xref: /openbmc/linux/fs/binfmt_elf.c (revision f9a82c48)
1 /*
2  * linux/fs/binfmt_elf.c
3  *
4  * These are the functions used to load ELF format executables as used
5  * on SVr4 machines.  Information on the format may be found in the book
6  * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
7  * Tools".
8  *
9  * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
10  */
11 
12 #include <linux/module.h>
13 #include <linux/kernel.h>
14 #include <linux/fs.h>
15 #include <linux/mm.h>
16 #include <linux/mman.h>
17 #include <linux/errno.h>
18 #include <linux/signal.h>
19 #include <linux/binfmts.h>
20 #include <linux/string.h>
21 #include <linux/file.h>
22 #include <linux/slab.h>
23 #include <linux/personality.h>
24 #include <linux/elfcore.h>
25 #include <linux/init.h>
26 #include <linux/highuid.h>
27 #include <linux/compiler.h>
28 #include <linux/highmem.h>
29 #include <linux/pagemap.h>
30 #include <linux/vmalloc.h>
31 #include <linux/security.h>
32 #include <linux/random.h>
33 #include <linux/elf.h>
34 #include <linux/elf-randomize.h>
35 #include <linux/utsname.h>
36 #include <linux/coredump.h>
37 #include <linux/sched.h>
38 #include <linux/sched/coredump.h>
39 #include <linux/sched/task_stack.h>
40 #include <linux/sched/cputime.h>
41 #include <linux/cred.h>
42 #include <linux/dax.h>
43 #include <linux/uaccess.h>
44 #include <asm/param.h>
45 #include <asm/page.h>
46 
47 #ifndef user_long_t
48 #define user_long_t long
49 #endif
50 #ifndef user_siginfo_t
51 #define user_siginfo_t siginfo_t
52 #endif
53 
54 /* That's for binfmt_elf_fdpic to deal with */
55 #ifndef elf_check_fdpic
56 #define elf_check_fdpic(ex) false
57 #endif
58 
59 static int load_elf_binary(struct linux_binprm *bprm);
60 
61 #ifdef CONFIG_USELIB
62 static int load_elf_library(struct file *);
63 #else
64 #define load_elf_library NULL
65 #endif
66 
67 /*
68  * If we don't support core dumping, then supply a NULL so we
69  * don't even try.
70  */
71 #ifdef CONFIG_ELF_CORE
72 static int elf_core_dump(struct coredump_params *cprm);
73 #else
74 #define elf_core_dump	NULL
75 #endif
76 
77 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
78 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
79 #else
80 #define ELF_MIN_ALIGN	PAGE_SIZE
81 #endif
82 
83 #ifndef ELF_CORE_EFLAGS
84 #define ELF_CORE_EFLAGS	0
85 #endif
86 
87 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
88 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
89 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
90 
91 static struct linux_binfmt elf_format = {
92 	.module		= THIS_MODULE,
93 	.load_binary	= load_elf_binary,
94 	.load_shlib	= load_elf_library,
95 	.core_dump	= elf_core_dump,
96 	.min_coredump	= ELF_EXEC_PAGESIZE,
97 };
98 
99 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
100 
101 static int set_brk(unsigned long start, unsigned long end, int prot)
102 {
103 	start = ELF_PAGEALIGN(start);
104 	end = ELF_PAGEALIGN(end);
105 	if (end > start) {
106 		/*
107 		 * Map the last of the bss segment.
108 		 * If the header is requesting these pages to be
109 		 * executable, honour that (ppc32 needs this).
110 		 */
111 		int error = vm_brk_flags(start, end - start,
112 				prot & PROT_EXEC ? VM_EXEC : 0);
113 		if (error)
114 			return error;
115 	}
116 	current->mm->start_brk = current->mm->brk = end;
117 	return 0;
118 }
119 
120 /* We need to explicitly zero any fractional pages
121    after the data section (i.e. bss).  This would
122    contain the junk from the file that should not
123    be in memory
124  */
125 static int padzero(unsigned long elf_bss)
126 {
127 	unsigned long nbyte;
128 
129 	nbyte = ELF_PAGEOFFSET(elf_bss);
130 	if (nbyte) {
131 		nbyte = ELF_MIN_ALIGN - nbyte;
132 		if (clear_user((void __user *) elf_bss, nbyte))
133 			return -EFAULT;
134 	}
135 	return 0;
136 }
137 
138 /* Let's use some macros to make this stack manipulation a little clearer */
139 #ifdef CONFIG_STACK_GROWSUP
140 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
141 #define STACK_ROUND(sp, items) \
142 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
143 #define STACK_ALLOC(sp, len) ({ \
144 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
145 	old_sp; })
146 #else
147 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
148 #define STACK_ROUND(sp, items) \
149 	(((unsigned long) (sp - items)) &~ 15UL)
150 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
151 #endif
152 
153 #ifndef ELF_BASE_PLATFORM
154 /*
155  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
156  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
157  * will be copied to the user stack in the same manner as AT_PLATFORM.
158  */
159 #define ELF_BASE_PLATFORM NULL
160 #endif
161 
162 static int
163 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
164 		unsigned long load_addr, unsigned long interp_load_addr)
165 {
166 	unsigned long p = bprm->p;
167 	int argc = bprm->argc;
168 	int envc = bprm->envc;
169 	elf_addr_t __user *sp;
170 	elf_addr_t __user *u_platform;
171 	elf_addr_t __user *u_base_platform;
172 	elf_addr_t __user *u_rand_bytes;
173 	const char *k_platform = ELF_PLATFORM;
174 	const char *k_base_platform = ELF_BASE_PLATFORM;
175 	unsigned char k_rand_bytes[16];
176 	int items;
177 	elf_addr_t *elf_info;
178 	int ei_index = 0;
179 	const struct cred *cred = current_cred();
180 	struct vm_area_struct *vma;
181 
182 	/*
183 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
184 	 * evictions by the processes running on the same package. One
185 	 * thing we can do is to shuffle the initial stack for them.
186 	 */
187 
188 	p = arch_align_stack(p);
189 
190 	/*
191 	 * If this architecture has a platform capability string, copy it
192 	 * to userspace.  In some cases (Sparc), this info is impossible
193 	 * for userspace to get any other way, in others (i386) it is
194 	 * merely difficult.
195 	 */
196 	u_platform = NULL;
197 	if (k_platform) {
198 		size_t len = strlen(k_platform) + 1;
199 
200 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
201 		if (__copy_to_user(u_platform, k_platform, len))
202 			return -EFAULT;
203 	}
204 
205 	/*
206 	 * If this architecture has a "base" platform capability
207 	 * string, copy it to userspace.
208 	 */
209 	u_base_platform = NULL;
210 	if (k_base_platform) {
211 		size_t len = strlen(k_base_platform) + 1;
212 
213 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
214 		if (__copy_to_user(u_base_platform, k_base_platform, len))
215 			return -EFAULT;
216 	}
217 
218 	/*
219 	 * Generate 16 random bytes for userspace PRNG seeding.
220 	 */
221 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
222 	u_rand_bytes = (elf_addr_t __user *)
223 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
224 	if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
225 		return -EFAULT;
226 
227 	/* Create the ELF interpreter info */
228 	elf_info = (elf_addr_t *)current->mm->saved_auxv;
229 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
230 #define NEW_AUX_ENT(id, val) \
231 	do { \
232 		elf_info[ei_index++] = id; \
233 		elf_info[ei_index++] = val; \
234 	} while (0)
235 
236 #ifdef ARCH_DLINFO
237 	/*
238 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
239 	 * AUXV.
240 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
241 	 * ARCH_DLINFO changes
242 	 */
243 	ARCH_DLINFO;
244 #endif
245 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
246 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
247 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
248 	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
249 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
250 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
251 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
252 	NEW_AUX_ENT(AT_FLAGS, 0);
253 	NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
254 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
255 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
256 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
257 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
258 	NEW_AUX_ENT(AT_SECURE, bprm->secureexec);
259 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
260 #ifdef ELF_HWCAP2
261 	NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
262 #endif
263 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
264 	if (k_platform) {
265 		NEW_AUX_ENT(AT_PLATFORM,
266 			    (elf_addr_t)(unsigned long)u_platform);
267 	}
268 	if (k_base_platform) {
269 		NEW_AUX_ENT(AT_BASE_PLATFORM,
270 			    (elf_addr_t)(unsigned long)u_base_platform);
271 	}
272 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
273 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
274 	}
275 #undef NEW_AUX_ENT
276 	/* AT_NULL is zero; clear the rest too */
277 	memset(&elf_info[ei_index], 0,
278 	       sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
279 
280 	/* And advance past the AT_NULL entry.  */
281 	ei_index += 2;
282 
283 	sp = STACK_ADD(p, ei_index);
284 
285 	items = (argc + 1) + (envc + 1) + 1;
286 	bprm->p = STACK_ROUND(sp, items);
287 
288 	/* Point sp at the lowest address on the stack */
289 #ifdef CONFIG_STACK_GROWSUP
290 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
291 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
292 #else
293 	sp = (elf_addr_t __user *)bprm->p;
294 #endif
295 
296 
297 	/*
298 	 * Grow the stack manually; some architectures have a limit on how
299 	 * far ahead a user-space access may be in order to grow the stack.
