xref: /linux/fs/binfmt_elf.c (revision c8b90d40d5bba8e6fba457b8a7c10d3c0d467e37)
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 /*
114  * We need to explicitly zero any trailing portion of the page that follows
115  * p_filesz when it ends before the page ends (e.g. bss), otherwise this
116  * memory will contain the junk from the file that should not be present.
117  */
118 static int padzero(unsigned long address)
119 {
120 	unsigned long nbyte;
121 
122 	nbyte = ELF_PAGEOFFSET(address);
123 	if (nbyte) {
124 		nbyte = ELF_MIN_ALIGN - nbyte;
125 		if (clear_user((void __user *)address, nbyte))
126 			return -EFAULT;
127 	}
128 	return 0;
129 }
130 
131 /* Let's use some macros to make this stack manipulation a little clearer */
132 #ifdef CONFIG_STACK_GROWSUP
133 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
134 #define STACK_ROUND(sp, items) \
135 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
136 #define STACK_ALLOC(sp, len) ({ \
137 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
138 	old_sp; })
139 #else
140 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
141 #define STACK_ROUND(sp, items) \
142 	(((unsigned long) (sp - items)) &~ 15UL)
143 #define STACK_ALLOC(sp, len) (sp -= len)
144 #endif
145 
146 #ifndef ELF_BASE_PLATFORM
147 /*
148  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
149  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
150  * will be copied to the user stack in the same manner as AT_PLATFORM.
151  */
152 #define ELF_BASE_PLATFORM NULL
153 #endif
154 
155 static int
156 create_elf_tables(struct linux_binprm *bprm, const struct elfhdr *exec,
157 		unsigned long interp_load_addr,
158 		unsigned long e_entry, unsigned long phdr_addr)
159 {
160 	struct mm_struct *mm = current->mm;
161 	unsigned long p = bprm->p;
162 	int argc = bprm->argc;
163 	int envc = bprm->envc;
164 	elf_addr_t __user *sp;
165 	elf_addr_t __user *u_platform;
166 	elf_addr_t __user *u_base_platform;
167 	elf_addr_t __user *u_rand_bytes;
168 	const char *k_platform = ELF_PLATFORM;
169 	const char *k_base_platform = ELF_BASE_PLATFORM;
170 	unsigned char k_rand_bytes[16];
171 	int items;
172 	elf_addr_t *elf_info;
173 	elf_addr_t flags = 0;
174 	int ei_index;
175 	const struct cred *cred = current_cred();
176 	struct vm_area_struct *vma;
177 
178 	/*
179 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
180 	 * evictions by the processes running on the same package. One
181 	 * thing we can do is to shuffle the initial stack for them.
182 	 */
183 
184 	p = arch_align_stack(p);
185 
186 	/*
187 	 * If this architecture has a platform capability string, copy it
188 	 * to userspace.  In some cases (Sparc), this info is impossible
189 	 * for userspace to get any other way, in others (i386) it is
190 	 * merely difficult.
191 	 */
192 	u_platform = NULL;
193 	if (k_platform) {
194 		size_t len = strlen(k_platform) + 1;
195 
196 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
197 		if (copy_to_user(u_platform, k_platform, len))
198 			return -EFAULT;
199 	}
200 
201 	/*
202 	 * If this architecture has a "base" platform capability
203 	 * string, copy it to userspace.
204 	 */
205 	u_base_platform = NULL;
206 	if (k_base_platform) {
207 		size_t len = strlen(k_base_platform) + 1;
208 
209 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
210 		if (copy_to_user(u_base_platform, k_base_platform, len))
211 			return -EFAULT;
212 	}
213 
214 	/*
215 	 * Generate 16 random bytes for userspace PRNG seeding.
216 	 */
217 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
218 	u_rand_bytes = (elf_addr_t __user *)
219 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
220 	if (copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
221 		return -EFAULT;
222 
223 	/* Create the ELF interpreter info */
224 	elf_info = (elf_addr_t *)mm->saved_auxv;
225 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
226 #define NEW_AUX_ENT(id, val) \
227 	do { \
228 		*elf_info++ = id; \
229 		*elf_info++ = val; \
230 	} while (0)
231 
232 #ifdef ARCH_DLINFO
233 	/*
234 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
235 	 * AUXV.
236 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
237 	 * ARCH_DLINFO changes
238 	 */
239 	ARCH_DLINFO;
240 #endif
241 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
242 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
243 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
244 	NEW_AUX_ENT(AT_PHDR, phdr_addr);
245 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
246 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
247 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
248 	if (bprm->interp_flags & BINPRM_FLAGS_PRESERVE_ARGV0)
249 		flags |= AT_FLAGS_PRESERVE_ARGV0;
250 	NEW_AUX_ENT(AT_FLAGS, flags);
251 	NEW_AUX_ENT(AT_ENTRY, e_entry);
252 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
253 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
254 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
255 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
256 	NEW_AUX_ENT(AT_SECURE, bprm->secureexec);
257 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
258 #ifdef ELF_HWCAP2
259 	NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
260 #endif
261 #ifdef ELF_HWCAP3
262 	NEW_AUX_ENT(AT_HWCAP3, ELF_HWCAP3);
263 #endif
264 #ifdef ELF_HWCAP4
265 	NEW_AUX_ENT(AT_HWCAP4, ELF_HWCAP4);
266 #endif
267 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
268 	if (k_platform) {
269 		NEW_AUX_ENT(AT_PLATFORM,
270 			    (elf_addr_t)(unsigned long)u_platform);
271 	}
272 	if (k_base_platform) {
273 		NEW_AUX_ENT(AT_BASE_PLATFORM,
274 			    (elf_addr_t)(unsigned long)u_base_platform);
275 	}
276 	if (bprm->have_execfd) {
277 		NEW_AUX_ENT(AT_EXECFD, bprm->execfd);
278 	}
279 #ifdef CONFIG_RSEQ
280 	NEW_AUX_ENT(AT_RSEQ_FEATURE_SIZE, offsetof(struct rseq, end));
281 	NEW_AUX_ENT(AT_RSEQ_ALIGN, __alignof__(struct rseq));
282 #endif
283 #undef NEW_AUX_ENT
284 	/* AT_NULL is zero; clear the rest too */
285 	memset(elf_info, 0, (char *)mm->saved_auxv +
286 			sizeof(mm->saved_auxv) - (char *)elf_info);
287 
288 	/* And advance past the AT_NULL entry.  */
289 	elf_info += 2;
290 
291 	ei_index = elf_info - (elf_addr_t *)mm->saved_auxv;
292 	sp = STACK_ADD(p, ei_index);
293 
294 	items = (argc + 1) + (envc + 1) + 1;
295 	bprm->p = STACK_ROUND(sp, items);
296 
297 	/* Point sp at the lowest address on the stack */
298 #ifdef CONFIG_STACK_GROWSUP
299 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
300 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
301 #else
302 	sp = (elf_addr_t __user *)bprm->p;
303 #endif
304 
305 
306 	/*
307 	 * Grow the stack manually; some architectures have a limit on how
308 	 * far ahead a user-space access may be in order to grow the stack.
309 	 */
310 	if (mmap_write_lock_killable(mm))
311 		return -EINTR;
312 	vma = find_extend_vma_locked(mm, bprm->p);
313 	mmap_write_unlock(mm);
314 	if (!vma)
315 		return -EFAULT;
316 
317 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
318 	if (put_user(argc, sp++))
319 		return -EFAULT;
320 
321 	/* Populate list of argv pointers back to argv strings. */
322 	p = mm->arg_end = mm->arg_start;
323 	while (argc-- > 0) {
324 		size_t len;
325 		if (put_user((elf_addr_t)p, sp++))
326 			return -EFAULT;
327 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
328 		if (!len || len > MAX_ARG_STRLEN)
329 			return -EINVAL;
330 		p += len;
331 	}
332 	if (put_user(0, sp++))
333 		return -EFAULT;
334 	mm->arg_end = p;
335 
336 	/* Populate list of envp pointers back to envp strings. */
337 	mm->env_end = mm->env_start = p;
338 	while (envc-- > 0) {
339 		size_t len;
340 		if (put_user((elf_addr_t)p, sp++))
341 			return -EFAULT;
342 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
343 		if (!len || len > MAX_ARG_STRLEN)
344 			return -EINVAL;
345 		p += len;
346 	}
347 	if (put_user(0, sp++))
348 		return -EFAULT;
349 	mm->env_end = p;
350 
351 	/* Put the elf_info on the stack in the right place.  */
352 	if (copy_to_user(sp, mm->saved_auxv, ei_index * sizeof(elf_addr_t)))
353 		return -EFAULT;
354 	return 0;
355 }
356 
357 /*
358  * Map "eppnt->p_filesz" bytes from "filep" offset "eppnt->p_offset"
359  * into memory at "addr". (Note that p_filesz is rounded up to the
360  * next page, so any extra bytes from the file must be wiped.)
