xref: /linux/fs/binfmt_elf.c (revision 3a39d672e7f48b8d6b91a09afa4b55352773b4b5)
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  */
padzero(unsigned long address)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
create_elf_tables(struct linux_binprm * bprm,const struct elfhdr * exec,unsigned long interp_load_addr,unsigned long e_entry,unsigned long phdr_addr)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 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
262 	if (k_platform) {
263 		NEW_AUX_ENT(AT_PLATFORM,
264 			    (elf_addr_t)(unsigned long)u_platform);
265 	}
266 	if (k_base_platform) {
267 		NEW_AUX_ENT(AT_BASE_PLATFORM,
268 			    (elf_addr_t)(unsigned long)u_base_platform);
269 	}
270 	if (bprm->have_execfd) {
271 		NEW_AUX_ENT(AT_EXECFD, bprm->execfd);
272 	}
273 #ifdef CONFIG_RSEQ
274 	NEW_AUX_ENT(AT_RSEQ_FEATURE_SIZE, offsetof(struct rseq, end));
275 	NEW_AUX_ENT(AT_RSEQ_ALIGN, __alignof__(struct rseq));
276 #endif
277 #undef NEW_AUX_ENT
278 	/* AT_NULL is zero; clear the rest too */
279 	memset(elf_info, 0, (char *)mm->saved_auxv +
280 			sizeof(mm->saved_auxv) - (char *)elf_info);
281 
282 	/* And advance past the AT_NULL entry.  */
283 	elf_info += 2;
284 
285 	ei_index = elf_info - (elf_addr_t *)mm->saved_auxv;
286 	sp = STACK_ADD(p, ei_index);
287 
288 	items = (argc + 1) + (envc + 1) + 1;
289 	bprm->p = STACK_ROUND(sp, items);
290 
291 	/* Point sp at the lowest address on the stack */
292 #ifdef CONFIG_STACK_GROWSUP
293 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
294 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
295 #else
296 	sp = (elf_addr_t __user *)bprm->p;
297 #endif
298 
299 
300 	/*
301 	 * Grow the stack manually; some architectures have a limit on how
302 	 * far ahead a user-space access may be in order to grow the stack.
303 	 */
304 	if (mmap_write_lock_killable(mm))
305 		return -EINTR;
306 	vma = find_extend_vma_locked(mm, bprm->p);
307 	mmap_write_unlock(mm);
308 	if (!vma)
309 		return -EFAULT;
310 
311 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
312 	if (put_user(argc, sp++))
313 		return -EFAULT;
314 
315 	/* Populate list of argv pointers back to argv strings. */
316 	p = mm->arg_end = mm->arg_start;
317 	while (argc-- > 0) {
318 		size_t len;
319 		if (put_user((elf_addr_t)p, sp++))
320 			return -EFAULT;
321 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
322 		if (!len || len > MAX_ARG_STRLEN)
323 			return -EINVAL;
324 		p += len;
325 	}
326 	if (put_user(0, sp++))
327 		return -EFAULT;
328 	mm->arg_end = p;
329 
330 	/* Populate list of envp pointers back to envp strings. */
331 	mm->env_end = mm->env_start = p;
332 	while (envc-- > 0) {
333 		size_t len;
334 		if (put_user((elf_addr_t)p, sp++))
335 			return -EFAULT;
336 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
337 		if (!len || len > MAX_ARG_STRLEN)
338 			return -EINVAL;
339 		p += len;
340 	}
341 	if (put_user(0, sp++))
342 		return -EFAULT;
343 	mm->env_end = p;
344 
345 	/* Put the elf_info on the stack in the right place.  */
346 	if (copy_to_user(sp, mm->saved_auxv, ei_index * sizeof(elf_addr_t)))
347 		return -EFAULT;
348 	return 0;
349 }
350 
351 /*
352  * Map "eppnt->p_filesz" bytes from "filep" offset "eppnt->p_offset"
353  * into memory at "addr". (Note that p_filesz is rounded up to the
354  * next page, so any extra bytes from the file must be wiped.)
355  */
elf_map(struct file * filep,unsigned long addr,const struct elf_phdr * eppnt,int prot,int type,unsigned long total_size)356 static unsigned long elf_map(struct file *filep, unsigned long addr,
357 		const struct elf_phdr *eppnt, int prot, int type,
358 		unsigned long total_size)
359 {
360 	unsigned long map_addr;
361 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
362 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
363 	addr = ELF_PAGESTART(addr);
364 	size = ELF_PAGEALIGN(size);
365 
366 	/* mmap() will return -EINVAL if given a zero size, but a
367 	 * segment with zero filesize is perfectly valid */
368 	if (!size)
369 		return addr;
370 
371 	/*
372 	* total_size is the size of the ELF (interpreter) image.
373 	* The _first_ mmap needs to know the full size, otherwise
374 	* randomization might put this image into an overlapping
375 	* position with the ELF binary image. (since size < total_size)
376 	* So we first map the 'big' image - and unmap the remainder at
377 	* the end. (which unmap is needed for ELF images with holes.)
378 	*/
379 	if (total_size) {
380 		total_size = ELF_PAGEALIGN(total_size);
381 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
382 		if (!BAD_ADDR(map_addr))
383 			vm_munmap(map_addr+size, total_size-size);
384 	} else
385 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
386 
387 	if ((type & MAP_FIXED_NOREPLACE) &&
388 	    PTR_ERR((void *)map_addr) == -EEXIST)
389 		pr_info("%d (%s): Uhuuh, elf segment at %px requested but the memory is mapped already\n",
390 			task_pid_nr(current), current->comm, (void *)addr);
391 
392 	return(map_addr);
393 }
394 
395 /*
396  * Map "eppnt->p_filesz" bytes from "filep" offset "eppnt->p_offset"
397  * into memory at "addr". Memory from "p_filesz" through "p_memsz"
398  * rounded up to the next page is zeroed.
399  */
elf_load(struct file * filep,unsigned long addr,const struct elf_phdr * eppnt,int prot,int type,unsigned long total_size)400 static unsigned long elf_load(struct file *filep, unsigned long addr,
401 		const struct elf_phdr *eppnt, int prot, int type,
402 		unsigned long total_size)
403 {
404 	unsigned long zero_start, zero_end;
405 	unsigned long map_addr;
406 
407 	if (eppnt->p_filesz) {
408 		map_addr = elf_map(filep, addr, eppnt, prot, type, total_size);
409 		if (BAD_ADDR(map_addr))
410 			return map_addr;
411 		if (eppnt->p_memsz > eppnt->p_filesz) {
412 			zero_start = map_addr + ELF_PAGEOFFSET(eppnt->p_vaddr) +
413 				eppnt->p_filesz;
414 			zero_end = map_addr + ELF_PAGEOFFSET(eppnt->p_vaddr) +
415 				eppnt->p_memsz;
416 
417 			/*
418 			 * Zero the end of the last mapped page but ignore
419 			 * any errors if the segment isn't writable.
420 			 */
421 			if (padzero(zero_start) && (prot & PROT_WRITE))
422 				return -EFAULT;
423 		}
424 	} else {
425 		map_addr = zero_start = ELF_PAGESTART(addr);
426 		zero_end = zero_start + ELF_PAGEOFFSET(eppnt->p_vaddr) +
427 			eppnt->p_memsz;
428 	}
429 	if (eppnt->p_memsz > eppnt->p_filesz) {
430 		/*
431 		 * Map the last of the segment.
432 		 * If the header is requesting these pages to be
433 		 * executable, honour that (ppc32 needs this).
434 		 */
435 		int error;
436 
437 		zero_start = ELF_PAGEALIGN(zero_start);
438 		zero_end = ELF_PAGEALIGN(zero_end);
439 
440 		error = vm_brk_flags(zero_start, zero_end - zero_start,
441 				     prot & PROT_EXEC ? VM_EXEC : 0);
442 		if (error)
443 			map_addr = error;
444 	}
445 	return map_addr;
446 }
447 
448 
total_mapping_size(const struct elf_phdr * phdr,int nr)449 static unsigned long total_mapping_size(const struct elf_phdr *phdr, int nr)
450 {
451 	elf_addr_t min_addr = -1;
452 	elf_addr_t max_addr = 0;
453 	bool pt_load = false;
454 	int i;
455 
456 	for (i = 0; i < nr; i++) {
457 		if (phdr[i].p_type == PT_LOAD) {
458 			min_addr = min(min_addr, ELF_PAGESTART(phdr[i].p_vaddr));
459 			max_addr = max(max_addr, phdr[i].p_vaddr + phdr[i].p_memsz);
460 			pt_load = true;
461 		}
462 	}
463 	return pt_load ? (max_addr - min_addr) : 0;
464 }
465 
elf_read(struct file * file,void * buf,size_t len,loff_t pos)466 static int elf_read(struct file *file, void *buf, size_t len, loff_t pos)
467 {
468 	ssize_t rv;
469 
470 	rv = kernel_read(file, buf, len, &pos);
471 	if (unlikely(rv != len)) {
472 		return (rv < 0) ? rv : -EIO;
473 	}
474 	return 0;
475 }
476 
maximum_alignment(struct elf_phdr * cmds,int nr)477 static unsigned long maximum_alignment(struct elf_phdr *cmds, int nr)
478 {
479 	unsigned long alignment = 0;
480 	int i;
481 
482 	for (i = 0; i < nr; i++) {
483 		if (cmds[i].p_type == PT_LOAD) {
484 			unsigned long p_align = cmds[i].p_align;
485 
486 			/* skip non-power of two alignments as invalid */
487 			if (!is_power_of_2(p_align))
488 				continue;
489 			alignment = max(alignment, p_align);
490 		}
491 	}
492 
493 	/* ensure we align to at least one page */
494 	return ELF_PAGEALIGN(alignment);
495 }
496 
497 /**
498  * load_elf_phdrs() - load ELF program headers
499  * @elf_ex:   ELF header of the binary whose program headers should be loaded
500  * @elf_file: the opened ELF binary file
501  *
502  * Loads ELF program headers from the binary file elf_file, which has the ELF
503  * header pointed to by elf_ex, into a newly allocated array. The caller is
504  * responsible for freeing the allocated data. Returns NULL upon failure.