300 	 */
301 	vma = find_extend_vma(current->mm, bprm->p);
302 	if (!vma)
303 		return -EFAULT;
304 
305 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
306 	if (__put_user(argc, sp++))
307 		return -EFAULT;
308 
309 	/* Populate list of argv pointers back to argv strings. */
310 	p = current->mm->arg_end = current->mm->arg_start;
311 	while (argc-- > 0) {
312 		size_t len;
313 		if (__put_user((elf_addr_t)p, sp++))
314 			return -EFAULT;
315 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
316 		if (!len || len > MAX_ARG_STRLEN)
317 			return -EINVAL;
318 		p += len;
319 	}
320 	if (__put_user(0, sp++))
321 		return -EFAULT;
322 	current->mm->arg_end = p;
323 
324 	/* Populate list of envp pointers back to envp strings. */
325 	current->mm->env_end = current->mm->env_start = p;
326 	while (envc-- > 0) {
327 		size_t len;
328 		if (__put_user((elf_addr_t)p, sp++))
329 			return -EFAULT;
330 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
331 		if (!len || len > MAX_ARG_STRLEN)
332 			return -EINVAL;
333 		p += len;
334 	}
335 	if (__put_user(0, sp++))
336 		return -EFAULT;
337 	current->mm->env_end = p;
338 
339 	/* Put the elf_info on the stack in the right place.  */
340 	if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
341 		return -EFAULT;
342 	return 0;
343 }
344 
345 #ifndef elf_map
346 
347 static unsigned long elf_map(struct file *filep, unsigned long addr,
348 		const struct elf_phdr *eppnt, int prot, int type,
349 		unsigned long total_size)
350 {
351 	unsigned long map_addr;
352 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
353 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
354 	addr = ELF_PAGESTART(addr);
355 	size = ELF_PAGEALIGN(size);
356 
357 	/* mmap() will return -EINVAL if given a zero size, but a
358 	 * segment with zero filesize is perfectly valid */
359 	if (!size)
360 		return addr;
361 
362 	/*
363 	* total_size is the size of the ELF (interpreter) image.
364 	* The _first_ mmap needs to know the full size, otherwise
365 	* randomization might put this image into an overlapping
366 	* position with the ELF binary image. (since size < total_size)
367 	* So we first map the 'big' image - and unmap the remainder at
368 	* the end. (which unmap is needed for ELF images with holes.)
369 	*/
370 	if (total_size) {
371 		total_size = ELF_PAGEALIGN(total_size);
372 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
373 		if (!BAD_ADDR(map_addr))
374 			vm_munmap(map_addr+size, total_size-size);
375 	} else
376 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
377 
378 	if ((type & MAP_FIXED_NOREPLACE) &&
379 	    PTR_ERR((void *)map_addr) == -EEXIST)
380 		pr_info("%d (%s): Uhuuh, elf segment at %px requested but the memory is mapped already\n",
381 			task_pid_nr(current), current->comm, (void *)addr);
382 
383 	return(map_addr);
384 }
385 
386 #endif /* !elf_map */
387 
388 static unsigned long total_mapping_size(const struct elf_phdr *cmds, int nr)
389 {
390 	int i, first_idx = -1, last_idx = -1;
391 
392 	for (i = 0; i < nr; i++) {
393 		if (cmds[i].p_type == PT_LOAD) {
394 			last_idx = i;
395 			if (first_idx == -1)
396 				first_idx = i;
397 		}
398 	}
399 	if (first_idx == -1)
400 		return 0;
401 
402 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
403 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
404 }
405 
406 /**
407  * load_elf_phdrs() - load ELF program headers
408  * @elf_ex:   ELF header of the binary whose program headers should be loaded
409  * @elf_file: the opened ELF binary file
410  *
411  * Loads ELF program headers from the binary file elf_file, which has the ELF
412  * header pointed to by elf_ex, into a newly allocated array. The caller is
413  * responsible for freeing the allocated data. Returns an ERR_PTR upon failure.
414  */
415 static struct elf_phdr *load_elf_phdrs(const struct elfhdr *elf_ex,
416 				       struct file *elf_file)
417 {
418 	struct elf_phdr *elf_phdata = NULL;
419 	int retval, err = -1;
420 	loff_t pos = elf_ex->e_phoff;
421 	unsigned int size;
422 
423 	/*
424 	 * If the size of this structure has changed, then punt, since
425 	 * we will be doing the wrong thing.
426 	 */
427 	if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
428 		goto out;
429 
430 	/* Sanity check the number of program headers... */
431 	/* ...and their total size. */
432 	size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
433 	if (size == 0 || size > 65536 || size > ELF_MIN_ALIGN)
434 		goto out;
435 
436 	elf_phdata = kmalloc(size, GFP_KERNEL);
437 	if (!elf_phdata)
438 		goto out;
439 
440 	/* Read in the program headers */
441 	retval = kernel_read(elf_file, elf_phdata, size, &pos);
442 	if (retval != size) {
443 		err = (retval < 0) ? retval : -EIO;
444 		goto out;
445 	}
446 
447 	/* Success! */
448 	err = 0;
449 out:
450 	if (err) {
451 		kfree(elf_phdata);
452 		elf_phdata = NULL;
453 	}
454 	return elf_phdata;
455 }
456 
457 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
458 
459 /**
460  * struct arch_elf_state - arch-specific ELF loading state
461  *
462  * This structure is used to preserve architecture specific data during
463  * the loading of an ELF file, throughout the checking of architecture
464  * specific ELF headers & through to the point where the ELF load is
465  * known to be proceeding (ie. SET_PERSONALITY).
466  *
467  * This implementation is a dummy for architectures which require no
468  * specific state.
469  */
470 struct arch_elf_state {
471 };
472 
473 #define INIT_ARCH_ELF_STATE {}
474 
475 /**
476  * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
477  * @ehdr:	The main ELF header
478  * @phdr:	The program header to check
479  * @elf:	The open ELF file
480  * @is_interp:	True if the phdr is from the interpreter of the ELF being
481  *		loaded, else false.
482  * @state:	Architecture-specific state preserved throughout the process
483  *		of loading the ELF.
484  *
485  * Inspects the program header phdr to validate its correctness and/or
486  * suitability for the system. Called once per ELF program header in the
487  * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
488  * interpreter.
489  *
490  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
491  *         with that return code.
492  */
493 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
494 				   struct elf_phdr *phdr,
495 				   struct file *elf, bool is_interp,
496 				   struct arch_elf_state *state)
497 {
498 	/* Dummy implementation, always proceed */
499 	return 0;
500 }
501 
502 /**
503  * arch_check_elf() - check an ELF executable
504  * @ehdr:	The main ELF header
505  * @has_interp:	True if the ELF has an interpreter, else false.
506  * @interp_ehdr: The interpreter's ELF header
507  * @state:	Architecture-specific state preserved throughout the process
508  *		of loading the ELF.
509  *
510  * Provides a final opportunity for architecture code to reject the loading
511  * of the ELF & cause an exec syscall to return an error. This is called after
512  * all program headers to be checked by arch_elf_pt_proc have been.
513  *
514  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
515  *         with that return code.
516  */
517 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
518 				 struct elfhdr *interp_ehdr,
519 				 struct arch_elf_state *state)
520 {
521 	/* Dummy implementation, always proceed */
522 	return 0;
523 }
524 
525 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
526 
527 /* This is much more generalized than the library routine read function,
528    so we keep this separate.  Technically the library read function
529    is only provided so that we can read a.out libraries that have
530    an ELF header */
531 
532 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
533 		struct file *interpreter, unsigned long *interp_map_addr,
534 		unsigned long no_base, struct elf_phdr *interp_elf_phdata)
535 {
536 	struct elf_phdr *eppnt;
537 	unsigned long load_addr = 0;
538 	int load_addr_set = 0;
539 	unsigned long last_bss = 0, elf_bss = 0;
540 	int bss_prot = 0;
541 	unsigned long error = ~0UL;
542 	unsigned long total_size;
543 	int i;
544 
545 	/* First of all, some simple consistency checks */
546 	if (interp_elf_ex->e_type != ET_EXEC &&
547 	    interp_elf_ex->e_type != ET_DYN)
548 		goto out;
549 	if (!elf_check_arch(interp_elf_ex) ||
550 	    elf_check_fdpic(interp_elf_ex))
551 		goto out;
552 	if (!interpreter->f_op->mmap)
553 		goto out;
554 
555 	total_size = total_mapping_size(interp_elf_phdata,
556 					interp_elf_ex->e_phnum);
557 	if (!total_size) {
558 		error = -EINVAL;
559 		goto out;
560 	}
561 
562 	eppnt = interp_elf_phdata;
563 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
564 		if (eppnt->p_type == PT_LOAD) {
565 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
566 			int elf_prot = 0;
567 			unsigned long vaddr = 0;
568 			unsigned long k, map_addr;
569 
570 			if (eppnt->p_flags & PF_R)
571 		    		elf_prot = PROT_READ;
572 			if (eppnt->p_flags & PF_W)
573 				elf_prot |= PROT_WRITE;
574 			if (eppnt->p_flags & PF_X)
575 				elf_prot |= PROT_EXEC;
576 			vaddr = eppnt->p_vaddr;
577 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
578 				elf_type |= MAP_FIXED_NOREPLACE;
579 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
580 				load_addr = -vaddr;
581 
582 			map_addr = elf_map(interpreter, load_addr + vaddr,
583 					eppnt, elf_prot, elf_type, total_size);
584 			total_size = 0;
585 			if (!*interp_map_addr)
586 				*interp_map_addr = map_addr;
587 			error = map_addr;
588 			if (BAD_ADDR(map_addr))
589 				goto out;
590 
591 			if (!load_addr_set &&
592 			    interp_elf_ex->e_type == ET_DYN) {
593 				load_addr = map_addr - ELF_PAGESTART(vaddr);
594 				load_addr_set = 1;
595 			}
596 
597 			/*
598 			 * Check to see if the section's size will overflow the
599 			 * allowed task size. Note that p_filesz must always be
600 			 * <= p_memsize so it's only necessary to check p_memsz.
601 			 */
602 			k = load_addr + eppnt->p_vaddr;
603 			if (BAD_ADDR(k) ||
604 			    eppnt->p_filesz > eppnt->p_memsz ||
605 			    eppnt->p_memsz > TASK_SIZE ||
606 			    TASK_SIZE - eppnt->p_memsz < k) {
607 				error = -ENOMEM;
608 				goto out;
609 			}
610 
611 			/*
612 			 * Find the end of the file mapping for this phdr, and
613 			 * keep track of the largest address we see for this.
614 			 */
615 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
616 			if (k > elf_bss)
617 				elf_bss = k;
618 
619 			/*
620 			 * Do the same thing for the memory mapping - between
621 			 * elf_bss and last_bss is the bss section.