361  */
362 static unsigned long elf_map(struct file *filep, unsigned long addr,
363 		const struct elf_phdr *eppnt, int prot, int type,
364 		unsigned long total_size)
365 {
366 	unsigned long map_addr;
367 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
368 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
369 	addr = ELF_PAGESTART(addr);
370 	size = ELF_PAGEALIGN(size);
371 
372 	/* mmap() will return -EINVAL if given a zero size, but a
373 	 * segment with zero filesize is perfectly valid */
374 	if (!size)
375 		return addr;
376 
377 	/*
378 	* total_size is the size of the ELF (interpreter) image.
379 	* The _first_ mmap needs to know the full size, otherwise
380 	* randomization might put this image into an overlapping
381 	* position with the ELF binary image. (since size < total_size)
382 	* So we first map the 'big' image - and unmap the remainder at
383 	* the end. (which unmap is needed for ELF images with holes.)
384 	*/
385 	if (total_size) {
386 		total_size = ELF_PAGEALIGN(total_size);
387 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
388 		if (!BAD_ADDR(map_addr))
389 			vm_munmap(map_addr+size, total_size-size);
390 	} else
391 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
392 
393 	if ((type & MAP_FIXED_NOREPLACE) &&
394 	    PTR_ERR((void *)map_addr) == -EEXIST)
395 		pr_info("%d (%s): Uhuuh, elf segment at %px requested but the memory is mapped already\n",
396 			task_pid_nr(current), current->comm, (void *)addr);
397 
398 	return(map_addr);
399 }
400 
401 /*
402  * Map "eppnt->p_filesz" bytes from "filep" offset "eppnt->p_offset"
403  * into memory at "addr". Memory from "p_filesz" through "p_memsz"
404  * rounded up to the next page is zeroed.
405  */
406 static unsigned long elf_load(struct file *filep, unsigned long addr,
407 		const struct elf_phdr *eppnt, int prot, int type,
408 		unsigned long total_size)
409 {
410 	unsigned long zero_start, zero_end;
411 	unsigned long map_addr;
412 
413 	if (eppnt->p_filesz) {
414 		map_addr = elf_map(filep, addr, eppnt, prot, type, total_size);
415 		if (BAD_ADDR(map_addr))
416 			return map_addr;
417 		if (eppnt->p_memsz > eppnt->p_filesz) {
418 			zero_start = map_addr + ELF_PAGEOFFSET(eppnt->p_vaddr) +
419 				eppnt->p_filesz;
420 			zero_end = map_addr + ELF_PAGEOFFSET(eppnt->p_vaddr) +
421 				eppnt->p_memsz;
422 
423 			/*
424 			 * Zero the end of the last mapped page but ignore
425 			 * any errors if the segment isn't writable.
426 			 */
427 			if (padzero(zero_start) && (prot & PROT_WRITE))
428 				return -EFAULT;
429 		}
430 	} else {
431 		map_addr = zero_start = ELF_PAGESTART(addr);
432 		zero_end = zero_start + ELF_PAGEOFFSET(eppnt->p_vaddr) +
433 			eppnt->p_memsz;
434 	}
435 	if (eppnt->p_memsz > eppnt->p_filesz) {
436 		/*
437 		 * Map the last of the segment.
438 		 * If the header is requesting these pages to be
439 		 * executable, honour that (ppc32 needs this).
440 		 */
441 		int error;
442 
443 		zero_start = ELF_PAGEALIGN(zero_start);
444 		zero_end = ELF_PAGEALIGN(zero_end);
445 
446 		error = vm_brk_flags(zero_start, zero_end - zero_start,
447 				     prot & PROT_EXEC ? VM_EXEC : 0);
448 		if (error)
449 			map_addr = error;
450 	}
451 	return map_addr;
452 }
453 
454 
455 static unsigned long total_mapping_size(const struct elf_phdr *phdr, int nr)
456 {
457 	elf_addr_t min_addr = -1;
458 	elf_addr_t max_addr = 0;
459 	bool pt_load = false;
460 	int i;
461 
462 	for (i = 0; i < nr; i++) {
463 		if (phdr[i].p_type == PT_LOAD) {
464 			min_addr = min(min_addr, ELF_PAGESTART(phdr[i].p_vaddr));
465 			max_addr = max(max_addr, phdr[i].p_vaddr + phdr[i].p_memsz);
466 			pt_load = true;
467 		}
468 	}
469 	return pt_load ? (max_addr - min_addr) : 0;
470 }
471 
472 static int elf_read(struct file *file, void *buf, size_t len, loff_t pos)
473 {
474 	ssize_t rv;
475 
476 	rv = kernel_read(file, buf, len, &pos);
477 	if (unlikely(rv != len)) {
478 		return (rv < 0) ? rv : -EIO;
479 	}
480 	return 0;
481 }
482 
483 static unsigned long maximum_alignment(struct elf_phdr *cmds, int nr)
484 {
485 	unsigned long alignment = 0;
486 	int i;
487 
488 	for (i = 0; i < nr; i++) {
489 		if (cmds[i].p_type == PT_LOAD) {
490 			unsigned long p_align = cmds[i].p_align;
491 
492 			/* skip non-power of two alignments as invalid */
493 			if (!is_power_of_2(p_align))
494 				continue;
495 			alignment = max(alignment, p_align);
496 		}
497 	}
498 
499 	/* ensure we align to at least one page */
500 	return ELF_PAGEALIGN(alignment);
501 }
502 
503 /**
504  * load_elf_phdrs() - load ELF program headers
505  * @elf_ex:   ELF header of the binary whose program headers should be loaded
506  * @elf_file: the opened ELF binary file
507  *
508  * Loads ELF program headers from the binary file elf_file, which has the ELF
509  * header pointed to by elf_ex, into a newly allocated array. The caller is
510  * responsible for freeing the allocated data. Returns NULL upon failure.
511  */
512 static struct elf_phdr *load_elf_phdrs(const struct elfhdr *elf_ex,
513 				       struct file *elf_file)
514 {
515 	struct elf_phdr *elf_phdata = NULL;
516 	int retval = -1;
517 	unsigned int size;
518 
519 	/*
520 	 * If the size of this structure has changed, then punt, since
521 	 * we will be doing the wrong thing.
522 	 */
523 	if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
524 		goto out;
525 
526 	/* Sanity check the number of program headers... */
527 	/* ...and their total size. */
528 	size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
529 	if (size == 0 || size > 65536 || size > ELF_MIN_ALIGN)
530 		goto out;
531 
532 	elf_phdata = kmalloc(size, GFP_KERNEL);
533 	if (!elf_phdata)
534 		goto out;
535 
536 	/* Read in the program headers */
537 	retval = elf_read(elf_file, elf_phdata, size, elf_ex->e_phoff);
538 
539 out:
540 	if (retval) {
541 		kfree(elf_phdata);
542 		elf_phdata = NULL;
543 	}
544 	return elf_phdata;
545 }
546 
547 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
548 
549 /**
550  * struct arch_elf_state - arch-specific ELF loading state
551  *
552  * This structure is used to preserve architecture specific data during
553  * the loading of an ELF file, throughout the checking of architecture
554  * specific ELF headers & through to the point where the ELF load is
555  * known to be proceeding (ie. SET_PERSONALITY).
556  *
557  * This implementation is a dummy for architectures which require no
558  * specific state.
559  */
560 struct arch_elf_state {
561 };
562 
563 #define INIT_ARCH_ELF_STATE {}
564 
565 /**
566  * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
567  * @ehdr:	The main ELF header
568  * @phdr:	The program header to check
569  * @elf:	The open ELF file
570  * @is_interp:	True if the phdr is from the interpreter of the ELF being
571  *		loaded, else false.
572  * @state:	Architecture-specific state preserved throughout the process
573  *		of loading the ELF.
574  *
575  * Inspects the program header phdr to validate its correctness and/or
576  * suitability for the system. Called once per ELF program header in the
577  * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
578  * interpreter.
579  *
580  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
581  *         with that return code.
582  */
583 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
584 				   struct elf_phdr *phdr,
585 				   struct file *elf, bool is_interp,
586 				   struct arch_elf_state *state)
587 {
588 	/* Dummy implementation, always proceed */
589 	return 0;
590 }
591 
592 /**
593  * arch_check_elf() - check an ELF executable
594  * @ehdr:	The main ELF header
595  * @has_interp:	True if the ELF has an interpreter, else false.
596  * @interp_ehdr: The interpreter's ELF header
597  * @state:	Architecture-specific state preserved throughout the process
598  *		of loading the ELF.
599  *
600  * Provides a final opportunity for architecture code to reject the loading
601  * of the ELF & cause an exec syscall to return an error. This is called after
602  * all program headers to be checked by arch_elf_pt_proc have been.
603  *
604  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
605  *         with that return code.