505  */
load_elf_phdrs(const struct elfhdr * elf_ex,struct file * elf_file)506 static struct elf_phdr *load_elf_phdrs(const struct elfhdr *elf_ex,
507 				       struct file *elf_file)
508 {
509 	struct elf_phdr *elf_phdata = NULL;
510 	int retval = -1;
511 	unsigned int size;
512 
513 	/*
514 	 * If the size of this structure has changed, then punt, since
515 	 * we will be doing the wrong thing.
516 	 */
517 	if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
518 		goto out;
519 
520 	/* Sanity check the number of program headers... */
521 	/* ...and their total size. */
522 	size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
523 	if (size == 0 || size > 65536 || size > ELF_MIN_ALIGN)
524 		goto out;
525 
526 	elf_phdata = kmalloc(size, GFP_KERNEL);
527 	if (!elf_phdata)
528 		goto out;
529 
530 	/* Read in the program headers */
531 	retval = elf_read(elf_file, elf_phdata, size, elf_ex->e_phoff);
532 
533 out:
534 	if (retval) {
535 		kfree(elf_phdata);
536 		elf_phdata = NULL;
537 	}
538 	return elf_phdata;
539 }
540 
541 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
542 
543 /**
544  * struct arch_elf_state - arch-specific ELF loading state
545  *
546  * This structure is used to preserve architecture specific data during
547  * the loading of an ELF file, throughout the checking of architecture
548  * specific ELF headers & through to the point where the ELF load is
549  * known to be proceeding (ie. SET_PERSONALITY).
550  *
551  * This implementation is a dummy for architectures which require no
552  * specific state.
553  */
554 struct arch_elf_state {
555 };
556 
557 #define INIT_ARCH_ELF_STATE {}
558 
559 /**
560  * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
561  * @ehdr:	The main ELF header
562  * @phdr:	The program header to check
563  * @elf:	The open ELF file
564  * @is_interp:	True if the phdr is from the interpreter of the ELF being
565  *		loaded, else false.
566  * @state:	Architecture-specific state preserved throughout the process
567  *		of loading the ELF.
568  *
569  * Inspects the program header phdr to validate its correctness and/or
570  * suitability for the system. Called once per ELF program header in the
571  * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
572  * interpreter.
573  *
574  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
575  *         with that return code.
576  */
arch_elf_pt_proc(struct elfhdr * ehdr,struct elf_phdr * phdr,struct file * elf,bool is_interp,struct arch_elf_state * state)577 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
578 				   struct elf_phdr *phdr,
579 				   struct file *elf, bool is_interp,
580 				   struct arch_elf_state *state)
581 {
582 	/* Dummy implementation, always proceed */
583 	return 0;
584 }
585 
586 /**
587  * arch_check_elf() - check an ELF executable
588  * @ehdr:	The main ELF header
589  * @has_interp:	True if the ELF has an interpreter, else false.
590  * @interp_ehdr: The interpreter's ELF header
591  * @state:	Architecture-specific state preserved throughout the process
592  *		of loading the ELF.
593  *
594  * Provides a final opportunity for architecture code to reject the loading
595  * of the ELF & cause an exec syscall to return an error. This is called after
596  * all program headers to be checked by arch_elf_pt_proc have been.
597  *
598  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
599  *         with that return code.
600  */
arch_check_elf(struct elfhdr * ehdr,bool has_interp,struct elfhdr * interp_ehdr,struct arch_elf_state * state)601 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
602 				 struct elfhdr *interp_ehdr,
603 				 struct arch_elf_state *state)
604 {
605 	/* Dummy implementation, always proceed */
606 	return 0;
607 }
608 
609 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
610 
make_prot(u32 p_flags,struct arch_elf_state * arch_state,bool has_interp,bool is_interp)611 static inline int make_prot(u32 p_flags, struct arch_elf_state *arch_state,
612 			    bool has_interp, bool is_interp)
613 {
614 	int prot = 0;
615 
616 	if (p_flags & PF_R)
617 		prot |= PROT_READ;
618 	if (p_flags & PF_W)
619 		prot |= PROT_WRITE;
620 	if (p_flags & PF_X)
621 		prot |= PROT_EXEC;
622 
623 	return arch_elf_adjust_prot(prot, arch_state, has_interp, is_interp);
624 }
625 
626 /* This is much more generalized than the library routine read function,
627    so we keep this separate.  Technically the library read function
628    is only provided so that we can read a.out libraries that have
629    an ELF header */
630 
load_elf_interp(struct elfhdr * interp_elf_ex,struct file * interpreter,unsigned long no_base,struct elf_phdr * interp_elf_phdata,struct arch_elf_state * arch_state)631 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
632 		struct file *interpreter,
633 		unsigned long no_base, struct elf_phdr *interp_elf_phdata,
634 		struct arch_elf_state *arch_state)
635 {
636 	struct elf_phdr *eppnt;
637 	unsigned long load_addr = 0;
638 	int load_addr_set = 0;
639 	unsigned long error = ~0UL;
640 	unsigned long total_size;
641 	int i;
642 
643 	/* First of all, some simple consistency checks */
644 	if (interp_elf_ex->e_type != ET_EXEC &&
645 	    interp_elf_ex->e_type != ET_DYN)
646 		goto out;
647 	if (!elf_check_arch(interp_elf_ex) ||
648 	    elf_check_fdpic(interp_elf_ex))
649 		goto out;
650 	if (!interpreter->f_op->mmap)
651 		goto out;
652 
653 	total_size = total_mapping_size(interp_elf_phdata,
654 					interp_elf_ex->e_phnum);
655 	if (!total_size) {
656 		error = -EINVAL;
657 		goto out;
658 	}
659 
660 	eppnt = interp_elf_phdata;
661 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
662 		if (eppnt->p_type == PT_LOAD) {
663 			int elf_type = MAP_PRIVATE;
664 			int elf_prot = make_prot(eppnt->p_flags, arch_state,
665 						 true, true);
666 			unsigned long vaddr = 0;
667 			unsigned long k, map_addr;
668 
669 			vaddr = eppnt->p_vaddr;
670 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
671 				elf_type |= MAP_FIXED;
672 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
673 				load_addr = -vaddr;
674 
675 			map_addr = elf_load(interpreter, load_addr + vaddr,
676 					eppnt, elf_prot, elf_type, total_size);
677 			total_size = 0;
678 			error = map_addr;
679 			if (BAD_ADDR(map_addr))
680 				goto out;
681 
682 			if (!load_addr_set &&
683 			    interp_elf_ex->e_type == ET_DYN) {
684 				load_addr = map_addr - ELF_PAGESTART(vaddr);
685 				load_addr_set = 1;
686 			}
687 
688 			/*
689 			 * Check to see if the section's size will overflow the
690 			 * allowed task size. Note that p_filesz must always be
691 			 * <= p_memsize so it's only necessary to check p_memsz.
692 			 */
693 			k = load_addr + eppnt->p_vaddr;
694 			if (BAD_ADDR(k) ||
695 			    eppnt->p_filesz > eppnt->p_memsz ||
696 			    eppnt->p_memsz > TASK_SIZE ||
697 			    TASK_SIZE - eppnt->p_memsz < k) {
698 				error = -ENOMEM;
699 				goto out;
700 			}
701 		}
702 	}
703 
704 	error = load_addr;
705 out:
706 	return error;
707 }
708 
709 /*
710  * These are the functions used to load ELF style executables and shared
711  * libraries.  There is no binary dependent code anywhere else.