622 			 */
623 			k = load_addr + eppnt->p_vaddr + eppnt->p_memsz;
624 			if (k > last_bss) {
625 				last_bss = k;
626 				bss_prot = elf_prot;
627 			}
628 		}
629 	}
630 
631 	/*
632 	 * Now fill out the bss section: first pad the last page from
633 	 * the file up to the page boundary, and zero it from elf_bss
634 	 * up to the end of the page.
635 	 */
636 	if (padzero(elf_bss)) {
637 		error = -EFAULT;
638 		goto out;
639 	}
640 	/*
641 	 * Next, align both the file and mem bss up to the page size,
642 	 * since this is where elf_bss was just zeroed up to, and where
643 	 * last_bss will end after the vm_brk_flags() below.
644 	 */
645 	elf_bss = ELF_PAGEALIGN(elf_bss);
646 	last_bss = ELF_PAGEALIGN(last_bss);
647 	/* Finally, if there is still more bss to allocate, do it. */
648 	if (last_bss > elf_bss) {
649 		error = vm_brk_flags(elf_bss, last_bss - elf_bss,
650 				bss_prot & PROT_EXEC ? VM_EXEC : 0);
651 		if (error)
652 			goto out;
653 	}
654 
655 	error = load_addr;
656 out:
657 	return error;
658 }
659 
660 /*
661  * These are the functions used to load ELF style executables and shared
662  * libraries.  There is no binary dependent code anywhere else.
663  */
664 
665 #ifndef STACK_RND_MASK
666 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))	/* 8MB of VA */
667 #endif
668 
669 static unsigned long randomize_stack_top(unsigned long stack_top)
670 {
671 	unsigned long random_variable = 0;
672 
673 	if (current->flags & PF_RANDOMIZE) {
674 		random_variable = get_random_long();
675 		random_variable &= STACK_RND_MASK;
676 		random_variable <<= PAGE_SHIFT;
677 	}
678 #ifdef CONFIG_STACK_GROWSUP
679 	return PAGE_ALIGN(stack_top) + random_variable;
680 #else
681 	return PAGE_ALIGN(stack_top) - random_variable;
682 #endif
683 }
684 
685 static int load_elf_binary(struct linux_binprm *bprm)
686 {
687 	struct file *interpreter = NULL; /* to shut gcc up */
688  	unsigned long load_addr = 0, load_bias = 0;
689 	int load_addr_set = 0;
690 	char * elf_interpreter = NULL;
691 	unsigned long error;
692 	struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
693 	unsigned long elf_bss, elf_brk;
694 	int bss_prot = 0;
695 	int retval, i;
696 	unsigned long elf_entry;
697 	unsigned long interp_load_addr = 0;
698 	unsigned long start_code, end_code, start_data, end_data;
699 	unsigned long reloc_func_desc __maybe_unused = 0;
700 	int executable_stack = EXSTACK_DEFAULT;
701 	struct pt_regs *regs = current_pt_regs();
702 	struct {
703 		struct elfhdr elf_ex;
704 		struct elfhdr interp_elf_ex;
705 	} *loc;
706 	struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
707 	loff_t pos;
708 
709 	loc = kmalloc(sizeof(*loc), GFP_KERNEL);
710 	if (!loc) {
711 		retval = -ENOMEM;
712 		goto out_ret;
713 	}
714 
715 	/* Get the exec-header */
716 	loc->elf_ex = *((struct elfhdr *)bprm->buf);
717 
718 	retval = -ENOEXEC;
719 	/* First of all, some simple consistency checks */
720 	if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
721 		goto out;
722 
723 	if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
724 		goto out;
725 	if (!elf_check_arch(&loc->elf_ex))
726 		goto out;
727 	if (elf_check_fdpic(&loc->elf_ex))
728 		goto out;
729 	if (!bprm->file->f_op->mmap)
730 		goto out;
731 
732 	elf_phdata = load_elf_phdrs(&loc->elf_ex, bprm->file);
733 	if (!elf_phdata)
734 		goto out;
735 
736 	elf_ppnt = elf_phdata;
737 	elf_bss = 0;
738 	elf_brk = 0;
739 
740 	start_code = ~0UL;
741 	end_code = 0;
742 	start_data = 0;
743 	end_data = 0;
744 
745 	for (i = 0; i < loc->elf_ex.e_phnum; i++) {
746 		if (elf_ppnt->p_type == PT_INTERP) {
747 			/* This is the program interpreter used for
748 			 * shared libraries - for now assume that this
749 			 * is an a.out format binary
750 			 */
751 			retval = -ENOEXEC;
752 			if (elf_ppnt->p_filesz > PATH_MAX ||
753 			    elf_ppnt->p_filesz < 2)
754 				goto out_free_ph;
755 
756 			retval = -ENOMEM;
757 			elf_interpreter = kmalloc(elf_ppnt->p_filesz,
758 						  GFP_KERNEL);
759 			if (!elf_interpreter)
760 				goto out_free_ph;
761 
762 			pos = elf_ppnt->p_offset;
763 			retval = kernel_read(bprm->file, elf_interpreter,
764 					     elf_ppnt->p_filesz, &pos);
765 			if (retval != elf_ppnt->p_filesz) {
766 				if (retval >= 0)
767 					retval = -EIO;
768 				goto out_free_interp;
769 			}
770 			/* make sure path is NULL terminated */
771 			retval = -ENOEXEC;
772 			if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
773 				goto out_free_interp;
774 
775 			interpreter = open_exec(elf_interpreter);
776 			retval = PTR_ERR(interpreter);
777 			if (IS_ERR(interpreter))
778 				goto out_free_interp;
779 
780 			/*
781 			 * If the binary is not readable then enforce
782 			 * mm->dumpable = 0 regardless of the interpreter's
783 			 * permissions.
784 			 */
785 			would_dump(bprm, interpreter);
786 
787 			/* Get the exec headers */
788 			pos = 0;
789 			retval = kernel_read(interpreter, &loc->interp_elf_ex,
790 					     sizeof(loc->interp_elf_ex), &pos);
791 			if (retval != sizeof(loc->interp_elf_ex)) {
792 				if (retval >= 0)
793 					retval = -EIO;
794 				goto out_free_dentry;
795 			}
796 
797 			break;
798 		}
799 		elf_ppnt++;
800 	}
801 
802 	elf_ppnt = elf_phdata;
803 	for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
804 		switch (elf_ppnt->p_type) {
805 		case PT_GNU_STACK:
806 			if (elf_ppnt->p_flags & PF_X)
807 				executable_stack = EXSTACK_ENABLE_X;
808 			else
809 				executable_stack = EXSTACK_DISABLE_X;
810 			break;
811 
812 		case PT_LOPROC ... PT_HIPROC:
813 			retval = arch_elf_pt_proc(&loc->elf_ex, elf_ppnt,
814 						  bprm->file, false,
815 						  &arch_state);
816 			if (retval)
817 				goto out_free_dentry;
818 			break;
819 		}
820 
821 	/* Some simple consistency checks for the interpreter */
822 	if (elf_interpreter) {
823 		retval = -ELIBBAD;
824 		/* Not an ELF interpreter */
825 		if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
826 			goto out_free_dentry;
827 		/* Verify the interpreter has a valid arch */
828 		if (!elf_check_arch(&loc->interp_elf_ex) ||
829 		    elf_check_fdpic(&loc->interp_elf_ex))
830 			goto out_free_dentry;
831 
832 		/* Load the interpreter program headers */
833 		interp_elf_phdata = load_elf_phdrs(&loc->interp_elf_ex,
834 						   interpreter);
835 		if (!interp_elf_phdata)
836 			goto out_free_dentry;
837 
838 		/* Pass PT_LOPROC..PT_HIPROC headers to arch code */
839 		elf_ppnt = interp_elf_phdata;
840 		for (i = 0; i < loc->interp_elf_ex.e_phnum; i++, elf_ppnt++)
841 			switch (elf_ppnt->p_type) {
842 			case PT_LOPROC ... PT_HIPROC:
843 				retval = arch_elf_pt_proc(&loc->interp_elf_ex,
844 							  elf_ppnt, interpreter,
845 							  true, &arch_state);
846 				if (retval)
847 					goto out_free_dentry;
848 				break;
849 			}
850 	}
851 
852 	/*
853 	 * Allow arch code to reject the ELF at this point, whilst it's
854 	 * still possible to return an error to the code that invoked
855 	 * the exec syscall.
856 	 */
857 	retval = arch_check_elf(&loc->elf_ex,
858 				!!interpreter, &loc->interp_elf_ex,
859 				&arch_state);
860 	if (retval)
861 		goto out_free_dentry;
862 
863 	/* Flush all traces of the currently running executable */
864 	retval = flush_old_exec(bprm);
865 	if (retval)
866 		goto out_free_dentry;
867 
868 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
869 	   may depend on the personality.  */
870 	SET_PERSONALITY2(loc->elf_ex, &arch_state);
871 	if (elf_read_implies_exec(loc->elf_ex, executable_stack))
872 		current->personality |= READ_IMPLIES_EXEC;
873 
874 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
875 		current->flags |= PF_RANDOMIZE;
876 
877 	setup_new_exec(bprm);
878 	install_exec_creds(bprm);
879 
880 	/* Do this so that we can load the interpreter, if need be.  We will
881 	   change some of these later */
882 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
883 				 executable_stack);
884 	if (retval < 0)
885 		goto out_free_dentry;
886 
887 	current->mm->start_stack = bprm->p;
888 
889 	/* Now we do a little grungy work by mmapping the ELF image into
890 	   the correct location in memory. */
891 	for(i = 0, elf_ppnt = elf_phdata;
892 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
893 		int elf_prot = 0, elf_flags, elf_fixed = MAP_FIXED_NOREPLACE;
894 		unsigned long k, vaddr;
895 		unsigned long total_size = 0;
896 
897 		if (elf_ppnt->p_type != PT_LOAD)
898 			continue;
899 
900 		if (unlikely (elf_brk > elf_bss)) {
901 			unsigned long nbyte;
902 
903 			/* There was a PT_LOAD segment with p_memsz > p_filesz
904 			   before this one. Map anonymous pages, if needed,
905 			   and clear the area.  */
906 			retval = set_brk(elf_bss + load_bias,
907 					 elf_brk + load_bias,
908 					 bss_prot);
909 			if (retval)
910 				goto out_free_dentry;
911 			nbyte = ELF_PAGEOFFSET(elf_bss);
912 			if (nbyte) {
913 				nbyte = ELF_MIN_ALIGN - nbyte;
914 				if (nbyte > elf_brk - elf_bss)
915 					nbyte = elf_brk - elf_bss;
916 				if (clear_user((void __user *)elf_bss +
917 							load_bias, nbyte)) {
918 					/*
919 					 * This bss-zeroing can fail if the ELF
920 					 * file specifies odd protections. So
921 					 * we don't check the return value
922 					 */
923 				}
924 			}
925 
926 			/*
927 			 * Some binaries have overlapping elf segments and then
928 			 * we have to forcefully map over an existing mapping
929 			 * e.g. over this newly established brk mapping.
