xref: /linux/fs/binfmt_elf.c (revision c1aac62f36c1e37ee81c9e09ee9ee733eef05dcb)
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/dax.h>
39 #include <linux/uaccess.h>
40 #include <asm/param.h>
41 #include <asm/page.h>
42 
43 #ifndef user_long_t
44 #define user_long_t long
45 #endif
46 #ifndef user_siginfo_t
47 #define user_siginfo_t siginfo_t
48 #endif
49 
50 static int load_elf_binary(struct linux_binprm *bprm);
51 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
52 				int, int, unsigned long);
53 
54 #ifdef CONFIG_USELIB
55 static int load_elf_library(struct file *);
56 #else
57 #define load_elf_library NULL
58 #endif
59 
60 /*
61  * If we don't support core dumping, then supply a NULL so we
62  * don't even try.
63  */
64 #ifdef CONFIG_ELF_CORE
65 static int elf_core_dump(struct coredump_params *cprm);
66 #else
67 #define elf_core_dump	NULL
68 #endif
69 
70 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
71 #define ELF_MIN_ALIGN	ELF_EXEC_PAGESIZE
72 #else
73 #define ELF_MIN_ALIGN	PAGE_SIZE
74 #endif
75 
76 #ifndef ELF_CORE_EFLAGS
77 #define ELF_CORE_EFLAGS	0
78 #endif
79 
80 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
81 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
82 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
83 
84 static struct linux_binfmt elf_format = {
85 	.module		= THIS_MODULE,
86 	.load_binary	= load_elf_binary,
87 	.load_shlib	= load_elf_library,
88 	.core_dump	= elf_core_dump,
89 	.min_coredump	= ELF_EXEC_PAGESIZE,
90 };
91 
92 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
93 
94 static int set_brk(unsigned long start, unsigned long end)
95 {
96 	start = ELF_PAGEALIGN(start);
97 	end = ELF_PAGEALIGN(end);
98 	if (end > start) {
99 		int error = vm_brk(start, end - start);
100 		if (error)
101 			return error;
102 	}
103 	current->mm->start_brk = current->mm->brk = end;
104 	return 0;
105 }
106 
107 /* We need to explicitly zero any fractional pages
108    after the data section (i.e. bss).  This would
109    contain the junk from the file that should not
110    be in memory
111  */
112 static int padzero(unsigned long elf_bss)
113 {
114 	unsigned long nbyte;
115 
116 	nbyte = ELF_PAGEOFFSET(elf_bss);
117 	if (nbyte) {
118 		nbyte = ELF_MIN_ALIGN - nbyte;
119 		if (clear_user((void __user *) elf_bss, nbyte))
120 			return -EFAULT;
121 	}
122 	return 0;
123 }
124 
125 /* Let's use some macros to make this stack manipulation a little clearer */
126 #ifdef CONFIG_STACK_GROWSUP
127 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
128 #define STACK_ROUND(sp, items) \
129 	((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
130 #define STACK_ALLOC(sp, len) ({ \
131 	elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
132 	old_sp; })
133 #else
134 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
135 #define STACK_ROUND(sp, items) \
136 	(((unsigned long) (sp - items)) &~ 15UL)
137 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
138 #endif
139 
140 #ifndef ELF_BASE_PLATFORM
141 /*
142  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
143  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
144  * will be copied to the user stack in the same manner as AT_PLATFORM.
145  */
146 #define ELF_BASE_PLATFORM NULL
147 #endif
148 
149 static int
150 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
151 		unsigned long load_addr, unsigned long interp_load_addr)
152 {
153 	unsigned long p = bprm->p;
154 	int argc = bprm->argc;
155 	int envc = bprm->envc;
156 	elf_addr_t __user *argv;
157 	elf_addr_t __user *envp;
158 	elf_addr_t __user *sp;
159 	elf_addr_t __user *u_platform;
160 	elf_addr_t __user *u_base_platform;
161 	elf_addr_t __user *u_rand_bytes;
162 	const char *k_platform = ELF_PLATFORM;
163 	const char *k_base_platform = ELF_BASE_PLATFORM;
164 	unsigned char k_rand_bytes[16];
165 	int items;
166 	elf_addr_t *elf_info;
167 	int ei_index = 0;
168 	const struct cred *cred = current_cred();
169 	struct vm_area_struct *vma;
170 
171 	/*
172 	 * In some cases (e.g. Hyper-Threading), we want to avoid L1
173 	 * evictions by the processes running on the same package. One
174 	 * thing we can do is to shuffle the initial stack for them.
175 	 */
176 
177 	p = arch_align_stack(p);
178 
179 	/*
180 	 * If this architecture has a platform capability string, copy it
181 	 * to userspace.  In some cases (Sparc), this info is impossible
182 	 * for userspace to get any other way, in others (i386) it is
183 	 * merely difficult.
184 	 */
185 	u_platform = NULL;
186 	if (k_platform) {
187 		size_t len = strlen(k_platform) + 1;
188 
189 		u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
190 		if (__copy_to_user(u_platform, k_platform, len))
191 			return -EFAULT;
192 	}
193 
194 	/*
195 	 * If this architecture has a "base" platform capability
196 	 * string, copy it to userspace.
197 	 */
198 	u_base_platform = NULL;
199 	if (k_base_platform) {
200 		size_t len = strlen(k_base_platform) + 1;
201 
202 		u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
203 		if (__copy_to_user(u_base_platform, k_base_platform, len))
204 			return -EFAULT;
205 	}
206 
207 	/*
208 	 * Generate 16 random bytes for userspace PRNG seeding.
209 	 */
210 	get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
211 	u_rand_bytes = (elf_addr_t __user *)
212 		       STACK_ALLOC(p, sizeof(k_rand_bytes));
213 	if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
214 		return -EFAULT;
215 
216 	/* Create the ELF interpreter info */
217 	elf_info = (elf_addr_t *)current->mm->saved_auxv;
218 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
219 #define NEW_AUX_ENT(id, val) \
220 	do { \
221 		elf_info[ei_index++] = id; \
222 		elf_info[ei_index++] = val; \
223 	} while (0)
224 
225 #ifdef ARCH_DLINFO
226 	/*
227 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
228 	 * AUXV.
229 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
230 	 * ARCH_DLINFO changes
231 	 */
232 	ARCH_DLINFO;
233 #endif
234 	NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
235 	NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
236 	NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
237 	NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
238 	NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
239 	NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
240 	NEW_AUX_ENT(AT_BASE, interp_load_addr);
241 	NEW_AUX_ENT(AT_FLAGS, 0);
242 	NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
243 	NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
244 	NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
245 	NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
246 	NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
247  	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
248 	NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
249 #ifdef ELF_HWCAP2
250 	NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
251 #endif
252 	NEW_AUX_ENT(AT_EXECFN, bprm->exec);
253 	if (k_platform) {
254 		NEW_AUX_ENT(AT_PLATFORM,
255 			    (elf_addr_t)(unsigned long)u_platform);
256 	}
257 	if (k_base_platform) {
258 		NEW_AUX_ENT(AT_BASE_PLATFORM,
259 			    (elf_addr_t)(unsigned long)u_base_platform);
260 	}
261 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
262 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
263 	}
264 #undef NEW_AUX_ENT
265 	/* AT_NULL is zero; clear the rest too */
266 	memset(&elf_info[ei_index], 0,
267 	       sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
268 
269 	/* And advance past the AT_NULL entry.  */
270 	ei_index += 2;
271 
272 	sp = STACK_ADD(p, ei_index);
273 
274 	items = (argc + 1) + (envc + 1) + 1;
275 	bprm->p = STACK_ROUND(sp, items);
276 
277 	/* Point sp at the lowest address on the stack */
278 #ifdef CONFIG_STACK_GROWSUP
279 	sp = (elf_addr_t __user *)bprm->p - items - ei_index;
280 	bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
281 #else
282 	sp = (elf_addr_t __user *)bprm->p;
283 #endif
284 
285 
286 	/*
287 	 * Grow the stack manually; some architectures have a limit on how
288 	 * far ahead a user-space access may be in order to grow the stack.
289 	 */
290 	vma = find_extend_vma(current->mm, bprm->p);
291 	if (!vma)
292 		return -EFAULT;
293 
294 	/* Now, let's put argc (and argv, envp if appropriate) on the stack */
295 	if (__put_user(argc, sp++))
296 		return -EFAULT;
297 	argv = sp;
298 	envp = argv + argc + 1;
299 
300 	/* Populate argv and envp */
301 	p = current->mm->arg_end = current->mm->arg_start;
302 	while (argc-- > 0) {
303 		size_t len;
304 		if (__put_user((elf_addr_t)p, argv++))
305 			return -EFAULT;
306 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
307 		if (!len || len > MAX_ARG_STRLEN)
308 			return -EINVAL;
309 		p += len;
310 	}
311 	if (__put_user(0, argv))
312 		return -EFAULT;
313 	current->mm->arg_end = current->mm->env_start = p;
314 	while (envc-- > 0) {
315 		size_t len;
316 		if (__put_user((elf_addr_t)p, envp++))
317 			return -EFAULT;
318 		len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
319 		if (!len || len > MAX_ARG_STRLEN)
320 			return -EINVAL;
321 		p += len;
322 	}
323 	if (__put_user(0, envp))
324 		return -EFAULT;
325 	current->mm->env_end = p;
326 
327 	/* Put the elf_info on the stack in the right place.  */
328 	sp = (elf_addr_t __user *)envp + 1;
329 	if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
330 		return -EFAULT;
331 	return 0;
332 }
333 
334 #ifndef elf_map
335 
336 static unsigned long elf_map(struct file *filep, unsigned long addr,
337 		struct elf_phdr *eppnt, int prot, int type,
338 		unsigned long total_size)
339 {
340 	unsigned long map_addr;
341 	unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
342 	unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
343 	addr = ELF_PAGESTART(addr);
344 	size = ELF_PAGEALIGN(size);
345 
346 	/* mmap() will return -EINVAL if given a zero size, but a
347 	 * segment with zero filesize is perfectly valid */
348 	if (!size)
349 		return addr;
350 
351 	/*
352 	* total_size is the size of the ELF (interpreter) image.
