xref: /openbmc/linux/fs/binfmt_elf.c (revision 8d7071af)
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/log2.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/errno.h>
20 #include <linux/signal.h>
21 #include <linux/binfmts.h>
22 #include <linux/string.h>
23 #include <linux/file.h>
24 #include <linux/slab.h>
25 #include <linux/personality.h>
26 #include <linux/elfcore.h>
27 #include <linux/init.h>
28 #include <linux/highuid.h>
29 #include <linux/compiler.h>
30 #include <linux/highmem.h>
31 #include <linux/hugetlb.h>
32 #include <linux/pagemap.h>
33 #include <linux/vmalloc.h>
34 #include <linux/security.h>
35 #include <linux/random.h>
36 #include <linux/elf.h>
37 #include <linux/elf-randomize.h>
38 #include <linux/utsname.h>
39 #include <linux/coredump.h>
40 #include <linux/sched.h>
41 #include <linux/sched/coredump.h>
42 #include <linux/sched/task_stack.h>
43 #include <linux/sched/cputime.h>
44 #include <linux/sizes.h>
45 #include <linux/types.h>
46 #include <linux/cred.h>
47 #include <linux/dax.h>
48 #include <linux/uaccess.h>
49 #include <linux/rseq.h>
50 #include <asm/param.h>
51 #include <asm/page.h>
52 
53 #ifndef ELF_COMPAT
54 #define ELF_COMPAT 0
55 #endif
56 
57 #ifndef user_long_t
58 #define user_long_t long
59 #endif
60 #ifndef user_siginfo_t
61 #define user_siginfo_t siginfo_t
62 #endif
63 
64 /* That's for binfmt_elf_fdpic to deal with */
65 #ifndef elf_check_fdpic
66 #define elf_check_fdpic(ex) false
67 #endif
68 
69 static int load_elf_binary(struct linux_binprm *bprm);
70 
71 #ifdef CONFIG_USELIB
72 static int load_elf_library(struct file *);
73 #else
74 #define load_elf_library NULL
75 #endif
76 
77 /*
78  * If we don't support core dumping, then supply a NULL so we
79  * don't even try.
80  */
81 #ifdef CONFIG_ELF_CORE
82 static int elf_core_dump(struct coredump_params *cprm);
83 #else
84 #define elf_core_dump	NULL
85 #endif
86 
87 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
88 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
89 #else
90 #define ELF_MIN_ALIGN	PAGE_SIZE
91 #endif
92 
93 #ifndef ELF_CORE_EFLAGS
94 #define ELF_CORE_EFLAGS	0
95 #endif
96 
97 #define ELF_PAGESTART(_v) ((_v) & ~(int)(ELF_MIN_ALIGN-1))
98 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
99 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
100 
101 static struct linux_binfmt elf_format = {
102 	.module		= THIS_MODULE,
103 	.load_binary	= load_elf_binary,
104 	.load_shlib	= load_elf_library,
105 #ifdef CONFIG_COREDUMP
106 	.core_dump	= elf_core_dump,
107 	.min_coredump	= ELF_EXEC_PAGESIZE,
108 #endif
109 };
110 
111 #define BAD_ADDR(x) (unlikely((unsigned long)(x) >= TASK_SIZE))
112 
113 static int set_brk(unsigned long start, unsigned long end, int prot)
114 {
115 	start = ELF_PAGEALIGN(start);
116 	end = ELF_PAGEALIGN(end);
117 	if (end > start) {
118 		/*
119 		 * Map the last of the bss segment.
120 		 * If the header is requesting these pages to be
121 		 * executable, honour that (ppc32 needs this).
122 		 */
123 		int error = vm_brk_flags(start, end - start,
124 				prot & PROT_EXEC ? VM_EXEC : 0);
125 		if (error)
126 			return error;
127 	}
128 	current->mm->start_brk = current->mm->brk = end;
129 	return 0;
130 }
131 
132 /* We need to explicitly zero any fractional pages
133    after the data section (i.e. bss).  This would
134    contain the junk from the file that should not
135    be in memory
136  */
137 static int padzero(unsigned long elf_bss)
138 {
139 	unsigned long nbyte;
140 
141 	nbyte = ELF_PAGEOFFSET(elf_bss);
142 	if (nbyte) {
143 		nbyte = ELF_MIN_ALIGN - nbyte;
144 		if (clear_user((void __user *) elf_bss, nbyte))
145 			return -EFAULT;
146 	}
147 	return 0;
148 }
149 
150 /* Let's use some macros to make this stack manipulation a little clearer */
151 #ifdef CONFIG_STACK_GROWSUP
152 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
153 #define STACK_ROUND(sp, items) \
154 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
155 #define STACK_ALLOC(sp, len) ({ \
156 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
157 	old_sp; })
158 #else
159 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
160 #define STACK_ROUND(sp, items) \
161 	(((unsigned long) (sp - items)) &~ 15UL)
162 #define STACK_ALLOC(sp, len) (sp -= len)
163 #endif
164 
165 #ifndef ELF_BASE_PLATFORM
166 /*
167  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
168  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
169  * will be copied to the user stack in the same manner as AT_PLATFORM.
170  */
171 #define ELF_BASE_PLATFORM NULL
172 #endif
173 
174 static int
175 create_elf_tables(struct linux_binprm *bprm, const struct elfhdr *exec,
176 		unsigned long interp_load_addr,
177 		unsigned long e_entry, unsigned long phdr_addr)
178 {
179 	struct mm_struct *mm = current->mm;
180 	unsigned long p = bprm->p;
181 	int argc = bprm->argc;
182 	int envc = bprm->envc;
183 	elf_addr_t __user *sp;
184 	elf_addr_t __user *u_platform;
185 	elf_addr_t __user *u_base_platform;
186 	elf_addr_t __user *u_rand_bytes;
187 	const char *k_platform = ELF_PLATFORM;
188 	const char *k_base_platform = ELF_BASE_PLATFORM;
189 	unsigned char k_rand_bytes[16];
190 	int items;
191 	elf_addr_t *elf_info;
192 	elf_addr_t flags = 0;
193 	int ei_index;
194 	const struct cred *cred = current_cred();
195 	struct vm_area_struct *vma;
196 
197 	/*
198 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
199 	 * evictions by the processes running on the same package. One
200 	 * thing we can do is to shuffle the initial stack for them.
201 	 */
202 
203 	p = arch_align_stack(p);
204 
205 	/*
206 	 * If this architecture has a platform capability string, copy it
207 	 * to userspace.  In some cases (Sparc), this info is impossible
208 	 * for userspace to get any other way, in others (i386) it is
209 	 * merely difficult.
210 	 */
211 	u_platform = NULL;
212 	if (k_platform) {
213 		size_t len = strlen(k_platform) + 1;
214 
215 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
216 		if (copy_to_user(u_platform, k_platform, len))
217 			return -EFAULT;
218 	}
219 
220 	/*
221 	 * If this architecture has a "base" platform capability
222 	 * string, copy it to userspace.
223 	 */
224 	u_base_platform = NULL;
225 	if (k_base_platform) {
226 		size_t len = strlen(k_base_platform) + 1;
227 
228 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
229 		if (copy_to_user(u_base_platform, k_base_platform, len))
230 			return -EFAULT;
231 	}
232 
233 	/*
234 	 * Generate 16 random bytes for userspace PRNG seeding.
235 	 */
236 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
237 	u_rand_bytes = (elf_addr_t __user *)
238 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
239 	if (copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
240 		return -EFAULT;
241 
242 	/* Create the ELF interpreter info */
243 	elf_info = (elf_addr_t *)mm->saved_auxv;
244 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
245 #define NEW_AUX_ENT(id, val) \
246 	do { \
247 		*elf_info++ = id; \
248 		*elf_info++ = val; \
249 	} while (0)
250 
251 #ifdef ARCH_DLINFO
252 	/*
253 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
254 	 * AUXV.
255 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
256 	 * ARCH_DLINFO changes
257 	 */
258 	ARCH_DLINFO;
259 #endif
260 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
261 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
262 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
263 	NEW_AUX_ENT(AT_PHDR, phdr_addr);
264 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
265 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
266 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
267 	if (bprm->interp_flags & BINPRM_FLAGS_PRESERVE_ARGV0)
268 		flags |= AT_FLAGS_PRESERVE_ARGV0;
269 	NEW_AUX_ENT(AT_FLAGS, flags);
270 	NEW_AUX_ENT(AT_ENTRY, e_entry);
271 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
272 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
273 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
274 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
275 	NEW_AUX_ENT(AT_SECURE, bprm->secureexec);
276 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
277 #ifdef ELF_HWCAP2
278 	NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
279 #endif
280 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
281 	if (k_platform) {
282 		NEW_AUX_ENT(AT_PLATFORM,
283 			    (elf_addr_t)(unsigned long)u_platform);
284 	}
285 	if (k_base_platform) {
286 		NEW_AUX_ENT(AT_BASE_PLATFORM,
287 			    (elf_addr_t)(unsigned long)u_base_platform);
288 	}
289 	if (bprm->have_execfd) {
290 		NEW_AUX_ENT(AT_EXECFD, bprm->execfd);
291 	}
292 #ifdef CONFIG_RSEQ
293 	NEW_AUX_ENT(AT_RSEQ_FEATURE_SIZE, offsetof(struct rseq, end));
294 	NEW_AUX_ENT(AT_RSEQ_ALIGN, __alignof__(struct rseq));
295 #endif
296 #undef NEW_AUX_ENT
297 	/* AT_NULL is zero; clear the rest too */
298 	memset(elf_info, 0, (char *)mm->saved_auxv +
299 			sizeof(mm->saved_auxv) - (char *)elf_info);
300 
301 	/* And advance past the AT_NULL entry.  */
302 	elf_info += 2;
303 
304 	ei_index = elf_info - (elf_addr_t *)mm->saved_auxv;
305 	sp = STACK_ADD(p, ei_index);
306 
307 	items = (argc + 1) + (envc + 1) + 1;
308 	bprm->p = STACK_ROUND(sp, items);
309 
310 	/* Point sp at the lowest address on the stack */
311 #ifdef CONFIG_STACK_GROWSUP
312 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
313 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
314 #else
315 	sp = (elf_addr_t __user *)bprm->p;
316 #endif
317 
318 
319 	/*
320 	 * Grow the stack manually; some architectures have a limit on how
321 	 * far ahead a user-space access may be in order to grow the stack.