930 			 */
931 			elf_fixed = MAP_FIXED;
932 		}
933 
934 		if (elf_ppnt->p_flags & PF_R)
935 			elf_prot |= PROT_READ;
936 		if (elf_ppnt->p_flags & PF_W)
937 			elf_prot |= PROT_WRITE;
938 		if (elf_ppnt->p_flags & PF_X)
939 			elf_prot |= PROT_EXEC;
940 
941 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
942 
943 		vaddr = elf_ppnt->p_vaddr;
944 		/*
945 		 * If we are loading ET_EXEC or we have already performed
946 		 * the ET_DYN load_addr calculations, proceed normally.
947 		 */
948 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
949 			elf_flags |= elf_fixed;
950 		} else if (loc->elf_ex.e_type == ET_DYN) {
951 			/*
952 			 * This logic is run once for the first LOAD Program
953 			 * Header for ET_DYN binaries to calculate the
954 			 * randomization (load_bias) for all the LOAD
955 			 * Program Headers, and to calculate the entire
956 			 * size of the ELF mapping (total_size). (Note that
957 			 * load_addr_set is set to true later once the
958 			 * initial mapping is performed.)
959 			 *
960 			 * There are effectively two types of ET_DYN
961 			 * binaries: programs (i.e. PIE: ET_DYN with INTERP)
962 			 * and loaders (ET_DYN without INTERP, since they
963 			 * _are_ the ELF interpreter). The loaders must
964 			 * be loaded away from programs since the program
965 			 * may otherwise collide with the loader (especially
966 			 * for ET_EXEC which does not have a randomized
967 			 * position). For example to handle invocations of
968 			 * "./ld.so someprog" to test out a new version of
969 			 * the loader, the subsequent program that the
970 			 * loader loads must avoid the loader itself, so
971 			 * they cannot share the same load range. Sufficient
972 			 * room for the brk must be allocated with the
973 			 * loader as well, since brk must be available with
974 			 * the loader.
975 			 *
976 			 * Therefore, programs are loaded offset from
977 			 * ELF_ET_DYN_BASE and loaders are loaded into the
978 			 * independently randomized mmap region (0 load_bias
979 			 * without MAP_FIXED).
980 			 */
981 			if (elf_interpreter) {
982 				load_bias = ELF_ET_DYN_BASE;
983 				if (current->flags & PF_RANDOMIZE)
984 					load_bias += arch_mmap_rnd();
985 				elf_flags |= elf_fixed;
986 			} else
987 				load_bias = 0;
988 
989 			/*
990 			 * Since load_bias is used for all subsequent loading
991 			 * calculations, we must lower it by the first vaddr
992 			 * so that the remaining calculations based on the
993 			 * ELF vaddrs will be correctly offset. The result
994 			 * is then page aligned.
995 			 */
996 			load_bias = ELF_PAGESTART(load_bias - vaddr);
997 
998 			total_size = total_mapping_size(elf_phdata,
999 							loc->elf_ex.e_phnum);
1000 			if (!total_size) {
1001 				retval = -EINVAL;
1002 				goto out_free_dentry;
1003 			}
1004 		}
1005 
1006 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
1007 				elf_prot, elf_flags, total_size);
1008 		if (BAD_ADDR(error)) {
1009 			retval = IS_ERR((void *)error) ?
1010 				PTR_ERR((void*)error) : -EINVAL;
1011 			goto out_free_dentry;
1012 		}
1013 
1014 		if (!load_addr_set) {
1015 			load_addr_set = 1;
1016 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
1017 			if (loc->elf_ex.e_type == ET_DYN) {
1018 				load_bias += error -
1019 				             ELF_PAGESTART(load_bias + vaddr);
1020 				load_addr += load_bias;
1021 				reloc_func_desc = load_bias;
1022 			}
1023 		}
1024 		k = elf_ppnt->p_vaddr;
1025 		if (k < start_code)
1026 			start_code = k;
1027 		if (start_data < k)
1028 			start_data = k;
1029 
1030 		/*
1031 		 * Check to see if the section's size will overflow the
1032 		 * allowed task size. Note that p_filesz must always be
1033 		 * <= p_memsz so it is only necessary to check p_memsz.
1034 		 */
1035 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
1036 		    elf_ppnt->p_memsz > TASK_SIZE ||
1037 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
1038 			/* set_brk can never work. Avoid overflows. */
1039 			retval = -EINVAL;
1040 			goto out_free_dentry;
1041 		}
1042 
1043 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
1044 
1045 		if (k > elf_bss)
1046 			elf_bss = k;
1047 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
1048 			end_code = k;
1049 		if (end_data < k)
1050 			end_data = k;
1051 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
1052 		if (k > elf_brk) {
1053 			bss_prot = elf_prot;
1054 			elf_brk = k;
1055 		}
1056 	}
1057 
1058 	loc->elf_ex.e_entry += load_bias;
1059 	elf_bss += load_bias;
1060 	elf_brk += load_bias;
1061 	start_code += load_bias;
1062 	end_code += load_bias;
1063 	start_data += load_bias;
1064 	end_data += load_bias;
1065 
1066 	/* Calling set_brk effectively mmaps the pages that we need
1067 	 * for the bss and break sections.  We must do this before
1068 	 * mapping in the interpreter, to make sure it doesn't wind
1069 	 * up getting placed where the bss needs to go.
1070 	 */
1071 	retval = set_brk(elf_bss, elf_brk, bss_prot);
1072 	if (retval)
1073 		goto out_free_dentry;
1074 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
1075 		retval = -EFAULT; /* Nobody gets to see this, but.. */
1076 		goto out_free_dentry;
1077 	}
1078 
1079 	if (elf_interpreter) {
1080 		unsigned long interp_map_addr = 0;
1081 
1082 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
1083 					    interpreter,
1084 					    &interp_map_addr,
1085 					    load_bias, interp_elf_phdata);
1086 		if (!IS_ERR((void *)elf_entry)) {
1087 			/*
1088 			 * load_elf_interp() returns relocation
1089 			 * adjustment
1090 			 */
1091 			interp_load_addr = elf_entry;
1092 			elf_entry += loc->interp_elf_ex.e_entry;
1093 		}
1094 		if (BAD_ADDR(elf_entry)) {
1095 			retval = IS_ERR((void *)elf_entry) ?
1096 					(int)elf_entry : -EINVAL;
1097 			goto out_free_dentry;
1098 		}
1099 		reloc_func_desc = interp_load_addr;
1100 
1101 		allow_write_access(interpreter);
1102 		fput(interpreter);
1103 		kfree(elf_interpreter);
1104 	} else {
1105 		elf_entry = loc->elf_ex.e_entry;
1106 		if (BAD_ADDR(elf_entry)) {
1107 			retval = -EINVAL;
1108 			goto out_free_dentry;
1109 		}
1110 	}
1111 
1112 	kfree(interp_elf_phdata);
1113 	kfree(elf_phdata);
1114 
1115 	set_binfmt(&elf_format);
1116 
1117 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1118 	retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
1119 	if (retval < 0)
1120 		goto out;
1121 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1122 
1123 	retval = create_elf_tables(bprm, &loc->elf_ex,
1124 			  load_addr, interp_load_addr);
1125 	if (retval < 0)
1126 		goto out;
1127 	/* N.B. passed_fileno might not be initialized? */
1128 	current->mm->end_code = end_code;
1129 	current->mm->start_code = start_code;
1130 	current->mm->start_data = start_data;
1131 	current->mm->end_data = end_data;
1132 	current->mm->start_stack = bprm->p;
1133 
1134 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1135 		current->mm->brk = current->mm->start_brk =
1136 			arch_randomize_brk(current->mm);
1137 #ifdef compat_brk_randomized
1138 		current->brk_randomized = 1;
1139 #endif
1140 	}
1141 
1142 	if (current->personality & MMAP_PAGE_ZERO) {
1143 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
1144 		   and some applications "depend" upon this behavior.
1145 		   Since we do not have the power to recompile these, we
1146 		   emulate the SVr4 behavior. Sigh. */
1147 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1148 				MAP_FIXED | MAP_PRIVATE, 0);
1149 	}
1150 
1151 #ifdef ELF_PLAT_INIT
1152 	/*
1153 	 * The ABI may specify that certain registers be set up in special
1154 	 * ways (on i386 %edx is the address of a DT_FINI function, for
1155 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
1156 	 * that the e_entry field is the address of the function descriptor
1157 	 * for the startup routine, rather than the address of the startup
1158 	 * routine itself.  This macro performs whatever initialization to
1159 	 * the regs structure is required as well as any relocations to the
1160 	 * function descriptor entries when executing dynamically links apps.