353 	* The _first_ mmap needs to know the full size, otherwise
354 	* randomization might put this image into an overlapping
355 	* position with the ELF binary image. (since size < total_size)
356 	* So we first map the 'big' image - and unmap the remainder at
357 	* the end. (which unmap is needed for ELF images with holes.)
358 	*/
359 	if (total_size) {
360 		total_size = ELF_PAGEALIGN(total_size);
361 		map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
362 		if (!BAD_ADDR(map_addr))
363 			vm_munmap(map_addr+size, total_size-size);
364 	} else
365 		map_addr = vm_mmap(filep, addr, size, prot, type, off);
366 
367 	return(map_addr);
368 }
369 
370 #endif /* !elf_map */
371 
372 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
373 {
374 	int i, first_idx = -1, last_idx = -1;
375 
376 	for (i = 0; i < nr; i++) {
377 		if (cmds[i].p_type == PT_LOAD) {
378 			last_idx = i;
379 			if (first_idx == -1)
380 				first_idx = i;
381 		}
382 	}
383 	if (first_idx == -1)
384 		return 0;
385 
386 	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
387 				ELF_PAGESTART(cmds[first_idx].p_vaddr);
388 }
389 
390 /**
391  * load_elf_phdrs() - load ELF program headers
392  * @elf_ex:   ELF header of the binary whose program headers should be loaded
393  * @elf_file: the opened ELF binary file
394  *
395  * Loads ELF program headers from the binary file elf_file, which has the ELF
396  * header pointed to by elf_ex, into a newly allocated array. The caller is
397  * responsible for freeing the allocated data. Returns an ERR_PTR upon failure.
398  */
399 static struct elf_phdr *load_elf_phdrs(struct elfhdr *elf_ex,
400 				       struct file *elf_file)
401 {
402 	struct elf_phdr *elf_phdata = NULL;
403 	int retval, size, err = -1;
404 
405 	/*
406 	 * If the size of this structure has changed, then punt, since
407 	 * we will be doing the wrong thing.
408 	 */
409 	if (elf_ex->e_phentsize != sizeof(struct elf_phdr))
410 		goto out;
411 
412 	/* Sanity check the number of program headers... */
413 	if (elf_ex->e_phnum < 1 ||
414 		elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
415 		goto out;
416 
417 	/* ...and their total size. */
418 	size = sizeof(struct elf_phdr) * elf_ex->e_phnum;
419 	if (size > ELF_MIN_ALIGN)
420 		goto out;
421 
422 	elf_phdata = kmalloc(size, GFP_KERNEL);
423 	if (!elf_phdata)
424 		goto out;
425 
426 	/* Read in the program headers */
427 	retval = kernel_read(elf_file, elf_ex->e_phoff,
428 			     (char *)elf_phdata, size);
429 	if (retval != size) {
430 		err = (retval < 0) ? retval : -EIO;
431 		goto out;
432 	}
433 
434 	/* Success! */
435 	err = 0;
436 out:
437 	if (err) {
438 		kfree(elf_phdata);
439 		elf_phdata = NULL;
440 	}
441 	return elf_phdata;
442 }
443 
444 #ifndef CONFIG_ARCH_BINFMT_ELF_STATE
445 
446 /**
447  * struct arch_elf_state - arch-specific ELF loading state
448  *
449  * This structure is used to preserve architecture specific data during
450  * the loading of an ELF file, throughout the checking of architecture
451  * specific ELF headers & through to the point where the ELF load is
452  * known to be proceeding (ie. SET_PERSONALITY).
453  *
454  * This implementation is a dummy for architectures which require no
455  * specific state.
456  */
457 struct arch_elf_state {
458 };
459 
460 #define INIT_ARCH_ELF_STATE {}
461 
462 /**
463  * arch_elf_pt_proc() - check a PT_LOPROC..PT_HIPROC ELF program header
464  * @ehdr:	The main ELF header
465  * @phdr:	The program header to check
466  * @elf:	The open ELF file
467  * @is_interp:	True if the phdr is from the interpreter of the ELF being
468  *		loaded, else false.
469  * @state:	Architecture-specific state preserved throughout the process
470  *		of loading the ELF.
471  *
472  * Inspects the program header phdr to validate its correctness and/or
473  * suitability for the system. Called once per ELF program header in the
474  * range PT_LOPROC to PT_HIPROC, for both the ELF being loaded and its
475  * interpreter.
476  *
477  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
478  *         with that return code.
479  */
480 static inline int arch_elf_pt_proc(struct elfhdr *ehdr,
481 				   struct elf_phdr *phdr,
482 				   struct file *elf, bool is_interp,
483 				   struct arch_elf_state *state)
484 {
485 	/* Dummy implementation, always proceed */
486 	return 0;
487 }
488 
489 /**
490  * arch_check_elf() - check an ELF executable
491  * @ehdr:	The main ELF header
492  * @has_interp:	True if the ELF has an interpreter, else false.
493  * @interp_ehdr: The interpreter's ELF header
494  * @state:	Architecture-specific state preserved throughout the process
495  *		of loading the ELF.
496  *
497  * Provides a final opportunity for architecture code to reject the loading
498  * of the ELF & cause an exec syscall to return an error. This is called after
499  * all program headers to be checked by arch_elf_pt_proc have been.
500  *
501  * Return: Zero to proceed with the ELF load, non-zero to fail the ELF load
502  *         with that return code.
503  */
504 static inline int arch_check_elf(struct elfhdr *ehdr, bool has_interp,
505 				 struct elfhdr *interp_ehdr,
506 				 struct arch_elf_state *state)
507 {
508 	/* Dummy implementation, always proceed */
509 	return 0;
510 }
511 
512 #endif /* !CONFIG_ARCH_BINFMT_ELF_STATE */
513 
514 /* This is much more generalized than the library routine read function,
515    so we keep this separate.  Technically the library read function
516    is only provided so that we can read a.out libraries that have
517    an ELF header */
518 
519 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
520 		struct file *interpreter, unsigned long *interp_map_addr,
521 		unsigned long no_base, struct elf_phdr *interp_elf_phdata)
522 {
523 	struct elf_phdr *eppnt;
524 	unsigned long load_addr = 0;
525 	int load_addr_set = 0;
526 	unsigned long last_bss = 0, elf_bss = 0;
527 	unsigned long error = ~0UL;
528 	unsigned long total_size;
529 	int i;
530 
531 	/* First of all, some simple consistency checks */
532 	if (interp_elf_ex->e_type != ET_EXEC &&
533 	    interp_elf_ex->e_type != ET_DYN)
534 		goto out;
535 	if (!elf_check_arch(interp_elf_ex))
536 		goto out;
537 	if (!interpreter->f_op->mmap)
538 		goto out;
539 
540 	total_size = total_mapping_size(interp_elf_phdata,
541 					interp_elf_ex->e_phnum);
542 	if (!total_size) {
543 		error = -EINVAL;
544 		goto out;
545 	}
546 
547 	eppnt = interp_elf_phdata;
548 	for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
549 		if (eppnt->p_type == PT_LOAD) {
550 			int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
551 			int elf_prot = 0;
552 			unsigned long vaddr = 0;
553 			unsigned long k, map_addr;
554 
555 			if (eppnt->p_flags & PF_R)
556 		    		elf_prot = PROT_READ;
557 			if (eppnt->p_flags & PF_W)
558 				elf_prot |= PROT_WRITE;
559 			if (eppnt->p_flags & PF_X)
560 				elf_prot |= PROT_EXEC;
561 			vaddr = eppnt->p_vaddr;
562 			if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
563 				elf_type |= MAP_FIXED;
564 			else if (no_base && interp_elf_ex->e_type == ET_DYN)
565 				load_addr = -vaddr;
566 
567 			map_addr = elf_map(interpreter, load_addr + vaddr,
568 					eppnt, elf_prot, elf_type, total_size);
569 			total_size = 0;
570 			if (!*interp_map_addr)
571 				*interp_map_addr = map_addr;
572 			error = map_addr;
573 			if (BAD_ADDR(map_addr))
574 				goto out;
575 
576 			if (!load_addr_set &&
577 			    interp_elf_ex->e_type == ET_DYN) {
578 				load_addr = map_addr - ELF_PAGESTART(vaddr);
579 				load_addr_set = 1;
580 			}
581 
582 			/*
583 			 * Check to see if the section's size will overflow the
584 			 * allowed task size. Note that p_filesz must always be
585 			 * <= p_memsize so it's only necessary to check p_memsz.
586 			 */
587 			k = load_addr + eppnt->p_vaddr;
588 			if (BAD_ADDR(k) ||
589 			    eppnt->p_filesz > eppnt->p_memsz ||
590 			    eppnt->p_memsz > TASK_SIZE ||
591 			    TASK_SIZE - eppnt->p_memsz < k) {
592 				error = -ENOMEM;
593 				goto out;
594 			}
595 
596 			/*
597 			 * Find the end of the file mapping for this phdr, and
598 			 * keep track of the largest address we see for this.
599 			 */
600 			k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
601 			if (k > elf_bss)
602 				elf_bss = k;
603 
604 			/*
605 			 * Do the same thing for the memory mapping - between
606 			 * elf_bss and last_bss is the bss section.
607 			 */
608 			k = load_addr + eppnt->p_vaddr + eppnt->p_memsz;
609 			if (k > last_bss)
610 				last_bss = k;
611 		}
612 	}
613 
614 	/*
615 	 * Now fill out the bss section: first pad the last page from
616 	 * the file up to the page boundary, and zero it from elf_bss
617 	 * up to the end of the page.