606  */
607 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
608 				 struct elfhdr *interp_ehdr,
609 				 struct arch_elf_state *state)
610 {
611 	/* Dummy implementation, always proceed */
612 	return 0;
613 }
614 
615 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
616 
617 static inline int make_prot(u32 p_flags, struct arch_elf_state *arch_state,
618 			    bool has_interp, bool is_interp)
619 {
620 	int prot = 0;
621 
622 	if (p_flags & PF_R)
623 		prot |= PROT_READ;
624 	if (p_flags & PF_W)
625 		prot |= PROT_WRITE;
626 	if (p_flags & PF_X)
627 		prot |= PROT_EXEC;
628 
629 	return arch_elf_adjust_prot(prot, arch_state, has_interp, is_interp);
630 }
631 
632 /* This is much more generalized than the library routine read function,
633    so we keep this separate.  Technically the library read function
634    is only provided so that we can read a.out libraries that have
635    an ELF header */
636 
637 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
638 		struct file *interpreter,
639 		unsigned long no_base, struct elf_phdr *interp_elf_phdata,
640 		struct arch_elf_state *arch_state)
641 {
642 	struct elf_phdr *eppnt;
643 	unsigned long load_addr = 0;
644 	int load_addr_set = 0;
645 	unsigned long error = ~0UL;
646 	unsigned long total_size;
647 	int i;
648 
649 	/* First of all, some simple consistency checks */
650 	if (interp_elf_ex->e_type != ET_EXEC &&
651 	    interp_elf_ex->e_type != ET_DYN)
652 		goto out;
653 	if (!elf_check_arch(interp_elf_ex) ||
654 	    elf_check_fdpic(interp_elf_ex))
655 		goto out;
656 	if (!interpreter->f_op->mmap)
657 		goto out;
658 
659 	total_size = total_mapping_size(interp_elf_phdata,
660 					interp_elf_ex->e_phnum);
661 	if (!total_size) {
662 		error = -EINVAL;
663 		goto out;
664 	}
665 
666 	eppnt = interp_elf_phdata;
667 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
668 		if (eppnt->p_type == PT_LOAD) {
669 			int elf_type = MAP_PRIVATE;
670 			int elf_prot = make_prot(eppnt->p_flags, arch_state,
671 						 true, true);
672 			unsigned long vaddr = 0;
673 			unsigned long k, map_addr;
674 
675 			vaddr = eppnt->p_vaddr;
676 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
677 				elf_type |= MAP_FIXED;
678 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
679 				load_addr = -vaddr;
680 
681 			map_addr = elf_load(interpreter, load_addr + vaddr,
682 					eppnt, elf_prot, elf_type, total_size);
683 			total_size = 0;
684 			error = map_addr;
685 			if (BAD_ADDR(map_addr))
686 				goto out;
687 
688 			if (!load_addr_set &&
689 			    interp_elf_ex->e_type == ET_DYN) {
690 				load_addr = map_addr - ELF_PAGESTART(vaddr);
691 				load_addr_set = 1;
692 			}
693 
694 			/*
695 			 * Check to see if the section's size will overflow the
696 			 * allowed task size. Note that p_filesz must always be
697 			 * <= p_memsize so it's only necessary to check p_memsz.
698 			 */
699 			k = load_addr + eppnt->p_vaddr;
700 			if (BAD_ADDR(k) ||
701 			    eppnt->p_filesz > eppnt->p_memsz ||
702 			    eppnt->p_memsz > TASK_SIZE ||
703 			    TASK_SIZE - eppnt->p_memsz < k) {
704 				error = -ENOMEM;
705 				goto out;
706 			}
707 		}
708 	}
709 
710 	error = load_addr;
711 out:
712 	return error;
713 }
714 
715 /*
716  * These are the functions used to load ELF style executables and shared
717  * libraries.  There is no binary dependent code anywhere else.
718  */
719 
720 static int parse_elf_property(const char *data, size_t *off, size_t datasz,
721 			      struct arch_elf_state *arch,
722 			      bool have_prev_type, u32 *prev_type)
723 {
724 	size_t o, step;
725 	const struct gnu_property *pr;
726 	int ret;
727 
728 	if (*off == datasz)
729 		return -ENOENT;
730 
731 	if (WARN_ON_ONCE(*off > datasz || *off % ELF_GNU_PROPERTY_ALIGN))
732 		return -EIO;
733 	o = *off;
734 	datasz -= *off;
735 
736 	if (datasz < sizeof(*pr))
737 		return -ENOEXEC;
738 	pr = (const struct gnu_property *)(data + o);
739 	o += sizeof(*pr);
740 	datasz -= sizeof(*pr);
741 
742 	if (pr->pr_datasz > datasz)
743 		return -ENOEXEC;
744 
745 	WARN_ON_ONCE(o % ELF_GNU_PROPERTY_ALIGN);
746 	step = round_up(pr->pr_datasz, ELF_GNU_PROPERTY_ALIGN);
747 	if (step > datasz)
748 		return -ENOEXEC;
749 
750 	/* Properties are supposed to be unique and sorted on pr_type: */
751 	if (have_prev_type && pr->pr_type <= *prev_type)
752 		return -ENOEXEC;
753 	*prev_type = pr->pr_type;
754 
755 	ret = arch_parse_elf_property(pr->pr_type, data + o,
756 				      pr->pr_datasz, ELF_COMPAT, arch);
757 	if (ret)
758 		return ret;
759 
760 	*off = o + step;
761 	return 0;
762 }
763 
764 #define NOTE_DATA_SZ SZ_1K
765 #define GNU_PROPERTY_TYPE_0_NAME "GNU"
766 #define NOTE_NAME_SZ (sizeof(GNU_PROPERTY_TYPE_0_NAME))
767 
768 static int parse_elf_properties(struct file *f, const struct elf_phdr *phdr,
769 				struct arch_elf_state *arch)
770 {
771 	union {
772 		struct elf_note nhdr;
773 		char data[NOTE_DATA_SZ];
774 	} note;
775 	loff_t pos;
776 	ssize_t n;
777 	size_t off, datasz;
778 	int ret;
779 	bool have_prev_type;
780 	u32 prev_type;
781 
782 	if (!IS_ENABLED(CONFIG_ARCH_USE_GNU_PROPERTY) || !phdr)
783 		return 0;
784 
785 	/* load_elf_binary() shouldn't call us unless this is true... */
786 	if (WARN_ON_ONCE(phdr->p_type != PT_GNU_PROPERTY))
787 		return -ENOEXEC;
788 
789 	/* If the properties are crazy large, that's too bad (for now): */
790 	if (phdr->p_filesz > sizeof(note))
791 		return -ENOEXEC;
792 
793 	pos = phdr->p_offset;
794 	n = kernel_read(f, &note, phdr->p_filesz, &pos);
795 
796 	BUILD_BUG_ON(sizeof(note) < sizeof(note.nhdr) + NOTE_NAME_SZ);
797 	if (n < 0 || n < sizeof(note.nhdr) + NOTE_NAME_SZ)
798 		return -EIO;
799 
800 	if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
801 	    note.nhdr.n_namesz != NOTE_NAME_SZ ||
802 	    strncmp(note.data + sizeof(note.nhdr),
803 		    GNU_PROPERTY_TYPE_0_NAME, n - sizeof(note.nhdr)))
804 		return -ENOEXEC;
805 
806 	off = round_up(sizeof(note.nhdr) + NOTE_NAME_SZ,
807 		       ELF_GNU_PROPERTY_ALIGN);
808 	if (off > n)
809 		return -ENOEXEC;
810 
811 	if (note.nhdr.n_descsz > n - off)
812 		return -ENOEXEC;
813 	datasz = off + note.nhdr.n_descsz;
814 
815 	have_prev_type = false;
816 	do {
817 		ret = parse_elf_property(note.data, &off, datasz, arch,
818 					 have_prev_type, &prev_type);
819 		have_prev_type = true;
820 	} while (!ret);
821 
822 	return ret == -ENOENT ? 0 : ret;
823 }
824 
825 static int load_elf_binary(struct linux_binprm *bprm)
826 {
827 	struct file *interpreter = NULL; /* to shut gcc up */
828 	unsigned long load_bias = 0, phdr_addr = 0;
829 	int first_pt_load = 1;
830 	unsigned long error;
831 	struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
832 	struct elf_phdr *elf_property_phdata = NULL;
833 	unsigned long elf_brk;
834 	int retval, i;
835 	unsigned long elf_entry;
836 	unsigned long e_entry;
837 	unsigned long interp_load_addr = 0;
838 	unsigned long start_code, end_code, start_data, end_data;
839 	unsigned long reloc_func_desc __maybe_unused = 0;
840 	int executable_stack = EXSTACK_DEFAULT;
841 	struct elfhdr *elf_ex = (struct elfhdr *)bprm->buf;
842 	struct elfhdr *interp_elf_ex = NULL;
843 	struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
844 	struct mm_struct *mm;
845 	struct pt_regs *regs;
846 
847 	retval = -ENOEXEC;
848 	/* First of all, some simple consistency checks */
849 	if (memcmp(elf_ex->e_ident, ELFMAG, SELFMAG) != 0)
850 		goto out;
851 
852 	if (elf_ex->e_type != ET_EXEC && elf_ex->e_type != ET_DYN)
853 		goto out;
854 	if (!elf_check_arch(elf_ex))
855 		goto out;
856 	if (elf_check_fdpic(elf_ex))
857 		goto out;
858 	if (!bprm->file->f_op->mmap)
859 		goto out;
860 
861 	elf_phdata = load_elf_phdrs(elf_ex, bprm->file);
862 	if (!elf_phdata)
863 		goto out;
864 
865 	elf_ppnt = elf_phdata;
866 	for (i = 0; i < elf_ex->e_phnum; i++, elf_ppnt++) {
867 		char *elf_interpreter;
868 
869 		if (elf_ppnt->p_type == PT_GNU_PROPERTY) {
870 			elf_property_phdata = elf_ppnt;
871 			continue;
872 		}
873 
874 		if (elf_ppnt->p_type != PT_INTERP)
875 			continue;
876 
877 		/*
878 		 * This is the program interpreter used for shared libraries -
879 		 * for now assume that this is an a.out format binary.