712  */
713 
parse_elf_property(const char * data,size_t * off,size_t datasz,struct arch_elf_state * arch,bool have_prev_type,u32 * prev_type)714 static int parse_elf_property(const char *data, size_t *off, size_t datasz,
715 			      struct arch_elf_state *arch,
716 			      bool have_prev_type, u32 *prev_type)
717 {
718 	size_t o, step;
719 	const struct gnu_property *pr;
720 	int ret;
721 
722 	if (*off == datasz)
723 		return -ENOENT;
724 
725 	if (WARN_ON_ONCE(*off > datasz || *off % ELF_GNU_PROPERTY_ALIGN))
726 		return -EIO;
727 	o = *off;
728 	datasz -= *off;
729 
730 	if (datasz < sizeof(*pr))
731 		return -ENOEXEC;
732 	pr = (const struct gnu_property *)(data + o);
733 	o += sizeof(*pr);
734 	datasz -= sizeof(*pr);
735 
736 	if (pr->pr_datasz > datasz)
737 		return -ENOEXEC;
738 
739 	WARN_ON_ONCE(o % ELF_GNU_PROPERTY_ALIGN);
740 	step = round_up(pr->pr_datasz, ELF_GNU_PROPERTY_ALIGN);
741 	if (step > datasz)
742 		return -ENOEXEC;
743 
744 	/* Properties are supposed to be unique and sorted on pr_type: */
745 	if (have_prev_type && pr->pr_type <= *prev_type)
746 		return -ENOEXEC;
747 	*prev_type = pr->pr_type;
748 
749 	ret = arch_parse_elf_property(pr->pr_type, data + o,
750 				      pr->pr_datasz, ELF_COMPAT, arch);
751 	if (ret)
752 		return ret;
753 
754 	*off = o + step;
755 	return 0;
756 }
757 
758 #define NOTE_DATA_SZ SZ_1K
759 #define GNU_PROPERTY_TYPE_0_NAME "GNU"
760 #define NOTE_NAME_SZ (sizeof(GNU_PROPERTY_TYPE_0_NAME))
761 
parse_elf_properties(struct file * f,const struct elf_phdr * phdr,struct arch_elf_state * arch)762 static int parse_elf_properties(struct file *f, const struct elf_phdr *phdr,
763 				struct arch_elf_state *arch)
764 {
765 	union {
766 		struct elf_note nhdr;
767 		char data[NOTE_DATA_SZ];
768 	} note;
769 	loff_t pos;
770 	ssize_t n;
771 	size_t off, datasz;
772 	int ret;
773 	bool have_prev_type;
774 	u32 prev_type;
775 
776 	if (!IS_ENABLED(CONFIG_ARCH_USE_GNU_PROPERTY) || !phdr)
777 		return 0;
778 
779 	/* load_elf_binary() shouldn't call us unless this is true... */
780 	if (WARN_ON_ONCE(phdr->p_type != PT_GNU_PROPERTY))
781 		return -ENOEXEC;
782 
783 	/* If the properties are crazy large, that's too bad (for now): */
784 	if (phdr->p_filesz > sizeof(note))
785 		return -ENOEXEC;
786 
787 	pos = phdr->p_offset;
788 	n = kernel_read(f, &note, phdr->p_filesz, &pos);
789 
790 	BUILD_BUG_ON(sizeof(note) < sizeof(note.nhdr) + NOTE_NAME_SZ);
791 	if (n < 0 || n < sizeof(note.nhdr) + NOTE_NAME_SZ)
792 		return -EIO;
793 
794 	if (note.nhdr.n_type != NT_GNU_PROPERTY_TYPE_0 ||
795 	    note.nhdr.n_namesz != NOTE_NAME_SZ ||
796 	    strncmp(note.data + sizeof(note.nhdr),
797 		    GNU_PROPERTY_TYPE_0_NAME, n - sizeof(note.nhdr)))
798 		return -ENOEXEC;
799 
800 	off = round_up(sizeof(note.nhdr) + NOTE_NAME_SZ,
801 		       ELF_GNU_PROPERTY_ALIGN);
802 	if (off > n)
803 		return -ENOEXEC;
804 
805 	if (note.nhdr.n_descsz > n - off)
806 		return -ENOEXEC;
807 	datasz = off + note.nhdr.n_descsz;
808 
809 	have_prev_type = false;
810 	do {
811 		ret = parse_elf_property(note.data, &off, datasz, arch,
812 					 have_prev_type, &prev_type);
813 		have_prev_type = true;
814 	} while (!ret);
815 
816 	return ret == -ENOENT ? 0 : ret;
817 }
818 
load_elf_binary(struct linux_binprm * bprm)819 static int load_elf_binary(struct linux_binprm *bprm)
820 {
821 	struct file *interpreter = NULL; /* to shut gcc up */
822 	unsigned long load_bias = 0, phdr_addr = 0;
823 	int first_pt_load = 1;
824 	unsigned long error;
825 	struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
826 	struct elf_phdr *elf_property_phdata = NULL;
827 	unsigned long elf_brk;
828 	int retval, i;
829 	unsigned long elf_entry;
830 	unsigned long e_entry;
831 	unsigned long interp_load_addr = 0;
832 	unsigned long start_code, end_code, start_data, end_data;
833 	unsigned long reloc_func_desc __maybe_unused = 0;
834 	int executable_stack = EXSTACK_DEFAULT;
835 	struct elfhdr *elf_ex = (struct elfhdr *)bprm->buf;
836 	struct elfhdr *interp_elf_ex = NULL;
837 	struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
838 	struct mm_struct *mm;
839 	struct pt_regs *regs;
840 
841 	retval = -ENOEXEC;
842 	/* First of all, some simple consistency checks */
843 	if (memcmp(elf_ex->e_ident, ELFMAG, SELFMAG) != 0)
844 		goto out;
845 
846 	if (elf_ex->e_type != ET_EXEC && elf_ex->e_type != ET_DYN)
847 		goto out;
848 	if (!elf_check_arch(elf_ex))
849 		goto out;
850 	if (elf_check_fdpic(elf_ex))
851 		goto out;
852 	if (!bprm->file->f_op->mmap)
853 		goto out;
854 
855 	elf_phdata = load_elf_phdrs(elf_ex, bprm->file);
856 	if (!elf_phdata)
857 		goto out;
858 
859 	elf_ppnt = elf_phdata;
860 	for (i = 0; i < elf_ex->e_phnum; i++, elf_ppnt++) {
861 		char *elf_interpreter;
862 
863 		if (elf_ppnt->p_type == PT_GNU_PROPERTY) {
864 			elf_property_phdata = elf_ppnt;
865 			continue;
866 		}
867 
868 		if (elf_ppnt->p_type != PT_INTERP)
869 			continue;
870 
871 		/*
872 		 * This is the program interpreter used for shared libraries -
873 		 * for now assume that this is an a.out format binary.
874 		 */
875 		retval = -ENOEXEC;
876 		if (elf_ppnt->p_filesz > PATH_MAX || elf_ppnt->p_filesz < 2)
877 			goto out_free_ph;
878 
879 		retval = -ENOMEM;
880 		elf_interpreter = kmalloc(elf_ppnt->p_filesz, GFP_KERNEL);
881 		if (!elf_interpreter)
882 			goto out_free_ph;
883 
884 		retval = elf_read(bprm->file, elf_interpreter, elf_ppnt->p_filesz,
885 				  elf_ppnt->p_offset);
886 		if (retval < 0)
887 			goto out_free_interp;
888 		/* make sure path is NULL terminated */
889 		retval = -ENOEXEC;
890 		if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
891 			goto out_free_interp;
892 
893 		interpreter = open_exec(elf_interpreter);
894 		kfree(elf_interpreter);
895 		retval = PTR_ERR(interpreter);
896 		if (IS_ERR(interpreter))
897 			goto out_free_ph;
898 
899 		/*
900 		 * If the binary is not readable then enforce mm->dumpable = 0
901 		 * regardless of the interpreter's permissions.
902 		 */
903 		would_dump(bprm, interpreter);
904 
905 		interp_elf_ex = kmalloc(sizeof(*interp_elf_ex), GFP_KERNEL);
906 		if (!interp_elf_ex) {
907 			retval = -ENOMEM;
908 			goto out_free_file;
909 		}
910 
911 		/* Get the exec headers */
912 		retval = elf_read(interpreter, interp_elf_ex,
913 				  sizeof(*interp_elf_ex), 0);
914 		if (retval < 0)
915 			goto out_free_dentry;
916 
917 		break;
918 
919 out_free_interp:
920 		kfree(elf_interpreter);
921 		goto out_free_ph;
922 	}
923 
924 	elf_ppnt = elf_phdata;
925 	for (i = 0; i < elf_ex->e_phnum; i++, elf_ppnt++)
926 		switch (elf_ppnt->p_type) {
927 		case PT_GNU_STACK:
928 			if (elf_ppnt->p_flags & PF_X)
929 				executable_stack = EXSTACK_ENABLE_X;
930 			else
931 				executable_stack = EXSTACK_DISABLE_X;
932 			break;
933 
934 		case PT_LOPROC ... PT_HIPROC:
935 			retval = arch_elf_pt_proc(elf_ex, elf_ppnt,
936 						  bprm->file, false,
937 						  &arch_state);
938 			if (retval)
939 				goto out_free_dentry;
940 			break;
941 		}
942 
943 	/* Some simple consistency checks for the interpreter */
944 	if (interpreter) {
945 		retval = -ELIBBAD;
946 		/* Not an ELF interpreter */
947 		if (memcmp(interp_elf_ex->e_ident, ELFMAG, SELFMAG) != 0)
948 			goto out_free_dentry;
949 		/* Verify the interpreter has a valid arch */
950 		if (!elf_check_arch(interp_elf_ex) ||
951 		    elf_check_fdpic(interp_elf_ex))
952 			goto out_free_dentry;
953 
954 		/* Load the interpreter program headers */
955 		interp_elf_phdata = load_elf_phdrs(interp_elf_ex,
956 						   interpreter);
957 		if (!interp_elf_phdata)
958 			goto out_free_dentry;
959 
960 		/* Pass PT_LOPROC..PT_HIPROC headers to arch code */
961 		elf_property_phdata = NULL;
962 		elf_ppnt = interp_elf_phdata;
963 		for (i = 0; i < interp_elf_ex->e_phnum; i++, elf_ppnt++)
964 			switch (elf_ppnt->p_type) {
965 			case PT_GNU_PROPERTY:
966 				elf_property_phdata = elf_ppnt;
967 				break;
968 
969 			case PT_LOPROC ... PT_HIPROC:
970 				retval = arch_elf_pt_proc(interp_elf_ex,
971 							  elf_ppnt, interpreter,
972 							  true, &arch_state);
973 				if (retval)
974 					goto out_free_dentry;
975 				break;
976 			}
977 	}
978 
979 	retval = parse_elf_properties(interpreter ?: bprm->file,
980 				      elf_property_phdata, &arch_state);
981 	if (retval)
982 		goto out_free_dentry;
983 
984 	/*
985 	 * Allow arch code to reject the ELF at this point, whilst it's
986 	 * still possible to return an error to the code that invoked
987 	 * the exec syscall.