618 	 */
619 	if (padzero(elf_bss)) {
620 		error = -EFAULT;
621 		goto out;
622 	}
623 	/*
624 	 * Next, align both the file and mem bss up to the page size,
625 	 * since this is where elf_bss was just zeroed up to, and where
626 	 * last_bss will end after the vm_brk() below.
627 	 */
628 	elf_bss = ELF_PAGEALIGN(elf_bss);
629 	last_bss = ELF_PAGEALIGN(last_bss);
630 	/* Finally, if there is still more bss to allocate, do it. */
631 	if (last_bss > elf_bss) {
632 		error = vm_brk(elf_bss, last_bss - elf_bss);
633 		if (error)
634 			goto out;
635 	}
636 
637 	error = load_addr;
638 out:
639 	return error;
640 }
641 
642 /*
643  * These are the functions used to load ELF style executables and shared
644  * libraries.  There is no binary dependent code anywhere else.
645  */
646 
647 #ifndef STACK_RND_MASK
648 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))	/* 8MB of VA */
649 #endif
650 
651 static unsigned long randomize_stack_top(unsigned long stack_top)
652 {
653 	unsigned long random_variable = 0;
654 
655 	if ((current->flags & PF_RANDOMIZE) &&
656 		!(current->personality & ADDR_NO_RANDOMIZE)) {
657 		random_variable = get_random_long();
658 		random_variable &= STACK_RND_MASK;
659 		random_variable <<= PAGE_SHIFT;
660 	}
661 #ifdef CONFIG_STACK_GROWSUP
662 	return PAGE_ALIGN(stack_top) + random_variable;
663 #else
664 	return PAGE_ALIGN(stack_top) - random_variable;
665 #endif
666 }
667 
668 static int load_elf_binary(struct linux_binprm *bprm)
669 {
670 	struct file *interpreter = NULL; /* to shut gcc up */
671  	unsigned long load_addr = 0, load_bias = 0;
672 	int load_addr_set = 0;
673 	char * elf_interpreter = NULL;
674 	unsigned long error;
675 	struct elf_phdr *elf_ppnt, *elf_phdata, *interp_elf_phdata = NULL;
676 	unsigned long elf_bss, elf_brk;
677 	int retval, i;
678 	unsigned long elf_entry;
679 	unsigned long interp_load_addr = 0;
680 	unsigned long start_code, end_code, start_data, end_data;
681 	unsigned long reloc_func_desc __maybe_unused = 0;
682 	int executable_stack = EXSTACK_DEFAULT;
683 	struct pt_regs *regs = current_pt_regs();
684 	struct {
685 		struct elfhdr elf_ex;
686 		struct elfhdr interp_elf_ex;
687 	} *loc;
688 	struct arch_elf_state arch_state = INIT_ARCH_ELF_STATE;
689 
690 	loc = kmalloc(sizeof(*loc), GFP_KERNEL);
691 	if (!loc) {
692 		retval = -ENOMEM;
693 		goto out_ret;
694 	}
695 
696 	/* Get the exec-header */
697 	loc->elf_ex = *((struct elfhdr *)bprm->buf);
698 
699 	retval = -ENOEXEC;
700 	/* First of all, some simple consistency checks */
701 	if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
702 		goto out;
703 
704 	if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
705 		goto out;
706 	if (!elf_check_arch(&loc->elf_ex))
707 		goto out;
708 	if (!bprm->file->f_op->mmap)
709 		goto out;
710 
711 	elf_phdata = load_elf_phdrs(&loc->elf_ex, bprm->file);
712 	if (!elf_phdata)
713 		goto out;
714 
715 	elf_ppnt = elf_phdata;
716 	elf_bss = 0;
717 	elf_brk = 0;
718 
719 	start_code = ~0UL;
720 	end_code = 0;
721 	start_data = 0;
722 	end_data = 0;
723 
724 	for (i = 0; i < loc->elf_ex.e_phnum; i++) {
725 		if (elf_ppnt->p_type == PT_INTERP) {
726 			/* This is the program interpreter used for
727 			 * shared libraries - for now assume that this
728 			 * is an a.out format binary
729 			 */
730 			retval = -ENOEXEC;
731 			if (elf_ppnt->p_filesz > PATH_MAX ||
732 			    elf_ppnt->p_filesz < 2)
733 				goto out_free_ph;
734 
735 			retval = -ENOMEM;
736 			elf_interpreter = kmalloc(elf_ppnt->p_filesz,
737 						  GFP_KERNEL);
738 			if (!elf_interpreter)
739 				goto out_free_ph;
740 
741 			retval = kernel_read(bprm->file, elf_ppnt->p_offset,
742 					     elf_interpreter,
743 					     elf_ppnt->p_filesz);
744 			if (retval != elf_ppnt->p_filesz) {
745 				if (retval >= 0)
746 					retval = -EIO;
747 				goto out_free_interp;
748 			}
749 			/* make sure path is NULL terminated */
750 			retval = -ENOEXEC;
751 			if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
752 				goto out_free_interp;
753 
754 			interpreter = open_exec(elf_interpreter);
755 			retval = PTR_ERR(interpreter);
756 			if (IS_ERR(interpreter))
757 				goto out_free_interp;
758 
759 			/*
760 			 * If the binary is not readable then enforce
761 			 * mm->dumpable = 0 regardless of the interpreter's
762 			 * permissions.
763 			 */
764 			would_dump(bprm, interpreter);
765 
766 			/* Get the exec headers */
767 			retval = kernel_read(interpreter, 0,
768 					     (void *)&loc->interp_elf_ex,
769 					     sizeof(loc->interp_elf_ex));
770 			if (retval != sizeof(loc->interp_elf_ex)) {
771 				if (retval >= 0)
772 					retval = -EIO;
773 				goto out_free_dentry;
774 			}
775 
776 			break;
777 		}
778 		elf_ppnt++;
779 	}
780 
781 	elf_ppnt = elf_phdata;
782 	for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
783 		switch (elf_ppnt->p_type) {
784 		case PT_GNU_STACK:
785 			if (elf_ppnt->p_flags & PF_X)
786 				executable_stack = EXSTACK_ENABLE_X;
787 			else
788 				executable_stack = EXSTACK_DISABLE_X;
789 			break;
790 
791 		case PT_LOPROC ... PT_HIPROC:
792 			retval = arch_elf_pt_proc(&loc->elf_ex, elf_ppnt,
793 						  bprm->file, false,
794 						  &arch_state);
795 			if (retval)
796 				goto out_free_dentry;
797 			break;
798 		}
799 
800 	/* Some simple consistency checks for the interpreter */
801 	if (elf_interpreter) {
802 		retval = -ELIBBAD;
803 		/* Not an ELF interpreter */
804 		if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
805 			goto out_free_dentry;
806 		/* Verify the interpreter has a valid arch */
807 		if (!elf_check_arch(&loc->interp_elf_ex))
808 			goto out_free_dentry;
809 
810 		/* Load the interpreter program headers */
811 		interp_elf_phdata = load_elf_phdrs(&loc->interp_elf_ex,
812 						   interpreter);
813 		if (!interp_elf_phdata)
814 			goto out_free_dentry;
815 
816 		/* Pass PT_LOPROC..PT_HIPROC headers to arch code */
817 		elf_ppnt = interp_elf_phdata;
818 		for (i = 0; i < loc->interp_elf_ex.e_phnum; i++, elf_ppnt++)
819 			switch (elf_ppnt->p_type) {
820 			case PT_LOPROC ... PT_HIPROC:
821 				retval = arch_elf_pt_proc(&loc->interp_elf_ex,
822 							  elf_ppnt, interpreter,
823 							  true, &arch_state);
824 				if (retval)
825 					goto out_free_dentry;
826 				break;
827 			}
828 	}
829 
830 	/*
831 	 * Allow arch code to reject the ELF at this point, whilst it's
832 	 * still possible to return an error to the code that invoked
833 	 * the exec syscall.
834 	 */
835 	retval = arch_check_elf(&loc->elf_ex,
836 				!!interpreter, &loc->interp_elf_ex,
837 				&arch_state);
838 	if (retval)
839 		goto out_free_dentry;
840 
841 	/* Flush all traces of the currently running executable */
842 	retval = flush_old_exec(bprm);
843 	if (retval)
844 		goto out_free_dentry;
845 
846 	/* Do this immediately, since STACK_TOP as used in setup_arg_pages
847 	   may depend on the personality.  */
848 	SET_PERSONALITY2(loc->elf_ex, &arch_state);
849 	if (elf_read_implies_exec(loc->elf_ex, executable_stack))
850 		current->personality |= READ_IMPLIES_EXEC;
851 
852 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
853 		current->flags |= PF_RANDOMIZE;
854 
855 	setup_new_exec(bprm);
856 	install_exec_creds(bprm);
857 
858 	/* Do this so that we can load the interpreter, if need be.  We will
859 	   change some of these later */
860 	retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
861 				 executable_stack);
862 	if (retval < 0)
863 		goto out_free_dentry;
864 
865 	current->mm->start_stack = bprm->p;
866 
867 	/* Now we do a little grungy work by mmapping the ELF image into
868 	   the correct location in memory. */
869 	for(i = 0, elf_ppnt = elf_phdata;
870 	    i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
871 		int elf_prot = 0, elf_flags;
872 		unsigned long k, vaddr;
873 		unsigned long total_size = 0;
874 
875 		if (elf_ppnt->p_type != PT_LOAD)
876 			continue;
877 
878 		if (unlikely (elf_brk > elf_bss)) {
879 			unsigned long nbyte;
880 
881 			/* There was a PT_LOAD segment with p_memsz > p_filesz
882 			   before this one. Map anonymous pages, if needed,
883 			   and clear the area.  */
884 			retval = set_brk(elf_bss + load_bias,
885 					 elf_brk + load_bias);
886 			if (retval)
887 				goto out_free_dentry;
888 			nbyte = ELF_PAGEOFFSET(elf_bss);
889 			if (nbyte) {
890 				nbyte = ELF_MIN_ALIGN - nbyte;
891 				if (nbyte > elf_brk - elf_bss)
892 					nbyte = elf_brk - elf_bss;
893 				if (clear_user((void __user *)elf_bss +
894 							load_bias, nbyte)) {
895 					/*
896 					 * This bss-zeroing can fail if the ELF
897 					 * file specifies odd protections. So
898 					 * we don't check the return value
899 					 */
900 				}
901 			}
902 		}
903 
904 		if (elf_ppnt->p_flags & PF_R)
905 			elf_prot |= PROT_READ;
906 		if (elf_ppnt->p_flags & PF_W)
907 			elf_prot |= PROT_WRITE;
908 		if (elf_ppnt->p_flags & PF_X)
909 			elf_prot |= PROT_EXEC;
910 
911 		elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
912 
913 		vaddr = elf_ppnt->p_vaddr;
914 		if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
915 			elf_flags |= MAP_FIXED;
916 		} else if (loc->elf_ex.e_type == ET_DYN) {
917 			/* Try and get dynamic programs out of the way of the
918 			 * default mmap base, as well as whatever program they
919 			 * might try to exec.  This is because the brk will
920 			 * follow the loader, and is not movable.  */
921 			load_bias = ELF_ET_DYN_BASE - vaddr;
922 			if (current->flags & PF_RANDOMIZE)
923 				load_bias += arch_mmap_rnd();
924 			load_bias = ELF_PAGESTART(load_bias);
925 			total_size = total_mapping_size(elf_phdata,
926 							loc->elf_ex.e_phnum);
927 			if (!total_size) {
928 				retval = -EINVAL;
929 				goto out_free_dentry;
930 			}
931 		}
932 
933 		error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
934 				elf_prot, elf_flags, total_size);
935 		if (BAD_ADDR(error)) {
936 			retval = IS_ERR((void *)error) ?