322 	 */
323 	if (mmap_write_lock_killable(mm))
324 		return -EINTR;
325 	vma = find_extend_vma_locked(mm, bprm->p);
326 	mmap_write_unlock(mm);
327 	if (!vma)
328 		return -EFAULT;
329 
330 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
331 	if (put_user(argc, sp++))
332 		return -EFAULT;
333 
334 	/* Populate list of argv pointers back to argv strings. */
335 	p = mm->arg_end = mm->arg_start;
336 	while (argc-- > 0) {
337 		size_t len;
338 		if (put_user((elf_addr_t)p, sp++))
339 			return -EFAULT;
340 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
341 		if (!len || len > MAX_ARG_STRLEN)
342 			return -EINVAL;
343 		p += len;
344 	}
345 	if (put_user(0, sp++))
346 		return -EFAULT;
347 	mm->arg_end = p;
348 
349 	/* Populate list of envp pointers back to envp strings. */
350 	mm->env_end = mm->env_start = p;
351 	while (envc-- > 0) {
352 		size_t len;
353 		if (put_user((elf_addr_t)p, sp++))
354 			return -EFAULT;
355 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
356 		if (!len || len > MAX_ARG_STRLEN)
357 			return -EINVAL;
358 		p += len;
359 	}
360 	if (put_user(0, sp++))
361 		return -EFAULT;
362 	mm->env_end = p;
363 
364 	/* Put the elf_info on the stack in the right place.  */
365 	if (copy_to_user(sp, mm->saved_auxv, ei_index * sizeof(elf_addr_t)))
366 		return -EFAULT;
367 	return 0;
368 }
369 
370 static unsigned long elf_map(struct file *filep, unsigned long addr,
371 		const struct elf_phdr *eppnt, int prot, int type,
372 		unsigned long total_size)
373 {
374 	unsigned long map_addr;
375 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
376 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
377 	addr = ELF_PAGESTART(addr);
378 	size = ELF_PAGEALIGN(size);
379 
380 	/* mmap() will return -EINVAL if given a zero size, but a
381 	 * segment with zero filesize is perfectly valid */
382 	if (!size)
383 		return addr;
384 
385 	/*
386 	* total_size is the size of the ELF (interpreter) image.
387 	* The _first_ mmap needs to know the full size, otherwise
388 	* randomization might put this image into an overlapping
389 	* position with the ELF binary image. (since size < total_size)
390 	* So we first map the 'big' image - and unmap the remainder at
391 	* the end. (which unmap is needed for ELF images with holes.)
392 	*/
393 	if (total_size) {
394 		total_size = ELF_PAGEALIGN(total_size);
395 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
396 		if (!BAD_ADDR(map_addr))
397 			vm_munmap(map_addr+size, total_size-size);
398 	} else
399 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
400 
401 	if ((type & MAP_FIXED_NOREPLACE) &&
402 	    PTR_ERR((void *)map_addr) == -EEXIST)
403 		pr_info("%d (%s): Uhuuh, elf segment at %px requested but the memory is mapped already\n",
404 			task_pid_nr(current), current->comm, (void *)addr);
405 
406 	return(map_addr);
407 }
408 
409 static unsigned long total_mapping_size(const struct elf_phdr *phdr, int nr)
410 {
411 	elf_addr_t min_addr = -1;
412 	elf_addr_t max_addr = 0;
413 	bool pt_load = false;
414 	int i;
415 
416 	for (i = 0; i < nr; i++) {
417 		if (phdr[i].p_type == PT_LOAD) {
418 			min_addr = min(min_addr, ELF_PAGESTART(phdr[i].p_vaddr));
419 			max_addr = max(max_addr, phdr[i].p_vaddr + phdr[i].p_memsz);
420 			pt_load = true;
421 		}
422 	}
423 	return pt_load ? (max_addr - min_addr) : 0;
424 }
425 
426 static int elf_read(struct file *file, void *buf, size_t len, loff_t pos)
427 {
428 	ssize_t rv;
429 
430 	rv = kernel_read(file, buf, len, &pos);
431 	if (unlikely(rv != len)) {
432 		return (rv < 0) ? rv : -EIO;
433 	}
434 	return 0;
435 }
436 
437 static unsigned long maximum_alignment(struct elf_phdr *cmds, int nr)
438 {
439 	unsigned long alignment = 0;
440 	int i;
441 
442 	for (i = 0; i < nr; i++) {
443 		if (cmds[i].p_type == PT_LOAD) {
444 			unsigned long p_align = cmds[i].p_align;
445 
446 			/* skip non-power of two alignments as invalid */
447 			if (!is_power_of_2(p_align))
448 				continue;
449 			alignment = max(alignment, p_align);
450 		}
451 	}
452 
453 	/* ensure we align to at least one page */
454 	return ELF_PAGEALIGN(alignment);
455 }
456 
457 /**
458  * load_elf_phdrs() - load ELF program headers
459  * @elf_ex:   ELF header of the binary whose program headers should be loaded
460  * @elf_file: the opened ELF binary file
461  *
462  * Loads ELF program headers from the binary file elf_file, which has the ELF
463  * header pointed to by elf_ex, into a newly allocated array. The caller is
464  * responsible for freeing the allocated data. Returns NULL upon failure.
465  */
466 static struct elf_phdr *load_elf_phdrs(const struct elfhdr *elf_ex,
467 				       struct file *elf_file)
468 {
469 	struct elf_phdr *elf_phdata = NULL;
470 	int retval = -1;
471 	unsigned int size;
472 
473 	/*
474 	 * If the size of this structure has changed, then punt, since
475 	 * we will be doing the wrong thing.
476 	 */
477 	if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
478 		goto out;
479 
480 	/* Sanity check the number of program headers... */
481 	/* ...and their total size. */
482 	size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
483 	if (size == 0 || size > 65536 || size > ELF_MIN_ALIGN)
484 		goto out;
485 
486 	elf_phdata = kmalloc(size, GFP_KERNEL);
487 	if (!elf_phdata)
488 		goto out;
489 
490 	/* Read in the program headers */
491 	retval = elf_read(elf_file, elf_phdata, size, elf_ex->e_phoff);
492 
493 out:
494 	if (retval) {
495 		kfree(elf_phdata);
496 		elf_phdata = NULL;
497 	}
498 	return elf_phdata;
499 }
500 
501 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
502 
503 /**
504  * struct arch_elf_state - arch-specific ELF loading state
505  *
506  * This structure is used to preserve architecture specific data during
507  * the loading of an ELF file, throughout the checking of architecture
508  * specific ELF headers & through to the point where the ELF load is
509  * known to be proceeding (ie. SET_PERSONALITY).
510  *
511  * This implementation is a dummy for architectures which require no
512  * specific state.
513  */
514 struct arch_elf_state {
515 };
516 
517 #define INIT_ARCH_ELF_STATE {}
518 
519 /**
520  * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
521  * @ehdr:	The main ELF header
522  * @phdr:	The program header to check
523  * @elf:	The open ELF file
524  * @is_interp:	True if the phdr is from the interpreter of the ELF being
525  *		loaded, else false.
526  * @state:	Architecture-specific state preserved throughout the process
527  *		of loading the ELF.
528  *
529  * Inspects the program header phdr to validate its correctness and/or
530  * suitability for the system. Called once per ELF program header in the
531  * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
532  * interpreter.
533  *
534  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
535  *         with that return code.
536  */
537 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
538 				   struct elf_phdr *phdr,
539 				   struct file *elf, bool is_interp,
540 				   struct arch_elf_state *state)
541 {
542 	/* Dummy implementation, always proceed */
543 	return 0;
544 }
545 
546 /**
547  * arch_check_elf() - check an ELF executable
548  * @ehdr:	The main ELF header
549  * @has_interp:	True if the ELF has an interpreter, else false.
550  * @interp_ehdr: The interpreter's ELF header
551  * @state:	Architecture-specific state preserved throughout the process
552  *		of loading the ELF.
553  *
554  * Provides a final opportunity for architecture code to reject the loading
555  * of the ELF & cause an exec syscall to return an error. This is called after
556  * all program headers to be checked by arch_elf_pt_proc have been.
557  *
558  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
559  *         with that return code.