988 	 */
989 	retval = arch_check_elf(elf_ex,
990 				!!interpreter, interp_elf_ex,
991 				&arch_state);
992 	if (retval)
993 		goto out_free_dentry;
994 
995 	/* Flush all traces of the currently running executable */
996 	retval = begin_new_exec(bprm);
997 	if (retval)
998 		goto out_free_dentry;
999 
1000 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
1001 	   may depend on the personality.  */
1002 	SET_PERSONALITY2(*elf_ex, &arch_state);
1003 	if (elf_read_implies_exec(*elf_ex, executable_stack))
1004 		current->personality |= READ_IMPLIES_EXEC;
1005 
1006 	const int snapshot_randomize_va_space = READ_ONCE(randomize_va_space);
1007 	if (!(current->personality & ADDR_NO_RANDOMIZE) && snapshot_randomize_va_space)
1008 		current->flags |= PF_RANDOMIZE;
1009 
1010 	setup_new_exec(bprm);
1011 
1012 	/* Do this so that we can load the interpreter, if need be.  We will
1013 	   change some of these later */
1014 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
1015 				 executable_stack);
1016 	if (retval < 0)
1017 		goto out_free_dentry;
1018 
1019 	elf_brk = 0;
1020 
1021 	start_code = ~0UL;
1022 	end_code = 0;
1023 	start_data = 0;
1024 	end_data = 0;
1025 
1026 	/* Now we do a little grungy work by mmapping the ELF image into
1027 	   the correct location in memory. */
1028 	for(i = 0, elf_ppnt = elf_phdata;
1029 	    i < elf_ex->e_phnum; i++, elf_ppnt++) {
1030 		int elf_prot, elf_flags;
1031 		unsigned long k, vaddr;
1032 		unsigned long total_size = 0;
1033 		unsigned long alignment;
1034 
1035 		if (elf_ppnt->p_type != PT_LOAD)
1036 			continue;
1037 
1038 		elf_prot = make_prot(elf_ppnt->p_flags, &arch_state,
1039 				     !!interpreter, false);
1040 
1041 		elf_flags = MAP_PRIVATE;
1042 
1043 		vaddr = elf_ppnt->p_vaddr;
1044 		/*
1045 		 * The first time through the loop, first_pt_load is true:
1046 		 * layout will be calculated. Once set, use MAP_FIXED since
1047 		 * we know we've already safely mapped the entire region with
1048 		 * MAP_FIXED_NOREPLACE in the once-per-binary logic following.
1049 		 */
1050 		if (!first_pt_load) {
1051 			elf_flags |= MAP_FIXED;
1052 		} else if (elf_ex->e_type == ET_EXEC) {
1053 			/*
1054 			 * This logic is run once for the first LOAD Program
1055 			 * Header for ET_EXEC binaries. No special handling
1056 			 * is needed.
1057 			 */
1058 			elf_flags |= MAP_FIXED_NOREPLACE;
1059 		} else if (elf_ex->e_type == ET_DYN) {
1060 			/*
1061 			 * This logic is run once for the first LOAD Program
1062 			 * Header for ET_DYN binaries to calculate the
1063 			 * randomization (load_bias) for all the LOAD
1064 			 * Program Headers.
1065 			 */
1066 
1067 			/*
1068 			 * Calculate the entire size of the ELF mapping
1069 			 * (total_size), used for the initial mapping,
1070 			 * due to load_addr_set which is set to true later
1071 			 * once the initial mapping is performed.
1072 			 *
1073 			 * Note that this is only sensible when the LOAD
1074 			 * segments are contiguous (or overlapping). If
1075 			 * used for LOADs that are far apart, this would
1076 			 * cause the holes between LOADs to be mapped,
1077 			 * running the risk of having the mapping fail,
1078 			 * as it would be larger than the ELF file itself.
1079 			 *
1080 			 * As a result, only ET_DYN does this, since
1081 			 * some ET_EXEC (e.g. ia64) may have large virtual
1082 			 * memory holes between LOADs.
1083 			 *
1084 			 */
1085 			total_size = total_mapping_size(elf_phdata,
1086 							elf_ex->e_phnum);
1087 			if (!total_size) {
1088 				retval = -EINVAL;
1089 				goto out_free_dentry;
1090 			}
1091 
1092 			/* Calculate any requested alignment. */
1093 			alignment = maximum_alignment(elf_phdata, elf_ex->e_phnum);
1094 
1095 			/*
1096 			 * There are effectively two types of ET_DYN
1097 			 * binaries: programs (i.e. PIE: ET_DYN with PT_INTERP)
1098 			 * and loaders (ET_DYN without PT_INTERP, since they
1099 			 * _are_ the ELF interpreter). The loaders must
1100 			 * be loaded away from programs since the program
1101 			 * may otherwise collide with the loader (especially
1102 			 * for ET_EXEC which does not have a randomized
1103 			 * position). For example to handle invocations of
1104 			 * "./ld.so someprog" to test out a new version of
1105 			 * the loader, the subsequent program that the
1106 			 * loader loads must avoid the loader itself, so
1107 			 * they cannot share the same load range. Sufficient
1108 			 * room for the brk must be allocated with the
1109 			 * loader as well, since brk must be available with
1110 			 * the loader.
1111 			 *
1112 			 * Therefore, programs are loaded offset from
1113 			 * ELF_ET_DYN_BASE and loaders are loaded into the
1114 			 * independently randomized mmap region (0 load_bias
1115 			 * without MAP_FIXED nor MAP_FIXED_NOREPLACE).
1116 			 */
1117 			if (interpreter) {
1118 				/* On ET_DYN with PT_INTERP, we do the ASLR. */
1119 				load_bias = ELF_ET_DYN_BASE;
1120 				if (current->flags & PF_RANDOMIZE)
1121 					load_bias += arch_mmap_rnd();
1122 				/* Adjust alignment as requested. */
1123 				if (alignment)
1124 					load_bias &= ~(alignment - 1);
1125 				elf_flags |= MAP_FIXED_NOREPLACE;
1126 			} else {
1127 				/*
1128 				 * For ET_DYN without PT_INTERP, we rely on
1129 				 * the architectures's (potentially ASLR) mmap
1130 				 * base address (via a load_bias of 0).
1131 				 *
1132 				 * When a large alignment is requested, we
1133 				 * must do the allocation at address "0" right
1134 				 * now to discover where things will load so
1135 				 * that we can adjust the resulting alignment.
1136 				 * In this case (load_bias != 0), we can use
1137 				 * MAP_FIXED_NOREPLACE to make sure the mapping
1138 				 * doesn't collide with anything.
1139 				 */
1140 				if (alignment > ELF_MIN_ALIGN) {
1141 					load_bias = elf_load(bprm->file, 0, elf_ppnt,
1142 							     elf_prot, elf_flags, total_size);
1143 					if (BAD_ADDR(load_bias)) {
1144 						retval = IS_ERR_VALUE(load_bias) ?
1145 							 PTR_ERR((void*)load_bias) : -EINVAL;
1146 						goto out_free_dentry;
1147 					}
1148 					vm_munmap(load_bias, total_size);
1149 					/* Adjust alignment as requested. */
1150 					if (alignment)
1151 						load_bias &= ~(alignment - 1);
1152 					elf_flags |= MAP_FIXED_NOREPLACE;
1153 				} else
1154 					load_bias = 0;
1155 			}
1156 
1157 			/*
1158 			 * Since load_bias is used for all subsequent loading
1159 			 * calculations, we must lower it by the first vaddr
1160 			 * so that the remaining calculations based on the
1161 			 * ELF vaddrs will be correctly offset. The result
1162 			 * is then page aligned.