1161 	 */
1162 	ELF_PLAT_INIT(regs, reloc_func_desc);
1163 #endif
1164 
1165 	finalize_exec(bprm);
1166 	start_thread(regs, elf_entry, bprm->p);
1167 	retval = 0;
1168 out:
1169 	kfree(loc);
1170 out_ret:
1171 	return retval;
1172 
1173 	/* error cleanup */
1174 out_free_dentry:
1175 	kfree(interp_elf_phdata);
1176 	allow_write_access(interpreter);
1177 	if (interpreter)
1178 		fput(interpreter);
1179 out_free_interp:
1180 	kfree(elf_interpreter);
1181 out_free_ph:
1182 	kfree(elf_phdata);
1183 	goto out;
1184 }
1185 
1186 #ifdef CONFIG_USELIB
1187 /* This is really simpleminded and specialized - we are loading an
1188    a.out library that is given an ELF header. */
1189 static int load_elf_library(struct file *file)
1190 {
1191 	struct elf_phdr *elf_phdata;
1192 	struct elf_phdr *eppnt;
1193 	unsigned long elf_bss, bss, len;
1194 	int retval, error, i, j;
1195 	struct elfhdr elf_ex;
1196 	loff_t pos = 0;
1197 
1198 	error = -ENOEXEC;
1199 	retval = kernel_read(file, &elf_ex, sizeof(elf_ex), &pos);
1200 	if (retval != sizeof(elf_ex))
1201 		goto out;
1202 
1203 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1204 		goto out;
1205 
1206 	/* First of all, some simple consistency checks */
1207 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1208 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1209 		goto out;
1210 	if (elf_check_fdpic(&elf_ex))
1211 		goto out;
1212 
1213 	/* Now read in all of the header information */
1214 
1215 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1216 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1217 
1218 	error = -ENOMEM;
1219 	elf_phdata = kmalloc(j, GFP_KERNEL);
1220 	if (!elf_phdata)
1221 		goto out;
1222 
1223 	eppnt = elf_phdata;
1224 	error = -ENOEXEC;
1225 	pos =  elf_ex.e_phoff;
1226 	retval = kernel_read(file, eppnt, j, &pos);
1227 	if (retval != j)
1228 		goto out_free_ph;
1229 
1230 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1231 		if ((eppnt + i)->p_type == PT_LOAD)
1232 			j++;
1233 	if (j != 1)
1234 		goto out_free_ph;
1235 
1236 	while (eppnt->p_type != PT_LOAD)
1237 		eppnt++;
1238 
1239 	/* Now use mmap to map the library into memory. */
1240 	error = vm_mmap(file,
1241 			ELF_PAGESTART(eppnt->p_vaddr),
1242 			(eppnt->p_filesz +
1243 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1244 			PROT_READ | PROT_WRITE | PROT_EXEC,
1245 			MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_DENYWRITE,
1246 			(eppnt->p_offset -
1247 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1248 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1249 		goto out_free_ph;
1250 
1251 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1252 	if (padzero(elf_bss)) {
1253 		error = -EFAULT;
1254 		goto out_free_ph;
1255 	}
1256 
1257 	len = ELF_PAGEALIGN(eppnt->p_filesz + eppnt->p_vaddr);
1258 	bss = ELF_PAGEALIGN(eppnt->p_memsz + eppnt->p_vaddr);
1259 	if (bss > len) {
1260 		error = vm_brk(len, bss - len);
1261 		if (error)
1262 			goto out_free_ph;
1263 	}
1264 	error = 0;
1265 
1266 out_free_ph:
1267 	kfree(elf_phdata);
1268 out:
1269 	return error;
1270 }
1271 #endif /* #ifdef CONFIG_USELIB */
1272 
1273 #ifdef CONFIG_ELF_CORE
1274 /*
1275  * ELF core dumper
1276  *
1277  * Modelled on fs/exec.c:aout_core_dump()
1278  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1279  */
1280 
1281 /*
1282  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1283  * that are useful for post-mortem analysis are included in every core dump.
1284  * In that way we ensure that the core dump is fully interpretable later
1285  * without matching up the same kernel and hardware config to see what PC values
1286  * meant. These special mappings include - vDSO, vsyscall, and other
1287  * architecture specific mappings
1288  */
1289 static bool always_dump_vma(struct vm_area_struct *vma)
1290 {
1291 	/* Any vsyscall mappings? */
1292 	if (vma == get_gate_vma(vma->vm_mm))
1293 		return true;
1294 
1295 	/*
1296 	 * Assume that all vmas with a .name op should always be dumped.
1297 	 * If this changes, a new vm_ops field can easily be added.
1298 	 */
1299 	if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1300 		return true;
1301 
1302 	/*
1303 	 * arch_vma_name() returns non-NULL for special architecture mappings,
1304 	 * such as vDSO sections.
1305 	 */
1306 	if (arch_vma_name(vma))
1307 		return true;
1308 
1309 	return false;
1310 }
1311 
1312 /*
1313  * Decide what to dump of a segment, part, all or none.
1314  */
1315 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1316 				   unsigned long mm_flags)
1317 {
1318 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1319 
1320 	/* always dump the vdso and vsyscall sections */
1321 	if (always_dump_vma(vma))
1322 		goto whole;
1323 
1324 	if (vma->vm_flags & VM_DONTDUMP)
1325 		return 0;
1326 
1327 	/* support for DAX */
1328 	if (vma_is_dax(vma)) {
1329 		if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1330 			goto whole;
1331 		if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1332 			goto whole;
1333 		return 0;
1334 	}
1335 
1336 	/* Hugetlb memory check */
1337 	if (vma->vm_flags & VM_HUGETLB) {
1338 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1339 			goto whole;
1340 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1341 			goto whole;
1342 		return 0;
1343 	}
1344 
1345 	/* Do not dump I/O mapped devices or special mappings */
1346 	if (vma->vm_flags & VM_IO)
1347 		return 0;
1348 
1349 	/* By default, dump shared memory if mapped from an anonymous file. */
1350 	if (vma->vm_flags & VM_SHARED) {
1351 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1352 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1353 			goto whole;
1354 		return 0;
1355 	}
1356 
1357 	/* Dump segments that have been written to.  */
1358 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1359 		goto whole;
1360 	if (vma->vm_file == NULL)
1361 		return 0;
1362 
1363 	if (FILTER(MAPPED_PRIVATE))
1364 		goto whole;
1365 
1366 	/*
1367 	 * If this looks like the beginning of a DSO or executable mapping,
1368 	 * check for an ELF header.  If we find one, dump the first page to
1369 	 * aid in determining what was mapped here.
1370 	 */
1371 	if (FILTER(ELF_HEADERS) &&
1372 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1373 		u32 __user *header = (u32 __user *) vma->vm_start;
1374 		u32 word;
1375 		mm_segment_t fs = get_fs();
1376 		/*
1377 		 * Doing it this way gets the constant folded by GCC.
1378 		 */
1379 		union {
1380 			u32 cmp;
1381 			char elfmag[SELFMAG];
1382 		} magic;
1383 		BUILD_BUG_ON(SELFMAG != sizeof word);
1384 		magic.elfmag[EI_MAG0] = ELFMAG0;
1385 		magic.elfmag[EI_MAG1] = ELFMAG1;
1386 		magic.elfmag[EI_MAG2] = ELFMAG2;
1387 		magic.elfmag[EI_MAG3] = ELFMAG3;
1388 		/*
1389 		 * Switch to the user "segment" for get_user(),
1390 		 * then put back what elf_core_dump() had in place.
1391 		 */
1392 		set_fs(USER_DS);
1393 		if (unlikely(get_user(word, header)))
1394 			word = 0;
1395 		set_fs(fs);
1396 		if (word == magic.cmp)
1397 			return PAGE_SIZE;
1398 	}
1399 
1400 #undef	FILTER
1401 
1402 	return 0;
1403 
1404 whole:
1405 	return vma->vm_end - vma->vm_start;
1406 }
1407 
1408 /* An ELF note in memory */
1409 struct memelfnote
1410 {
1411 	const char *name;
1412 	int type;
1413 	unsigned int datasz;
1414 	void *data;
1415 };
1416 
1417 static int notesize(struct memelfnote *en)
1418 {
1419 	int sz;
1420 
1421 	sz = sizeof(struct elf_note);
1422 	sz += roundup(strlen(en->name) + 1, 4);
1423 	sz += roundup(en->datasz, 4);
1424 
1425 	return sz;
1426 }
1427 
1428 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1429 {
1430 	struct elf_note en;
1431 	en.n_namesz = strlen(men->name) + 1;
1432 	en.n_descsz = men->datasz;
1433 	en.n_type = men->type;
1434 
1435 	return dump_emit(cprm, &en, sizeof(en)) &&
1436 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1437 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1438 }
1439 
1440 static void fill_elf_header(struct elfhdr *elf, int segs,
1441 			    u16 machine, u32 flags)
1442 {
1443 	memset(elf, 0, sizeof(*elf));
1444 
1445 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1446 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1447 	elf->e_ident[EI_DATA] = ELF_DATA;
1448 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1449 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1450 
1451 	elf->e_type = ET_CORE;
1452 	elf->e_machine = machine;
1453 	elf->e_version = EV_CURRENT;
1454 	elf->e_phoff = sizeof(struct elfhdr);
1455 	elf->e_flags = flags;
1456 	elf->e_ehsize = sizeof(struct elfhdr);
1457 	elf->e_phentsize = sizeof(struct elf_phdr);
1458 	elf->e_phnum = segs;
1459 
1460 	return;
1461 }
1462 
1463 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1464 {
1465 	phdr->p_type = PT_NOTE;
1466 	phdr->p_offset = offset;
1467 	phdr->p_vaddr = 0;
1468 	phdr->p_paddr = 0;
1469 	phdr->p_filesz = sz;
1470 	phdr->p_memsz = 0;
1471 	phdr->p_flags = 0;
1472 	phdr->p_align = 0;
1473 	return;
1474 }
1475 
1476 static void fill_note(struct memelfnote *note, const char *name, int type,
1477 		unsigned int sz, void *data)
1478 {
1479 	note->name = name;
1480 	note->type = type;
1481 	note->datasz = sz;
1482 	note->data = data;
1483 	return;
1484 }
1485 
1486 /*
1487  * fill up all the fields in prstatus from the given task struct, except
1488  * registers which need to be filled up separately.