560  */
561 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
562 				 struct elfhdr *interp_ehdr,
563 				 struct arch_elf_state *state)
564 {
565 	/* Dummy implementation, always proceed */
566 	return 0;
567 }
568 
569 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
570 
571 static inline int make_prot(u32 p_flags, struct arch_elf_state *arch_state,
572 			    bool has_interp, bool is_interp)
573 {
574 	int prot = 0;
575 
576 	if (p_flags & PF_R)
577 		prot |= PROT_READ;
578 	if (p_flags & PF_W)
579 		prot |= PROT_WRITE;
580 	if (p_flags & PF_X)
581 		prot |= PROT_EXEC;
582 
583 	return arch_elf_adjust_prot(prot, arch_state, has_interp, is_interp);
584 }
585 
586 /* This is much more generalized than the library routine read function,
587    so we keep this separate.  Technically the library read function
588    is only provided so that we can read a.out libraries that have
589    an ELF header */
590 
591 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
592 		struct file *interpreter,
593 		unsigned long no_base, struct elf_phdr *interp_elf_phdata,
594 		struct arch_elf_state *arch_state)
595 {
596 	struct elf_phdr *eppnt;
597 	unsigned long load_addr = 0;
598 	int load_addr_set = 0;
599 	unsigned long last_bss = 0, elf_bss = 0;
600 	int bss_prot = 0;
601 	unsigned long error = ~0UL;
602 	unsigned long total_size;
603 	int i;
604 
605 	/* First of all, some simple consistency checks */
606 	if (interp_elf_ex->e_type != ET_EXEC &&
607 	    interp_elf_ex->e_type != ET_DYN)
608 		goto out;
609 	if (!elf_check_arch(interp_elf_ex) ||
610 	    elf_check_fdpic(interp_elf_ex))
611 		goto out;
612 	if (!interpreter->f_op->mmap)
613 		goto out;
614 
615 	total_size = total_mapping_size(interp_elf_phdata,
616 					interp_elf_ex->e_phnum);
617 	if (!total_size) {
618 		error = -EINVAL;
619 		goto out;
620 	}
621 
622 	eppnt = interp_elf_phdata;
623 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
624 		if (eppnt->p_type == PT_LOAD) {
625 			int elf_type = MAP_PRIVATE;
626 			int elf_prot = make_prot(eppnt->p_flags, arch_state,
627 						 true, true);
628 			unsigned long vaddr = 0;
629 			unsigned long k, map_addr;
630 
631 			vaddr = eppnt->p_vaddr;
632 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
633 				elf_type |= MAP_FIXED;
634 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
635 				load_addr = -vaddr;
636 
637 			map_addr = elf_map(interpreter, load_addr + vaddr,
638 					eppnt, elf_prot, elf_type, total_size);
639 			total_size = 0;
640 			error = map_addr;
641 			if (BAD_ADDR(map_addr))
642 				goto out;
643 
644 			if (!load_addr_set &&
645 			    interp_elf_ex->e_type == ET_DYN) {
646 				load_addr = map_addr - ELF_PAGESTART(vaddr);
647 				load_addr_set = 1;
648 			}
649 
650 			/*
651 			 * Check to see if the section's size will overflow the
652 			 * allowed task size. Note that p_filesz must always be
653 			 * <= p_memsize so it's only necessary to check p_memsz.
654 			 */
655 			k = load_addr + eppnt->p_vaddr;
656 			if (BAD_ADDR(k) ||
657 			    eppnt->p_filesz > eppnt->p_memsz ||
658 			    eppnt->p_memsz > TASK_SIZE ||
659 			    TASK_SIZE - eppnt->p_memsz < k) {
660 				error = -ENOMEM;
661 				goto out;
662 			}
663 
664 			/*
665 			 * Find the end of the file mapping for this phdr, and
666 			 * keep track of the largest address we see for this.
667 			 */
668 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
669 			if (k > elf_bss)
670 				elf_bss = k;
671 
672 			/*
673 			 * Do the same thing for the memory mapping - between
674 			 * elf_bss and last_bss is the bss section.
675 			 */
676 			k = load_addr + eppnt->p_vaddr + eppnt->p_memsz;
677 			if (k > last_bss) {
678 				last_bss = k;
679 				bss_prot = elf_prot;
680 			}
681 		}
682 	}
683 
684 	/*
685 	 * Now fill out the bss section: first pad the last page from
686 	 * the file up to the page boundary, and zero it from elf_bss
687 	 * up to the end of the page.
688 	 */
689 	if (padzero(elf_bss)) {
690 		error = -EFAULT;
691 		goto out;
692 	}
693 	/*
694 	 * Next, align both the file and mem bss up to the page size,
695 	 * since this is where elf_bss was just zeroed up to, and where
696 	 * last_bss will end after the vm_brk_flags() below.
697 	 */
698 	elf_bss = ELF_PAGEALIGN(elf_bss);
699 	last_bss = ELF_PAGEALIGN(last_bss);
700 	/* Finally, if there is still more bss to allocate, do it. */
701 	if (last_bss > elf_bss) {
702 		error = vm_brk_flags(elf_bss, last_bss - elf_bss,
703 				bss_prot & PROT_EXEC ? VM_EXEC : 0);
704 		if (error)
705 			goto out;
706 	}
707 
708 	error = load_addr;
709 out:
710 	return error;
711 }
712 
713 /*
714  * These are the functions used to load ELF style executables and shared
715  * libraries.  There is no binary dependent code anywhere else.
716  */
717 
718 static int parse_elf_property(const char *data, size_t *off, size_t datasz,
719 			      struct arch_elf_state *arch,
720 			      bool have_prev_type, u32 *prev_type)
721 {
722 	size_t o, step;
723 	const struct gnu_property *pr;
724 	int ret;
725 
726 	if (*off == datasz)
727 		return -ENOENT;
728 
729 	if (WARN_ON_ONCE(*off > datasz || *off % ELF_GNU_PROPERTY_ALIGN))
730 		return -EIO;
731 	o = *off;
732 	datasz -= *off;
733 
734 	if (datasz < sizeof(*pr))
735 		return -ENOEXEC;
736 	pr = (const struct gnu_property *)(data + o);
737 	o += sizeof(*pr);
738 	datasz -= sizeof(*pr);
739 
740 	if (pr->pr_datasz > datasz)
741 		return -ENOEXEC;
742 
743 	WARN_ON_ONCE(o % ELF_GNU_PROPERTY_ALIGN);
744 	step = round_up(pr->pr_datasz, ELF_GNU_PROPERTY_ALIGN);
745 	if (step > datasz)
746 		return -ENOEXEC;
747 
748 	/* Properties are supposed to be unique and sorted on pr_type: */
749 	if (have_prev_type && pr->pr_type <= *prev_type)
750 		return -ENOEXEC;
751 	*prev_type = pr->pr_type;
752 
753 	ret = arch_parse_elf_property(pr->pr_type, data + o,
754 				      pr->pr_datasz, ELF_COMPAT, arch);
755 	if (ret)
756 		return ret;
757 
758 	*off = o + step;
759 	return 0;
760 }
761 
762 #define NOTE_DATA_SZ SZ_1K
763 #define GNU_PROPERTY_TYPE_0_NAME "GNU"
764 #define NOTE_NAME_SZ (sizeof(GNU_PROPERTY_TYPE_0_NAME))
765 
766 static int parse_elf_properties(struct file *f, const struct elf_phdr *phdr,
767 				struct arch_elf_state *arch)
768 {
769 	union {
770 		struct elf_note nhdr;
771 		char data[NOTE_DATA_SZ];
772 	} note;
773 	loff_t pos;
774 	ssize_t n;
775 	size_t off, datasz;
776 	int ret;
777 	bool have_prev_type;
778 	u32 prev_type;
779 
780 	if (!IS_ENABLED(CONFIG_ARCH_USE_GNU_PROPERTY) || !phdr)
781 		return 0;
782 
783 	/* load_elf_binary() shouldn't call us unless this is true... */
784 	if (WARN_ON_ONCE(phdr->p_type != PT_GNU_PROPERTY))
785 		return -ENOEXEC;
786 
787 	/* If the properties are crazy large, that's too bad (for now): */
788 	if (phdr->p_filesz > sizeof(note))
789 		return -ENOEXEC;
790 
791 	pos = phdr->p_offset;
792 	n = kernel_read(f, &note, phdr->p_filesz, &pos);
793 
794 	BUILD_BUG_ON(sizeof(note) < sizeof(note.nhdr) + NOTE_NAME_SZ);
795 	if (n < 0 || n < sizeof(note.nhdr) + NOTE_NAME_SZ)
796 		return -EIO;
797 
798 	if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
799 	    note.nhdr.n_namesz != NOTE_NAME_SZ ||
800 	    strncmp(note.data + sizeof(note.nhdr),
801 		    GNU_PROPERTY_TYPE_0_NAME, n - sizeof(note.nhdr)))
802 		return -ENOEXEC;
803 
804 	off = round_up(sizeof(note.nhdr) + NOTE_NAME_SZ,
805 		       ELF_GNU_PROPERTY_ALIGN);
806 	if (off > n)
807 		return -ENOEXEC;
808 
809 	if (note.nhdr.n_descsz > n - off)
810 		return -ENOEXEC;
811 	datasz = off + note.nhdr.n_descsz;
812 
813 	have_prev_type = false;
814 	do {
815 		ret = parse_elf_property(note.data, &off, datasz, arch,
816 					 have_prev_type, &prev_type);
817 		have_prev_type = true;
818 	} while (!ret);
819 
820 	return ret == -ENOENT ? 0 : ret;
821 }
822 
823 static int load_elf_binary(struct linux_binprm *bprm)
824 {
825 	struct file *interpreter = NULL; /* to shut gcc up */
826 	unsigned long load_bias = 0, phdr_addr = 0;
827 	int first_pt_load = 1;
828 	unsigned long error;
829 	struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
830 	struct elf_phdr *elf_property_phdata = NULL;
831 	unsigned long elf_bss, elf_brk;
832 	int bss_prot = 0;
833 	int retval, i;
834 	unsigned long elf_entry;
835 	unsigned long e_entry;
836 	unsigned long interp_load_addr = 0;
837 	unsigned long start_code, end_code, start_data, end_data;
838 	unsigned long reloc_func_desc __maybe_unused = 0;
839 	int executable_stack = EXSTACK_DEFAULT;
840 	struct elfhdr *elf_ex = (struct elfhdr *)bprm->buf;
841 	struct elfhdr *interp_elf_ex = NULL;
842 	struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
843 	struct mm_struct *mm;
844 	struct pt_regs *regs;
845 
846 	retval = -ENOEXEC;
847 	/* First of all, some simple consistency checks */
848 	if (memcmp(elf_ex->e_ident, ELFMAG, SELFMAG) != 0)
849 		goto out;
850 
851 	if (elf_ex->e_type != ET_EXEC && elf_ex->e_type != ET_DYN)
852 		goto out;
853 	if (!elf_check_arch(elf_ex))
854 		goto out;
855 	if (elf_check_fdpic(elf_ex))
856 		goto out;
857 	if (!bprm->file->f_op->mmap)
858 		goto out;
859 
860 	elf_phdata = load_elf_phdrs(elf_ex, bprm->file);
861 	if (!elf_phdata)
862 		goto out;
863 
864 	elf_ppnt = elf_phdata;
865 	for (i = 0; i < elf_ex->e_phnum; i++, elf_ppnt++) {
866 		char *elf_interpreter;
867 
868 		if (elf_ppnt->p_type == PT_GNU_PROPERTY) {
869 			elf_property_phdata = elf_ppnt;
870 			continue;
871 		}
872 
873 		if (elf_ppnt->p_type != PT_INTERP)
874 			continue;
875 
876 		/*
877 		 * This is the program interpreter used for shared libraries -
878 		 * for now assume that this is an a.out format binary.