937 				PTR_ERR((void*)error) : -EINVAL;
938 			goto out_free_dentry;
939 		}
940 
941 		if (!load_addr_set) {
942 			load_addr_set = 1;
943 			load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
944 			if (loc->elf_ex.e_type == ET_DYN) {
945 				load_bias += error -
946 				             ELF_PAGESTART(load_bias + vaddr);
947 				load_addr += load_bias;
948 				reloc_func_desc = load_bias;
949 			}
950 		}
951 		k = elf_ppnt->p_vaddr;
952 		if (k < start_code)
953 			start_code = k;
954 		if (start_data < k)
955 			start_data = k;
956 
957 		/*
958 		 * Check to see if the section's size will overflow the
959 		 * allowed task size. Note that p_filesz must always be
960 		 * <= p_memsz so it is only necessary to check p_memsz.
961 		 */
962 		if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
963 		    elf_ppnt->p_memsz > TASK_SIZE ||
964 		    TASK_SIZE - elf_ppnt->p_memsz < k) {
965 			/* set_brk can never work. Avoid overflows. */
966 			retval = -EINVAL;
967 			goto out_free_dentry;
968 		}
969 
970 		k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
971 
972 		if (k > elf_bss)
973 			elf_bss = k;
974 		if ((elf_ppnt->p_flags & PF_X) && end_code < k)
975 			end_code = k;
976 		if (end_data < k)
977 			end_data = k;
978 		k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
979 		if (k > elf_brk)
980 			elf_brk = k;
981 	}
982 
983 	loc->elf_ex.e_entry += load_bias;
984 	elf_bss += load_bias;
985 	elf_brk += load_bias;
986 	start_code += load_bias;
987 	end_code += load_bias;
988 	start_data += load_bias;
989 	end_data += load_bias;
990 
991 	/* Calling set_brk effectively mmaps the pages that we need
992 	 * for the bss and break sections.  We must do this before
993 	 * mapping in the interpreter, to make sure it doesn't wind
994 	 * up getting placed where the bss needs to go.
995 	 */
996 	retval = set_brk(elf_bss, elf_brk);
997 	if (retval)
998 		goto out_free_dentry;
999 	if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
1000 		retval = -EFAULT; /* Nobody gets to see this, but.. */
1001 		goto out_free_dentry;
1002 	}
1003 
1004 	if (elf_interpreter) {
1005 		unsigned long interp_map_addr = 0;
1006 
1007 		elf_entry = load_elf_interp(&loc->interp_elf_ex,
1008 					    interpreter,
1009 					    &interp_map_addr,
1010 					    load_bias, interp_elf_phdata);
1011 		if (!IS_ERR((void *)elf_entry)) {
1012 			/*
1013 			 * load_elf_interp() returns relocation
1014 			 * adjustment
1015 			 */
1016 			interp_load_addr = elf_entry;
1017 			elf_entry += loc->interp_elf_ex.e_entry;
1018 		}
1019 		if (BAD_ADDR(elf_entry)) {
1020 			retval = IS_ERR((void *)elf_entry) ?
1021 					(int)elf_entry : -EINVAL;
1022 			goto out_free_dentry;
1023 		}
1024 		reloc_func_desc = interp_load_addr;
1025 
1026 		allow_write_access(interpreter);
1027 		fput(interpreter);
1028 		kfree(elf_interpreter);
1029 	} else {
1030 		elf_entry = loc->elf_ex.e_entry;
1031 		if (BAD_ADDR(elf_entry)) {
1032 			retval = -EINVAL;
1033 			goto out_free_dentry;
1034 		}
1035 	}
1036 
1037 	kfree(interp_elf_phdata);
1038 	kfree(elf_phdata);
1039 
1040 	set_binfmt(&elf_format);
1041 
1042 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1043 	retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
1044 	if (retval < 0)
1045 		goto out;
1046 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1047 
1048 	retval = create_elf_tables(bprm, &loc->elf_ex,
1049 			  load_addr, interp_load_addr);
1050 	if (retval < 0)
1051 		goto out;
1052 	/* N.B. passed_fileno might not be initialized? */
1053 	current->mm->end_code = end_code;
1054 	current->mm->start_code = start_code;
1055 	current->mm->start_data = start_data;
1056 	current->mm->end_data = end_data;
1057 	current->mm->start_stack = bprm->p;
1058 
1059 	if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
1060 		current->mm->brk = current->mm->start_brk =
1061 			arch_randomize_brk(current->mm);
1062 #ifdef compat_brk_randomized
1063 		current->brk_randomized = 1;
1064 #endif
1065 	}
1066 
1067 	if (current->personality & MMAP_PAGE_ZERO) {
1068 		/* Why this, you ask???  Well SVr4 maps page 0 as read-only,
1069 		   and some applications "depend" upon this behavior.
1070 		   Since we do not have the power to recompile these, we
1071 		   emulate the SVr4 behavior. Sigh. */
1072 		error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1073 				MAP_FIXED | MAP_PRIVATE, 0);
1074 	}
1075 
1076 #ifdef ELF_PLAT_INIT
1077 	/*
1078 	 * The ABI may specify that certain registers be set up in special
1079 	 * ways (on i386 %edx is the address of a DT_FINI function, for
1080 	 * example.  In addition, it may also specify (eg, PowerPC64 ELF)
1081 	 * that the e_entry field is the address of the function descriptor
1082 	 * for the startup routine, rather than the address of the startup
1083 	 * routine itself.  This macro performs whatever initialization to
1084 	 * the regs structure is required as well as any relocations to the
1085 	 * function descriptor entries when executing dynamically links apps.
1086 	 */
1087 	ELF_PLAT_INIT(regs, reloc_func_desc);
1088 #endif
1089 
1090 	start_thread(regs, elf_entry, bprm->p);
1091 	retval = 0;
1092 out:
1093 	kfree(loc);
1094 out_ret:
1095 	return retval;
1096 
1097 	/* error cleanup */
1098 out_free_dentry:
1099 	kfree(interp_elf_phdata);
1100 	allow_write_access(interpreter);
1101 	if (interpreter)
1102 		fput(interpreter);
1103 out_free_interp:
1104 	kfree(elf_interpreter);
1105 out_free_ph:
1106 	kfree(elf_phdata);
1107 	goto out;
1108 }
1109 
1110 #ifdef CONFIG_USELIB
1111 /* This is really simpleminded and specialized - we are loading an
1112    a.out library that is given an ELF header. */
1113 static int load_elf_library(struct file *file)
1114 {
1115 	struct elf_phdr *elf_phdata;
1116 	struct elf_phdr *eppnt;
1117 	unsigned long elf_bss, bss, len;
1118 	int retval, error, i, j;
1119 	struct elfhdr elf_ex;
1120 
1121 	error = -ENOEXEC;
1122 	retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1123 	if (retval != sizeof(elf_ex))
1124 		goto out;
1125 
1126 	if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1127 		goto out;
1128 
1129 	/* First of all, some simple consistency checks */
1130 	if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1131 	    !elf_check_arch(&elf_ex) || !file->f_op->mmap)
1132 		goto out;
1133 
1134 	/* Now read in all of the header information */
1135 
1136 	j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1137 	/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1138 
1139 	error = -ENOMEM;
1140 	elf_phdata = kmalloc(j, GFP_KERNEL);
1141 	if (!elf_phdata)
1142 		goto out;
1143 
1144 	eppnt = elf_phdata;
1145 	error = -ENOEXEC;
1146 	retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1147 	if (retval != j)
1148 		goto out_free_ph;
1149 
1150 	for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1151 		if ((eppnt + i)->p_type == PT_LOAD)
1152 			j++;
1153 	if (j != 1)
1154 		goto out_free_ph;
1155 
1156 	while (eppnt->p_type != PT_LOAD)
1157 		eppnt++;
1158 
1159 	/* Now use mmap to map the library into memory. */
1160 	error = vm_mmap(file,
1161 			ELF_PAGESTART(eppnt->p_vaddr),
1162 			(eppnt->p_filesz +
1163 			 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1164 			PROT_READ | PROT_WRITE | PROT_EXEC,
1165 			MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1166 			(eppnt->p_offset -
1167 			 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1168 	if (error != ELF_PAGESTART(eppnt->p_vaddr))
1169 		goto out_free_ph;
1170 
1171 	elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1172 	if (padzero(elf_bss)) {
1173 		error = -EFAULT;
1174 		goto out_free_ph;
1175 	}
1176 
1177 	len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1178 			    ELF_MIN_ALIGN - 1);
1179 	bss = eppnt->p_memsz + eppnt->p_vaddr;
1180 	if (bss > len) {
1181 		error = vm_brk(len, bss - len);
1182 		if (error)
1183 			goto out_free_ph;
1184 	}
1185 	error = 0;
1186 
1187 out_free_ph:
1188 	kfree(elf_phdata);
1189 out:
1190 	return error;
1191 }
1192 #endif /* #ifdef CONFIG_USELIB */
1193 
1194 #ifdef CONFIG_ELF_CORE
1195 /*
1196  * ELF core dumper
1197  *
1198  * Modelled on fs/exec.c:aout_core_dump()
1199  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1200  */
1201 
1202 /*
1203  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1204  * that are useful for post-mortem analysis are included in every core dump.