1163 			 */
1164 			load_bias = ELF_PAGESTART(load_bias - vaddr);
1165 		}
1166 
1167 		error = elf_load(bprm->file, load_bias + vaddr, elf_ppnt,
1168 				elf_prot, elf_flags, total_size);
1169 		if (BAD_ADDR(error)) {
1170 			retval = IS_ERR_VALUE(error) ?
1171 				PTR_ERR((void*)error) : -EINVAL;
1172 			goto out_free_dentry;
1173 		}
1174 
1175 		if (first_pt_load) {
1176 			first_pt_load = 0;
1177 			if (elf_ex->e_type == ET_DYN) {
1178 				load_bias += error -
1179 				             ELF_PAGESTART(load_bias + vaddr);
1180 				reloc_func_desc = load_bias;
1181 			}
1182 		}
1183 
1184 		/*
1185 		 * Figure out which segment in the file contains the Program
1186 		 * Header table, and map to the associated memory address.
1187 		 */
1188 		if (elf_ppnt->p_offset <= elf_ex->e_phoff &&
1189 		    elf_ex->e_phoff < elf_ppnt->p_offset + elf_ppnt->p_filesz) {
1190 			phdr_addr = elf_ex->e_phoff - elf_ppnt->p_offset +
1191 				    elf_ppnt->p_vaddr;
1192 		}
1193 
1194 		k = elf_ppnt->p_vaddr;
1195 		if ((elf_ppnt->p_flags & PF_X) && k < start_code)
1196 			start_code = k;
1197 		if (start_data < k)
1198 			start_data = k;
1199 
1200 		/*
1201 		 * Check to see if the section's size will overflow the
1202 		 * allowed task size. Note that p_filesz must always be
1203 		 * <= p_memsz so it is only necessary to check p_memsz.
1204 		 */
1205 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
1206 		    elf_ppnt->p_memsz > TASK_SIZE ||
1207 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
1208 			/* set_brk can never work. Avoid overflows. */
1209 			retval = -EINVAL;
1210 			goto out_free_dentry;
1211 		}
1212 
1213 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
1214 
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 			elf_brk = k;
1222 	}
1223 
1224 	e_entry = elf_ex->e_entry + load_bias;
1225 	phdr_addr += load_bias;
1226 	elf_brk += load_bias;
1227 	start_code += load_bias;
1228 	end_code += load_bias;
1229 	start_data += load_bias;
1230 	end_data += load_bias;
1231 
1232 	current->mm->start_brk = current->mm->brk = ELF_PAGEALIGN(elf_brk);
1233 
1234 	if (interpreter) {
1235 		elf_entry = load_elf_interp(interp_elf_ex,
1236 					    interpreter,
1237 					    load_bias, interp_elf_phdata,
1238 					    &arch_state);
1239 		if (!IS_ERR_VALUE(elf_entry)) {
1240 			/*
1241 			 * load_elf_interp() returns relocation
1242 			 * adjustment
1243 			 */
1244 			interp_load_addr = elf_entry;
1245 			elf_entry += interp_elf_ex->e_entry;
1246 		}
1247 		if (BAD_ADDR(elf_entry)) {
1248 			retval = IS_ERR_VALUE(elf_entry) ?
1249 					(int)elf_entry : -EINVAL;
1250 			goto out_free_dentry;
1251 		}
1252 		reloc_func_desc = interp_load_addr;
1253 
1254 		fput(interpreter);
1255 
1256 		kfree(interp_elf_ex);
1257 		kfree(interp_elf_phdata);
1258 	} else {
1259 		elf_entry = e_entry;
1260 		if (BAD_ADDR(elf_entry)) {
1261 			retval = -EINVAL;
1262 			goto out_free_dentry;
1263 		}
1264 	}
1265 
1266 	kfree(elf_phdata);
1267 
1268 	set_binfmt(&elf_format);
1269 
1270 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1271 	retval = ARCH_SETUP_ADDITIONAL_PAGES(bprm, elf_ex, !!interpreter);
1272 	if (retval < 0)
1273 		goto out;
1274 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1275 
1276 	retval = create_elf_tables(bprm, elf_ex, interp_load_addr,
1277 				   e_entry, phdr_addr);
1278 	if (retval < 0)
1279 		goto out;
1280 
1281 	mm = current->mm;
1282 	mm->end_code = end_code;
1283 	mm->start_code = start_code;
1284 	mm->start_data = start_data;
1285 	mm->end_data = end_data;
1286 	mm->start_stack = bprm->p;
1287 
1288 	if ((current->flags & PF_RANDOMIZE) && (snapshot_randomize_va_space > 1)) {
1289 		/*
1290 		 * For architectures with ELF randomization, when executing
1291 		 * a loader directly (i.e. no interpreter listed in ELF
1292 		 * headers), move the brk area out of the mmap region
1293 		 * (since it grows up, and may collide early with the stack
1294 		 * growing down), and into the unused ELF_ET_DYN_BASE region.
1295 		 */
1296 		if (IS_ENABLED(CONFIG_ARCH_HAS_ELF_RANDOMIZE) &&
1297 		    elf_ex->e_type == ET_DYN && !interpreter) {
1298 			mm->brk = mm->start_brk = ELF_ET_DYN_BASE;
1299 		} else {
1300 			/* Otherwise leave a gap between .bss and brk. */
1301 			mm->brk = mm->start_brk = mm->brk + PAGE_SIZE;
1302 		}
1303 
1304 		mm->brk = mm->start_brk = arch_randomize_brk(mm);
1305 #ifdef compat_brk_randomized
1306 		current->brk_randomized = 1;
1307 #endif
1308 	}
1309 
1310 	if (current->personality & MMAP_PAGE_ZERO) {
1311 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
1312 		   and some applications "depend" upon this behavior.
1313 		   Since we do not have the power to recompile these, we
1314 		   emulate the SVr4 behavior. Sigh. */
1315 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1316 				MAP_FIXED | MAP_PRIVATE, 0);
1317 
1318 		retval = do_mseal(0, PAGE_SIZE, 0);
1319 		if (retval)
1320 			pr_warn_ratelimited("pid=%d, couldn't seal address 0, ret=%d.\n",
1321 					    task_pid_nr(current), retval);
1322 	}
1323 
1324 	regs = current_pt_regs();
1325 #ifdef ELF_PLAT_INIT
1326 	/*
1327 	 * The ABI may specify that certain registers be set up in special
1328 	 * ways (on i386 %edx is the address of a DT_FINI function, for
1329 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
1330 	 * that the e_entry field is the address of the function descriptor
1331 	 * for the startup routine, rather than the address of the startup
1332 	 * routine itself.  This macro performs whatever initialization to
1333 	 * the regs structure is required as well as any relocations to the
1334 	 * function descriptor entries when executing dynamically links apps.