880 		 */
881 		retval = -ENOEXEC;
882 		if (elf_ppnt->p_filesz > PATH_MAX || elf_ppnt->p_filesz < 2)
883 			goto out_free_ph;
884 
885 		retval = -ENOMEM;
886 		elf_interpreter = kmalloc(elf_ppnt->p_filesz, GFP_KERNEL);
887 		if (!elf_interpreter)
888 			goto out_free_ph;
889 
890 		retval = elf_read(bprm->file, elf_interpreter, elf_ppnt->p_filesz,
891 				  elf_ppnt->p_offset);
892 		if (retval < 0)
893 			goto out_free_interp;
894 		/* make sure path is NULL terminated */
895 		retval = -ENOEXEC;
896 		if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
897 			goto out_free_interp;
898 
899 		interpreter = open_exec(elf_interpreter);
900 		kfree(elf_interpreter);
901 		retval = PTR_ERR(interpreter);
902 		if (IS_ERR(interpreter))
903 			goto out_free_ph;
904 
905 		/*
906 		 * If the binary is not readable then enforce mm->dumpable = 0
907 		 * regardless of the interpreter's permissions.
908 		 */
909 		would_dump(bprm, interpreter);
910 
911 		interp_elf_ex = kmalloc(sizeof(*interp_elf_ex), GFP_KERNEL);
912 		if (!interp_elf_ex) {
913 			retval = -ENOMEM;
914 			goto out_free_file;
915 		}
916 
917 		/* Get the exec headers */
918 		retval = elf_read(interpreter, interp_elf_ex,
919 				  sizeof(*interp_elf_ex), 0);
920 		if (retval < 0)
921 			goto out_free_dentry;
922 
923 		break;
924 
925 out_free_interp:
926 		kfree(elf_interpreter);
927 		goto out_free_ph;
928 	}
929 
930 	elf_ppnt = elf_phdata;
931 	for (i = 0; i < elf_ex->e_phnum; i++, elf_ppnt++)
932 		switch (elf_ppnt->p_type) {
933 		case PT_GNU_STACK:
934 			if (elf_ppnt->p_flags & PF_X)
935 				executable_stack = EXSTACK_ENABLE_X;
936 			else
937 				executable_stack = EXSTACK_DISABLE_X;
938 			break;
939 
940 		case PT_LOPROC ... PT_HIPROC:
941 			retval = arch_elf_pt_proc(elf_ex, elf_ppnt,
942 						  bprm->file, false,
943 						  &arch_state);
944 			if (retval)
945 				goto out_free_dentry;
946 			break;
947 		}
948 
949 	/* Some simple consistency checks for the interpreter */
950 	if (interpreter) {
951 		retval = -ELIBBAD;
952 		/* Not an ELF interpreter */
953 		if (memcmp(interp_elf_ex->e_ident, ELFMAG, SELFMAG) != 0)
954 			goto out_free_dentry;
955 		/* Verify the interpreter has a valid arch */
956 		if (!elf_check_arch(interp_elf_ex) ||
957 		    elf_check_fdpic(interp_elf_ex))
958 			goto out_free_dentry;
959 
960 		/* Load the interpreter program headers */
961 		interp_elf_phdata = load_elf_phdrs(interp_elf_ex,
962 						   interpreter);
963 		if (!interp_elf_phdata)
964 			goto out_free_dentry;
965 
966 		/* Pass PT_LOPROC..PT_HIPROC headers to arch code */
967 		elf_property_phdata = NULL;
968 		elf_ppnt = interp_elf_phdata;
969 		for (i = 0; i < interp_elf_ex->e_phnum; i++, elf_ppnt++)
970 			switch (elf_ppnt->p_type) {
971 			case PT_GNU_PROPERTY:
972 				elf_property_phdata = elf_ppnt;
973 				break;
974 
975 			case PT_LOPROC ... PT_HIPROC:
976 				retval = arch_elf_pt_proc(interp_elf_ex,
977 							  elf_ppnt, interpreter,
978 							  true, &arch_state);
979 				if (retval)
980 					goto out_free_dentry;
981 				break;
982 			}
983 	}
984 
985 	retval = parse_elf_properties(interpreter ?: bprm->file,
986 				      elf_property_phdata, &arch_state);
987 	if (retval)
988 		goto out_free_dentry;
989 
990 	/*
991 	 * Allow arch code to reject the ELF at this point, whilst it's
992 	 * still possible to return an error to the code that invoked
993 	 * the exec syscall.
994 	 */
995 	retval = arch_check_elf(elf_ex,
996 				!!interpreter, interp_elf_ex,
997 				&arch_state);
998 	if (retval)
999 		goto out_free_dentry;
1000 
1001 	/* Flush all traces of the currently running executable */
1002 	retval = begin_new_exec(bprm);
1003 	if (retval)
1004 		goto out_free_dentry;
1005 
1006 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
1007 	   may depend on the personality.  */
1008 	SET_PERSONALITY2(*elf_ex, &arch_state);
1009 	if (elf_read_implies_exec(*elf_ex, executable_stack))
1010 		current->personality |= READ_IMPLIES_EXEC;
1011 
1012 	const int snapshot_randomize_va_space = READ_ONCE(randomize_va_space);
1013 	if (!(current->personality & ADDR_NO_RANDOMIZE) && snapshot_randomize_va_space)
1014 		current->flags |= PF_RANDOMIZE;
1015 
1016 	setup_new_exec(bprm);
1017 
1018 	/* Do this so that we can load the interpreter, if need be.  We will
1019 	   change some of these later */
1020 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
1021 				 executable_stack);
1022 	if (retval < 0)
1023 		goto out_free_dentry;
1024 
1025 	elf_brk = 0;
1026 
1027 	start_code = ~0UL;
1028 	end_code = 0;
1029 	start_data = 0;
1030 	end_data = 0;
1031 
1032 	/* Now we do a little grungy work by mmapping the ELF image into
1033 	   the correct location in memory. */
1034 	for(i = 0, elf_ppnt = elf_phdata;
1035 	    i < elf_ex->e_phnum; i++, elf_ppnt++) {
1036 		int elf_prot, elf_flags;
1037 		unsigned long k, vaddr;
1038 		unsigned long total_size = 0;
1039 		unsigned long alignment;
1040 
1041 		if (elf_ppnt->p_type != PT_LOAD)
1042 			continue;
1043 
1044 		elf_prot = make_prot(elf_ppnt->p_flags, &arch_state,
1045 				     !!interpreter, false);
1046 
1047 		elf_flags = MAP_PRIVATE;
1048 
1049 		vaddr = elf_ppnt->p_vaddr;
1050 		/*
1051 		 * The first time through the loop, first_pt_load is true:
1052 		 * layout will be calculated. Once set, use MAP_FIXED since
1053 		 * we know we've already safely mapped the entire region with
1054 		 * MAP_FIXED_NOREPLACE in the once-per-binary logic following.
1055 		 */
1056 		if (!first_pt_load) {
1057 			elf_flags |= MAP_FIXED;
1058 		} else if (elf_ex->e_type == ET_EXEC) {
1059 			/*
1060 			 * This logic is run once for the first LOAD Program
1061 			 * Header for ET_EXEC binaries. No special handling
1062 			 * is needed.
1063 			 */
1064 			elf_flags |= MAP_FIXED_NOREPLACE;
1065 		} else if (elf_ex->e_type == ET_DYN) {
1066 			/*
1067 			 * This logic is run once for the first LOAD Program
1068 			 * Header for ET_DYN binaries to calculate the
1069 			 * randomization (load_bias) for all the LOAD
1070 			 * Program Headers.
1071 			 */
1072 
1073 			/*
1074 			 * Calculate the entire size of the ELF mapping
1075 			 * (total_size), used for the initial mapping,
1076 			 * due to load_addr_set which is set to true later
1077 			 * once the initial mapping is performed.
1078 			 *
1079 			 * Note that this is only sensible when the LOAD
1080 			 * segments are contiguous (or overlapping). If
1081 			 * used for LOADs that are far apart, this would
1082 			 * cause the holes between LOADs to be mapped,
1083 			 * running the risk of having the mapping fail,
1084 			 * as it would be larger than the ELF file itself.
1085 			 *
1086 			 * As a result, only ET_DYN does this, since
1087 			 * some ET_EXEC (e.g. ia64) may have large virtual
1088 			 * memory holes between LOADs.