1489  */
1490 static void fill_prstatus(struct elf_prstatus *prstatus,
1491 		struct task_struct *p, long signr)
1492 {
1493 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1494 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1495 	prstatus->pr_sighold = p->blocked.sig[0];
1496 	rcu_read_lock();
1497 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1498 	rcu_read_unlock();
1499 	prstatus->pr_pid = task_pid_vnr(p);
1500 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1501 	prstatus->pr_sid = task_session_vnr(p);
1502 	if (thread_group_leader(p)) {
1503 		struct task_cputime cputime;
1504 
1505 		/*
1506 		 * This is the record for the group leader.  It shows the
1507 		 * group-wide total, not its individual thread total.
1508 		 */
1509 		thread_group_cputime(p, &cputime);
1510 		prstatus->pr_utime = ns_to_timeval(cputime.utime);
1511 		prstatus->pr_stime = ns_to_timeval(cputime.stime);
1512 	} else {
1513 		u64 utime, stime;
1514 
1515 		task_cputime(p, &utime, &stime);
1516 		prstatus->pr_utime = ns_to_timeval(utime);
1517 		prstatus->pr_stime = ns_to_timeval(stime);
1518 	}
1519 
1520 	prstatus->pr_cutime = ns_to_timeval(p->signal->cutime);
1521 	prstatus->pr_cstime = ns_to_timeval(p->signal->cstime);
1522 }
1523 
1524 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1525 		       struct mm_struct *mm)
1526 {
1527 	const struct cred *cred;
1528 	unsigned int i, len;
1529 
1530 	/* first copy the parameters from user space */
1531 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1532 
1533 	len = mm->arg_end - mm->arg_start;
1534 	if (len >= ELF_PRARGSZ)
1535 		len = ELF_PRARGSZ-1;
1536 	if (copy_from_user(&psinfo->pr_psargs,
1537 		           (const char __user *)mm->arg_start, len))
1538 		return -EFAULT;
1539 	for(i = 0; i < len; i++)
1540 		if (psinfo->pr_psargs[i] == 0)
1541 			psinfo->pr_psargs[i] = ' ';
1542 	psinfo->pr_psargs[len] = 0;
1543 
1544 	rcu_read_lock();
1545 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1546 	rcu_read_unlock();
1547 	psinfo->pr_pid = task_pid_vnr(p);
1548 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1549 	psinfo->pr_sid = task_session_vnr(p);
1550 
1551 	i = p->state ? ffz(~p->state) + 1 : 0;
1552 	psinfo->pr_state = i;
1553 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1554 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1555 	psinfo->pr_nice = task_nice(p);
1556 	psinfo->pr_flag = p->flags;
1557 	rcu_read_lock();
1558 	cred = __task_cred(p);
1559 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1560 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1561 	rcu_read_unlock();
1562 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1563 
1564 	return 0;
1565 }
1566 
1567 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1568 {
1569 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1570 	int i = 0;
1571 	do
1572 		i += 2;
1573 	while (auxv[i - 2] != AT_NULL);
1574 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1575 }
1576 
1577 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1578 		const kernel_siginfo_t *siginfo)
1579 {
1580 	mm_segment_t old_fs = get_fs();
1581 	set_fs(KERNEL_DS);
1582 	copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1583 	set_fs(old_fs);
1584 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1585 }
1586 
1587 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1588 /*
1589  * Format of NT_FILE note:
1590  *
1591  * long count     -- how many files are mapped
1592  * long page_size -- units for file_ofs
1593  * array of [COUNT] elements of
1594  *   long start
1595  *   long end
1596  *   long file_ofs
1597  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1598  */
1599 static int fill_files_note(struct memelfnote *note)
1600 {
1601 	struct vm_area_struct *vma;
1602 	unsigned count, size, names_ofs, remaining, n;
1603 	user_long_t *data;
1604 	user_long_t *start_end_ofs;
1605 	char *name_base, *name_curpos;
1606 
1607 	/* *Estimated* file count and total data size needed */
1608 	count = current->mm->map_count;
1609 	if (count > UINT_MAX / 64)
1610 		return -EINVAL;
1611 	size = count * 64;
1612 
1613 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1614  alloc:
1615 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1616 		return -EINVAL;
1617 	size = round_up(size, PAGE_SIZE);
1618 	data = kvmalloc(size, GFP_KERNEL);
1619 	if (ZERO_OR_NULL_PTR(data))
1620 		return -ENOMEM;
1621 
1622 	start_end_ofs = data + 2;
1623 	name_base = name_curpos = ((char *)data) + names_ofs;
1624 	remaining = size - names_ofs;
1625 	count = 0;
1626 	for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1627 		struct file *file;
1628 		const char *filename;
1629 
1630 		file = vma->vm_file;
1631 		if (!file)
1632 			continue;
1633 		filename = file_path(file, name_curpos, remaining);
1634 		if (IS_ERR(filename)) {
1635 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1636 				kvfree(data);
1637 				size = size * 5 / 4;
1638 				goto alloc;
1639 			}
1640 			continue;
1641 		}
1642 
1643 		/* file_path() fills at the end, move name down */
1644 		/* n = strlen(filename) + 1: */
1645 		n = (name_curpos + remaining) - filename;
1646 		remaining = filename - name_curpos;
1647 		memmove(name_curpos, filename, n);
1648 		name_curpos += n;
1649 
1650 		*start_end_ofs++ = vma->vm_start;
1651 		*start_end_ofs++ = vma->vm_end;
1652 		*start_end_ofs++ = vma->vm_pgoff;
1653 		count++;
1654 	}
1655 
1656 	/* Now we know exact count of files, can store it */
1657 	data[0] = count;
1658 	data[1] = PAGE_SIZE;
1659 	/*
1660 	 * Count usually is less than current->mm->map_count,
1661 	 * we need to move filenames down.
1662 	 */
1663 	n = current->mm->map_count - count;
1664 	if (n != 0) {
1665 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1666 		memmove(name_base - shift_bytes, name_base,
1667 			name_curpos - name_base);
1668 		name_curpos -= shift_bytes;
1669 	}
1670 
1671 	size = name_curpos - (char *)data;
1672 	fill_note(note, "CORE", NT_FILE, size, data);
1673 	return 0;
1674 }
1675 
1676 #ifdef CORE_DUMP_USE_REGSET
1677 #include <linux/regset.h>
1678 
1679 struct elf_thread_core_info {
1680 	struct elf_thread_core_info *next;
1681 	struct task_struct *task;
1682 	struct elf_prstatus prstatus;
1683 	struct memelfnote notes[0];
1684 };
1685 
1686 struct elf_note_info {
1687 	struct elf_thread_core_info *thread;
1688 	struct memelfnote psinfo;
1689 	struct memelfnote signote;
1690 	struct memelfnote auxv;
1691 	struct memelfnote files;
1692 	user_siginfo_t csigdata;
1693 	size_t size;
1694 	int thread_notes;
1695 };
1696 
1697 /*
1698  * When a regset has a writeback hook, we call it on each thread before
1699  * dumping user memory.  On register window machines, this makes sure the
1700  * user memory backing the register data is up to date before we read it.
1701  */
1702 static void do_thread_regset_writeback(struct task_struct *task,
1703 				       const struct user_regset *regset)
1704 {
1705 	if (regset->writeback)
1706 		regset->writeback(task, regset, 1);
1707 }
1708 
1709 #ifndef PRSTATUS_SIZE
1710 #define PRSTATUS_SIZE(S, R) sizeof(S)
1711 #endif
1712 
1713 #ifndef SET_PR_FPVALID
1714 #define SET_PR_FPVALID(S, V, R) ((S)->pr_fpvalid = (V))
1715 #endif
1716 
1717 static int fill_thread_core_info(struct elf_thread_core_info *t,
1718 				 const struct user_regset_view *view,
1719 				 long signr, size_t *total)
1720 {
1721 	unsigned int i;
1722 	unsigned int regset0_size = regset_size(t->task, &view->regsets[0]);
1723 
1724 	/*
1725 	 * NT_PRSTATUS is the one special case, because the regset data
1726 	 * goes into the pr_reg field inside the note contents, rather
1727 	 * than being the whole note contents.  We fill the reset in here.
1728 	 * We assume that regset 0 is NT_PRSTATUS.
1729 	 */
1730 	fill_prstatus(&t->prstatus, t->task, signr);
1731 	(void) view->regsets[0].get(t->task, &view->regsets[0], 0, regset0_size,
1732 				    &t->prstatus.pr_reg, NULL);
1733 
1734 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1735 		  PRSTATUS_SIZE(t->prstatus, regset0_size), &t->prstatus);
1736 	*total += notesize(&t->notes[0]);
1737 
1738 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1739 
1740 	/*
1741 	 * Each other regset might generate a note too.  For each regset
1742 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1743 	 * all zero and we'll know to skip writing it later.
1744 	 */
1745 	for (i = 1; i < view->n; ++i) {
1746 		const struct user_regset *regset = &view->regsets[i];
1747 		do_thread_regset_writeback(t->task, regset);
1748 		if (regset->core_note_type && regset->get &&
1749 		    (!regset->active || regset->active(t->task, regset) > 0)) {
1750 			int ret;
1751 			size_t size = regset_size(t->task, regset);
1752 			void *data = kmalloc(size, GFP_KERNEL);
1753 			if (unlikely(!data))
1754 				return 0;
1755 			ret = regset->get(t->task, regset,
1756 					  0, size, data, NULL);
1757 			if (unlikely(ret))
1758 				kfree(data);
1759 			else {
1760 				if (regset->core_note_type != NT_PRFPREG)
1761 					fill_note(&t->notes[i], "LINUX",
1762 						  regset->core_note_type,
1763 						  size, data);
1764 				else {
1765 					SET_PR_FPVALID(&t->prstatus,
1766 							1, regset0_size);
1767 					fill_note(&t->notes[i], "CORE",
1768 						  NT_PRFPREG, size, data);
1769 				}
1770 				*total += notesize(&t->notes[i]);
1771 			}
1772 		}
1773 	}
1774 
1775 	return 1;
1776 }
1777 
1778 static int fill_note_info(struct elfhdr *elf, int phdrs,
1779 			  struct elf_note_info *info,
1780 			  const kernel_siginfo_t *siginfo, struct pt_regs *regs)
1781 {
1782 	struct task_struct *dump_task = current;
1783 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1784 	struct elf_thread_core_info *t;
1785 	struct elf_prpsinfo *psinfo;
1786 	struct core_thread *ct;
1787 	unsigned int i;
1788 
1789 	info->size = 0;
1790 	info->thread = NULL;
1791 
1792 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1793 	if (psinfo == NULL) {
1794 		info->psinfo.data = NULL; /* So we don't free this wrongly */
1795 		return 0;
1796 	}
1797 
1798 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1799 
1800 	/*
1801 	 * Figure out how many notes we're going to need for each thread.