1335 	 */
1336 	ELF_PLAT_INIT(regs, reloc_func_desc);
1337 #endif
1338 
1339 	finalize_exec(bprm);
1340 	START_THREAD(elf_ex, regs, elf_entry, bprm->p);
1341 	retval = 0;
1342 out:
1343 	return retval;
1344 
1345 	/* error cleanup */
1346 out_free_dentry:
1347 	kfree(interp_elf_ex);
1348 	kfree(interp_elf_phdata);
1349 out_free_file:
1350 	if (interpreter)
1351 		fput(interpreter);
1352 out_free_ph:
1353 	kfree(elf_phdata);
1354 	goto out;
1355 }
1356 
1357 #ifdef CONFIG_USELIB
1358 /* This is really simpleminded and specialized - we are loading an
1359    a.out library that is given an ELF header. */
load_elf_library(struct file * file)1360 static int load_elf_library(struct file *file)
1361 {
1362 	struct elf_phdr *elf_phdata;
1363 	struct elf_phdr *eppnt;
1364 	int retval, error, i, j;
1365 	struct elfhdr elf_ex;
1366 
1367 	error = -ENOEXEC;
1368 	retval = elf_read(file, &elf_ex, sizeof(elf_ex), 0);
1369 	if (retval < 0)
1370 		goto out;
1371 
1372 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1373 		goto out;
1374 
1375 	/* First of all, some simple consistency checks */
1376 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1377 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1378 		goto out;
1379 	if (elf_check_fdpic(&elf_ex))
1380 		goto out;
1381 
1382 	/* Now read in all of the header information */
1383 
1384 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1385 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1386 
1387 	error = -ENOMEM;
1388 	elf_phdata = kmalloc(j, GFP_KERNEL);
1389 	if (!elf_phdata)
1390 		goto out;
1391 
1392 	eppnt = elf_phdata;
1393 	error = -ENOEXEC;
1394 	retval = elf_read(file, eppnt, j, elf_ex.e_phoff);
1395 	if (retval < 0)
1396 		goto out_free_ph;
1397 
1398 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1399 		if ((eppnt + i)->p_type == PT_LOAD)
1400 			j++;
1401 	if (j != 1)
1402 		goto out_free_ph;
1403 
1404 	while (eppnt->p_type != PT_LOAD)
1405 		eppnt++;
1406 
1407 	/* Now use mmap to map the library into memory. */
1408 	error = elf_load(file, ELF_PAGESTART(eppnt->p_vaddr),
1409 			eppnt,
1410 			PROT_READ | PROT_WRITE | PROT_EXEC,
1411 			MAP_FIXED_NOREPLACE | MAP_PRIVATE,
1412 			0);
1413 
1414 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1415 		goto out_free_ph;
1416 
1417 	error = 0;
1418 
1419 out_free_ph:
1420 	kfree(elf_phdata);
1421 out:
1422 	return error;
1423 }
1424 #endif /* #ifdef CONFIG_USELIB */
1425 
1426 #ifdef CONFIG_ELF_CORE
1427 /*
1428  * ELF core dumper
1429  *
1430  * Modelled on fs/exec.c:aout_core_dump()
1431  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1432  */
1433 
1434 /* An ELF note in memory */
1435 struct memelfnote
1436 {
1437 	const char *name;
1438 	int type;
1439 	unsigned int datasz;
1440 	void *data;
1441 };
1442 
notesize(struct memelfnote * en)1443 static int notesize(struct memelfnote *en)
1444 {
1445 	int sz;
1446 
1447 	sz = sizeof(struct elf_note);
1448 	sz += roundup(strlen(en->name) + 1, 4);
1449 	sz += roundup(en->datasz, 4);
1450 
1451 	return sz;
1452 }
1453 
writenote(struct memelfnote * men,struct coredump_params * cprm)1454 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1455 {
1456 	struct elf_note en;
1457 	en.n_namesz = strlen(men->name) + 1;
1458 	en.n_descsz = men->datasz;
1459 	en.n_type = men->type;
1460 
1461 	return dump_emit(cprm, &en, sizeof(en)) &&
1462 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1463 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1464 }
1465 
fill_elf_header(struct elfhdr * elf,int segs,u16 machine,u32 flags)1466 static void fill_elf_header(struct elfhdr *elf, int segs,
1467 			    u16 machine, u32 flags)
1468 {
1469 	memset(elf, 0, sizeof(*elf));
1470 
1471 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1472 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1473 	elf->e_ident[EI_DATA] = ELF_DATA;
1474 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1475 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1476 
1477 	elf->e_type = ET_CORE;
1478 	elf->e_machine = machine;
1479 	elf->e_version = EV_CURRENT;
1480 	elf->e_phoff = sizeof(struct elfhdr);
1481 	elf->e_flags = flags;
1482 	elf->e_ehsize = sizeof(struct elfhdr);
1483 	elf->e_phentsize = sizeof(struct elf_phdr);
1484 	elf->e_phnum = segs;
1485 }
1486 
fill_elf_note_phdr(struct elf_phdr * phdr,int sz,loff_t offset)1487 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1488 {
1489 	phdr->p_type = PT_NOTE;
1490 	phdr->p_offset = offset;
1491 	phdr->p_vaddr = 0;
1492 	phdr->p_paddr = 0;
1493 	phdr->p_filesz = sz;
1494 	phdr->p_memsz = 0;
1495 	phdr->p_flags = 0;
1496 	phdr->p_align = 4;
1497 }
1498 
fill_note(struct memelfnote * note,const char * name,int type,unsigned int sz,void * data)1499 static void fill_note(struct memelfnote *note, const char *name, int type,
1500 		unsigned int sz, void *data)
1501 {
1502 	note->name = name;
1503 	note->type = type;
1504 	note->datasz = sz;
1505 	note->data = data;
1506 }
1507 
1508 /*
1509  * fill up all the fields in prstatus from the given task struct, except
1510  * registers which need to be filled up separately.
1511  */
fill_prstatus(struct elf_prstatus_common * prstatus,struct task_struct * p,long signr)1512 static void fill_prstatus(struct elf_prstatus_common *prstatus,
1513 		struct task_struct *p, long signr)
1514 {
1515 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1516 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1517 	prstatus->pr_sighold = p->blocked.sig[0];
1518 	rcu_read_lock();
1519 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1520 	rcu_read_unlock();
1521 	prstatus->pr_pid = task_pid_vnr(p);
1522 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1523 	prstatus->pr_sid = task_session_vnr(p);
1524 	if (thread_group_leader(p)) {
1525 		struct task_cputime cputime;
1526 
1527 		/*
1528 		 * This is the record for the group leader.  It shows the
1529 		 * group-wide total, not its individual thread total.
1530 		 */
1531 		thread_group_cputime(p, &cputime);
1532 		prstatus->pr_utime = ns_to_kernel_old_timeval(cputime.utime);
1533 		prstatus->pr_stime = ns_to_kernel_old_timeval(cputime.stime);
1534 	} else {
1535 		u64 utime, stime;
1536 
1537 		task_cputime(p, &utime, &stime);
1538 		prstatus->pr_utime = ns_to_kernel_old_timeval(utime);
1539 		prstatus->pr_stime = ns_to_kernel_old_timeval(stime);
1540 	}
1541 
1542 	prstatus->pr_cutime = ns_to_kernel_old_timeval(p->signal->cutime);
1543 	prstatus->pr_cstime = ns_to_kernel_old_timeval(p->signal->cstime);
1544 }
1545 
fill_psinfo(struct elf_prpsinfo * psinfo,struct task_struct * p,struct mm_struct * mm)1546 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1547 		       struct mm_struct *mm)
1548 {
1549 	const struct cred *cred;
1550 	unsigned int i, len;
1551 	unsigned int state;
1552 
1553 	/* first copy the parameters from user space */
1554 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1555 
1556 	len = mm->arg_end - mm->arg_start;
1557 	if (len >= ELF_PRARGSZ)
1558 		len = ELF_PRARGSZ-1;
1559 	if (copy_from_user(&psinfo->pr_psargs,
1560 		           (const char __user *)mm->arg_start, len))
1561 		return -EFAULT;
1562 	for(i = 0; i < len; i++)
1563 		if (psinfo->pr_psargs[i] == 0)
1564 			psinfo->pr_psargs[i] = ' ';
1565 	psinfo->pr_psargs[len] = 0;
1566 
1567 	rcu_read_lock();
1568 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1569 	rcu_read_unlock();
1570 	psinfo->pr_pid = task_pid_vnr(p);
1571 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1572 	psinfo->pr_sid = task_session_vnr(p);
1573 
1574 	state = READ_ONCE(p->__state);
1575 	i = state ? ffz(~state) + 1 : 0;
1576 	psinfo->pr_state = i;
1577 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1578 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1579 	psinfo->pr_nice = task_nice(p);
1580 	psinfo->pr_flag = p->flags;
1581 	rcu_read_lock();
1582 	cred = __task_cred(p);
1583 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1584 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1585 	rcu_read_unlock();
1586 	get_task_comm(psinfo->pr_fname, p);
1587 
1588 	return 0;
1589 }
1590 
fill_auxv_note(struct memelfnote * note,struct mm_struct * mm)1591 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1592 {
1593 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1594 	int i = 0;
1595 	do
1596 		i += 2;
1597 	while (auxv[i - 2] != AT_NULL);
1598 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1599 }
1600 
fill_siginfo_note(struct memelfnote * note,user_siginfo_t * csigdata,const kernel_siginfo_t * siginfo)1601 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1602 		const kernel_siginfo_t *siginfo)
1603 {
1604 	copy_siginfo_to_external(csigdata, siginfo);
1605 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1606 }
1607 
1608 /*
1609  * Format of NT_FILE note:
1610  *
1611  * long count     -- how many files are mapped
1612  * long page_size -- units for file_ofs
1613  * array of [COUNT] elements of
1614  *   long start
1615  *   long end
1616  *   long file_ofs
1617  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1618  */
fill_files_note(struct memelfnote * note,struct coredump_params * cprm)1619 static int fill_files_note(struct memelfnote *note, struct coredump_params *cprm)
1620 {
1621 	unsigned count, size, names_ofs, remaining, n;
1622 	user_long_t *data;
1623 	user_long_t *start_end_ofs;
1624 	char *name_base, *name_curpos;
1625 	int i;
1626 
1627 	/* *Estimated* file count and total data size needed */
1628 	count = cprm->vma_count;
1629 	if (count > UINT_MAX / 64)
1630 		return -EINVAL;
1631 	size = count * 64;
1632 
1633 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1634  alloc:
1635 	/* paranoia check */
1636 	if (size >= core_file_note_size_limit) {
1637 		pr_warn_once("coredump Note size too large: %u (does kernel.core_file_note_size_limit sysctl need adjustment?\n",
1638 			      size);
1639 		return -EINVAL;
1640 	}
1641 	size = round_up(size, PAGE_SIZE);
1642 	/*
1643 	 * "size" can be 0 here legitimately.
1644 	 * Let it ENOMEM and omit NT_FILE section which will be empty anyway.
