNEL);
1762 			if (unlikely(!data))
1763 				return 0;
1764 			ret = regset->get(t->task, regset,
1765 					  0, size, data, NULL);
1766 			if (unlikely(ret))
1767 				kfree(data);
1768 			else {
1769 				if (regset->core_note_type != NT_PRFPREG)
1770 					fill_note(&t->notes[i], "LINUX",
1771 						  regset->core_note_type,
1772 						  size, data);
1773 				else {
1774 					SET_PR_FPVALID(&t->prstatus,
1775 							1, regset0_size);
1776 					fill_note(&t->notes[i], "CORE",
1777 						  NT_PRFPREG, size, data);
1778 				}
1779 				*total += notesize(&t->notes[i]);
1780 			}
1781 		}
1782 	}
1783 
1784 	return 1;
1785 }
1786 
1787 static int fill_note_info(struct elfhdr *elf, int phdrs,
1788 			  struct elf_note_info *info,
1789 			  const kernel_siginfo_t *siginfo, struct pt_regs *regs)
1790 {
1791 	struct task_struct *dump_task = current;
1792 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1793 	struct elf_thread_core_info *t;
1794 	struct elf_prpsinfo *psinfo;
1795 	struct core_thread *ct;
1796 	unsigned int i;
1797 
1798 	info->size = 0;
1799 	info->thread = NULL;
1800 
1801 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1802 	if (psinfo == NULL) {
1803 		info->psinfo.data = NULL; /* So we don't free this wrongly */
1804 		return 0;
1805 	}
1806 
1807 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1808 
1809 	/*
1810 	 * Figure out how many notes we're going to need for each thread.
1811 	 */
1812 	info->thread_notes = 0;
1813 	for (i = 0; i < view->n; ++i)
1814 		if (view->regsets[i].core_note_type != 0)
1815 			++info->thread_notes;
1816 
1817 	/*
1818 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1819 	 * since it is our one special case.
1820 	 */
1821 	if (unlikely(info->thread_notes == 0) ||
1822 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1823 		WARN_ON(1);
1824 		return 0;
1825 	}
1826 
1827 	/*
1828 	 * Initialize the ELF file header.
1829 	 */
1830 	fill_elf_header(elf, phdrs,
1831 			view->e_machine, view->e_flags);
1832 
1833 	/*
1834 	 * Allocate a structure for each thread.
1835 	 */
1836 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1837 		t = kzalloc(offsetof(struct elf_thread_core_info,
1838 				     notes[info->thread_notes]),
1839 			    GFP_KERNEL);
1840 		if (unlikely(!t))
1841 			return 0;
1842 
1843 		t->task = ct->task;
1844 		if (ct->task == dump_task || !info->thread) {
1845 			t->next = info->thread;
1846 			info->thread = t;
1847 		} else {
1848 			/*
1849 			 * Make sure to keep the original task at
1850 			 * the head of the list.
1851 			 */
1852 			t->next = info->thread->next;
1853 			info->thread->next = t;
1854 		}
1855 	}
1856 
1857 	/*
1858 	 * Now fill in each thread's information.
1859 	 */
1860 	for (t = info->thread; t != NULL; t = t->next)
1861 		if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1862 			return 0;
1863 
1864 	/*
1865 	 * Fill in the two process-wide notes.
1866 	 */
1867 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1868 	info->size += notesize(&info->psinfo);
1869 
1870 	fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1871 	info->size += notesize(&info->signote);
1872 
1873 	fill_auxv_note(&info->auxv, current->mm);
1874 	info->size += notesize(&info->auxv);
1875 
1876 	if (fill_files_note(&info->files) == 0)
1877 		info->size += notesize(&info->files);
1878 
1879 	return 1;
1880 }
1881 
1882 static size_t get_note_info_size(struct elf_note_info *info)
1883 {
1884 	return info->size;
1885 }
1886 
1887 /*
1888  * Write all the notes for each thread.  When writing the first thread, the
1889  * process-wide notes are interleaved after the first thread-specific note.