879 		 */
880 		retval = -ENOEXEC;
881 		if (elf_ppnt->p_filesz > PATH_MAX || elf_ppnt->p_filesz < 2)
882 			goto out_free_ph;
883 
884 		retval = -ENOMEM;
885 		elf_interpreter = kmalloc(elf_ppnt->p_filesz, GFP_KERNEL);
886 		if (!elf_interpreter)
887 			goto out_free_ph;
888 
889 		retval = elf_read(bprm->file, elf_interpreter, elf_ppnt->p_filesz,
890 				  elf_ppnt->p_offset);
891 		if (retval < 0)
892 			goto out_free_interp;
893 		/* make sure path is NULL terminated */
894 		retval = -ENOEXEC;
895 		if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
896 			goto out_free_interp;
897 
898 		interpreter = open_exec(elf_interpreter);
899 		kfree(elf_interpreter);
900 		retval = PTR_ERR(interpreter);
901 		if (IS_ERR(interpreter))
902 			goto out_free_ph;
903 
904 		/*
905 		 * If the binary is not readable then enforce mm->dumpable = 0
906 		 * regardless of the interpreter's permissions.
907 		 */
908 		would_dump(bprm, interpreter);
909 
910 		interp_elf_ex = kmalloc(sizeof(*interp_elf_ex), GFP_KERNEL);
911 		if (!interp_elf_ex) {
912 			retval = -ENOMEM;
913 			goto out_free_file;
914 		}
915 
916 		/* Get the exec headers */
917 		retval = elf_read(interpreter, interp_elf_ex,
918 				  sizeof(*interp_elf_ex), 0);
919 		if (retval < 0)
920 			goto out_free_dentry;
921 
922 		break;
923 
924 out_free_interp:
925 		kfree(elf_interpreter);
926 		goto out_free_ph;
927 	}
928 
929 	elf_ppnt = elf_phdata;
930 	for (i = 0; i < elf_ex->e_phnum; i++, elf_ppnt++)
931 		switch (elf_ppnt->p_type) {
932 		case PT_GNU_STACK:
933 			if (elf_ppnt->p_flags & PF_X)
934 				executable_stack = EXSTACK_ENABLE_X;
935 			else
936 				executable_stack = EXSTACK_DISABLE_X;
937 			break;
938 
939 		case PT_LOPROC ... PT_HIPROC:
940 			retval = arch_elf_pt_proc(elf_ex, elf_ppnt,
941 						  bprm->file, false,
942 						  &arch_state);
943 			if (retval)
944 				goto out_free_dentry;
945 			break;
946 		}
947 
948 	/* Some simple consistency checks for the interpreter */
949 	if (interpreter) {
950 		retval = -ELIBBAD;
951 		/* Not an ELF interpreter */
952 		if (memcmp(interp_elf_ex->e_ident, ELFMAG, SELFMAG) != 0)
953 			goto out_free_dentry;
954 		/* Verify the interpreter has a valid arch */
955 		if (!elf_check_arch(interp_elf_ex) ||
956 		    elf_check_fdpic(interp_elf_ex))
957 			goto out_free_dentry;
958 
959 		/* Load the interpreter program headers */
960 		interp_elf_phdata = load_elf_phdrs(interp_elf_ex,
961 						   interpreter);
962 		if (!interp_elf_phdata)
963 			goto out_free_dentry;
964 
965 		/* Pass PT_LOPROC..PT_HIPROC headers to arch code */
966 		elf_property_phdata = NULL;
967 		elf_ppnt = interp_elf_phdata;
968 		for (i = 0; i < interp_elf_ex->e_phnum; i++, elf_ppnt++)
969 			switch (elf_ppnt->p_type) {
970 			case PT_GNU_PROPERTY:
971 				elf_property_phdata = elf_ppnt;
972 				break;
973 
974 			case PT_LOPROC ... PT_HIPROC:
975 				retval = arch_elf_pt_proc(interp_elf_ex,
976 							  elf_ppnt, interpreter,
977 							  true, &arch_state);
978 				if (retval)
979 					goto out_free_dentry;
980 				break;
981 			}
982 	}
983 
984 	retval = parse_elf_properties(interpreter ?: bprm->file,
985 				      elf_property_phdata, &arch_state);
986 	if (retval)
987 		goto out_free_dentry;
988 
989 	/*
990 	 * Allow arch code to reject the ELF at this point, whilst it's
991 	 * still possible to return an error to the code that invoked
992 	 * the exec syscall.
993 	 */
994 	retval = arch_check_elf(elf_ex,
995 				!!interpreter, interp_elf_ex,
996 				&arch_state);
997 	if (retval)
998 		goto out_free_dentry;
999 
1000 	/* Flush all traces of the currently running executable */
1001 	retval = begin_new_exec(bprm);
1002 	if (retval)
1003 		goto out_free_dentry;
1004 
1005 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
1006 	   may depend on the personality.  */
1007 	SET_PERSONALITY2(*elf_ex, &arch_state);
1008 	if (elf_read_implies_exec(*elf_ex, executable_stack))
1009 		current->personality |= READ_IMPLIES_EXEC;
1010 
1011 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1012 		current->flags |= PF_RANDOMIZE;
1013 
1014 	setup_new_exec(bprm);
1015 
1016 	/* Do this so that we can load the interpreter, if need be.  We will
1017 	   change some of these later */
1018 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
1019 				 executable_stack);
1020 	if (retval < 0)
1021 		goto out_free_dentry;
1022 
1023 	elf_bss = 0;
1024 	elf_brk = 0;
1025 
1026 	start_code = ~0UL;
1027 	end_code = 0;
1028 	start_data = 0;
1029 	end_data = 0;
1030 
1031 	/* Now we do a little grungy work by mmapping the ELF image into
1032 	   the correct location in memory. */
1033 	for(i = 0, elf_ppnt = elf_phdata;
1034 	    i < elf_ex->e_phnum; i++, elf_ppnt++) {
1035 		int elf_prot, elf_flags;
1036 		unsigned long k, vaddr;
1037 		unsigned long total_size = 0;
1038 		unsigned long alignment;
1039 
1040 		if (elf_ppnt->p_type != PT_LOAD)
1041 			continue;
1042 
1043 		if (unlikely (elf_brk > elf_bss)) {
1044 			unsigned long nbyte;
1045 
1046 			/* There was a PT_LOAD segment with p_memsz > p_filesz
1047 			   before this one. Map anonymous pages, if needed,
1048 			   and clear the area.  */
1049 			retval = set_brk(elf_bss + load_bias,
1050 					 elf_brk + load_bias,
1051 					 bss_prot);
1052 			if (retval)
1053 				goto out_free_dentry;
1054 			nbyte = ELF_PAGEOFFSET(elf_bss);
1055 			if (nbyte) {
1056 				nbyte = ELF_MIN_ALIGN - nbyte;
1057 				if (nbyte > elf_brk - elf_bss)
1058 					nbyte = elf_brk - elf_bss;
1059 				if (clear_user((void __user *)elf_bss +
1060 							load_bias, nbyte)) {
1061 					/*
1062 					 * This bss-zeroing can fail if the ELF
1063 					 * file specifies odd protections. So
1064 					 * we don't check the return value
1065 					 */
1066 				}
1067 			}
1068 		}
1069 
1070 		elf_prot = make_prot(elf_ppnt->p_flags, &arch_state,
1071 				     !!interpreter, false);
1072 
1073 		elf_flags = MAP_PRIVATE;
1074 
1075 		vaddr = elf_ppnt->p_vaddr;
1076 		/*
1077 		 * The first time through the loop, first_pt_load is true:
1078 		 * layout will be calculated. Once set, use MAP_FIXED since
1079 		 * we know we've already safely mapped the entire region with
1080 		 * MAP_FIXED_NOREPLACE in the once-per-binary logic following.
1081 		 */
1082 		if (!first_pt_load) {
1083 			elf_flags |= MAP_FIXED;
1084 		} else if (elf_ex->e_type == ET_EXEC) {
1085 			/*
1086 			 * This logic is run once for the first LOAD Program
1087 			 * Header for ET_EXEC binaries. No special handling
1088 			 * is needed.