1645 	 */
1646 	data = kvmalloc(size, GFP_KERNEL);
1647 	if (ZERO_OR_NULL_PTR(data))
1648 		return -ENOMEM;
1649 
1650 	start_end_ofs = data + 2;
1651 	name_base = name_curpos = ((char *)data) + names_ofs;
1652 	remaining = size - names_ofs;
1653 	count = 0;
1654 	for (i = 0; i < cprm->vma_count; i++) {
1655 		struct core_vma_metadata *m = &cprm->vma_meta[i];
1656 		struct file *file;
1657 		const char *filename;
1658 
1659 		file = m->file;
1660 		if (!file)
1661 			continue;
1662 		filename = file_path(file, name_curpos, remaining);
1663 		if (IS_ERR(filename)) {
1664 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1665 				kvfree(data);
1666 				size = size * 5 / 4;
1667 				goto alloc;
1668 			}
1669 			continue;
1670 		}
1671 
1672 		/* file_path() fills at the end, move name down */
1673 		/* n = strlen(filename) + 1: */
1674 		n = (name_curpos + remaining) - filename;
1675 		remaining = filename - name_curpos;
1676 		memmove(name_curpos, filename, n);
1677 		name_curpos += n;
1678 
1679 		*start_end_ofs++ = m->start;
1680 		*start_end_ofs++ = m->end;
1681 		*start_end_ofs++ = m->pgoff;
1682 		count++;
1683 	}
1684 
1685 	/* Now we know exact count of files, can store it */
1686 	data[0] = count;
1687 	data[1] = PAGE_SIZE;
1688 	/*
1689 	 * Count usually is less than mm->map_count,
1690 	 * we need to move filenames down.
1691 	 */
1692 	n = cprm->vma_count - count;
1693 	if (n != 0) {
1694 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1695 		memmove(name_base - shift_bytes, name_base,
1696 			name_curpos - name_base);
1697 		name_curpos -= shift_bytes;
1698 	}
1699 
1700 	size = name_curpos - (char *)data;
1701 	fill_note(note, "CORE", NT_FILE, size, data);
1702 	return 0;
1703 }
1704 
1705 #include <linux/regset.h>
1706 
1707 struct elf_thread_core_info {
1708 	struct elf_thread_core_info *next;
1709 	struct task_struct *task;
1710 	struct elf_prstatus prstatus;
1711 	struct memelfnote notes[];
1712 };
1713 
1714 struct elf_note_info {
1715 	struct elf_thread_core_info *thread;
1716 	struct memelfnote psinfo;
1717 	struct memelfnote signote;
1718 	struct memelfnote auxv;
1719 	struct memelfnote files;
1720 	user_siginfo_t csigdata;
1721 	size_t size;
1722 	int thread_notes;
1723 };
1724 
1725 #ifdef CORE_DUMP_USE_REGSET
1726 /*
1727  * When a regset has a writeback hook, we call it on each thread before
1728  * dumping user memory.  On register window machines, this makes sure the
1729  * user memory backing the register data is up to date before we read it.
1730  */
do_thread_regset_writeback(struct task_struct * task,const struct user_regset * regset)1731 static void do_thread_regset_writeback(struct task_struct *task,
1732 				       const struct user_regset *regset)
1733 {
1734 	if (regset->writeback)
1735 		regset->writeback(task, regset, 1);
1736 }
1737 
1738 #ifndef PRSTATUS_SIZE
1739 #define PRSTATUS_SIZE sizeof(struct elf_prstatus)
1740 #endif
1741 
1742 #ifndef SET_PR_FPVALID
1743 #define SET_PR_FPVALID(S) ((S)->pr_fpvalid = 1)
1744 #endif
1745 
fill_thread_core_info(struct elf_thread_core_info * t,const struct user_regset_view * view,long signr,struct elf_note_info * info)1746 static int fill_thread_core_info(struct elf_thread_core_info *t,
1747 				 const struct user_regset_view *view,
1748 				 long signr, struct elf_note_info *info)
1749 {
1750 	unsigned int note_iter, view_iter;
1751 
1752 	/*
1753 	 * NT_PRSTATUS is the one special case, because the regset data
1754 	 * goes into the pr_reg field inside the note contents, rather
1755 	 * than being the whole note contents.  We fill the regset in here.
1756 	 * We assume that regset 0 is NT_PRSTATUS.
1757 	 */
1758 	fill_prstatus(&t->prstatus.common, t->task, signr);
1759 	regset_get(t->task, &view->regsets[0],
1760 		   sizeof(t->prstatus.pr_reg), &t->prstatus.pr_reg);
1761 
1762 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1763 		  PRSTATUS_SIZE, &t->prstatus);
1764 	info->size += notesize(&t->notes[0]);
1765 
1766 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1767 
1768 	/*
1769 	 * Each other regset might generate a note too.  For each regset
1770 	 * that has no core_note_type or is inactive, skip it.
1771 	 */
1772 	note_iter = 1;
1773 	for (view_iter = 1; view_iter < view->n; ++view_iter) {
1774 		const struct user_regset *regset = &view->regsets[view_iter];
1775 		int note_type = regset->core_note_type;
1776 		bool is_fpreg = note_type == NT_PRFPREG;
1777 		void *data;
1778 		int ret;
1779 
1780 		do_thread_regset_writeback(t->task, regset);
1781 		if (!note_type) // not for coredumps
1782 			continue;
1783 		if (regset->active && regset->active(t->task, regset) <= 0)
1784 			continue;
1785 
1786 		ret = regset_get_alloc(t->task, regset, ~0U, &data);
1787 		if (ret < 0)
1788 			continue;
1789 
1790 		if (WARN_ON_ONCE(note_iter >= info->thread_notes))
1791 			break;
1792 
1793 		if (is_fpreg)
1794 			SET_PR_FPVALID(&t->prstatus);
1795 
1796 		fill_note(&t->notes[note_iter], is_fpreg ? "CORE" : "LINUX",
1797 			  note_type, ret, data);
1798 
1799 		info->size += notesize(&t->notes[note_iter]);
1800 		note_iter++;
1801 	}
1802 
1803 	return 1;
1804 }
1805 #else
fill_thread_core_info(struct elf_thread_core_info * t,const struct user_regset_view * view,long signr,struct elf_note_info * info)1806 static int fill_thread_core_info(struct elf_thread_core_info *t,
1807 				 const struct user_regset_view *view,
1808 				 long signr, struct elf_note_info *info)
1809 {
1810 	struct task_struct *p = t->task;
1811 	elf_fpregset_t *fpu;
1812 
1813 	fill_prstatus(&t->prstatus.common, p, signr);
1814 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1815 
1816 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1817 		  &(t->prstatus));
1818 	info->size += notesize(&t->notes[0]);
1819 
1820 	fpu = kzalloc(sizeof(elf_fpregset_t), GFP_KERNEL);
1821 	if (!fpu || !elf_core_copy_task_fpregs(p, fpu)) {
1822 		kfree(fpu);
1823 		return 1;
1824 	}
1825 
1826 	t->prstatus.pr_fpvalid = 1;
1827 	fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1828 	info->size += notesize(&t->notes[1]);
1829 
1830 	return 1;
1831 }
1832 #endif
1833 
fill_note_info(struct elfhdr * elf,int phdrs,struct elf_note_info * info,struct coredump_params * cprm)1834 static int fill_note_info(struct elfhdr *elf, int phdrs,
1835 			  struct elf_note_info *info,
1836 			  struct coredump_params *cprm)
1837 {
1838 	struct task_struct *dump_task = current;
1839 	const struct user_regset_view *view;
1840 	struct elf_thread_core_info *t;
1841 	struct elf_prpsinfo *psinfo;
1842 	struct core_thread *ct;
1843 
1844 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1845 	if (!psinfo)
1846 		return 0;
1847 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1848 
1849 #ifdef CORE_DUMP_USE_REGSET
1850 	view = task_user_regset_view(dump_task);
1851 
1852 	/*
1853 	 * Figure out how many notes we're going to need for each thread.
1854 	 */
1855 	info->thread_notes = 0;
1856 	for (int i = 0; i < view->n; ++i)
1857 		if (view->regsets[i].core_note_type != 0)
1858 			++info->thread_notes;
1859 
1860 	/*
1861 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1862 	 * since it is our one special case.
1863 	 */
1864 	if (unlikely(info->thread_notes == 0) ||
1865 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1866 		WARN_ON(1);
1867 		return 0;
1868 	}
1869 
1870 	/*
1871 	 * Initialize the ELF file header.
1872 	 */
1873 	fill_elf_header(elf, phdrs,
1874 			view->e_machine, view->e_flags);
1875 #else
1876 	view = NULL;
1877 	info->thread_notes = 2;
1878 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1879 #endif
1880 
1881 	/*
1882 	 * Allocate a structure for each thread.
1883 	 */
1884 	info->thread = kzalloc(offsetof(struct elf_thread_core_info,
1885 				     notes[info->thread_notes]),
1886 			    GFP_KERNEL);
1887 	if (unlikely(!info->thread))
1888 		return 0;
1889 
1890 	info->thread->task = dump_task;
1891 	for (ct = dump_task->signal->core_state->dumper.next; ct; ct = ct->next) {
1892 		t = kzalloc(offsetof(struct elf_thread_core_info,
1893 				     notes[info->thread_notes]),
1894 			    GFP_KERNEL);
1895 		if (unlikely(!t))
1896 			return 0;
1897 
1898 		t->task = ct->task;
1899 		t->next = info->thread->next;
1900 		info->thread->next = t;
1901 	}
1902 
1903 	/*
1904 	 * Now fill in each thread's information.