1890  */
1891 static int write_note_info(struct elf_note_info *info,
1892 			   struct coredump_params *cprm)
1893 {
1894 	bool first = true;
1895 	struct elf_thread_core_info *t = info->thread;
1896 
1897 	do {
1898 		int i;
1899 
1900 		if (!writenote(&t->notes[0], cprm))
1901 			return 0;
1902 
1903 		if (first && !writenote(&info->psinfo, cprm))
1904 			return 0;
1905 		if (first && !writenote(&info->signote, cprm))
1906 			return 0;
1907 		if (first && !writenote(&info->auxv, cprm))
1908 			return 0;
1909 		if (first && info->files.data &&
1910 				!writenote(&info->files, cprm))
1911 			return 0;
1912 
1913 		for (i = 1; i < info->thread_notes; ++i)
1914 			if (t->notes[i].data &&
1915 			    !writenote(&t->notes[i], cprm))
1916 				return 0;
1917 
1918 		first = false;
1919 		t = t->next;
1920 	} while (t);
1921 
1922 	return 1;
1923 }
1924 
1925 static void free_note_info(struct elf_note_info *info)
1926 {
1927 	struct elf_thread_core_info *threads = info->thread;
1928 	while (threads) {
1929 		unsigned int i;
1930 		struct elf_thread_core_info *t = threads;
1931 		threads = t->next;
1932 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1933 		for (i = 1; i < info->thread_notes; ++i)
1934 			kfree(t->notes[i].data);
1935 		kfree(t);
1936 	}
1937 	kfree(info->psinfo.data);
1938 	kvfree(info->files.data);
1939 }
1940 
1941 #else
1942 
1943 /* Here is the structure in which status of each thread is captured. */
1944 struct elf_thread_status
1945 {
1946 	struct list_head list;
1947 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1948 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1949 	struct task_struct *thread;
1950 #ifdef ELF_CORE_COPY_XFPREGS
1951 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1952 #endif
1953 	struct memelfnote notes[3];
1954 	int num_notes;
1955 };
1956 
1957 /*
1958  * In order to add the specific thread information for the elf file format,
1959  * we need to keep a linked list of every threads pr_status and then create
1960  * a single section for them in the final core file.
1961  */
1962 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1963 {
1964 	int sz = 0;
1965 	struct task_struct *p = t->thread;
1966 	t->num_notes = 0;
1967 
1968 	fill_prstatus(&t->prstatus, p, signr);
1969 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1970 
1971 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1972 		  &(t->prstatus));
1973 	t->num_notes++;
1974 	sz += notesize(&t->notes[0]);
1975 
1976 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1977 								&t->fpu))) {
1978 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1979 			  &(t->fpu));
1980 		t->num_notes++;
1981 		sz += notesize(&t->notes[1]);
1982 	}
1983 
1984 #ifdef ELF_CORE_COPY_XFPREGS
1985 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1986 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1987 			  sizeof(t->xfpu), &t->xfpu);
1988 		t->num_notes++;
1989 		sz += notesize(&t->notes[2]);
1990 	}
1991 #endif
1992 	return sz;
1993 }
1994 
1995 struct elf_note_info {
1996 	struct memelfnote *notes;
1997 	struct memelfnote *notes_files;
1998 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1999 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
2000 	struct list_head thread_list;
2001 	elf_fpregset_t *fpu;
2002 #ifdef ELF_CORE_COPY_XFPREGS
2003 	elf_fpxregset_t *xfpu;
2004 #endif
2005 	user_siginfo_t csigdata;
2006 	int thread_status_size;
2007 	int numnote;
2008 };
2009 
2010 static int elf_note_info_init(struct elf_note_info *info)
2011 {
2012 	memset(info, 0, sizeof(*info));
2013 	INIT_LIST_HEAD(&info->thread_list);
2014 
2015 	/* Allocate space for ELF notes */
2016 	info->notes = kmalloc_array(8, sizeof(struct memelfnote), GFP_KERNEL);
2017 	if (!info->notes)
2018 		return 0;
2019 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
2020 	if (!info->psinfo)
2021 		return 0;
2022 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
2023 	if (!info->prstatus)
2024 		return 0;
2025 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
2026 	if (!info->fpu)
2027 		return 0;
2028 #ifdef ELF_CORE_COPY_XFPREGS
2029 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
2030 	if (!info->xfpu)
2031 		return 0;
2032 #endif
2033 	return 1;
2034 }
2035 
2036 static int fill_note_info(struct elfhdr *elf, int phdrs,
2037 			  struct elf_note_info *info,
2038 			  const kernel_siginfo_t *siginfo, struct pt_regs *regs)
2039 {
2040 	struct core_thread *ct;
2041 	struct elf_thread_status *ets;
2042 
2043 	if (!elf_note_info_init(info))
2044 		return 0;
2045 
2046 	for (ct = current->mm->core_state->dumper.next;
2047 					ct; ct = ct->next) {
2048 		ets = kzalloc(sizeof(*ets), GFP_KERNEL);
2049 		if (!ets)
2050 			return 0;
2051 
2052 		ets->thread = ct->task;
2053 		list_add(&ets->list, &info->thread_list);
2054 	}
2055 
2056 	list_for_each_entry(ets, &info->thread_list, list) {
2057 		int sz;
2058 
2059 		sz = elf_dump_thread_status(siginfo->si_signo, ets);
2060 		info->thread_status_size += sz;
2061 	}
2062 	/* now collect the dump for the current */
2063 	memset(info->prstatus, 0, sizeof(*info->prstatus));
2064 	fill_prstatus(info->prstatus, current, siginfo->si_signo);
2065 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
2066 
2067 	/* Set up header */
2068 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
2069 
2070 	/*
2071 	 * Set up the notes in similar form to SVR4 core dumps made
2072 	 * with info from their /proc.