1089 			 */
1090 			elf_flags |= MAP_FIXED_NOREPLACE;
1091 		} else if (elf_ex->e_type == ET_DYN) {
1092 			/*
1093 			 * This logic is run once for the first LOAD Program
1094 			 * Header for ET_DYN binaries to calculate the
1095 			 * randomization (load_bias) for all the LOAD
1096 			 * Program Headers.
1097 			 *
1098 			 * There are effectively two types of ET_DYN
1099 			 * binaries: programs (i.e. PIE: ET_DYN with INTERP)
1100 			 * and loaders (ET_DYN without INTERP, since they
1101 			 * _are_ the ELF interpreter). The loaders must
1102 			 * be loaded away from programs since the program
1103 			 * may otherwise collide with the loader (especially
1104 			 * for ET_EXEC which does not have a randomized
1105 			 * position). For example to handle invocations of
1106 			 * "./ld.so someprog" to test out a new version of
1107 			 * the loader, the subsequent program that the
1108 			 * loader loads must avoid the loader itself, so
1109 			 * they cannot share the same load range. Sufficient
1110 			 * room for the brk must be allocated with the
1111 			 * loader as well, since brk must be available with
1112 			 * the loader.
1113 			 *
1114 			 * Therefore, programs are loaded offset from
1115 			 * ELF_ET_DYN_BASE and loaders are loaded into the
1116 			 * independently randomized mmap region (0 load_bias
1117 			 * without MAP_FIXED nor MAP_FIXED_NOREPLACE).
1118 			 */
1119 			if (interpreter) {
1120 				load_bias = ELF_ET_DYN_BASE;
1121 				if (current->flags & PF_RANDOMIZE)
1122 					load_bias += arch_mmap_rnd();
1123 				alignment = maximum_alignment(elf_phdata, elf_ex->e_phnum);
1124 				if (alignment)
1125 					load_bias &= ~(alignment - 1);
1126 				elf_flags |= MAP_FIXED_NOREPLACE;
1127 			} else
1128 				load_bias = 0;
1129 
1130 			/*
1131 			 * Since load_bias is used for all subsequent loading
1132 			 * calculations, we must lower it by the first vaddr
1133 			 * so that the remaining calculations based on the
1134 			 * ELF vaddrs will be correctly offset. The result
1135 			 * is then page aligned.
1136 			 */
1137 			load_bias = ELF_PAGESTART(load_bias - vaddr);
1138 
1139 			/*
1140 			 * Calculate the entire size of the ELF mapping
1141 			 * (total_size), used for the initial mapping,
1142 			 * due to load_addr_set which is set to true later
1143 			 * once the initial mapping is performed.
1144 			 *
1145 			 * Note that this is only sensible when the LOAD
1146 			 * segments are contiguous (or overlapping). If
1147 			 * used for LOADs that are far apart, this would
1148 			 * cause the holes between LOADs to be mapped,
1149 			 * running the risk of having the mapping fail,
1150 			 * as it would be larger than the ELF file itself.
1151 			 *
1152 			 * As a result, only ET_DYN does this, since
1153 			 * some ET_EXEC (e.g. ia64) may have large virtual
1154 			 * memory holes between LOADs.
1155 			 *
1156 			 */
1157 			total_size = total_mapping_size(elf_phdata,
1158 							elf_ex->e_phnum);
1159 			if (!total_size) {
1160 				retval = -EINVAL;
1161 				goto out_free_dentry;
1162 			}
1163 		}
1164 
1165 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
1166 				elf_prot, elf_flags, total_size);
1167 		if (BAD_ADDR(error)) {
1168 			retval = IS_ERR_VALUE(error) ?
1169 				PTR_ERR((void*)error) : -EINVAL;
1170 			goto out_free_dentry;
1171 		}
1172 
1173 		if (first_pt_load) {
1174 			first_pt_load = 0;
1175 			if (elf_ex->e_type == ET_DYN) {
1176 				load_bias += error -
1177 				             ELF_PAGESTART(load_bias + vaddr);
1178 				reloc_func_desc = load_bias;
1179 			}
1180 		}
1181 
1182 		/*
1183 		 * Figure out which segment in the file contains the Program
1184 		 * Header table, and map to the associated memory address.
1185 		 */
1186 		if (elf_ppnt->p_offset <= elf_ex->e_phoff &&
1187 		    elf_ex->e_phoff < elf_ppnt->p_offset + elf_ppnt->p_filesz) {
1188 			phdr_addr = elf_ex->e_phoff - elf_ppnt->p_offset +
1189 				    elf_ppnt->p_vaddr;
1190 		}
1191 
1192 		k = elf_ppnt->p_vaddr;
1193 		if ((elf_ppnt->p_flags & PF_X) && k < start_code)
1194 			start_code = k;
1195 		if (start_data < k)
1196 			start_data = k;
1197 
1198 		/*
1199 		 * Check to see if the section's size will overflow the
1200 		 * allowed task size. Note that p_filesz must always be
1201 		 * <= p_memsz so it is only necessary to check p_memsz.
1202 		 */
1203 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
1204 		    elf_ppnt->p_memsz > TASK_SIZE ||
1205 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
1206 			/* set_brk can never work. Avoid overflows. */
1207 			retval = -EINVAL;
1208 			goto out_free_dentry;
1209 		}
1210 
1211 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
1212 
1213 		if (k > elf_bss)
1214 			elf_bss = k;
1215 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
1216 			end_code = k;
1217 		if (end_data < k)
1218 			end_data = k;
1219 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
1220 		if (k > elf_brk) {
1221 			bss_prot = elf_prot;
1222 			elf_brk = k;
1223 		}
1224 	}
1225 
1226 	e_entry = elf_ex->e_entry + load_bias;
1227 	phdr_addr += load_bias;
1228 	elf_bss += load_bias;
1229 	elf_brk += load_bias;
1230 	start_code += load_bias;
1231 	end_code += load_bias;
1232 	start_data += load_bias;
1233 	end_data += load_bias;
1234 
1235 	/* Calling set_brk effectively mmaps the pages that we need
1236 	 * for the bss and break sections.  We must do this before
1237 	 * mapping in the interpreter, to make sure it doesn't wind
1238 	 * up getting placed where the bss needs to go.
1239 	 */
1240 	retval = set_brk(elf_bss, elf_brk, bss_prot);
1241 	if (retval)
1242 		goto out_free_dentry;
1243 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
1244 		retval = -EFAULT; /* Nobody gets to see this, but.. */
1245 		goto out_free_dentry;
1246 	}
1247 
1248 	if (interpreter) {
1249 		elf_entry = load_elf_interp(interp_elf_ex,
1250 					    interpreter,
1251 					    load_bias, interp_elf_phdata,
1252 					    &arch_state);
1253 		if (!IS_ERR_VALUE(elf_entry)) {
1254 			/*
1255 			 * load_elf_interp() returns relocation
1256 			 * adjustment
1257 			 */
1258 			interp_load_addr = elf_entry;
1259 			elf_entry += interp_elf_ex->e_entry;
1260 		}
1261 		if (BAD_ADDR(elf_entry)) {
1262 			retval = IS_ERR_VALUE(elf_entry) ?
1263 					(int)elf_entry : -EINVAL;
1264 			goto out_free_dentry;
1265 		}
1266 		reloc_func_desc = interp_load_addr;
1267 
1268 		allow_write_access(interpreter);
1269 		fput(interpreter);
1270 
1271 		kfree(interp_elf_ex);
1272 		kfree(interp_elf_phdata);
1273 	} else {
1274 		elf_entry = e_entry;
1275 		if (BAD_ADDR(elf_entry)) {
1276 			retval = -EINVAL;
1277 			goto out_free_dentry;
1278 		}
1279 	}
1280 
1281 	kfree(elf_phdata);
1282 
1283 	set_binfmt(&elf_format);
1284 
1285 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1286 	retval = ARCH_SETUP_ADDITIONAL_PAGES(bprm, elf_ex, !!interpreter);
1287 	if (retval < 0)
1288 		goto out;
1289 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1290 
1291 	retval = create_elf_tables(bprm, elf_ex, interp_load_addr,
1292 				   e_entry, phdr_addr);
1293 	if (retval < 0)
1294 		goto out;
1295 
1296 	mm = current->mm;
1297 	mm->end_code = end_code;
1298 	mm->start_code = start_code;
1299 	mm->start_data = start_data;
1300 	mm->end_data = end_data;
1301 	mm->start_stack = bprm->p;
1302 
1303 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1304 		/*
1305 		 * For architectures with ELF randomization, when executing
1306 		 * a loader directly (i.e. no interpreter listed in ELF
1307 		 * headers), move the brk area out of the mmap region
1308 		 * (since it grows up, and may collide early with the stack
1309 		 * growing down), and into the unused ELF_ET_DYN_BASE region.
1310 		 */
1311 		if (IS_ENABLED(CONFIG_ARCH_HAS_ELF_RANDOMIZE) &&
1312 		    elf_ex->e_type == ET_DYN && !interpreter) {
1313 			mm->brk = mm->start_brk = ELF_ET_DYN_BASE;
1314 		}
1315 
1316 		mm->brk = mm->start_brk = arch_randomize_brk(mm);
1317 #ifdef compat_brk_randomized
1318 		current->brk_randomized = 1;
1319 #endif
1320 	}
1321 
1322 	if (current->personality & MMAP_PAGE_ZERO) {
1323 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
1324 		   and some applications "depend" upon this behavior.