1802 	 */
1803 	info->thread_notes = 0;
1804 	for (i = 0; i < view->n; ++i)
1805 		if (view->regsets[i].core_note_type != 0)
1806 			++info->thread_notes;
1807 
1808 	/*
1809 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1810 	 * since it is our one special case.
1811 	 */
1812 	if (unlikely(info->thread_notes == 0) ||
1813 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1814 		WARN_ON(1);
1815 		return 0;
1816 	}
1817 
1818 	/*
1819 	 * Initialize the ELF file header.
1820 	 */
1821 	fill_elf_header(elf, phdrs,
1822 			view->e_machine, view->e_flags);
1823 
1824 	/*
1825 	 * Allocate a structure for each thread.
1826 	 */
1827 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1828 		t = kzalloc(offsetof(struct elf_thread_core_info,
1829 				     notes[info->thread_notes]),
1830 			    GFP_KERNEL);
1831 		if (unlikely(!t))
1832 			return 0;
1833 
1834 		t->task = ct->task;
1835 		if (ct->task == dump_task || !info->thread) {
1836 			t->next = info->thread;
1837 			info->thread = t;
1838 		} else {
1839 			/*
1840 			 * Make sure to keep the original task at
1841 			 * the head of the list.
1842 			 */
1843 			t->next = info->thread->next;
1844 			info->thread->next = t;
1845 		}
1846 	}
1847 
1848 	/*
1849 	 * Now fill in each thread's information.
1850 	 */
1851 	for (t = info->thread; t != NULL; t = t->next)
1852 		if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1853 			return 0;
1854 
1855 	/*
1856 	 * Fill in the two process-wide notes.
1857 	 */
1858 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1859 	info->size += notesize(&info->psinfo);
1860 
1861 	fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1862 	info->size += notesize(&info->signote);
1863 
1864 	fill_auxv_note(&info->auxv, current->mm);
1865 	info->size += notesize(&info->auxv);
1866 
1867 	if (fill_files_note(&info->files) == 0)
1868 		info->size += notesize(&info->files);
1869 
1870 	return 1;
1871 }
1872 
1873 static size_t get_note_info_size(struct elf_note_info *info)
1874 {
1875 	return info->size;
1876 }
1877 
1878 /*
1879  * Write all the notes for each thread.  When writing the first thread, the
1880  * process-wide notes are interleaved after the first thread-specific note.
1881  */
1882 static int write_note_info(struct elf_note_info *info,
1883 			   struct coredump_params *cprm)
1884 {
1885 	bool first = true;
1886 	struct elf_thread_core_info *t = info->thread;
1887 
1888 	do {
1889 		int i;
1890 
1891 		if (!writenote(&t->notes[0], cprm))
1892 			return 0;
1893 
1894 		if (first && !writenote(&info->psinfo, cprm))
1895 			return 0;
1896 		if (first && !writenote(&info->signote, cprm))
1897 			return 0;
1898 		if (first && !writenote(&info->auxv, cprm))
1899 			return 0;
1900 		if (first && info->files.data &&
1901 				!writenote(&info->files, cprm))
1902 			return 0;
1903 
1904 		for (i = 1; i < info->thread_notes; ++i)
1905 			if (t->notes[i].data &&
1906 			    !writenote(&t->notes[i], cprm))
1907 				return 0;
1908 
1909 		first = false;
1910 		t = t->next;
1911 	} while (t);
1912 
1913 	return 1;
1914 }
1915 
1916 static void free_note_info(struct elf_note_info *info)
1917 {
1918 	struct elf_thread_core_info *threads = info->thread;
1919 	while (threads) {
1920 		unsigned int i;
1921 		struct elf_thread_core_info *t = threads;
1922 		threads = t->next;
1923 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1924 		for (i = 1; i < info->thread_notes; ++i)
1925 			kfree(t->notes[i].data);
1926 		kfree(t);
1927 	}
1928 	kfree(info->psinfo.data);
1929 	kvfree(info->files.data);
1930 }
1931 
1932 #else
1933 
1934 /* Here is the structure in which status of each thread is captured. */
1935 struct elf_thread_status
1936 {
1937 	struct list_head list;
1938 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1939 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1940 	struct task_struct *thread;
1941 #ifdef ELF_CORE_COPY_XFPREGS
1942 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1943 #endif
1944 	struct memelfnote notes[3];
1945 	int num_notes;
1946 };
1947 
1948 /*
1949  * In order to add the specific thread information for the elf file format,
1950  * we need to keep a linked list of every threads pr_status and then create
1951  * a single section for them in the final core file.
1952  */
1953 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1954 {
1955 	int sz = 0;
1956 	struct task_struct *p = t->thread;
1957 	t->num_notes = 0;
1958 
1959 	fill_prstatus(&t->prstatus, p, signr);
1960 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1961 
1962 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1963 		  &(t->prstatus));
1964 	t->num_notes++;
1965 	sz += notesize(&t->notes[0]);
1966 
1967 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1968 								&t->fpu))) {
1969 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1970 			  &(t->fpu));
1971 		t->num_notes++;
1972 		sz += notesize(&t->notes[1]);
1973 	}
1974 
1975 #ifdef ELF_CORE_COPY_XFPREGS
1976 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1977 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1978 			  sizeof(t->xfpu), &t->xfpu);
1979 		t->num_notes++;
1980 		sz += notesize(&t->notes[2]);
1981 	}
1982 #endif
1983 	return sz;
1984 }
1985 
1986 struct elf_note_info {
1987 	struct memelfnote *notes;
1988 	struct memelfnote *notes_files;
1989 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1990 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1991 	struct list_head thread_list;
1992 	elf_fpregset_t *fpu;
1993 #ifdef ELF_CORE_COPY_XFPREGS
1994 	elf_fpxregset_t *xfpu;
1995 #endif
1996 	user_siginfo_t csigdata;
1997 	int thread_status_size;
1998 	int numnote;
1999 };
2000 
2001 static int elf_note_info_init(struct elf_note_info *info)
2002 {
2003 	memset(info, 0, sizeof(*info));
2004 	INIT_LIST_HEAD(&info->thread_list);
2005 
2006 	/* Allocate space for ELF notes */
2007 	info->notes = kmalloc_array(8, sizeof(struct memelfnote), GFP_KERNEL);
2008 	if (!info->notes)
2009 		return 0;
2010 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
2011 	if (!info->psinfo)
2012 		return 0;
2013 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
2014 	if (!info->prstatus)
2015 		return 0;
2016 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
2017 	if (!info->fpu)
2018 		return 0;
2019 #ifdef ELF_CORE_COPY_XFPREGS
2020 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
2021 	if (!info->xfpu)
2022 		return 0;
2023 #endif
2024 	return 1;
2025 }
2026 
2027 static int fill_note_info(struct elfhdr *elf, int phdrs,
2028 			  struct elf_note_info *info,
2029 			  const kernel_siginfo_t *siginfo, struct pt_regs *regs)
2030 {
2031 	struct core_thread *ct;
2032 	struct elf_thread_status *ets;
2033 
2034 	if (!elf_note_info_init(info))
2035 		return 0;
2036 
2037 	for (ct = current->mm->core_state->dumper.next;
2038 					ct; ct = ct->next) {
2039 		ets = kzalloc(sizeof(*ets), GFP_KERNEL);
2040 		if (!ets)
2041 			return 0;
2042 
2043 		ets->thread = ct->task;
2044 		list_add(&ets->list, &info->thread_list);
2045 	}
2046 
2047 	list_for_each_entry(ets, &info->thread_list, list) {
2048 		int sz;
2049 
2050 		sz = elf_dump_thread_status(siginfo->si_signo, ets);
2051 		info->thread_status_size += sz;
2052 	}
2053 	/* now collect the dump for the current */
2054 	memset(info->prstatus, 0, sizeof(*info->prstatus));
2055 	fill_prstatus(info->prstatus, current, siginfo->si_signo);
2056 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
2057 
2058 	/* Set up header */
2059 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
2060 
2061 	/*
2062 	 * Set up the notes in similar form to SVR4 core dumps made
2063 	 * with info from their /proc.