2073 	 */
2074 
2075 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
2076 		  sizeof(*info->prstatus), info->prstatus);
2077 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
2078 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
2079 		  sizeof(*info->psinfo), info->psinfo);
2080 
2081 	fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
2082 	fill_auxv_note(info->notes + 3, current->mm);
2083 	info->numnote = 4;
2084 
2085 	if (fill_files_note(info->notes + info->numnote) == 0) {
2086 		info->notes_files = info->notes + info->numnote;
2087 		info->numnote++;
2088 	}
2089 
2090 	/* Try to dump the FPU. */
2091 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2092 							       info->fpu);
2093 	if (info->prstatus->pr_fpvalid)
2094 		fill_note(info->notes + info->numnote++,
2095 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2096 #ifdef ELF_CORE_COPY_XFPREGS
2097 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
2098 		fill_note(info->notes + info->numnote++,
2099 			  "LINUX", ELF_CORE_XFPREG_TYPE,
2100 			  sizeof(*info->xfpu), info->xfpu);
2101 #endif
2102 
2103 	return 1;
2104 }
2105 
2106 static size_t get_note_info_size(struct elf_note_info *info)
2107 {
2108 	int sz = 0;
2109 	int i;
2110 
2111 	for (i = 0; i < info->numnote; i++)
2112 		sz += notesize(info->notes + i);
2113 
2114 	sz += info->thread_status_size;
2115 
2116 	return sz;
2117 }
2118 
2119 static int write_note_info(struct elf_note_info *info,
2120 			   struct coredump_params *cprm)
2121 {
2122 	struct elf_thread_status *ets;
2123 	int i;
2124 
2125 	for (i = 0; i < info->numnote; i++)
2126 		if (!writenote(info->notes + i, cprm))
2127 			return 0;
2128 
2129 	/* write out the thread status notes section */
2130 	list_for_each_entry(ets, &info->thread_list, list) {
2131 		for (i = 0; i < ets->num_notes; i++)
2132 			if (!writenote(&ets->notes[i], cprm))
2133 				return 0;
2134 	}
2135 
2136 	return 1;
2137 }
2138 
2139 static void free_note_info(struct elf_note_info *info)
2140 {
2141 	while (!list_empty(&info->thread_list)) {
2142 		struct list_head *tmp = info->thread_list.next;
2143 		list_del(tmp);
2144 		kfree(list_entry(tmp, struct elf_thread_status, list));
2145 	}
2146 
2147 	/* Free data possibly allocated by fill_files_note(): */
2148 	if (info->notes_files)
2149 		kvfree(info->notes_files->data);
2150 
2151 	kfree(info->prstatus);
2152 	kfree(info->psinfo);
2153 	kfree(info->notes);
2154 	kfree(info->fpu);
2155 #ifdef ELF_CORE_COPY_XFPREGS
2156 	kfree(info->xfpu);
2157 #endif
2158 }
2159 
2160 #endif
2161 
2162 static struct vm_area_struct *first_vma(struct task_struct *tsk,
2163 					struct vm_area_struct *gate_vma)
2164 {
2165 	struct vm_area_struct *ret = tsk->mm->mmap;
2166 
2167 	if (ret)
2168 		return ret;
2169 	return gate_vma;
2170 }
2171 /*
2172  * Helper function for iterating across a vma list.  It ensures that the caller
2173  * will visit `gate_vma' prior to terminating the search.