1325 		   Since we do not have the power to recompile these, we
1326 		   emulate the SVr4 behavior. Sigh. */
1327 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1328 				MAP_FIXED | MAP_PRIVATE, 0);
1329 	}
1330 
1331 	regs = current_pt_regs();
1332 #ifdef ELF_PLAT_INIT
1333 	/*
1334 	 * The ABI may specify that certain registers be set up in special
1335 	 * ways (on i386 %edx is the address of a DT_FINI function, for
1336 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
1337 	 * that the e_entry field is the address of the function descriptor
1338 	 * for the startup routine, rather than the address of the startup
1339 	 * routine itself.  This macro performs whatever initialization to
1340 	 * the regs structure is required as well as any relocations to the
1341 	 * function descriptor entries when executing dynamically links apps.
1342 	 */
1343 	ELF_PLAT_INIT(regs, reloc_func_desc);
1344 #endif
1345 
1346 	finalize_exec(bprm);
1347 	START_THREAD(elf_ex, regs, elf_entry, bprm->p);
1348 	retval = 0;
1349 out:
1350 	return retval;
1351 
1352 	/* error cleanup */
1353 out_free_dentry:
1354 	kfree(interp_elf_ex);
1355 	kfree(interp_elf_phdata);
1356 out_free_file:
1357 	allow_write_access(interpreter);
1358 	if (interpreter)
1359 		fput(interpreter);
1360 out_free_ph:
1361 	kfree(elf_phdata);
1362 	goto out;
1363 }
1364 
1365 #ifdef CONFIG_USELIB
1366 /* This is really simpleminded and specialized - we are loading an
1367    a.out library that is given an ELF header. */
1368 static int load_elf_library(struct file *file)
1369 {
1370 	struct elf_phdr *elf_phdata;
1371 	struct elf_phdr *eppnt;
1372 	unsigned long elf_bss, bss, len;
1373 	int retval, error, i, j;
1374 	struct elfhdr elf_ex;
1375 
1376 	error = -ENOEXEC;
1377 	retval = elf_read(file, &elf_ex, sizeof(elf_ex), 0);
1378 	if (retval < 0)
1379 		goto out;
1380 
1381 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1382 		goto out;
1383 
1384 	/* First of all, some simple consistency checks */
1385 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1386 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1387 		goto out;
1388 	if (elf_check_fdpic(&elf_ex))
1389 		goto out;
1390 
1391 	/* Now read in all of the header information */
1392 
1393 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1394 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1395 
1396 	error = -ENOMEM;
1397 	elf_phdata = kmalloc(j, GFP_KERNEL);
1398 	if (!elf_phdata)
1399 		goto out;
1400 
1401 	eppnt = elf_phdata;
1402 	error = -ENOEXEC;
1403 	retval = elf_read(file, eppnt, j, elf_ex.e_phoff);
1404 	if (retval < 0)
1405 		goto out_free_ph;
1406 
1407 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1408 		if ((eppnt + i)->p_type == PT_LOAD)
1409 			j++;
1410 	if (j != 1)
1411 		goto out_free_ph;
1412 
1413 	while (eppnt->p_type != PT_LOAD)
1414 		eppnt++;
1415 
1416 	/* Now use mmap to map the library into memory. */
1417 	error = vm_mmap(file,
1418 			ELF_PAGESTART(eppnt->p_vaddr),
1419 			(eppnt->p_filesz +
1420 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1421 			PROT_READ | PROT_WRITE | PROT_EXEC,
1422 			MAP_FIXED_NOREPLACE | MAP_PRIVATE,
1423 			(eppnt->p_offset -
1424 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1425 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1426 		goto out_free_ph;
1427 
1428 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1429 	if (padzero(elf_bss)) {
1430 		error = -EFAULT;
1431 		goto out_free_ph;
1432 	}
1433 
1434 	len = ELF_PAGEALIGN(eppnt->p_filesz + eppnt->p_vaddr);
1435 	bss = ELF_PAGEALIGN(eppnt->p_memsz + eppnt->p_vaddr);
1436 	if (bss > len) {
1437 		error = vm_brk(len, bss - len);
1438 		if (error)
1439 			goto out_free_ph;
1440 	}
1441 	error = 0;
1442 
1443 out_free_ph:
1444 	kfree(elf_phdata);
1445 out:
1446 	return error;
1447 }
1448 #endif /* #ifdef CONFIG_USELIB */
1449 
1450 #ifdef CONFIG_ELF_CORE
1451 /*
1452  * ELF core dumper
1453  *
1454  * Modelled on fs/exec.c:aout_core_dump()
1455  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1456  */
1457 
1458 /* An ELF note in memory */
1459 struct memelfnote
1460 {
1461 	const char *name;
1462 	int type;
1463 	unsigned int datasz;
1464 	void *data;
1465 };
1466 
1467 static int notesize(struct memelfnote *en)
1468 {
1469 	int sz;
1470 
1471 	sz = sizeof(struct elf_note);
1472 	sz += roundup(strlen(en->name) + 1, 4);
1473 	sz += roundup(en->datasz, 4);
1474 
1475 	return sz;
1476 }
1477 
1478 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1479 {
1480 	struct elf_note en;
1481 	en.n_namesz = strlen(men->name) + 1;
1482 	en.n_descsz = men->datasz;
1483 	en.n_type = men->type;
1484 
1485 	return dump_emit(cprm, &en, sizeof(en)) &&
1486 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1487 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1488 }
1489 
1490 static void fill_elf_header(struct elfhdr *elf, int segs,
1491 			    u16 machine, u32 flags)
1492 {
1493 	memset(elf, 0, sizeof(*elf));
1494 
1495 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1496 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1497 	elf->e_ident[EI_DATA] = ELF_DATA;
1498 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1499 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1500 
1501 	elf->e_type = ET_CORE;
1502 	elf->e_machine = machine;
1503 	elf->e_version = EV_CURRENT;
1504 	elf->e_phoff = sizeof(struct elfhdr);
1505 	elf->e_flags = flags;
1506 	elf->e_ehsize = sizeof(struct elfhdr);
1507 	elf->e_phentsize = sizeof(struct elf_phdr);
1508 	elf->e_phnum = segs;
1509 }
1510 
1511 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1512 {
1513 	phdr->p_type = PT_NOTE;
1514 	phdr->p_offset = offset;
1515 	phdr->p_vaddr = 0;
1516 	phdr->p_paddr = 0;
1517 	phdr->p_filesz = sz;
1518 	phdr->p_memsz = 0;
1519 	phdr->p_flags = 0;
1520 	phdr->p_align = 0;
1521 }
1522 
1523 static void fill_note(struct memelfnote *note, const char *name, int type,
1524 		unsigned int sz, void *data)
1525 {
1526 	note->name = name;
1527 	note->type = type;
1528 	note->datasz = sz;
1529 	note->data = data;
1530 }
1531 
1532 /*
1533  * fill up all the fields in prstatus from the given task struct, except
1534  * registers which need to be filled up separately.
1535  */
1536 static void fill_prstatus(struct elf_prstatus_common *prstatus,
1537 		struct task_struct *p, long signr)
1538 {
1539 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1540 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1541 	prstatus->pr_sighold = p->blocked.sig[0];
1542 	rcu_read_lock();
1543 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1544 	rcu_read_unlock();
1545 	prstatus->pr_pid = task_pid_vnr(p);
1546 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1547 	prstatus->pr_sid = task_session_vnr(p);
1548 	if (thread_group_leader(p)) {
1549 		struct task_cputime cputime;
1550 
1551 		/*
1552 		 * This is the record for the group leader.  It shows the
1553 		 * group-wide total, not its individual thread total.
1554 		 */
1555 		thread_group_cputime(p, &cputime);
1556 		prstatus->pr_utime = ns_to_kernel_old_timeval(cputime.utime);
1557 		prstatus->pr_stime = ns_to_kernel_old_timeval(cputime.stime);
1558 	} else {
1559 		u64 utime, stime;
1560 
1561 		task_cputime(p, &utime, &stime);
1562 		prstatus->pr_utime = ns_to_kernel_old_timeval(utime);
1563 		prstatus->pr_stime = ns_to_kernel_old_timeval(stime);
1564 	}
1565 
1566 	prstatus->pr_cutime = ns_to_kernel_old_timeval(p->signal->cutime);
1567 	prstatus->pr_cstime = ns_to_kernel_old_timeval(p->signal->cstime);
1568 }
1569 
1570 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1571 		       struct mm_struct *mm)
1572 {
1573 	const struct cred *cred;
1574 	unsigned int i, len;
1575 	unsigned int state;
1576 
1577 	/* first copy the parameters from user space */
1578 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1579 
1580 	len = mm->arg_end - mm->arg_start;
1581 	if (len >= ELF_PRARGSZ)
1582 		len = ELF_PRARGSZ-1;
1583 	if (copy_from_user(&psinfo->pr_psargs,
1584 		           (const char __user *)mm->arg_start, len))
1585 		return -EFAULT;
1586 	for(i = 0; i < len; i++)
1587 		if (psinfo->pr_psargs[i] == 0)
1588 			psinfo->pr_psargs[i] = ' ';
1589 	psinfo->pr_psargs[len] = 0;
1590 
1591 	rcu_read_lock();
1592 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1593 	rcu_read_unlock();
1594 	psinfo->pr_pid = task_pid_vnr(p);
1595 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1596 	psinfo->pr_sid = task_session_vnr(p);
1597 
1598 	state = READ_ONCE(p->__state);
1599 	i = state ? ffz(~state) + 1 : 0;
1600 	psinfo->pr_state = i;
1601 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1602 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1603 	psinfo->pr_nice = task_nice(p);
1604 	psinfo->pr_flag = p->flags;
1605 	rcu_read_lock();
1606 	cred = __task_cred(p);
1607 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1608 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1609 	rcu_read_unlock();
1610 	get_task_comm(psinfo->pr_fname, p);
1611 
1612 	return 0;
1613 }
1614 
1615 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1616 {
1617 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1618 	int i = 0;
1619 	do
1620 		i += 2;
1621 	while (auxv[i - 2] != AT_NULL);
1622 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1623 }
1624 
1625 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1626 		const kernel_siginfo_t *siginfo)
1627 {
1628 	copy_siginfo_to_external(csigdata, siginfo);
1629 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1630 }
1631 
1632 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1633 /*
1634  * Format of NT_FILE note:
1635  *
1636  * long count     -- how many files are mapped
1637  * long page_size -- units for file_ofs
1638  * array of [COUNT] elements of
1639  *   long start
1640  *   long end
1641  *   long file_ofs
1642  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1643  */
1644 static int fill_files_note(struct memelfnote *note, struct coredump_params *cprm)
1645 {
1646 	unsigned count, size, names_ofs, remaining, n;
1647 	user_long_t *data;
1648 	user_long_t *start_end_ofs;
1649 	char *name_base, *name_curpos;
1650 	int i;
1651 
1652 	/* *Estimated* file count and total data size needed */
1653 	count = cprm->vma_count;
1654 	if (count > UINT_MAX / 64)
1655 		return -EINVAL;
1656 	size = count * 64;
1657 
1658 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1659  alloc:
1660 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1661 		return -EINVAL;
1662 	size = round_up(size, PAGE_SIZE);
1663 	/*
1664 	 * "size" can be 0 here legitimately.