2064 	 */
2065 
2066 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
2067 		  sizeof(*info->prstatus), info->prstatus);
2068 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
2069 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
2070 		  sizeof(*info->psinfo), info->psinfo);
2071 
2072 	fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
2073 	fill_auxv_note(info->notes + 3, current->mm);
2074 	info->numnote = 4;
2075 
2076 	if (fill_files_note(info->notes + info->numnote) == 0) {
2077 		info->notes_files = info->notes + info->numnote;
2078 		info->numnote++;
2079 	}
2080 
2081 	/* Try to dump the FPU. */
2082 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2083 							       info->fpu);
2084 	if (info->prstatus->pr_fpvalid)
2085 		fill_note(info->notes + info->numnote++,
2086 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2087 #ifdef ELF_CORE_COPY_XFPREGS
2088 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
2089 		fill_note(info->notes + info->numnote++,
2090 			  "LINUX", ELF_CORE_XFPREG_TYPE,
2091 			  sizeof(*info->xfpu), info->xfpu);
2092 #endif
2093 
2094 	return 1;
2095 }
2096 
2097 static size_t get_note_info_size(struct elf_note_info *info)
2098 {
2099 	int sz = 0;
2100 	int i;
2101 
2102 	for (i = 0; i < info->numnote; i++)
2103 		sz += notesize(info->notes + i);
2104 
2105 	sz += info->thread_status_size;
2106 
2107 	return sz;
2108 }
2109 
2110 static int write_note_info(struct elf_note_info *info,
2111 			   struct coredump_params *cprm)
2112 {
2113 	struct elf_thread_status *ets;
2114 	int i;
2115 
2116 	for (i = 0; i < info->numnote; i++)
2117 		if (!writenote(info->notes + i, cprm))
2118 			return 0;
2119 
2120 	/* write out the thread status notes section */
2121 	list_for_each_entry(ets, &info->thread_list, list) {
2122 		for (i = 0; i < ets->num_notes; i++)
2123 			if (!writenote(&ets->notes[i], cprm))
2124 				return 0;
2125 	}
2126 
2127 	return 1;
2128 }
2129 
2130 static void free_note_info(struct elf_note_info *info)
2131 {
2132 	while (!list_empty(&info->thread_list)) {
2133 		struct list_head *tmp = info->thread_list.next;
2134 		list_del(tmp);
2135 		kfree(list_entry(tmp, struct elf_thread_status, list));
2136 	}
2137 
2138 	/* Free data possibly allocated by fill_files_note(): */
2139 	if (info->notes_files)
2140 		kvfree(info->notes_files->data);
2141 
2142 	kfree(info->prstatus);
2143 	kfree(info->psinfo);
2144 	kfree(info->notes);
2145 	kfree(info->fpu);
2146 #ifdef ELF_CORE_COPY_XFPREGS
2147 	kfree(info->xfpu);
2148 #endif
2149 }
2150 
2151 #endif
2152 
2153 static struct vm_area_struct *first_vma(struct task_struct *tsk,
2154 					struct vm_area_struct *gate_vma)
2155 {
2156 	struct vm_area_struct *ret = tsk->mm->mmap;
2157 
2158 	if (ret)
2159 		return ret;
2160 	return gate_vma;
2161 }
2162 /*
2163  * Helper function for iterating across a vma list.  It ensures that the caller
2164  * will visit `gate_vma' prior to terminating the search.
2165  */
2166 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2167 					struct vm_area_struct *gate_vma)
2168 {
2169 	struct vm_area_struct *ret;
2170 
2171 	ret = this_vma->vm_next;
2172 	if (ret)
2173 		return ret;
2174 	if (this_vma == gate_vma)
2175 		return NULL;
2176 	return gate_vma;
2177 }
2178 
2179 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2180 			     elf_addr_t e_shoff, int segs)
2181 {
2182 	elf->e_shoff = e_shoff;
2183 	elf->e_shentsize = sizeof(*shdr4extnum);
2184 	elf->e_shnum = 1;
2185 	elf->e_shstrndx = SHN_UNDEF;
2186 
2187 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2188 
2189 	shdr4extnum->sh_type = SHT_NULL;
2190 	shdr4extnum->sh_size = elf->e_shnum;
2191 	shdr4extnum->sh_link = elf->e_shstrndx;
2192 	shdr4extnum->sh_info = segs;
2193 }
2194 
2195 /*
2196  * Actual dumper
2197  *
2198  * This is a two-pass process; first we find the offsets of the bits,
2199  * and then they are actually written out.  If we run out of core limit
2200  * we just truncate.
2201  */
2202 static int elf_core_dump(struct coredump_params *cprm)
2203 {
2204 	int has_dumped = 0;
2205 	mm_segment_t fs;
2206 	int segs, i;
2207 	size_t vma_data_size = 0;
2208 	struct vm_area_struct *vma, *gate_vma;
2209 	struct elfhdr *elf = NULL;
2210 	loff_t offset = 0, dataoff;
2211 	struct elf_note_info info = { };
2212 	struct elf_phdr *phdr4note = NULL;
2213 	struct elf_shdr *shdr4extnum = NULL;
2214 	Elf_Half e_phnum;
2215 	elf_addr_t e_shoff;
2216 	elf_addr_t *vma_filesz = NULL;
2217 
2218 	/*
2219 	 * We no longer stop all VM operations.
2220 	 *
2221 	 * This is because those proceses that could possibly change map_count
2222 	 * or the mmap / vma pages are now blocked in do_exit on current
2223 	 * finishing this core dump.
2224 	 *
2225 	 * Only ptrace can touch these memory addresses, but it doesn't change
2226 	 * the map_count or the pages allocated. So no possibility of crashing
2227 	 * exists while dumping the mm->vm_next areas to the core file.
2228 	 */
2229 
2230 	/* alloc memory for large data structures: too large to be on stack */
2231 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2232 	if (!elf)
2233 		goto out;
2234 	/*
2235 	 * The number of segs are recored into ELF header as 16bit value.
2236 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2237 	 */
2238 	segs = current->mm->map_count;
2239 	segs += elf_core_extra_phdrs();
2240 
2241 	gate_vma = get_gate_vma(current->mm);
2242 	if (gate_vma != NULL)
2243 		segs++;
2244 
2245 	/* for notes section */
2246 	segs++;
2247 
2248 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2249 	 * this, kernel supports extended numbering. Have a look at
2250 	 * include/linux/elf.h for further information. */
2251 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2252 
2253 	/*
2254 	 * Collect all the non-memory information about the process for the
2255 	 * notes.  This also sets up the file header.
2256 	 */
2257 	if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2258 		goto cleanup;
2259 
2260 	has_dumped = 1;
2261 
2262 	fs = get_fs();
2263 	set_fs(KERNEL_DS);
2264 
2265 	offset += sizeof(*elf);				/* Elf header */
2266 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2267 
2268 	/* Write notes phdr entry */
2269 	{
2270 		size_t sz = get_note_info_size(&info);
2271 
2272 		sz += elf_coredump_extra_notes_size();
2273 
2274 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2275 		if (!phdr4note)
2276 			goto end_coredump;
2277 
2278 		fill_elf_note_phdr(phdr4note, sz, offset);
2279 		offset += sz;
2280 	}
2281 
2282 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2283 
2284 	if (segs - 1 > ULONG_MAX / sizeof(*vma_filesz))
2285 		goto end_coredump;
2286 	vma_filesz = kvmalloc(array_size(sizeof(*vma_filesz), (segs - 1)),
2287 			      GFP_KERNEL);
2288 	if (ZERO_OR_NULL_PTR(vma_filesz))
2289 		goto end_coredump;
2290 
2291 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2292 			vma = next_vma(vma, gate_vma)) {
2293 		unsigned long dump_size;
2294 
2295 		dump_size = vma_dump_size(vma, cprm->mm_flags);
2296 		vma_filesz[i++] = dump_size;
2297 		vma_data_size += dump_size;
2298 	}
2299 
2300 	offset += vma_data_size;
2301 	offset += elf_core_extra_data_size();
2302 	e_shoff = offset;
2303 
2304 	if (e_phnum == PN_XNUM) {
2305 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2306 		if (!shdr4extnum)
2307 			goto end_coredump;
2308 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2309 	}
2310 
2311 	offset = dataoff;
2312 
2313 	if (!dump_emit(cprm, elf, sizeof(*elf)))
2314 		goto end_coredump;
2315 
2316 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2317 		goto end_coredump;
2318 
2319 	/* Write program headers for segments dump */
2320 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2321 			vma = next_vma(vma, gate_vma)) {
2322 		struct elf_phdr phdr;
2323 
2324 		phdr.p_type = PT_LOAD;
2325 		phdr.p_offset = offset;
2326 		phdr.p_vaddr = vma->vm_start;
2327 		phdr.p_paddr = 0;
2328 		phdr.p_filesz = vma_filesz[i++];
2329 		phdr.p_memsz = vma->vm_end - vma->vm_start;
2330 		offset += phdr.p_filesz;
2331 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2332 		if (vma->vm_flags & VM_WRITE)
2333 			phdr.p_flags |= PF_W;
2334 		if (vma->vm_flags & VM_EXEC)
2335 			phdr.p_flags |= PF_X;
2336 		phdr.p_align = ELF_EXEC_PAGESIZE;
2337 
2338 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2339 			goto end_coredump;
2340 	}
2341 
2342 	if (!elf_core_write_extra_phdrs(cprm, offset))
2343 		goto end_coredump;
2344 
2345  	/* write out the notes section */
2346 	if (!write_note_info(&info, cprm))
2347 		goto end_coredump;
2348 
2349 	if (elf_coredump_extra_notes_write(cprm))
2350 		goto end_coredump;
2351 
2352 	/* Align to page */
2353 	if (!dump_skip(cprm, dataoff - cprm->pos))
2354 		goto end_coredump;
2355 
2356 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2357 			vma = next_vma(vma, gate_vma)) {
2358 		unsigned long addr;
2359 		unsigned long end;
2360 
2361 		end = vma->vm_start + vma_filesz[i++];
2362 
2363 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2364 			struct page *page;
2365 			int stop;
2366 
2367 			page = get_dump_page(addr);
2368 			if (page) {
2369 				void *kaddr = kmap(page);
2370 				stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2371 				kunmap(page);
2372 				put_page(page);
2373 			} else
2374 				stop = !dump_skip(cprm, PAGE_SIZE);
2375 			if (stop)
2376 				goto end_coredump;
2377 		}
2378 	}
2379 	dump_truncate(cprm);
2380 
2381 	if (!elf_core_write_extra_data(cprm))
2382 		goto end_coredump;
2383 
2384 	if (e_phnum == PN_XNUM) {
2385 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2386 			goto end_coredump;
2387 	}
2388 
2389 end_coredump:
2390 	set_fs(fs);
2391 
2392 cleanup:
2393 	free_note_info(&info);
2394 	kfree(shdr4extnum);
2395 	kvfree(vma_filesz);
2396 	kfree(phdr4note);
2397 	kfree(elf);
2398 out:
2399 	return has_dumped;
2400 }
2401 
2402 #endif		/* CONFIG_ELF_CORE */
2403 
2404 static int __init init_elf_binfmt(void)
2405 {
2406 	register_binfmt(&elf_format);
2407 	return 0;
2408 }
2409 
2410 static void __exit exit_elf_binfmt(void)
2411 {
2412 	/* Remove the COFF and ELF loaders. */
2413 	unregister_binfmt(&elf_format);
2414 }
2415 
2416 core_initcall(init_elf_binfmt);
2417 module_exit(exit_elf_binfmt);
2418 MODULE_LICENSE("GPL");
2419