2174  */
2175 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2176 					struct vm_area_struct *gate_vma)
2177 {
2178 	struct vm_area_struct *ret;
2179 
2180 	ret = this_vma->vm_next;
2181 	if (ret)
2182 		return ret;
2183 	if (this_vma == gate_vma)
2184 		return NULL;
2185 	return gate_vma;
2186 }
2187 
2188 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2189 			     elf_addr_t e_shoff, int segs)
2190 {
2191 	elf->e_shoff = e_shoff;
2192 	elf->e_shentsize = sizeof(*shdr4extnum);
2193 	elf->e_shnum = 1;
2194 	elf->e_shstrndx = SHN_UNDEF;
2195 
2196 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2197 
2198 	shdr4extnum->sh_type = SHT_NULL;
2199 	shdr4extnum->sh_size = elf->e_shnum;
2200 	shdr4extnum->sh_link = elf->e_shstrndx;
2201 	shdr4extnum->sh_info = segs;
2202 }
2203 
2204 /*
2205  * Actual dumper
2206  *
2207  * This is a two-pass process; first we find the offsets of the bits,
2208  * and then they are actually written out.  If we run out of core limit
2209  * we just truncate.
2210  */
2211 static int elf_core_dump(struct coredump_params *cprm)
2212 {
2213 	int has_dumped = 0;
2214 	mm_segment_t fs;
2215 	int segs, i;
2216 	size_t vma_data_size = 0;
2217 	struct vm_area_struct *vma, *gate_vma;
2218 	struct elfhdr *elf = NULL;
2219 	loff_t offset = 0, dataoff;
2220 	struct elf_note_info info = { };
2221 	struct elf_phdr *phdr4note = NULL;
2222 	struct elf_shdr *shdr4extnum = NULL;
2223 	Elf_Half e_phnum;
2224 	elf_addr_t e_shoff;
2225 	elf_addr_t *vma_filesz = NULL;
2226 
2227 	/*
2228 	 * We no longer stop all VM operations.
2229 	 *
2230 	 * This is because those proceses that could possibly change map_count
2231 	 * or the mmap / vma pages are now blocked in do_exit on current
2232 	 * finishing this core dump.
2233 	 *
2234 	 * Only ptrace can touch these memory addresses, but it doesn't change
2235 	 * the map_count or the pages allocated. So no possibility of crashing
2236 	 * exists while dumping the mm->vm_next areas to the core file.
2237 	 */
2238 
2239 	/* alloc memory for large data structures: too large to be on stack */
2240 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2241 	if (!elf)
2242 		goto out;
2243 	/*
2244 	 * The number of segs are recored into ELF header as 16bit value.
2245 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2246 	 */
2247 	segs = current->mm->map_count;
2248 	segs += elf_core_extra_phdrs();
2249 
2250 	gate_vma = get_gate_vma(current->mm);
2251 	if (gate_vma != NULL)
2252 		segs++;
2253 
2254 	/* for notes section */
2255 	segs++;
2256 
2257 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2258 	 * this, kernel supports extended numbering. Have a look at
2259 	 * include/linux/elf.h for further information. */
2260 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2261 
2262 	/*
2263 	 * Collect all the non-memory information about the process for the
2264 	 * notes.  This also sets up the file header.