1205  * In that way we ensure that the core dump is fully interpretable later
1206  * without matching up the same kernel and hardware config to see what PC values
1207  * meant. These special mappings include - vDSO, vsyscall, and other
1208  * architecture specific mappings
1209  */
1210 static bool always_dump_vma(struct vm_area_struct *vma)
1211 {
1212 	/* Any vsyscall mappings? */
1213 	if (vma == get_gate_vma(vma->vm_mm))
1214 		return true;
1215 
1216 	/*
1217 	 * Assume that all vmas with a .name op should always be dumped.
1218 	 * If this changes, a new vm_ops field can easily be added.
1219 	 */
1220 	if (vma->vm_ops && vma->vm_ops->name && vma->vm_ops->name(vma))
1221 		return true;
1222 
1223 	/*
1224 	 * arch_vma_name() returns non-NULL for special architecture mappings,
1225 	 * such as vDSO sections.
1226 	 */
1227 	if (arch_vma_name(vma))
1228 		return true;
1229 
1230 	return false;
1231 }
1232 
1233 /*
1234  * Decide what to dump of a segment, part, all or none.
1235  */
1236 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1237 				   unsigned long mm_flags)
1238 {
1239 #define FILTER(type)	(mm_flags & (1UL << MMF_DUMP_##type))
1240 
1241 	/* always dump the vdso and vsyscall sections */
1242 	if (always_dump_vma(vma))
1243 		goto whole;
1244 
1245 	if (vma->vm_flags & VM_DONTDUMP)
1246 		return 0;
1247 
1248 	/* support for DAX */
1249 	if (vma_is_dax(vma)) {
1250 		if ((vma->vm_flags & VM_SHARED) && FILTER(DAX_SHARED))
1251 			goto whole;
1252 		if (!(vma->vm_flags & VM_SHARED) && FILTER(DAX_PRIVATE))
1253 			goto whole;
1254 		return 0;
1255 	}
1256 
1257 	/* Hugetlb memory check */
1258 	if (vma->vm_flags & VM_HUGETLB) {
1259 		if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1260 			goto whole;
1261 		if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1262 			goto whole;
1263 		return 0;
1264 	}
1265 
1266 	/* Do not dump I/O mapped devices or special mappings */
1267 	if (vma->vm_flags & VM_IO)
1268 		return 0;
1269 
1270 	/* By default, dump shared memory if mapped from an anonymous file. */
1271 	if (vma->vm_flags & VM_SHARED) {
1272 		if (file_inode(vma->vm_file)->i_nlink == 0 ?
1273 		    FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1274 			goto whole;
1275 		return 0;
1276 	}
1277 
1278 	/* Dump segments that have been written to.  */
1279 	if (vma->anon_vma && FILTER(ANON_PRIVATE))
1280 		goto whole;
1281 	if (vma->vm_file == NULL)
1282 		return 0;
1283 
1284 	if (FILTER(MAPPED_PRIVATE))
1285 		goto whole;
1286 
1287 	/*
1288 	 * If this looks like the beginning of a DSO or executable mapping,
1289 	 * check for an ELF header.  If we find one, dump the first page to
1290 	 * aid in determining what was mapped here.
1291 	 */
1292 	if (FILTER(ELF_HEADERS) &&
1293 	    vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1294 		u32 __user *header = (u32 __user *) vma->vm_start;
1295 		u32 word;
1296 		mm_segment_t fs = get_fs();
1297 		/*
1298 		 * Doing it this way gets the constant folded by GCC.
1299 		 */
1300 		union {
1301 			u32 cmp;
1302 			char elfmag[SELFMAG];
1303 		} magic;
1304 		BUILD_BUG_ON(SELFMAG != sizeof word);
1305 		magic.elfmag[EI_MAG0] = ELFMAG0;
1306 		magic.elfmag[EI_MAG1] = ELFMAG1;
1307 		magic.elfmag[EI_MAG2] = ELFMAG2;
1308 		magic.elfmag[EI_MAG3] = ELFMAG3;
1309 		/*
1310 		 * Switch to the user "segment" for get_user(),
1311 		 * then put back what elf_core_dump() had in place.
1312 		 */
1313 		set_fs(USER_DS);
1314 		if (unlikely(get_user(word, header)))
1315 			word = 0;
1316 		set_fs(fs);
1317 		if (word == magic.cmp)
1318 			return PAGE_SIZE;
1319 	}
1320 
1321 #undef	FILTER
1322 
1323 	return 0;
1324 
1325 whole:
1326 	return vma->vm_end - vma->vm_start;
1327 }
1328 
1329 /* An ELF note in memory */
1330 struct memelfnote
1331 {
1332 	const char *name;
1333 	int type;
1334 	unsigned int datasz;
1335 	void *data;
1336 };
1337 
1338 static int notesize(struct memelfnote *en)
1339 {
1340 	int sz;
1341 
1342 	sz = sizeof(struct elf_note);
1343 	sz += roundup(strlen(en->name) + 1, 4);
1344 	sz += roundup(en->datasz, 4);
1345 
1346 	return sz;
1347 }
1348 
1349 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1350 {
1351 	struct elf_note en;
1352 	en.n_namesz = strlen(men->name) + 1;
1353 	en.n_descsz = men->datasz;
1354 	en.n_type = men->type;
1355 
1356 	return dump_emit(cprm, &en, sizeof(en)) &&
1357 	    dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1358 	    dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1359 }
1360 
1361 static void fill_elf_header(struct elfhdr *elf, int segs,
1362 			    u16 machine, u32 flags)
1363 {
1364 	memset(elf, 0, sizeof(*elf));
1365 
1366 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1367 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1368 	elf->e_ident[EI_DATA] = ELF_DATA;
1369 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1370 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1371 
1372 	elf->e_type = ET_CORE;
1373 	elf->e_machine = machine;
1374 	elf->e_version = EV_CURRENT;
1375 	elf->e_phoff = sizeof(struct elfhdr);
1376 	elf->e_flags = flags;
1377 	elf->e_ehsize = sizeof(struct elfhdr);
1378 	elf->e_phentsize = sizeof(struct elf_phdr);
1379 	elf->e_phnum = segs;
1380 
1381 	return;
1382 }
1383 
1384 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1385 {
1386 	phdr->p_type = PT_NOTE;
1387 	phdr->p_offset = offset;
1388 	phdr->p_vaddr = 0;
1389 	phdr->p_paddr = 0;
1390 	phdr->p_filesz = sz;
1391 	phdr->p_memsz = 0;
1392 	phdr->p_flags = 0;
1393 	phdr->p_align = 0;
1394 	return;
1395 }
1396 
1397 static void fill_note(struct memelfnote *note, const char *name, int type,
1398 		unsigned int sz, void *data)
1399 {
1400 	note->name = name;
1401 	note->type = type;
1402 	note->datasz = sz;
1403 	note->data = data;
1404 	return;
1405 }
1406 
1407 /*
1408  * fill up all the fields in prstatus from the given task struct, except
1409  * registers which need to be filled up separately.
1410  */
1411 static void fill_prstatus(struct elf_prstatus *prstatus,
1412 		struct task_struct *p, long signr)
1413 {
1414 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1415 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1416 	prstatus->pr_sighold = p->blocked.sig[0];
1417 	rcu_read_lock();
1418 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1419 	rcu_read_unlock();
1420 	prstatus->pr_pid = task_pid_vnr(p);
1421 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1422 	prstatus->pr_sid = task_session_vnr(p);
1423 	if (thread_group_leader(p)) {
1424 		struct task_cputime cputime;
1425 
1426 		/*
1427 		 * This is the record for the group leader.  It shows the
1428 		 * group-wide total, not its individual thread total.