1665 	 * Let it ENOMEM and omit NT_FILE section which will be empty anyway.
1666 	 */
1667 	data = kvmalloc(size, GFP_KERNEL);
1668 	if (ZERO_OR_NULL_PTR(data))
1669 		return -ENOMEM;
1670 
1671 	start_end_ofs = data + 2;
1672 	name_base = name_curpos = ((char *)data) + names_ofs;
1673 	remaining = size - names_ofs;
1674 	count = 0;
1675 	for (i = 0; i < cprm->vma_count; i++) {
1676 		struct core_vma_metadata *m = &cprm->vma_meta[i];
1677 		struct file *file;
1678 		const char *filename;
1679 
1680 		file = m->file;
1681 		if (!file)
1682 			continue;
1683 		filename = file_path(file, name_curpos, remaining);
1684 		if (IS_ERR(filename)) {
1685 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1686 				kvfree(data);
1687 				size = size * 5 / 4;
1688 				goto alloc;
1689 			}
1690 			continue;
1691 		}
1692 
1693 		/* file_path() fills at the end, move name down */
1694 		/* n = strlen(filename) + 1: */
1695 		n = (name_curpos + remaining) - filename;
1696 		remaining = filename - name_curpos;
1697 		memmove(name_curpos, filename, n);
1698 		name_curpos += n;
1699 
1700 		*start_end_ofs++ = m->start;
1701 		*start_end_ofs++ = m->end;
1702 		*start_end_ofs++ = m->pgoff;
1703 		count++;
1704 	}
1705 
1706 	/* Now we know exact count of files, can store it */
1707 	data[0] = count;
1708 	data[1] = PAGE_SIZE;
1709 	/*
1710 	 * Count usually is less than mm->map_count,
1711 	 * we need to move filenames down.
1712 	 */
1713 	n = cprm->vma_count - count;
1714 	if (n != 0) {
1715 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1716 		memmove(name_base - shift_bytes, name_base,
1717 			name_curpos - name_base);
1718 		name_curpos -= shift_bytes;
1719 	}
1720 
1721 	size = name_curpos - (char *)data;
1722 	fill_note(note, "CORE", NT_FILE, size, data);
1723 	return 0;
1724 }
1725 
1726 #include <linux/regset.h>
1727 
1728 struct elf_thread_core_info {
1729 	struct elf_thread_core_info *next;
1730 	struct task_struct *task;
1731 	struct elf_prstatus prstatus;
1732 	struct memelfnote notes[];
1733 };
1734 
1735 struct elf_note_info {
1736 	struct elf_thread_core_info *thread;
1737 	struct memelfnote psinfo;
1738 	struct memelfnote signote;
1739 	struct memelfnote auxv;
1740 	struct memelfnote files;
1741 	user_siginfo_t csigdata;
1742 	size_t size;
1743 	int thread_notes;
1744 };
1745 
1746 #ifdef CORE_DUMP_USE_REGSET
1747 /*
1748  * When a regset has a writeback hook, we call it on each thread before
1749  * dumping user memory.  On register window machines, this makes sure the
1750  * user memory backing the register data is up to date before we read it.
1751  */
1752 static void do_thread_regset_writeback(struct task_struct *task,
1753 				       const struct user_regset *regset)
1754 {
1755 	if (regset->writeback)
1756 		regset->writeback(task, regset, 1);
1757 }
1758 
1759 #ifndef PRSTATUS_SIZE
1760 #define PRSTATUS_SIZE sizeof(struct elf_prstatus)
1761 #endif
1762 
1763 #ifndef SET_PR_FPVALID
1764 #define SET_PR_FPVALID(S) ((S)->pr_fpvalid = 1)
1765 #endif
1766 
1767 static int fill_thread_core_info(struct elf_thread_core_info *t,
1768 				 const struct user_regset_view *view,
1769 				 long signr, struct elf_note_info *info)
1770 {
1771 	unsigned int note_iter, view_iter;
1772 
1773 	/*
1774 	 * NT_PRSTATUS is the one special case, because the regset data
1775 	 * goes into the pr_reg field inside the note contents, rather
1776 	 * than being the whole note contents.  We fill the reset in here.
1777 	 * We assume that regset 0 is NT_PRSTATUS.
1778 	 */
1779 	fill_prstatus(&t->prstatus.common, t->task, signr);
1780 	regset_get(t->task, &view->regsets[0],
1781 		   sizeof(t->prstatus.pr_reg), &t->prstatus.pr_reg);
1782 
1783 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1784 		  PRSTATUS_SIZE, &t->prstatus);
1785 	info->size += notesize(&t->notes[0]);
1786 
1787 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1788 
1789 	/*
1790 	 * Each other regset might generate a note too.  For each regset
1791 	 * that has no core_note_type or is inactive, skip it.
1792 	 */
1793 	note_iter = 1;
1794 	for (view_iter = 1; view_iter < view->n; ++view_iter) {
1795 		const struct user_regset *regset = &view->regsets[view_iter];
1796 		int note_type = regset->core_note_type;
1797 		bool is_fpreg = note_type == NT_PRFPREG;
1798 		void *data;
1799 		int ret;
1800 
1801 		do_thread_regset_writeback(t->task, regset);
1802 		if (!note_type) // not for coredumps
1803 			continue;
1804 		if (regset->active && regset->active(t->task, regset) <= 0)
1805 			continue;
1806 
1807 		ret = regset_get_alloc(t->task, regset, ~0U, &data);
1808 		if (ret < 0)
1809 			continue;
1810 
1811 		if (WARN_ON_ONCE(note_iter >= info->thread_notes))
1812 			break;
1813 
1814 		if (is_fpreg)
1815 			SET_PR_FPVALID(&t->prstatus);
1816 
1817 		fill_note(&t->notes[note_iter], is_fpreg ? "CORE" : "LINUX",
1818 			  note_type, ret, data);
1819 
1820 		info->size += notesize(&t->notes[note_iter]);
1821 		note_iter++;
1822 	}
1823 
1824 	return 1;
1825 }
1826 #else
1827 static int fill_thread_core_info(struct elf_thread_core_info *t,
1828 				 const struct user_regset_view *view,
1829 				 long signr, struct elf_note_info *info)
1830 {
1831 	struct task_struct *p = t->task;
1832 	elf_fpregset_t *fpu;
1833 
1834 	fill_prstatus(&t->prstatus.common, p, signr);
1835 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1836 
1837 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1838 		  &(t->prstatus));
1839 	info->size += notesize(&t->notes[0]);
1840 
1841 	fpu = kzalloc(sizeof(elf_fpregset_t), GFP_KERNEL);
1842 	if (!fpu || !elf_core_copy_task_fpregs(p, fpu)) {
1843 		kfree(fpu);
1844 		return 1;
1845 	}
1846 
1847 	t->prstatus.pr_fpvalid = 1;
1848 	fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1849 	info->size += notesize(&t->notes[1]);
1850 
1851 	return 1;
1852 }
1853 #endif
1854 
1855 static int fill_note_info(struct elfhdr *elf, int phdrs,
1856 			  struct elf_note_info *info,
1857 			  struct coredump_params *cprm)
1858 {
1859 	struct task_struct *dump_task = current;
1860 	const struct user_regset_view *view;
1861 	struct elf_thread_core_info *t;
1862 	struct elf_prpsinfo *psinfo;
1863 	struct core_thread *ct;
1864 
1865 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1866 	if (!psinfo)
1867 		return 0;
1868 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1869 
1870 #ifdef CORE_DUMP_USE_REGSET
1871 	view = task_user_regset_view(dump_task);
1872 
1873 	/*
1874 	 * Figure out how many notes we're going to need for each thread.
1875 	 */
1876 	info->thread_notes = 0;
1877 	for (int i = 0; i < view->n; ++i)
1878 		if (view->regsets[i].core_note_type != 0)
1879 			++info->thread_notes;
1880 
1881 	/*
1882 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1883 	 * since it is our one special case.
1884 	 */
1885 	if (unlikely(info->thread_notes == 0) ||
1886 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1887 		WARN_ON(1);
1888 		return 0;
1889 	}
1890 
1891 	/*
1892 	 * Initialize the ELF file header.