2265 	 */
2266 	if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2267 		goto cleanup;
2268 
2269 	has_dumped = 1;
2270 
2271 	fs = get_fs();
2272 	set_fs(KERNEL_DS);
2273 
2274 	offset += sizeof(*elf);				/* Elf header */
2275 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2276 
2277 	/* Write notes phdr entry */
2278 	{
2279 		size_t sz = get_note_info_size(&info);
2280 
2281 		sz += elf_coredump_extra_notes_size();
2282 
2283 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2284 		if (!phdr4note)
2285 			goto end_coredump;
2286 
2287 		fill_elf_note_phdr(phdr4note, sz, offset);
2288 		offset += sz;
2289 	}
2290 
2291 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2292 
2293 	if (segs - 1 > ULONG_MAX / sizeof(*vma_filesz))
2294 		goto end_coredump;
2295 	vma_filesz = kvmalloc(array_size(sizeof(*vma_filesz), (segs - 1)),
2296 			      GFP_KERNEL);
2297 	if (ZERO_OR_NULL_PTR(vma_filesz))
2298 		goto end_coredump;
2299 
2300 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2301 			vma = next_vma(vma, gate_vma)) {
2302 		unsigned long dump_size;
2303 
2304 		dump_size = vma_dump_size(vma, cprm->mm_flags);
2305 		vma_filesz[i++] = dump_size;
2306 		vma_data_size += dump_size;
2307 	}
2308 
2309 	offset += vma_data_size;
2310 	offset += elf_core_extra_data_size();
2311 	e_shoff = offset;
2312 
2313 	if (e_phnum == PN_XNUM) {
2314 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2315 		if (!shdr4extnum)
2316 			goto end_coredump;
2317 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2318 	}
2319 
2320 	offset = dataoff;
2321 
2322 	if (!dump_emit(cprm, elf, sizeof(*elf)))
2323 		goto end_coredump;
2324 
2325 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2326 		goto end_coredump;
2327 
2328 	/* Write program headers for segments dump */
2329 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2330 			vma = next_vma(vma, gate_vma)) {
2331 		struct elf_phdr phdr;
2332 
2333 		phdr.p_type = PT_LOAD;
2334 		phdr.p_offset = offset;
2335 		phdr.p_vaddr = vma->vm_start;
2336 		phdr.p_paddr = 0;
2337 		phdr.p_filesz = vma_filesz[i++];
2338 		phdr.p_memsz = vma->vm_end - vma->vm_start;
2339 		offset += phdr.p_filesz;
2340 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2341 		if (vma->vm_flags & VM_WRITE)
2342 			phdr.p_flags |= PF_W;
2343 		if (vma->vm_flags & VM_EXEC)
2344 			phdr.p_flags |= PF_X;
2345 		phdr.p_align = ELF_EXEC_PAGESIZE;
2346 
2347 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2348 			goto end_coredump;
2349 	}
2350 
2351 	if (!elf_core_write_extra_phdrs(cprm, offset))
2352 		goto end_coredump;
2353 
2354  	/* write out the notes section */
2355 	if (!write_note_info(&info, cprm))
2356 		goto end_coredump;
2357 
2358 	if (elf_coredump_extra_notes_write(cprm))
2359 		goto end_coredump;
2360 
2361 	/* Align to page */
2362 	if (!dump_skip(cprm, dataoff - cprm->pos))
2363 		goto end_coredump;
2364 
2365 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2366 			vma = next_vma(vma, gate_vma)) {
2367 		unsigned long addr;
2368 		unsigned long end;
2369 
2370 		end = vma->vm_start + vma_filesz[i++];
2371 
2372 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2373 			struct page *page;
2374 			int stop;
2375 
2376 			page = get_dump_page(addr);
2377 			if (page) {
2378 				void *kaddr = kmap(page);
2379 				stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2380 				kunmap(page);
2381 				put_page(page);
2382 			} else
2383 				stop = !dump_skip(cprm, PAGE_SIZE);
2384 			if (stop)
2385 				goto end_coredump;
2386 		}
2387 	}
2388 	dump_truncate(cprm);
2389 
2390 	if (!elf_core_write_extra_data(cprm))
2391 		goto end_coredump;
2392 
2393 	if (e_phnum == PN_XNUM) {
2394 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2395 			goto end_coredump;
2396 	}
2397 
2398 end_coredump:
2399 	set_fs(fs);
2400 
2401 cleanup:
2402 	free_note_info(&info);
2403 	kfree(shdr4extnum);
2404 	kvfree(vma_filesz);
2405 	kfree(phdr4note);
2406 	kfree(elf);
2407 out:
2408 	return has_dumped;
2409 }
2410 
2411 #endif		/* CONFIG_ELF_CORE */
2412 
2413 static int __init init_elf_binfmt(void)
2414 {
2415 	register_binfmt(&elf_format);
2416 	return 0;
2417 }
2418 
2419 static void __exit exit_elf_binfmt(void)
2420 {
2421 	/* Remove the COFF and ELF loaders. */
2422 	unregister_binfmt(&elf_format);
2423 }
2424 
2425 core_initcall(init_elf_binfmt);
2426 module_exit(exit_elf_binfmt);
2427 MODULE_LICENSE("GPL");
2428