1089 			 *
1090 			 */
1091 			total_size = total_mapping_size(elf_phdata,
1092 							elf_ex->e_phnum);
1093 			if (!total_size) {
1094 				retval = -EINVAL;
1095 				goto out_free_dentry;
1096 			}
1097 
1098 			/* Calculate any requested alignment. */
1099 			alignment = maximum_alignment(elf_phdata, elf_ex->e_phnum);
1100 
1101 			/*
1102 			 * There are effectively two types of ET_DYN
1103 			 * binaries: programs (i.e. PIE: ET_DYN with PT_INTERP)
1104 			 * and loaders (ET_DYN without PT_INTERP, since they
1105 			 * _are_ the ELF interpreter). The loaders must
1106 			 * be loaded away from programs since the program
1107 			 * may otherwise collide with the loader (especially
1108 			 * for ET_EXEC which does not have a randomized
1109 			 * position). For example to handle invocations of
1110 			 * "./ld.so someprog" to test out a new version of
1111 			 * the loader, the subsequent program that the
1112 			 * loader loads must avoid the loader itself, so
1113 			 * they cannot share the same load range. Sufficient
1114 			 * room for the brk must be allocated with the
1115 			 * loader as well, since brk must be available with
1116 			 * the loader.
1117 			 *
1118 			 * Therefore, programs are loaded offset from
1119 			 * ELF_ET_DYN_BASE and loaders are loaded into the
1120 			 * independently randomized mmap region (0 load_bias
1121 			 * without MAP_FIXED nor MAP_FIXED_NOREPLACE).
1122 			 */
1123 			if (interpreter) {
1124 				/* On ET_DYN with PT_INTERP, we do the ASLR. */
1125 				load_bias = ELF_ET_DYN_BASE;
1126 				if (current->flags & PF_RANDOMIZE)
1127 					load_bias += arch_mmap_rnd();
1128 				/* Adjust alignment as requested. */
1129 				if (alignment)
1130 					load_bias &= ~(alignment - 1);
1131 				elf_flags |= MAP_FIXED_NOREPLACE;
1132 			} else {
1133 				/*
1134 				 * For ET_DYN without PT_INTERP, we rely on
1135 				 * the architectures's (potentially ASLR) mmap
1136 				 * base address (via a load_bias of 0).
1137 				 *
1138 				 * When a large alignment is requested, we
1139 				 * must do the allocation at address "0" right
1140 				 * now to discover where things will load so
1141 				 * that we can adjust the resulting alignment.
1142 				 * In this case (load_bias != 0), we can use
1143 				 * MAP_FIXED_NOREPLACE to make sure the mapping
1144 				 * doesn't collide with anything.
1145 				 */
1146 				if (alignment > ELF_MIN_ALIGN) {
1147 					load_bias = elf_load(bprm->file, 0, elf_ppnt,
1148 							     elf_prot, elf_flags, total_size);
1149 					if (BAD_ADDR(load_bias)) {
1150 						retval = IS_ERR_VALUE(load_bias) ?
1151 							 PTR_ERR((void*)load_bias) : -EINVAL;
1152 						goto out_free_dentry;
1153 					}
1154 					vm_munmap(load_bias, total_size);
1155 					/* Adjust alignment as requested. */
1156 					if (alignment)
1157 						load_bias &= ~(alignment - 1);
1158 					elf_flags |= MAP_FIXED_NOREPLACE;
1159 				} else
1160 					load_bias = 0;
1161 			}
1162 
1163 			/*
1164 			 * Since load_bias is used for all subsequent loading
1165 			 * calculations, we must lower it by the first vaddr
1166 			 * so that the remaining calculations based on the
1167 			 * ELF vaddrs will be correctly offset. The result
1168 			 * is then page aligned.
1169 			 */
1170 			load_bias = ELF_PAGESTART(load_bias - vaddr);
1171 		}
1172 
1173 		error = elf_load(bprm->file, load_bias + vaddr, elf_ppnt,
1174 				elf_prot, elf_flags, total_size);
1175 		if (BAD_ADDR(error)) {
1176 			retval = IS_ERR_VALUE(error) ?
1177 				PTR_ERR((void*)error) : -EINVAL;
1178 			goto out_free_dentry;
1179 		}
1180 
1181 		if (first_pt_load) {
1182 			first_pt_load = 0;
1183 			if (elf_ex->e_type == ET_DYN) {
1184 				load_bias += error -
1185 				             ELF_PAGESTART(load_bias + vaddr);
1186 				reloc_func_desc = load_bias;
1187 			}
1188 		}
1189 
1190 		/*
1191 		 * Figure out which segment in the file contains the Program
1192 		 * Header table, and map to the associated memory address.
1193 		 */
1194 		if (elf_ppnt->p_offset <= elf_ex->e_phoff &&
1195 		    elf_ex->e_phoff < elf_ppnt->p_offset + elf_ppnt->p_filesz) {
1196 			phdr_addr = elf_ex->e_phoff - elf_ppnt->p_offset +
1197 				    elf_ppnt->p_vaddr;
1198 		}
1199 
1200 		k = elf_ppnt->p_vaddr;
1201 		if ((elf_ppnt->p_flags & PF_X) && k < start_code)
1202 			start_code = k;
1203 		if (start_data < k)
1204 			start_data = k;
1205 
1206 		/*
1207 		 * Check to see if the section's size will overflow the
1208 		 * allowed task size. Note that p_filesz must always be
1209 		 * <= p_memsz so it is only necessary to check p_memsz.
1210 		 */
1211 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
1212 		    elf_ppnt->p_memsz > TASK_SIZE ||
1213 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
1214 			/* set_brk can never work. Avoid overflows. */
1215 			retval = -EINVAL;
1216 			goto out_free_dentry;
1217 		}
1218 
1219 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
1220 
1221 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
1222 			end_code = k;
1223 		if (end_data < k)
1224 			end_data = k;
1225 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
1226 		if (k > elf_brk)
1227 			elf_brk = k;
1228 	}
1229 
1230 	e_entry = elf_ex->e_entry + load_bias;
1231 	phdr_addr += load_bias;
1232 	elf_brk += load_bias;
1233 	start_code += load_bias;
1234 	end_code += load_bias;
1235 	start_data += load_bias;
1236 	end_data += load_bias;
1237 
1238 	current->mm->start_brk = current->mm->brk = ELF_PAGEALIGN(elf_brk);
1239 
1240 	if (interpreter) {
1241 		elf_entry = load_elf_interp(interp_elf_ex,
1242 					    interpreter,
1243 					    load_bias, interp_elf_phdata,
1244 					    &arch_state);
1245 		if (!IS_ERR_VALUE(elf_entry)) {
1246 			/*
1247 			 * load_elf_interp() returns relocation
1248 			 * adjustment
1249 			 */
1250 			interp_load_addr = elf_entry;
1251 			elf_entry += interp_elf_ex->e_entry;
1252 		}
1253 		if (BAD_ADDR(elf_entry)) {
1254 			retval = IS_ERR_VALUE(elf_entry) ?
1255 					(int)elf_entry : -EINVAL;
1256 			goto out_free_dentry;
1257 		}
1258 		reloc_func_desc = interp_load_addr;
1259 
1260 		allow_write_access(interpreter);
1261 		fput(interpreter);
1262 
1263 		kfree(interp_elf_ex);
1264 		kfree(interp_elf_phdata);
1265 	} else {
1266 		elf_entry = e_entry;
1267 		if (BAD_ADDR(elf_entry)) {
1268 			retval = -EINVAL;
1269 			goto out_free_dentry;
1270 		}
1271 	}
1272 
1273 	kfree(elf_phdata);
1274 
1275 	set_binfmt(&elf_format);
1276 
1277 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1278 	retval = ARCH_SETUP_ADDITIONAL_PAGES(bprm, elf_ex, !!interpreter);
1279 	if (retval < 0)
1280 		goto out;
1281 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1282 
1283 	retval = create_elf_tables(bprm, elf_ex, interp_load_addr,
1284 				   e_entry, phdr_addr);
1285 	if (retval < 0)
1286 		goto out;
1287 
1288 	mm = current->mm;
1289 	mm->end_code = end_code;
1290 	mm->start_code = start_code;
1291 	mm->start_data = start_data;
1292 	mm->end_data = end_data;
1293 	mm->start_stack = bprm->p;
1294 
1295 	if ((current->flags & PF_RANDOMIZE) && (snapshot_randomize_va_space > 1)) {
1296 		/*
1297 		 * For architectures with ELF randomization, when executing
1298 		 * a loader directly (i.e. no interpreter listed in ELF
1299 		 * headers), move the brk area out of the mmap region
1300 		 * (since it grows up, and may collide early with the stack
1301 		 * growing down), and into the unused ELF_ET_DYN_BASE region.
1302 		 */
1303 		if (IS_ENABLED(CONFIG_ARCH_HAS_ELF_RANDOMIZE) &&
1304 		    elf_ex->e_type == ET_DYN && !interpreter) {
1305 			mm->brk = mm->start_brk = ELF_ET_DYN_BASE;
1306 		} else {
1307 			/* Otherwise leave a gap between .bss and brk. */
1308 			mm->brk = mm->start_brk = mm->brk + PAGE_SIZE;
1309 		}
1310 
1311 		mm->brk = mm->start_brk = arch_randomize_brk(mm);
1312 #ifdef compat_brk_randomized
1313 		current->brk_randomized = 1;
1314 #endif
1315 	}
1316 
1317 	if (current->personality & MMAP_PAGE_ZERO) {
1318 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
1319 		   and some applications "depend" upon this behavior.