1429 		 */
1430 		thread_group_cputime(p, &cputime);
1431 		prstatus->pr_utime = ns_to_timeval(cputime.utime);
1432 		prstatus->pr_stime = ns_to_timeval(cputime.stime);
1433 	} else {
1434 		u64 utime, stime;
1435 
1436 		task_cputime(p, &utime, &stime);
1437 		prstatus->pr_utime = ns_to_timeval(utime);
1438 		prstatus->pr_stime = ns_to_timeval(stime);
1439 	}
1440 
1441 	prstatus->pr_cutime = ns_to_timeval(p->signal->cutime);
1442 	prstatus->pr_cstime = ns_to_timeval(p->signal->cstime);
1443 }
1444 
1445 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1446 		       struct mm_struct *mm)
1447 {
1448 	const struct cred *cred;
1449 	unsigned int i, len;
1450 
1451 	/* first copy the parameters from user space */
1452 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1453 
1454 	len = mm->arg_end - mm->arg_start;
1455 	if (len >= ELF_PRARGSZ)
1456 		len = ELF_PRARGSZ-1;
1457 	if (copy_from_user(&psinfo->pr_psargs,
1458 		           (const char __user *)mm->arg_start, len))
1459 		return -EFAULT;
1460 	for(i = 0; i < len; i++)
1461 		if (psinfo->pr_psargs[i] == 0)
1462 			psinfo->pr_psargs[i] = ' ';
1463 	psinfo->pr_psargs[len] = 0;
1464 
1465 	rcu_read_lock();
1466 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1467 	rcu_read_unlock();
1468 	psinfo->pr_pid = task_pid_vnr(p);
1469 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1470 	psinfo->pr_sid = task_session_vnr(p);
1471 
1472 	i = p->state ? ffz(~p->state) + 1 : 0;
1473 	psinfo->pr_state = i;
1474 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1475 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1476 	psinfo->pr_nice = task_nice(p);
1477 	psinfo->pr_flag = p->flags;
1478 	rcu_read_lock();
1479 	cred = __task_cred(p);
1480 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1481 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1482 	rcu_read_unlock();
1483 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1484 
1485 	return 0;
1486 }
1487 
1488 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1489 {
1490 	elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1491 	int i = 0;
1492 	do
1493 		i += 2;
1494 	while (auxv[i - 2] != AT_NULL);
1495 	fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1496 }
1497 
1498 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1499 		const siginfo_t *siginfo)
1500 {
1501 	mm_segment_t old_fs = get_fs();
1502 	set_fs(KERNEL_DS);
1503 	copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1504 	set_fs(old_fs);
1505 	fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1506 }
1507 
1508 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1509 /*
1510  * Format of NT_FILE note:
1511  *
1512  * long count     -- how many files are mapped
1513  * long page_size -- units for file_ofs
1514  * array of [COUNT] elements of
1515  *   long start
1516  *   long end
1517  *   long file_ofs
1518  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1519  */
1520 static int fill_files_note(struct memelfnote *note)
1521 {
1522 	struct vm_area_struct *vma;
1523 	unsigned count, size, names_ofs, remaining, n;
1524 	user_long_t *data;
1525 	user_long_t *start_end_ofs;
1526 	char *name_base, *name_curpos;
1527 
1528 	/* *Estimated* file count and total data size needed */
1529 	count = current->mm->map_count;
1530 	size = count * 64;
1531 
1532 	names_ofs = (2 + 3 * count) * sizeof(data[0]);
1533  alloc:
1534 	if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1535 		return -EINVAL;
1536 	size = round_up(size, PAGE_SIZE);
1537 	data = vmalloc(size);
1538 	if (!data)
1539 		return -ENOMEM;
1540 
1541 	start_end_ofs = data + 2;
1542 	name_base = name_curpos = ((char *)data) + names_ofs;
1543 	remaining = size - names_ofs;
1544 	count = 0;
1545 	for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1546 		struct file *file;
1547 		const char *filename;
1548 
1549 		file = vma->vm_file;
1550 		if (!file)
1551 			continue;
1552 		filename = file_path(file, name_curpos, remaining);
1553 		if (IS_ERR(filename)) {
1554 			if (PTR_ERR(filename) == -ENAMETOOLONG) {
1555 				vfree(data);
1556 				size = size * 5 / 4;
1557 				goto alloc;
1558 			}
1559 			continue;
1560 		}
1561 
1562 		/* file_path() fills at the end, move name down */
1563 		/* n = strlen(filename) + 1: */
1564 		n = (name_curpos + remaining) - filename;
1565 		remaining = filename - name_curpos;
1566 		memmove(name_curpos, filename, n);
1567 		name_curpos += n;
1568 
1569 		*start_end_ofs++ = vma->vm_start;
1570 		*start_end_ofs++ = vma->vm_end;
1571 		*start_end_ofs++ = vma->vm_pgoff;
1572 		count++;
1573 	}
1574 
1575 	/* Now we know exact count of files, can store it */
1576 	data[0] = count;
1577 	data[1] = PAGE_SIZE;
1578 	/*
1579 	 * Count usually is less than current->mm->map_count,
1580 	 * we need to move filenames down.
1581 	 */
1582 	n = current->mm->map_count - count;
1583 	if (n != 0) {
1584 		unsigned shift_bytes = n * 3 * sizeof(data[0]);
1585 		memmove(name_base - shift_bytes, name_base,
1586 			name_curpos - name_base);
1587 		name_curpos -= shift_bytes;
1588 	}
1589 
1590 	size = name_curpos - (char *)data;
1591 	fill_note(note, "CORE", NT_FILE, size, data);
1592 	return 0;
1593 }
1594 
1595 #ifdef CORE_DUMP_USE_REGSET
1596 #include <linux/regset.h>
1597 
1598 struct elf_thread_core_info {
1599 	struct elf_thread_core_info *next;
1600 	struct task_struct *task;
1601 	struct elf_prstatus prstatus;
1602 	struct memelfnote notes[0];
1603 };
1604 
1605 struct elf_note_info {
1606 	struct elf_thread_core_info *thread;
1607 	struct memelfnote psinfo;
1608 	struct memelfnote signote;
1609 	struct memelfnote auxv;
1610 	struct memelfnote files;
1611 	user_siginfo_t csigdata;
1612 	size_t size;
1613 	int thread_notes;
1614 };
1615 
1616 /*
1617  * When a regset has a writeback hook, we call it on each thread before
1618  * dumping user memory.  On register window machines, this makes sure the
1619  * user memory backing the register data is up to date before we read it.
1620  */
1621 static void do_thread_regset_writeback(struct task_struct *task,
1622 				       const struct user_regset *regset)
1623 {
1624 	if (regset->writeback)
1625 		regset->writeback(task, regset, 1);
1626 }
1627 
1628 #ifndef PRSTATUS_SIZE
1629 #define PRSTATUS_SIZE(S, R) sizeof(S)
1630 #endif
1631 
1632 #ifndef SET_PR_FPVALID
1633 #define SET_PR_FPVALID(S, V, R) ((S)->pr_fpvalid = (V))
1634 #endif
1635 
1636 static int fill_thread_core_info(struct elf_thread_core_info *t,
1637 				 const struct user_regset_view *view,
1638 				 long signr, size_t *total)
1639 {
1640 	unsigned int i;
1641 	unsigned int regset_size = view->regsets[0].n * view->regsets[0].size;
1642 
1643 	/*
1644 	 * NT_PRSTATUS is the one special case, because the regset data
1645 	 * goes into the pr_reg field inside the note contents, rather
1646 	 * than being the whole note contents.  We fill the reset in here.
1647 	 * We assume that regset 0 is NT_PRSTATUS.
1648 	 */
1649 	fill_prstatus(&t->prstatus, t->task, signr);
1650 	(void) view->regsets[0].get(t->task, &view->regsets[0], 0, regset_size,
1651 				    &t->prstatus.pr_reg, NULL);
1652 
1653 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1654 		  PRSTATUS_SIZE(t->prstatus, regset_size), &t->prstatus);
1655 	*total += notesize(&t->notes[0]);
1656 
1657 	do_thread_regset_writeback(t->task, &view->regsets[0]);
1658 
1659 	/*
1660 	 * Each other regset might generate a note too.  For each regset
1661 	 * that has no core_note_type or is inactive, we leave t->notes[i]
1662 	 * all zero and we'll know to skip writing it later.
1663 	 */
1664 	for (i = 1; i < view->n; ++i) {
1665 		const struct user_regset *regset = &view->regsets[i];
1666 		do_thread_regset_writeback(t->task, regset);
1667 		if (regset->core_note_type && regset->get &&
1668 		    (!regset->active || regset->active(t->task, regset))) {
1669 			int ret;
1670 			size_t size = regset->n * regset->size;
1671 			void *data = kmalloc(size, GFP_KERNEL);
1672 			if (unlikely(!data))
1673 				return 0;
1674 			ret = regset->get(t->task, regset,
1675 					  0, size, data, NULL);
1676 			if (unlikely(ret))
1677 				kfree(data);
1678 			else {
1679 				if (regset->core_note_type != NT_PRFPREG)
1680 					fill_note(&t->notes[i], "LINUX",
1681 						  regset->core_note_type,
1682 						  size, data);
1683 				else {
1684 					SET_PR_FPVALID(&t->prstatus,
1685 							1, regset_size);
1686 					fill_note(&t->notes[i], "CORE",
1687 						  NT_PRFPREG, size, data);
1688 				}
1689 				*total += notesize(&t->notes[i]);
1690 			}
1691 		}
1692 	}
1693 
1694 	return 1;
1695 }
1696 
1697 static int fill_note_info(struct elfhdr *elf, int phdrs,
1698 			  struct elf_note_info *info,
1699 			  const siginfo_t *siginfo, struct pt_regs *regs)
1700 {
1701 	struct task_struct *dump_task = current;
1702 	const struct user_regset_view *view = task_user_regset_view(dump_task);
1703 	struct elf_thread_core_info *t;
1704 	struct elf_prpsinfo *psinfo;
1705 	struct core_thread *ct;
1706 	unsigned int i;
1707 
1708 	info->size = 0;
1709 	info->thread = NULL;
1710 
1711 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1712 	if (psinfo == NULL) {
1713 		info->psinfo.data = NULL; /* So we don't free this wrongly */
1714 		return 0;
1715 	}
1716 
1717 	fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1718 
1719 	/*
1720 	 * Figure out how many notes we're going to need for each thread.
1721 	 */
1722 	info->thread_notes = 0;
1723 	for (i = 0; i < view->n; ++i)
1724 		if (view->regsets[i].core_note_type != 0)
1725 			++info->thread_notes;
1726 
1727 	/*
1728 	 * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1729 	 * since it is our one special case.
1730 	 */
1731 	if (unlikely(info->thread_notes == 0) ||
1732 	    unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1733 		WARN_ON(1);
1734 		return 0;
1735 	}
1736 
1737 	/*
1738 	 * Initialize the ELF file header.