1893 	 */
1894 	fill_elf_header(elf, phdrs,
1895 			view->e_machine, view->e_flags);
1896 #else
1897 	view = NULL;
1898 	info->thread_notes = 2;
1899 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1900 #endif
1901 
1902 	/*
1903 	 * Allocate a structure for each thread.
1904 	 */
1905 	info->thread = kzalloc(offsetof(struct elf_thread_core_info,
1906 				     notes[info->thread_notes]),
1907 			    GFP_KERNEL);
1908 	if (unlikely(!info->thread))
1909 		return 0;
1910 
1911 	info->thread->task = dump_task;
1912 	for (ct = dump_task->signal->core_state->dumper.next; ct; ct = ct->next) {
1913 		t = kzalloc(offsetof(struct elf_thread_core_info,
1914 				     notes[info->thread_notes]),
1915 			    GFP_KERNEL);
1916 		if (unlikely(!t))
1917 			return 0;
1918 
1919 		t->task = ct->task;
1920 		t->next = info->thread->next;
1921 		info->thread->next = t;
1922 	}
1923 
1924 	/*
1925 	 * Now fill in each thread's information.
1926 	 */
1927 	for (t = info->thread; t != NULL; t = t->next)
1928 		if (!fill_thread_core_info(t, view, cprm->siginfo->si_signo, info))
1929 			return 0;
1930 
1931 	/*
1932 	 * Fill in the two process-wide notes.
1933 	 */
1934 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1935 	info->size += notesize(&info->psinfo);
1936 
1937 	fill_siginfo_note(&info->signote, &info->csigdata, cprm->siginfo);
1938 	info->size += notesize(&info->signote);
1939 
1940 	fill_auxv_note(&info->auxv, current->mm);
1941 	info->size += notesize(&info->auxv);
1942 
1943 	if (fill_files_note(&info->files, cprm) == 0)
1944 		info->size += notesize(&info->files);
1945 
1946 	return 1;
1947 }
1948 
1949 /*
1950  * Write all the notes for each thread.  When writing the first thread, the
1951  * process-wide notes are interleaved after the first thread-specific note.
1952  */
1953 static int write_note_info(struct elf_note_info *info,
1954 			   struct coredump_params *cprm)
1955 {
1956 	bool first = true;
1957 	struct elf_thread_core_info *t = info->thread;
1958 
1959 	do {
1960 		int i;
1961 
1962 		if (!writenote(&t->notes[0], cprm))
1963 			return 0;
1964 
1965 		if (first && !writenote(&info->psinfo, cprm))
1966 			return 0;
1967 		if (first && !writenote(&info->signote, cprm))
1968 			return 0;
1969 		if (first && !writenote(&info->auxv, cprm))
1970 			return 0;
1971 		if (first && info->files.data &&
1972 				!writenote(&info->files, cprm))
1973 			return 0;
1974 
1975 		for (i = 1; i < info->thread_notes; ++i)
1976 			if (t->notes[i].data &&
1977 			    !writenote(&t->notes[i], cprm))
1978 				return 0;
1979 
1980 		first = false;
1981 		t = t->next;
1982 	} while (t);
1983 
1984 	return 1;
1985 }
1986 
1987 static void free_note_info(struct elf_note_info *info)
1988 {
1989 	struct elf_thread_core_info *threads = info->thread;
1990 	while (threads) {
1991 		unsigned int i;
1992 		struct elf_thread_core_info *t = threads;
1993 		threads = t->next;
1994 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1995 		for (i = 1; i < info->thread_notes; ++i)
1996 			kfree(t->notes[i].data);
1997 		kfree(t);
1998 	}
1999 	kfree(info->psinfo.data);
2000 	kvfree(info->files.data);
2001 }
2002 
2003 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2004 			     elf_addr_t e_shoff, int segs)
2005 {
2006 	elf->e_shoff = e_shoff;
2007 	elf->e_shentsize = sizeof(*shdr4extnum);
2008 	elf->e_shnum = 1;
2009 	elf->e_shstrndx = SHN_UNDEF;
2010 
2011 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2012 
2013 	shdr4extnum->sh_type = SHT_NULL;
2014 	shdr4extnum->sh_size = elf->e_shnum;
2015 	shdr4extnum->sh_link = elf->e_shstrndx;
2016 	shdr4extnum->sh_info = segs;
2017 }
2018 
2019 /*
2020  * Actual dumper
2021  *
2022  * This is a two-pass process; first we find the offsets of the bits,
2023  * and then they are actually written out.  If we run out of core limit
2024  * we just truncate.
2025  */
2026 static int elf_core_dump(struct coredump_params *cprm)
2027 {
2028 	int has_dumped = 0;
2029 	int segs, i;
2030 	struct elfhdr elf;
2031 	loff_t offset = 0, dataoff;
2032 	struct elf_note_info info = { };
2033 	struct elf_phdr *phdr4note = NULL;
2034 	struct elf_shdr *shdr4extnum = NULL;
2035 	Elf_Half e_phnum;
2036 	elf_addr_t e_shoff;
2037 
2038 	/*
2039 	 * The number of segs are recored into ELF header as 16bit value.
2040 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2041 	 */
2042 	segs = cprm->vma_count + elf_core_extra_phdrs(cprm);
2043 
2044 	/* for notes section */
2045 	segs++;
2046 
2047 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2048 	 * this, kernel supports extended numbering. Have a look at
2049 	 * include/linux/elf.h for further information. */
2050 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2051 
2052 	/*
2053 	 * Collect all the non-memory information about the process for the
2054 	 * notes.  This also sets up the file header.
2055 	 */
2056 	if (!fill_note_info(&elf, e_phnum, &info, cprm))
2057 		goto end_coredump;
2058 
2059 	has_dumped = 1;
2060 
2061 	offset += sizeof(elf);				/* ELF header */
2062 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2063 
2064 	/* Write notes phdr entry */
2065 	{
2066 		size_t sz = info.size;
2067 
2068 		/* For cell spufs */
2069 		sz += elf_coredump_extra_notes_size();
2070 
2071 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2072 		if (!phdr4note)
2073 			goto end_coredump;
2074 
2075 		fill_elf_note_phdr(phdr4note, sz, offset);
2076 		offset += sz;
2077 	}
2078 
2079 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2080 
2081 	offset += cprm->vma_data_size;
2082 	offset += elf_core_extra_data_size(cprm);
2083 	e_shoff = offset;
2084 
2085 	if (e_phnum == PN_XNUM) {
2086 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2087 		if (!shdr4extnum)
2088 			goto end_coredump;
2089 		fill_extnum_info(&elf, shdr4extnum, e_shoff, segs);
2090 	}
2091 
2092 	offset = dataoff;
2093 
2094 	if (!dump_emit(cprm, &elf, sizeof(elf)))
2095 		goto end_coredump;
2096 
2097 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2098 		goto end_coredump;
2099 
2100 	/* Write program headers for segments dump */
2101 	for (i = 0; i < cprm->vma_count; i++) {
2102 		struct core_vma_metadata *meta = cprm->vma_meta + i;
2103 		struct elf_phdr phdr;
2104 
2105 		phdr.p_type = PT_LOAD;
2106 		phdr.p_offset = offset;
2107 		phdr.p_vaddr = meta->start;
2108 		phdr.p_paddr = 0;
2109 		phdr.p_filesz = meta->dump_size;
2110 		phdr.p_memsz = meta->end - meta->start;
2111 		offset += phdr.p_filesz;
2112 		phdr.p_flags = 0;
2113 		if (meta->flags & VM_READ)
2114 			phdr.p_flags |= PF_R;
2115 		if (meta->flags & VM_WRITE)
2116 			phdr.p_flags |= PF_W;
2117 		if (meta->flags & VM_EXEC)
2118 			phdr.p_flags |= PF_X;
2119 		phdr.p_align = ELF_EXEC_PAGESIZE;
2120 
2121 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2122 			goto end_coredump;
2123 	}
2124 
2125 	if (!elf_core_write_extra_phdrs(cprm, offset))
2126 		goto end_coredump;
2127 
2128 	/* write out the notes section */
2129 	if (!write_note_info(&info, cprm))
2130 		goto end_coredump;
2131 
2132 	/* For cell spufs */
2133 	if (elf_coredump_extra_notes_write(cprm))
2134 		goto end_coredump;
2135 
2136 	/* Align to page */
2137 	dump_skip_to(cprm, dataoff);
2138 
2139 	for (i = 0; i < cprm->vma_count; i++) {
2140 		struct core_vma_metadata *meta = cprm->vma_meta + i;
2141 
2142 		if (!dump_user_range(cprm, meta->start, meta->dump_size))
2143 			goto end_coredump;
2144 	}
2145 
2146 	if (!elf_core_write_extra_data(cprm))
2147 		goto end_coredump;
2148 
2149 	if (e_phnum == PN_XNUM) {
2150 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2151 			goto end_coredump;
2152 	}
2153 
2154 end_coredump:
2155 	free_note_info(&info);
2156 	kfree(shdr4extnum);
2157 	kfree(phdr4note);
2158 	return has_dumped;
2159 }
2160 
2161 #endif		/* CONFIG_ELF_CORE */
2162 
2163 static int __init init_elf_binfmt(void)
2164 {
2165 	register_binfmt(&elf_format);
2166 	return 0;
2167 }
2168 
2169 static void __exit exit_elf_binfmt(void)
2170 {
2171 	/* Remove the COFF and ELF loaders. */
2172 	unregister_binfmt(&elf_format);
2173 }
2174 
2175 core_initcall(init_elf_binfmt);
2176 module_exit(exit_elf_binfmt);
2177 
2178 #ifdef CONFIG_BINFMT_ELF_KUNIT_TEST
2179 #include "binfmt_elf_test.c"
2180 #endif
2181