1320 		   Since we do not have the power to recompile these, we
1321 		   emulate the SVr4 behavior. Sigh. */
1322 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1323 				MAP_FIXED | MAP_PRIVATE, 0);
1324 
1325 		retval = do_mseal(0, PAGE_SIZE, 0);
1326 		if (retval)
1327 			pr_warn_ratelimited("pid=%d, couldn't seal address 0, ret=%d.\n",
1328 					    task_pid_nr(current), retval);
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 	int retval, error, i, j;
1373 	struct elfhdr elf_ex;
1374 
1375 	error = -ENOEXEC;
1376 	retval = elf_read(file, &elf_ex, sizeof(elf_ex), 0);
1377 	if (retval < 0)
1378 		goto out;
1379 
1380 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1381 		goto out;
1382 
1383 	/* First of all, some simple consistency checks */
1384 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1385 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1386 		goto out;
1387 	if (elf_check_fdpic(&elf_ex))
1388 		goto out;
1389 
1390 	/* Now read in all of the header information */
1391 
1392 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1393 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1394 
1395 	error = -ENOMEM;
1396 	elf_phdata = kmalloc(j, GFP_KERNEL);
1397 	if (!elf_phdata)
1398 		goto out;
1399 
1400 	eppnt = elf_phdata;
1401 	error = -ENOEXEC;
1402 	retval = elf_read(file, eppnt, j, elf_ex.e_phoff);
1403 	if (retval < 0)
1404 		goto out_free_ph;
1405 
1406 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1407 		if ((eppnt + i)->p_type == PT_LOAD)
1408 			j++;
1409 	if (j != 1)
1410 		goto out_free_ph;
1411 
1412 	while (eppnt->p_type != PT_LOAD)
1413 		eppnt++;
1414 
1415 	/* Now use mmap to map the library into memory. */
1416 	error = elf_load(file, ELF_PAGESTART(eppnt->p_vaddr),
1417 			eppnt,
1418 			PROT_READ | PROT_WRITE | PROT_EXEC,
1419 			MAP_FIXED_NOREPLACE | MAP_PRIVATE,
1420 			0);
1421 
1422 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1423 		goto out_free_ph;
1424 
1425 	error = 0;
1426 
1427 out_free_ph:
1428 	kfree(elf_phdata);
1429 out:
1430 	return error;
1431 }
1432 #endif /* #ifdef CONFIG_USELIB */
1433 
1434 #ifdef CONFIG_ELF_CORE
1435 /*
1436  * ELF core dumper
1437  *
1438  * Modelled on fs/exec.c:aout_core_dump()
1439  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1440  */
1441 
1442 /* An ELF note in memory */
1443 struct memelfnote
1444 {
1445 	const char *name;
1446 	int type;
1447 	unsigned int datasz;
1448 	void *data;
1449 };
1450 
1451 static int notesize(struct memelfnote *en)
1452 {
1453 	int sz;
1454 
1455 	sz = sizeof(struct elf_note);
1456 	sz += roundup(strlen(en->name) + 1, 4);
1457 	sz += roundup(en->datasz, 4);
1458 
1459 	return sz;
1460 }
1461 
1462 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1463 {
1464 	struct elf_note en;
1465 	en.n_namesz = strlen(men->name) + 1;
1466 	en.n_descsz = men->datasz;
1467 	en.n_type = men->type;
1468 
1469 	return dump_emit(cprm, &en, sizeof(en)) &&
1470 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1471 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1472 }
1473 
1474 static void fill_elf_header(struct elfhdr *elf, int segs,
1475 			    u16 machine, u32 flags)
1476 {
1477 	memset(elf, 0, sizeof(*elf));
1478 
1479 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1480 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1481 	elf->e_ident[EI_DATA] = ELF_DATA;
1482 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1483 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1484 
1485 	elf->e_type = ET_CORE;
1486 	elf->e_machine = machine;
1487 	elf->e_version = EV_CURRENT;
1488 	elf->e_phoff = sizeof(struct elfhdr);
1489 	elf->e_flags = flags;
1490 	elf->e_ehsize = sizeof(struct elfhdr);
1491 	elf->e_phentsize = sizeof(struct elf_phdr);
1492 	elf->e_phnum = segs;
1493 }
1494 
1495 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1496 {
1497 	phdr->p_type = PT_NOTE;
1498 	phdr->p_offset = offset;
1499 	phdr->p_vaddr = 0;
1500 	phdr->p_paddr = 0;
1501 	phdr->p_filesz = sz;
1502 	phdr->p_memsz = 0;
1503 	phdr->p_flags = 0;
1504 	phdr->p_align = 4;
1505 }
1506 
1507 static void fill_note(struct memelfnote *note, const char *name, int type,
1508 		unsigned int sz, void *data)
1509 {
1510 	note->name = name;
1511 	note->type = type;
1512 	note->datasz = sz;
1513 	note->data = data;
1514 }
1515 
1516 /*
1517  * fill up all the fields in prstatus from the given task struct, except
1518  * registers which need to be filled up separately.
1519  */
1520 static void fill_prstatus(struct elf_prstatus_common *prstatus,
1521 		struct task_struct *p, long signr)
1522 {
1523 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1524 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1525 	prstatus->pr_sighold = p->blocked.sig[0];
1526 	rcu_read_lock();
1527 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1528 	rcu_read_unlock();
1529 	prstatus->pr_pid = task_pid_vnr(p);
1530 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1531 	prstatus->pr_sid = task_session_vnr(p);
1532 	if (thread_group_leader(p)) {
1533 		struct task_cputime cputime;
1534 
1535 		/*
1536 		 * This is the record for the group leader.  It shows the
1537 		 * group-wide total, not its individual thread total.
1538 		 */
1539 		thread_group_cputime(p, &cputime);
1540 		prstatus->pr_utime = ns_to_kernel_old_timeval(cputime.utime);
1541 		prstatus->pr_stime = ns_to_kernel_old_timeval(cputime.stime);
1542 	} else {
1543 		u64 utime, stime;
1544 
1545 		task_cputime(p, &utime, &stime);
1546 		prstatus->pr_utime = ns_to_kernel_old_timeval(utime);
1547 		prstatus->pr_stime = ns_to_kernel_old_timeval(stime);
1548 	}
1549 
1550 	prstatus->pr_cutime = ns_to_kernel_old_timeval(p->signal->cutime);
1551 	prstatus->pr_cstime = ns_to_kernel_old_timeval(p->signal->cstime);
1552 }
1553 
1554 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1555 		       struct mm_struct *mm)
1556 {
1557 	const struct cred *cred;
1558 	unsigned int i, len;
1559 	unsigned int state;
1560 
1561 	/* first copy the parameters from user space */
1562 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1563 
1564 	len = mm->arg_end - mm->arg_start;
1565 	if (len >= ELF_PRARGSZ)
1566 		len = ELF_PRARGSZ-1;
1567 	if (copy_from_user(&psinfo->pr_psargs,
1568 		           (const char __user *)mm->arg_start, len))
1569 		return -EFAULT;
1570 	for(i = 0; i < len; i++)
1571 		if (psinfo->pr_psargs[i] == 0)
1572 			psinfo->pr_psargs[i] = ' ';
1573 	psinfo->pr_psargs[len] = 0;
1574 
1575 	rcu_read_lock();
1576 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1577 	rcu_read_unlock();
1578 	psinfo->pr_pid = task_pid_vnr(p);
1579 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1580 	psinfo->pr_sid = task_session_vnr(p);
1581 
1582 	state = READ_ONCE(p->__state);
1583 	i = state ? ffz(~state) + 1 : 0;
1584 	psinfo->pr_state = i;
1585 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1586 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1587 	psinfo->pr_nice = task_nice(p);
1588 	psinfo->pr_flag = p->flags;
1589 	rcu_read_lock();
1590 	cred = __task_cred(p);
1591 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1592 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1593 	rcu_read_unlock();
1594 	get_task_comm(psinfo->pr_fname, p);
1595 
1596 	return 0;
1597 }
1598 
1599 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1600 {
1601 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1602 	int i = 0;
1603 	do
1604 		i += 2;
1605 	while (auxv[i - 2] != AT_NULL);
1606 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1607 }
1608 
1609 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1610 		const kernel_siginfo_t *siginfo)
1611 {
1612 	copy_siginfo_to_external(csigdata, siginfo);
1613 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1614 }
1615 
1616 /*
1617  * Format of NT_FILE note:
1618  *
1619  * long count     -- how many files are mapped
1620  * long page_size -- units for file_ofs
1621  * array of [COUNT] elements of
1622  *   long start
1623  *   long end
1624  *   long file_ofs
1625  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1626  */
1627 static int fill_files_note(struct memelfnote *note, struct coredump_params *cprm)
1628 {
1629 	unsigned count, size, names_ofs, remaining, n;
1630 	user_long_t *data;
1631 	user_long_t *start_end_ofs;
1632 	char *name_base, *name_curpos;
1633 	int i;
1634 
1635 	/* *Estimated* file count and total data size needed */
1636 	count = cprm->vma_count;
1637 	if (count > UINT_MAX / 64)
1638 		return -EINVAL;
1639 	size = count * 64;
1640 
1641 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1642  alloc:
1643 	/* paranoia check */
1644 	if (size >= core_file_note_size_limit) {
1645 		pr_warn_once("coredump Note size too large: %u (does kernel.core_file_note_size_limit sysctl need adjustment?\n",
1646 			      size);
1647 		return -EINVAL;
1648 	}
1649 	size = round_up(size, PAGE_SIZE);
1650 	/*
1651 	 * "size" can be 0 here legitimately.