1739 	 */
1740 	fill_elf_header(elf, phdrs,
1741 			view->e_machine, view->e_flags);
1742 
1743 	/*
1744 	 * Allocate a structure for each thread.
1745 	 */
1746 	for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1747 		t = kzalloc(offsetof(struct elf_thread_core_info,
1748 				     notes[info->thread_notes]),
1749 			    GFP_KERNEL);
1750 		if (unlikely(!t))
1751 			return 0;
1752 
1753 		t->task = ct->task;
1754 		if (ct->task == dump_task || !info->thread) {
1755 			t->next = info->thread;
1756 			info->thread = t;
1757 		} else {
1758 			/*
1759 			 * Make sure to keep the original task at
1760 			 * the head of the list.
1761 			 */
1762 			t->next = info->thread->next;
1763 			info->thread->next = t;
1764 		}
1765 	}
1766 
1767 	/*
1768 	 * Now fill in each thread's information.
1769 	 */
1770 	for (t = info->thread; t != NULL; t = t->next)
1771 		if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1772 			return 0;
1773 
1774 	/*
1775 	 * Fill in the two process-wide notes.
1776 	 */
1777 	fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1778 	info->size += notesize(&info->psinfo);
1779 
1780 	fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1781 	info->size += notesize(&info->signote);
1782 
1783 	fill_auxv_note(&info->auxv, current->mm);
1784 	info->size += notesize(&info->auxv);
1785 
1786 	if (fill_files_note(&info->files) == 0)
1787 		info->size += notesize(&info->files);
1788 
1789 	return 1;
1790 }
1791 
1792 static size_t get_note_info_size(struct elf_note_info *info)
1793 {
1794 	return info->size;
1795 }
1796 
1797 /*
1798  * Write all the notes for each thread.  When writing the first thread, the
1799  * process-wide notes are interleaved after the first thread-specific note.
1800  */
1801 static int write_note_info(struct elf_note_info *info,
1802 			   struct coredump_params *cprm)
1803 {
1804 	bool first = true;
1805 	struct elf_thread_core_info *t = info->thread;
1806 
1807 	do {
1808 		int i;
1809 
1810 		if (!writenote(&t->notes[0], cprm))
1811 			return 0;
1812 
1813 		if (first && !writenote(&info->psinfo, cprm))
1814 			return 0;
1815 		if (first && !writenote(&info->signote, cprm))
1816 			return 0;
1817 		if (first && !writenote(&info->auxv, cprm))
1818 			return 0;
1819 		if (first && info->files.data &&
1820 				!writenote(&info->files, cprm))
1821 			return 0;
1822 
1823 		for (i = 1; i < info->thread_notes; ++i)
1824 			if (t->notes[i].data &&
1825 			    !writenote(&t->notes[i], cprm))
1826 				return 0;
1827 
1828 		first = false;
1829 		t = t->next;
1830 	} while (t);
1831 
1832 	return 1;
1833 }
1834 
1835 static void free_note_info(struct elf_note_info *info)
1836 {
1837 	struct elf_thread_core_info *threads = info->thread;
1838 	while (threads) {
1839 		unsigned int i;
1840 		struct elf_thread_core_info *t = threads;
1841 		threads = t->next;
1842 		WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1843 		for (i = 1; i < info->thread_notes; ++i)
1844 			kfree(t->notes[i].data);
1845 		kfree(t);
1846 	}
1847 	kfree(info->psinfo.data);
1848 	vfree(info->files.data);
1849 }
1850 
1851 #else
1852 
1853 /* Here is the structure in which status of each thread is captured. */
1854 struct elf_thread_status
1855 {
1856 	struct list_head list;
1857 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1858 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1859 	struct task_struct *thread;
1860 #ifdef ELF_CORE_COPY_XFPREGS
1861 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1862 #endif
1863 	struct memelfnote notes[3];
1864 	int num_notes;
1865 };
1866 
1867 /*
1868  * In order to add the specific thread information for the elf file format,
1869  * we need to keep a linked list of every threads pr_status and then create
1870  * a single section for them in the final core file.
1871  */
1872 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1873 {
1874 	int sz = 0;
1875 	struct task_struct *p = t->thread;
1876 	t->num_notes = 0;
1877 
1878 	fill_prstatus(&t->prstatus, p, signr);
1879 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1880 
1881 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1882 		  &(t->prstatus));
1883 	t->num_notes++;
1884 	sz += notesize(&t->notes[0]);
1885 
1886 	if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1887 								&t->fpu))) {
1888 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1889 			  &(t->fpu));
1890 		t->num_notes++;
1891 		sz += notesize(&t->notes[1]);
1892 	}
1893 
1894 #ifdef ELF_CORE_COPY_XFPREGS
1895 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1896 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1897 			  sizeof(t->xfpu), &t->xfpu);
1898 		t->num_notes++;
1899 		sz += notesize(&t->notes[2]);
1900 	}
1901 #endif
1902 	return sz;
1903 }
1904 
1905 struct elf_note_info {
1906 	struct memelfnote *notes;
1907 	struct memelfnote *notes_files;
1908 	struct elf_prstatus *prstatus;	/* NT_PRSTATUS */
1909 	struct elf_prpsinfo *psinfo;	/* NT_PRPSINFO */
1910 	struct list_head thread_list;
1911 	elf_fpregset_t *fpu;
1912 #ifdef ELF_CORE_COPY_XFPREGS
1913 	elf_fpxregset_t *xfpu;
1914 #endif
1915 	user_siginfo_t csigdata;
1916 	int thread_status_size;
1917 	int numnote;
1918 };
1919 
1920 static int elf_note_info_init(struct elf_note_info *info)
1921 {
1922 	memset(info, 0, sizeof(*info));
1923 	INIT_LIST_HEAD(&info->thread_list);
1924 
1925 	/* Allocate space for ELF notes */
1926 	info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1927 	if (!info->notes)
1928 		return 0;
1929 	info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1930 	if (!info->psinfo)
1931 		return 0;
1932 	info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1933 	if (!info->prstatus)
1934 		return 0;
1935 	info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1936 	if (!info->fpu)
1937 		return 0;
1938 #ifdef ELF_CORE_COPY_XFPREGS
1939 	info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1940 	if (!info->xfpu)
1941 		return 0;
1942 #endif
1943 	return 1;
1944 }
1945 
1946 static int fill_note_info(struct elfhdr *elf, int phdrs,
1947 			  struct elf_note_info *info,
1948 			  const siginfo_t *siginfo, struct pt_regs *regs)
1949 {
1950 	struct list_head *t;
1951 	struct core_thread *ct;
1952 	struct elf_thread_status *ets;
1953 
1954 	if (!elf_note_info_init(info))
1955 		return 0;
1956 
1957 	for (ct = current->mm->core_state->dumper.next;
1958 					ct; ct = ct->next) {
1959 		ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1960 		if (!ets)
1961 			return 0;
1962 
1963 		ets->thread = ct->task;
1964 		list_add(&ets->list, &info->thread_list);
1965 	}
1966 
1967 	list_for_each(t, &info->thread_list) {
1968 		int sz;
1969 
1970 		ets = list_entry(t, struct elf_thread_status, list);
1971 		sz = elf_dump_thread_status(siginfo->si_signo, ets);
1972 		info->thread_status_size += sz;
1973 	}
1974 	/* now collect the dump for the current */
1975 	memset(info->prstatus, 0, sizeof(*info->prstatus));
1976 	fill_prstatus(info->prstatus, current, siginfo->si_signo);
1977 	elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1978 
1979 	/* Set up header */
1980 	fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1981 
1982 	/*
1983 	 * Set up the notes in similar form to SVR4 core dumps made
1984 	 * with info from their /proc.
1985 	 */
1986 
1987 	fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1988 		  sizeof(*info->prstatus), info->prstatus);
1989 	fill_psinfo(info->psinfo, current->group_leader, current->mm);
1990 	fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1991 		  sizeof(*info->psinfo), info->psinfo);
1992 
1993 	fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1994 	fill_auxv_note(info->notes + 3, current->mm);
1995 	info->numnote = 4;
1996 
1997 	if (fill_files_note(info->notes + info->numnote) == 0) {
1998 		info->notes_files = info->notes + info->numnote;
1999 		info->numnote++;
2000 	}
2001 
2002 	/* Try to dump the FPU. */
2003 	info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
2004 							       info->fpu);
2005 	if (info->prstatus->pr_fpvalid)
2006 		fill_note(info->notes + info->numnote++,
2007 			  "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
2008 #ifdef ELF_CORE_COPY_XFPREGS
2009 	if (elf_core_copy_task_xfpregs(current, info->xfpu))
2010 		fill_note(info->notes + info->numnote++,
2011 			  "LINUX", ELF_CORE_XFPREG_TYPE,
2012 			  sizeof(*info->xfpu), info->xfpu);
2013 #endif
2014 
2015 	return 1;
2016 }
2017 
2018 static size_t get_note_info_size(struct elf_note_info *info)
2019 {
2020 	int sz = 0;
2021 	int i;
2022 
2023 	for (i = 0; i < info->numnote; i++)
2024 		sz += notesize(info->notes + i);
2025 
2026 	sz += info->thread_status_size;
2027 
2028 	return sz;
2029 }
2030 
2031 static int write_note_info(struct elf_note_info *info,
2032 			   struct coredump_params *cprm)
2033 {
2034 	int i;
2035 	struct list_head *t;
2036 
2037 	for (i = 0; i < info->numnote; i++)
2038 		if (!writenote(info->notes + i, cprm))
2039 			return 0;
2040 
2041 	/* write out the thread status notes section */
2042 	list_for_each(t, &info->thread_list) {
2043 		struct elf_thread_status *tmp =
2044 				list_entry(t, struct elf_thread_status, list);
2045 
2046 		for (i = 0; i < tmp->num_notes; i++)
2047 			if (!writenote(&tmp->notes[i], cprm))
2048 				return 0;
2049 	}
2050 
2051 	return 1;
2052 }
2053 
2054 static void free_note_info(struct elf_note_info *info)
2055 {
2056 	while (!list_empty(&info->thread_list)) {
2057 		struct list_head *tmp = info->thread_list.next;
2058 		list_del(tmp);
2059 		kfree(list_entry(tmp, struct elf_thread_status, list));
2060 	}
2061 
2062 	/* Free data possibly allocated by fill_files_note(): */
2063 	if (info->notes_files)
2064 		vfree(info->notes_files->data);
2065 
2066 	kfree(info->prstatus);
2067 	kfree(info->psinfo);
2068 	kfree(info->notes);
2069 	kfree(info->fpu);
2070 #ifdef ELF_CORE_COPY_XFPREGS
2071 	kfree(info->xfpu);
2072 #endif
2073 }
2074 
2075 #endif
2076 
2077 static struct vm_area_struct *first_vma(struct task_struct *tsk,
2078 					struct vm_area_struct *gate_vma)
2079 {
2080 	struct vm_area_struct *ret = tsk->mm->mmap;
2081 
2082 	if (ret)
2083 		return ret;
2084 	return gate_vma;
2085 }
2086 /*
2087  * Helper function for iterating across a vma list.  It ensures that the caller
2088  * will visit `gate_vma' prior to terminating the search.