1905 	 */
1906 	for (t = info->thread; t != NULL; t = t->next)
1907 		if (!fill_thread_core_info(t, view, cprm->siginfo->si_signo, info))
1908 			return 0;
1909 
1910 	/*
1911 	 * Fill in the two process-wide notes.
1912 	 */
1913 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1914 	info->size += notesize(&info->psinfo);
1915 
1916 	fill_siginfo_note(&info->signote, &info->csigdata, cprm->siginfo);
1917 	info->size += notesize(&info->signote);
1918 
1919 	fill_auxv_note(&info->auxv, current->mm);
1920 	info->size += notesize(&info->auxv);
1921 
1922 	if (fill_files_note(&info->files, cprm) == 0)
1923 		info->size += notesize(&info->files);
1924 
1925 	return 1;
1926 }
1927 
1928 /*
1929  * Write all the notes for each thread.  When writing the first thread, the
1930  * process-wide notes are interleaved after the first thread-specific note.
1931  */
write_note_info(struct elf_note_info * info,struct coredump_params * cprm)1932 static int write_note_info(struct elf_note_info *info,
1933 			   struct coredump_params *cprm)
1934 {
1935 	bool first = true;
1936 	struct elf_thread_core_info *t = info->thread;
1937 
1938 	do {
1939 		int i;
1940 
1941 		if (!writenote(&t->notes[0], cprm))
1942 			return 0;
1943 
1944 		if (first && !writenote(&info->psinfo, cprm))
1945 			return 0;
1946 		if (first && !writenote(&info->signote, cprm))
1947 			return 0;
1948 		if (first && !writenote(&info->auxv, cprm))
1949 			return 0;
1950 		if (first && info->files.data &&
1951 				!writenote(&info->files, cprm))
1952 			return 0;
1953 
1954 		for (i = 1; i < info->thread_notes; ++i)
1955 			if (t->notes[i].data &&
1956 			    !writenote(&t->notes[i], cprm))
1957 				return 0;
1958 
1959 		first = false;
1960 		t = t->next;
1961 	} while (t);
1962 
1963 	return 1;
1964 }
1965 
free_note_info(struct elf_note_info * info)1966 static void free_note_info(struct elf_note_info *info)
1967 {
1968 	struct elf_thread_core_info *threads = info->thread;
1969 	while (threads) {
1970 		unsigned int i;
1971 		struct elf_thread_core_info *t = threads;
1972 		threads = t->next;
1973 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1974 		for (i = 1; i < info->thread_notes; ++i)
1975 			kvfree(t->notes[i].data);
1976 		kfree(t);
1977 	}
1978 	kfree(info->psinfo.data);
1979 	kvfree(info->files.data);
1980 }
1981 
fill_extnum_info(struct elfhdr * elf,struct elf_shdr * shdr4extnum,elf_addr_t e_shoff,int segs)1982 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1983 			     elf_addr_t e_shoff, int segs)
1984 {
1985 	elf->e_shoff = e_shoff;
1986 	elf->e_shentsize = sizeof(*shdr4extnum);
1987 	elf->e_shnum = 1;
1988 	elf->e_shstrndx = SHN_UNDEF;
1989 
1990 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1991 
1992 	shdr4extnum->sh_type = SHT_NULL;
1993 	shdr4extnum->sh_size = elf->e_shnum;
1994 	shdr4extnum->sh_link = elf->e_shstrndx;
1995 	shdr4extnum->sh_info = segs;
1996 }
1997 
1998 /*
1999  * Actual dumper
2000  *
2001  * This is a two-pass process; first we find the offsets of the bits,
2002  * and then they are actually written out.  If we run out of core limit
2003  * we just truncate.
2004  */
elf_core_dump(struct coredump_params * cprm)2005 static int elf_core_dump(struct coredump_params *cprm)
2006 {
2007 	int has_dumped = 0;
2008 	int segs, i;
2009 	struct elfhdr elf;
2010 	loff_t offset = 0, dataoff;
2011 	struct elf_note_info info = { };
2012 	struct elf_phdr *phdr4note = NULL;
2013 	struct elf_shdr *shdr4extnum = NULL;
2014 	Elf_Half e_phnum;
2015 	elf_addr_t e_shoff;
2016 
2017 	/*
2018 	 * The number of segs are recored into ELF header as 16bit value.
2019 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2020 	 */
2021 	segs = cprm->vma_count + elf_core_extra_phdrs(cprm);
2022 
2023 	/* for notes section */
2024 	segs++;
2025 
2026 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2027 	 * this, kernel supports extended numbering. Have a look at
2028 	 * include/linux/elf.h for further information. */
2029 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2030 
2031 	/*
2032 	 * Collect all the non-memory information about the process for the
2033 	 * notes.  This also sets up the file header.
2034 	 */
2035 	if (!fill_note_info(&elf, e_phnum, &info, cprm))
2036 		goto end_coredump;
2037 
2038 	has_dumped = 1;
2039 
2040 	offset += sizeof(elf);				/* ELF header */
2041 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2042 
2043 	/* Write notes phdr entry */
2044 	{
2045 		size_t sz = info.size;
2046 
2047 		/* For cell spufs and x86 xstate */
2048 		sz += elf_coredump_extra_notes_size();
2049 
2050 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2051 		if (!phdr4note)
2052 			goto end_coredump;
2053 
2054 		fill_elf_note_phdr(phdr4note, sz, offset);
2055 		offset += sz;
2056 	}
2057 
2058 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2059 
2060 	offset += cprm->vma_data_size;
2061 	offset += elf_core_extra_data_size(cprm);
2062 	e_shoff = offset;
2063 
2064 	if (e_phnum == PN_XNUM) {
2065 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2066 		if (!shdr4extnum)
2067 			goto end_coredump;
2068 		fill_extnum_info(&elf, shdr4extnum, e_shoff, segs);
2069 	}
2070 
2071 	offset = dataoff;
2072 
2073 	if (!dump_emit(cprm, &elf, sizeof(elf)))
2074 		goto end_coredump;
2075 
2076 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2077 		goto end_coredump;
2078 
2079 	/* Write program headers for segments dump */
2080 	for (i = 0; i < cprm->vma_count; i++) {
2081 		struct core_vma_metadata *meta = cprm->vma_meta + i;
2082 		struct elf_phdr phdr;
2083 
2084 		phdr.p_type = PT_LOAD;
2085 		phdr.p_offset = offset;
2086 		phdr.p_vaddr = meta->start;
2087 		phdr.p_paddr = 0;
2088 		phdr.p_filesz = meta->dump_size;
2089 		phdr.p_memsz = meta->end - meta->start;
2090 		offset += phdr.p_filesz;
2091 		phdr.p_flags = 0;
2092 		if (meta->flags & VM_READ)
2093 			phdr.p_flags |= PF_R;
2094 		if (meta->flags & VM_WRITE)
2095 			phdr.p_flags |= PF_W;
2096 		if (meta->flags & VM_EXEC)
2097 			phdr.p_flags |= PF_X;
2098 		phdr.p_align = ELF_EXEC_PAGESIZE;
2099 
2100 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2101 			goto end_coredump;
2102 	}
2103 
2104 	if (!elf_core_write_extra_phdrs(cprm, offset))
2105 		goto end_coredump;
2106 
2107 	/* write out the notes section */
2108 	if (!write_note_info(&info, cprm))
2109 		goto end_coredump;
2110 
2111 	/* For cell spufs and x86 xstate */
2112 	if (elf_coredump_extra_notes_write(cprm))
2113 		goto end_coredump;
2114 
2115 	/* Align to page */
2116 	dump_skip_to(cprm, dataoff);
2117 
2118 	for (i = 0; i < cprm->vma_count; i++) {
2119 		struct core_vma_metadata *meta = cprm->vma_meta + i;
2120 
2121 		if (!dump_user_range(cprm, meta->start, meta->dump_size))
2122 			goto end_coredump;
2123 	}
2124 
2125 	if (!elf_core_write_extra_data(cprm))
2126 		goto end_coredump;
2127 
2128 	if (e_phnum == PN_XNUM) {
2129 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2130 			goto end_coredump;
2131 	}
2132 
2133 end_coredump:
2134 	free_note_info(&info);
2135 	kfree(shdr4extnum);
2136 	kfree(phdr4note);
2137 	return has_dumped;
2138 }
2139 
2140 #endif		/* CONFIG_ELF_CORE */
2141 
init_elf_binfmt(void)2142 static int __init init_elf_binfmt(void)
2143 {
2144 	register_binfmt(&elf_format);
2145 	return 0;
2146 }
2147 
exit_elf_binfmt(void)2148 static void __exit exit_elf_binfmt(void)
2149 {
2150 	/* Remove the COFF and ELF loaders. */
2151 	unregister_binfmt(&elf_format);
2152 }
2153 
2154 core_initcall(init_elf_binfmt);
2155 module_exit(exit_elf_binfmt);
2156 
2157 #ifdef CONFIG_BINFMT_ELF_KUNIT_TEST
2158 #include "tests/binfmt_elf_kunit.c"
2159 #endif
2160