1652 	 * Let it ENOMEM and omit NT_FILE section which will be empty anyway.
1653 	 */
1654 	data = kvmalloc(size, GFP_KERNEL);
1655 	if (ZERO_OR_NULL_PTR(data))
1656 		return -ENOMEM;
1657 
1658 	start_end_ofs = data + 2;
1659 	name_base = name_curpos = ((char *)data) + names_ofs;
1660 	remaining = size - names_ofs;
1661 	count = 0;
1662 	for (i = 0; i < cprm->vma_count; i++) {
1663 		struct core_vma_metadata *m = &cprm->vma_meta[i];
1664 		struct file *file;
1665 		const char *filename;
1666 
1667 		file = m->file;
1668 		if (!file)
1669 			continue;
1670 		filename = file_path(file, name_curpos, remaining);
1671 		if (IS_ERR(filename)) {
1672 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1673 				kvfree(data);
1674 				size = size * 5 / 4;
1675 				goto alloc;
1676 			}
1677 			continue;
1678 		}
1679 
1680 		/* file_path() fills at the end, move name down */
1681 		/* n = strlen(filename) + 1: */
1682 		n = (name_curpos + remaining) - filename;
1683 		remaining = filename - name_curpos;
1684 		memmove(name_curpos, filename, n);
1685 		name_curpos += n;
1686 
1687 		*start_end_ofs++ = m->start;
1688 		*start_end_ofs++ = m->end;
1689 		*start_end_ofs++ = m->pgoff;
1690 		count++;
1691 	}
1692 
1693 	/* Now we know exact count of files, can store it */
1694 	data[0] = count;
1695 	data[1] = PAGE_SIZE;
1696 	/*
1697 	 * Count usually is less than mm->map_count,
1698 	 * we need to move filenames down.
1699 	 */
1700 	n = cprm->vma_count - count;
1701 	if (n != 0) {
1702 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1703 		memmove(name_base - shift_bytes, name_base,
1704 			name_curpos - name_base);
1705 		name_curpos -= shift_bytes;
1706 	}
1707 
1708 	size = name_curpos - (char *)data;
1709 	fill_note(note, "CORE", NT_FILE, size, data);
1710 	return 0;
1711 }
1712 
1713 #include <linux/regset.h>
1714 
1715 struct elf_thread_core_info {
1716 	struct elf_thread_core_info *next;
1717 	struct task_struct *task;
1718 	struct elf_prstatus prstatus;
1719 	struct memelfnote notes[];
1720 };
1721 
1722 struct elf_note_info {
1723 	struct elf_thread_core_info *thread;
1724 	struct memelfnote psinfo;
1725 	struct memelfnote signote;
1726 	struct memelfnote auxv;
1727 	struct memelfnote files;
1728 	user_siginfo_t csigdata;
1729 	size_t size;
1730 	int thread_notes;
1731 };
1732 
1733 #ifdef CORE_DUMP_USE_REGSET
1734 /*
1735  * When a regset has a writeback hook, we call it on each thread before
1736  * dumping user memory.  On register window machines, this makes sure the
1737  * user memory backing the register data is up to date before we read it.
1738  */
1739 static void do_thread_regset_writeback(struct task_struct *task,
1740 				       const struct user_regset *regset)
1741 {
1742 	if (regset->writeback)
1743 		regset->writeback(task, regset, 1);
1744 }
1745 
1746 #ifndef PRSTATUS_SIZE
1747 #define PRSTATUS_SIZE sizeof(struct elf_prstatus)
1748 #endif
1749 
1750 #ifndef SET_PR_FPVALID
1751 #define SET_PR_FPVALID(S) ((S)->pr_fpvalid = 1)
1752 #endif
1753 
1754 static int fill_thread_core_info(struct elf_thread_core_info *t,
1755 				 const struct user_regset_view *view,
1756 				 long signr, struct elf_note_info *info)
1757 {
1758 	unsigned int note_iter, view_iter;
1759 
1760 	/*
1761 	 * NT_PRSTATUS is the one special case, because the regset data
1762 	 * goes into the pr_reg field inside the note contents, rather
1763 	 * than being the whole note contents.  We fill the regset in here.
1764 	 * We assume that regset 0 is NT_PRSTATUS.
1765 	 */
1766 	fill_prstatus(&t->prstatus.common, t->task, signr);
1767 	regset_get(t->task, &view->regsets[0],
1768 		   sizeof(t->prstatus.pr_reg), &t->prstatus.pr_reg);
1769 
1770 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1771 		  PRSTATUS_SIZE, &t->prstatus);
1772 	info->size += notesize(&t->notes[0]);
1773 
1774 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1775 
1776 	/*
1777 	 * Each other regset might generate a note too.  For each regset
1778 	 * that has no core_note_type or is inactive, skip it.
1779 	 */
1780 	note_iter = 1;
1781 	for (view_iter = 1; view_iter < view->n; ++view_iter) {
1782 		const struct user_regset *regset = &view->regsets[view_iter];
1783 		int note_type = regset->core_note_type;
1784 		bool is_fpreg = note_type == NT_PRFPREG;
1785 		void *data;
1786 		int ret;
1787 
1788 		do_thread_regset_writeback(t->task, regset);
1789 		if (!note_type) // not for coredumps
1790 			continue;
1791 		if (regset->active && regset->active(t->task, regset) <= 0)
1792 			continue;
1793 
1794 		ret = regset_get_alloc(t->task, regset, ~0U, &data);
1795 		if (ret < 0)
1796 			continue;
1797 
1798 		if (WARN_ON_ONCE(note_iter >= info->thread_notes))
1799 			break;
1800 
1801 		if (is_fpreg)
1802 			SET_PR_FPVALID(&t->prstatus);
1803 
1804 		fill_note(&t->notes[note_iter], is_fpreg ? "CORE" : "LINUX",
1805 			  note_type, ret, data);
1806 
1807 		info->size += notesize(&t->notes[note_iter]);
1808 		note_iter++;
1809 	}
1810 
1811 	return 1;
1812 }
1813 #else
1814 static int fill_thread_core_info(struct elf_thread_core_info *t,
1815 				 const struct user_regset_view *view,
1816 				 long signr, struct elf_note_info *info)
1817 {
1818 	struct task_struct *p = t->task;
1819 	elf_fpregset_t *fpu;
1820 
1821 	fill_prstatus(&t->prstatus.common, p, signr);
1822 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1823 
1824 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1825 		  &(t->prstatus));
1826 	info->size += notesize(&t->notes[0]);
1827 
1828 	fpu = kzalloc(sizeof(elf_fpregset_t), GFP_KERNEL);
1829 	if (!fpu || !elf_core_copy_task_fpregs(p, fpu)) {
1830 		kfree(fpu);
1831 		return 1;
1832 	}
1833 
1834 	t->prstatus.pr_fpvalid = 1;
1835 	fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1836 	info->size += notesize(&t->notes[1]);
1837 
1838 	return 1;
1839 }
1840 #endif
1841 
1842 static int fill_note_info(struct elfhdr *elf, int phdrs,
1843 			  struct elf_note_info *info,
1844 			  struct coredump_params *cprm)
1845 {
1846 	struct task_struct *dump_task = current;
1847 	const struct user_regset_view *view;
1848 	struct elf_thread_core_info *t;
1849 	struct elf_prpsinfo *psinfo;
1850 	struct core_thread *ct;
1851 
1852 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1853 	if (!psinfo)
1854 		return 0;
1855 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1856 
1857 #ifdef CORE_DUMP_USE_REGSET
1858 	view = task_user_regset_view(dump_task);
1859 
1860 	/*
1861 	 * Figure out how many notes we're going to need for each thread.
1862 	 */
1863 	info->thread_notes = 0;
1864 	for (int i = 0; i < view->n; ++i)
1865 		if (view->regsets[i].core_note_type != 0)
1866 			++info->thread_notes;
1867 
1868 	/*
1869 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1870 	 * since it is our one special case.
1871 	 */
1872 	if (unlikely(info->thread_notes == 0) ||
1873 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1874 		WARN_ON(1);
1875 		return 0;
1876 	}
1877 
1878 	/*
1879 	 * Initialize the ELF file header.
1880 	 */
1881 	fill_elf_header(elf, phdrs,
1882 			view->e_machine, view->e_flags);
1883 #else
1884 	view = NULL;
1885 	info->thread_notes = 2;
1886 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1887 #endif
1888 
1889 	/*
1890 	 * Allocate a structure for each thread.