2089  */
2090 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
2091 					struct vm_area_struct *gate_vma)
2092 {
2093 	struct vm_area_struct *ret;
2094 
2095 	ret = this_vma->vm_next;
2096 	if (ret)
2097 		return ret;
2098 	if (this_vma == gate_vma)
2099 		return NULL;
2100 	return gate_vma;
2101 }
2102 
2103 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2104 			     elf_addr_t e_shoff, int segs)
2105 {
2106 	elf->e_shoff = e_shoff;
2107 	elf->e_shentsize = sizeof(*shdr4extnum);
2108 	elf->e_shnum = 1;
2109 	elf->e_shstrndx = SHN_UNDEF;
2110 
2111 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2112 
2113 	shdr4extnum->sh_type = SHT_NULL;
2114 	shdr4extnum->sh_size = elf->e_shnum;
2115 	shdr4extnum->sh_link = elf->e_shstrndx;
2116 	shdr4extnum->sh_info = segs;
2117 }
2118 
2119 /*
2120  * Actual dumper
2121  *
2122  * This is a two-pass process; first we find the offsets of the bits,
2123  * and then they are actually written out.  If we run out of core limit
2124  * we just truncate.
2125  */
2126 static int elf_core_dump(struct coredump_params *cprm)
2127 {
2128 	int has_dumped = 0;
2129 	mm_segment_t fs;
2130 	int segs, i;
2131 	size_t vma_data_size = 0;
2132 	struct vm_area_struct *vma, *gate_vma;
2133 	struct elfhdr *elf = NULL;
2134 	loff_t offset = 0, dataoff;
2135 	struct elf_note_info info = { };
2136 	struct elf_phdr *phdr4note = NULL;
2137 	struct elf_shdr *shdr4extnum = NULL;
2138 	Elf_Half e_phnum;
2139 	elf_addr_t e_shoff;
2140 	elf_addr_t *vma_filesz = NULL;
2141 
2142 	/*
2143 	 * We no longer stop all VM operations.
2144 	 *
2145 	 * This is because those proceses that could possibly change map_count
2146 	 * or the mmap / vma pages are now blocked in do_exit on current
2147 	 * finishing this core dump.
2148 	 *
2149 	 * Only ptrace can touch these memory addresses, but it doesn't change
2150 	 * the map_count or the pages allocated. So no possibility of crashing
2151 	 * exists while dumping the mm->vm_next areas to the core file.
2152 	 */
2153 
2154 	/* alloc memory for large data structures: too large to be on stack */
2155 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2156 	if (!elf)
2157 		goto out;
2158 	/*
2159 	 * The number of segs are recored into ELF header as 16bit value.
2160 	 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2161 	 */
2162 	segs = current->mm->map_count;
2163 	segs += elf_core_extra_phdrs();
2164 
2165 	gate_vma = get_gate_vma(current->mm);
2166 	if (gate_vma != NULL)
2167 		segs++;
2168 
2169 	/* for notes section */
2170 	segs++;
2171 
2172 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2173 	 * this, kernel supports extended numbering. Have a look at
2174 	 * include/linux/elf.h for further information. */
2175 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2176 
2177 	/*
2178 	 * Collect all the non-memory information about the process for the
2179 	 * notes.  This also sets up the file header.
2180 	 */
2181 	if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2182 		goto cleanup;
2183 
2184 	has_dumped = 1;
2185 
2186 	fs = get_fs();
2187 	set_fs(KERNEL_DS);
2188 
2189 	offset += sizeof(*elf);				/* Elf header */
2190 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
2191 
2192 	/* Write notes phdr entry */
2193 	{
2194 		size_t sz = get_note_info_size(&info);
2195 
2196 		sz += elf_coredump_extra_notes_size();
2197 
2198 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2199 		if (!phdr4note)
2200 			goto end_coredump;
2201 
2202 		fill_elf_note_phdr(phdr4note, sz, offset);
2203 		offset += sz;
2204 	}
2205 
2206 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2207 
2208 	if (segs - 1 > ULONG_MAX / sizeof(*vma_filesz))
2209 		goto end_coredump;
2210 	vma_filesz = vmalloc((segs - 1) * sizeof(*vma_filesz));
2211 	if (!vma_filesz)
2212 		goto end_coredump;
2213 
2214 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2215 			vma = next_vma(vma, gate_vma)) {
2216 		unsigned long dump_size;
2217 
2218 		dump_size = vma_dump_size(vma, cprm->mm_flags);
2219 		vma_filesz[i++] = dump_size;
2220 		vma_data_size += dump_size;
2221 	}
2222 
2223 	offset += vma_data_size;
2224 	offset += elf_core_extra_data_size();
2225 	e_shoff = offset;
2226 
2227 	if (e_phnum == PN_XNUM) {
2228 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2229 		if (!shdr4extnum)
2230 			goto end_coredump;
2231 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2232 	}
2233 
2234 	offset = dataoff;
2235 
2236 	if (!dump_emit(cprm, elf, sizeof(*elf)))
2237 		goto end_coredump;
2238 
2239 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
2240 		goto end_coredump;
2241 
2242 	/* Write program headers for segments dump */
2243 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2244 			vma = next_vma(vma, gate_vma)) {
2245 		struct elf_phdr phdr;
2246 
2247 		phdr.p_type = PT_LOAD;
2248 		phdr.p_offset = offset;
2249 		phdr.p_vaddr = vma->vm_start;
2250 		phdr.p_paddr = 0;
2251 		phdr.p_filesz = vma_filesz[i++];
2252 		phdr.p_memsz = vma->vm_end - vma->vm_start;
2253 		offset += phdr.p_filesz;
2254 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2255 		if (vma->vm_flags & VM_WRITE)
2256 			phdr.p_flags |= PF_W;
2257 		if (vma->vm_flags & VM_EXEC)
2258 			phdr.p_flags |= PF_X;
2259 		phdr.p_align = ELF_EXEC_PAGESIZE;
2260 
2261 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
2262 			goto end_coredump;
2263 	}
2264 
2265 	if (!elf_core_write_extra_phdrs(cprm, offset))
2266 		goto end_coredump;
2267 
2268  	/* write out the notes section */
2269 	if (!write_note_info(&info, cprm))
2270 		goto end_coredump;
2271 
2272 	if (elf_coredump_extra_notes_write(cprm))
2273 		goto end_coredump;
2274 
2275 	/* Align to page */
2276 	if (!dump_skip(cprm, dataoff - cprm->pos))
2277 		goto end_coredump;
2278 
2279 	for (i = 0, vma = first_vma(current, gate_vma); vma != NULL;
2280 			vma = next_vma(vma, gate_vma)) {
2281 		unsigned long addr;
2282 		unsigned long end;
2283 
2284 		end = vma->vm_start + vma_filesz[i++];
2285 
2286 		for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2287 			struct page *page;
2288 			int stop;
2289 
2290 			page = get_dump_page(addr);
2291 			if (page) {
2292 				void *kaddr = kmap(page);
2293 				stop = !dump_emit(cprm, kaddr, PAGE_SIZE);
2294 				kunmap(page);
2295 				put_page(page);
2296 			} else
2297 				stop = !dump_skip(cprm, PAGE_SIZE);
2298 			if (stop)
2299 				goto end_coredump;
2300 		}
2301 	}
2302 	dump_truncate(cprm);
2303 
2304 	if (!elf_core_write_extra_data(cprm))
2305 		goto end_coredump;
2306 
2307 	if (e_phnum == PN_XNUM) {
2308 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
2309 			goto end_coredump;
2310 	}
2311 
2312 end_coredump:
2313 	set_fs(fs);
2314 
2315 cleanup:
2316 	free_note_info(&info);
2317 	kfree(shdr4extnum);
2318 	vfree(vma_filesz);
2319 	kfree(phdr4note);
2320 	kfree(elf);
2321 out:
2322 	return has_dumped;
2323 }
2324 
2325 #endif		/* CONFIG_ELF_CORE */
2326 
2327 static int __init init_elf_binfmt(void)
2328 {
2329 	register_binfmt(&elf_format);
2330 	return 0;
2331 }
2332 
2333 static void __exit exit_elf_binfmt(void)
2334 {
2335 	/* Remove the COFF and ELF loaders. */
2336 	unregister_binfmt(&elf_format);
2337 }
2338 
2339 core_initcall(init_elf_binfmt);
2340 module_exit(exit_elf_binfmt);
2341 MODULE_LICENSE("GPL");
2342