1891 	 */
1892 	info->thread = kzalloc(offsetof(struct elf_thread_core_info,
1893 				     notes[info->thread_notes]),
1894 			    GFP_KERNEL);
1895 	if (unlikely(!info->thread))
1896 		return 0;
1897 
1898 	info->thread->task = dump_task;
1899 	for (ct = dump_task->signal->core_state->dumper.next; ct; ct = ct->next) {
1900 		t = kzalloc(offsetof(struct elf_thread_core_info,
1901 				     notes[info->thread_notes]),
1902 			    GFP_KERNEL);
1903 		if (unlikely(!t))
1904 			return 0;
1905 
1906 		t->task = ct->task;
1907 		t->next = info->thread->next;
1908 		info->thread->next = t;
1909 	}
1910 
1911 	/*
1912 	 * Now fill in each thread's information.
1913 	 */
1914 	for (t = info->thread; t != NULL; t = t->next)
1915 		if (!fill_thread_core_info(t, view, cprm->siginfo->si_signo, info))
1916 			return 0;
1917 
1918 	/*
1919 	 * Fill in the two process-wide notes.
1920 	 */
1921 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1922 	info->size += notesize(&info->psinfo);
1923 
1924 	fill_siginfo_note(&info->signote, &info->csigdata, cprm->siginfo);
1925 	info->size += notesize(&info->signote);
1926 
1927 	fill_auxv_note(&info->auxv, current->mm);
1928 	info->size += notesize(&info->auxv);
1929 
1930 	if (fill_files_note(&info->files, cprm) == 0)
1931 		info->size += notesize(&info->files);
1932 
1933 	return 1;
1934 }
1935 
1936 /*
1937  * Write all the notes for each thread.  When writing the first thread, the
1938  * process-wide notes are interleaved after the first thread-specific note.
1939  */
1940 static int write_note_info(struct elf_note_info *info,
1941 			   struct coredump_params *cprm)
1942 {
1943 	bool first = true;
1944 	struct elf_thread_core_info *t = info->thread;
1945 
1946 	do {
1947 		int i;
1948 
1949 		if (!writenote(&t->notes[0], cprm))
1950 			return 0;
1951 
1952 		if (first && !writenote(&info->psinfo, cprm))
1953 			return 0;
1954 		if (first && !writenote(&info->signote, cprm))
1955 			return 0;
1956 		if (first && !writenote(&info->auxv, cprm))
1957 			return 0;
1958 		if (first && info->files.data &&
1959 				!writenote(&info->files, cprm))
1960 			return 0;
1961 
1962 		for (i = 1; i < info->thread_notes; ++i)
1963 			if (t->notes[i].data &&
1964 			    !writenote(&t->notes[i], cprm))
1965 				return 0;
1966 
1967 		first = false;
1968 		t = t->next;
1969 	} while (t);
1970 
1971 	return 1;
1972 }
1973 
1974 static void free_note_info(struct elf_note_info *info)
1975 {
1976 	struct elf_thread_core_info *threads = info->thread;
1977 	while (threads) {
1978 		unsigned int i;
1979 		struct elf_thread_core_info *t = threads;
1980 		threads = t->next;
1981 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1982 		for (i = 1; i < info->thread_notes; ++i)
1983 			kvfree(t->notes[i].data);
1984 		kfree(t);
1985 	}
1986 	kfree(info->psinfo.data);
1987 	kvfree(info->files.data);
1988 }
1989 
1990 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1991 			     elf_addr_t e_shoff, int segs)
1992 {
1993 	elf->e_shoff = e_shoff;
1994 	elf->e_shentsize = sizeof(*shdr4extnum);
1995 	elf->e_shnum = 1;
1996 	elf->e_shstrndx = SHN_UNDEF;
1997 
1998 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1999 
2000 	shdr4extnum->sh_type = SHT_NULL;
2001 	shdr4extnum->sh_size = elf->e_shnum;
2002 	shdr4extnum->sh_link = elf->e_shstrndx;
2003 	shdr4extnum->sh_info = segs;
2004 }
2005 
2006 /*
2007  * Actual dumper
2008  *
2009  * This is a two-pass process; first we find the offsets of the bits,
2010  * and then they are actually written out.  If we run out of core limit
2011  * we just truncate.
2012  */
2013 static int elf_core_dump(struct coredump_params *cprm)
2014 {
2015 	int has_dumped = 0;
2016 	int segs, i;
2017 	struct elfhdr elf;
2018 	loff_t offset = 0, dataoff;
2019 	struct elf_note_info info = { };
2020 	struct elf_phdr *phdr4note = NULL;
2021 	struct elf_shdr *shdr4extnum = NULL;
2022 	Elf_Half e_phnum;
2023 	elf_addr_t e_shoff;
2024 
2025 	/*
2026 	 * The number of segs are recored into ELF header as 16bit value.
2027 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2028 	 */
2029 	segs = cprm->vma_count + elf_core_extra_phdrs(cprm);
2030 
2031 	/* for notes section */
2032 	segs++;
2033 
2034 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2035 	 * this, kernel supports extended numbering. Have a look at
2036 	 * include/linux/elf.h for further information. */
2037 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2038 
2039 	/*
2040 	 * Collect all the non-memory information about the process for the
2041 	 * notes.  This also sets up the file header.
2042 	 */
2043 	if (!fill_note_info(&elf, e_phnum, &info, cprm))
2044 		goto end_coredump;
2045 
2046 	has_dumped = 1;
2047 
2048 	offset += sizeof(elf);				/* ELF header */
2049 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2050 
2051 	/* Write notes phdr entry */
2052 	{
2053 		size_t sz = info.size;
2054 
2055 		/* For cell spufs and x86 xstate */
2056 		sz += elf_coredump_extra_notes_size();
2057 
2058 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2059 		if (!phdr4note)
2060 			goto end_coredump;
2061 
2062 		fill_elf_note_phdr(phdr4note, sz, offset);
2063 		offset += sz;
2064 	}
2065 
2066 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2067 
2068 	offset += cprm->vma_data_size;
2069 	offset += elf_core_extra_data_size(cprm);
2070 	e_shoff = offset;
2071 
2072 	if (e_phnum == PN_XNUM) {
2073 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2074 		if (!shdr4extnum)
2075 			goto end_coredump;
2076 		fill_extnum_info(&elf, shdr4extnum, e_shoff, segs);
2077 	}
2078 
2079 	offset = dataoff;
2080 
2081 	if (!dump_emit(cprm, &elf, sizeof(elf)))
2082 		goto end_coredump;
2083 
2084 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2085 		goto end_coredump;
2086 
2087 	/* Write program headers for segments dump */
2088 	for (i = 0; i < cprm->vma_count; i++) {
2089 		struct core_vma_metadata *meta = cprm->vma_meta + i;
2090 		struct elf_phdr phdr;
2091 
2092 		phdr.p_type = PT_LOAD;
2093 		phdr.p_offset = offset;
2094 		phdr.p_vaddr = meta->start;
2095 		phdr.p_paddr = 0;
2096 		phdr.p_filesz = meta->dump_size;
2097 		phdr.p_memsz = meta->end - meta->start;
2098 		offset += phdr.p_filesz;
2099 		phdr.p_flags = 0;
2100 		if (meta->flags & VM_READ)
2101 			phdr.p_flags |= PF_R;
2102 		if (meta->flags & VM_WRITE)
2103 			phdr.p_flags |= PF_W;
2104 		if (meta->flags & VM_EXEC)
2105 			phdr.p_flags |= PF_X;
2106 		phdr.p_align = ELF_EXEC_PAGESIZE;
2107 
2108 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2109 			goto end_coredump;
2110 	}
2111 
2112 	if (!elf_core_write_extra_phdrs(cprm, offset))
2113 		goto end_coredump;
2114 
2115 	/* write out the notes section */
2116 	if (!write_note_info(&info, cprm))
2117 		goto end_coredump;
2118 
2119 	/* For cell spufs and x86 xstate */
2120 	if (elf_coredump_extra_notes_write(cprm))
2121 		goto end_coredump;
2122 
2123 	/* Align to page */
2124 	dump_skip_to(cprm, dataoff);
2125 
2126 	for (i = 0; i < cprm->vma_count; i++) {
2127 		struct core_vma_metadata *meta = cprm->vma_meta + i;
2128 
2129 		if (!dump_user_range(cprm, meta->start, meta->dump_size))
2130 			goto end_coredump;
2131 	}
2132 
2133 	if (!elf_core_write_extra_data(cprm))
2134 		goto end_coredump;
2135 
2136 	if (e_phnum == PN_XNUM) {
2137 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2138 			goto end_coredump;
2139 	}
2140 
2141 end_coredump:
2142 	free_note_info(&info);
2143 	kfree(shdr4extnum);
2144 	kfree(phdr4note);
2145 	return has_dumped;
2146 }
2147 
2148 #endif		/* CONFIG_ELF_CORE */
2149 
2150 static int __init init_elf_binfmt(void)
2151 {
2152 	register_binfmt(&elf_format);
2153 	return 0;
2154 }
2155 
2156 static void __exit exit_elf_binfmt(void)
2157 {
2158 	/* Remove the COFF and ELF loaders. */
2159 	unregister_binfmt(&elf_format);
2160 }
2161 
2162 core_initcall(init_elf_binfmt);
2163 module_exit(exit_elf_binfmt);
2164 
2165 #ifdef CONFIG_BINFMT_ELF_KUNIT_TEST
2166 #include "tests/binfmt_elf_kunit.c"
2167 #endif
2168