xref: /linux/fs/binfmt_elf_fdpic.c (revision 987b741c52c7c6c68d46fbaeb95b8d1087f10b7f)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
3  *
4  * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  * Derived from binfmt_elf.c
7  */
8 
9 #include <linux/module.h>
10 
11 #include <linux/fs.h>
12 #include <linux/stat.h>
13 #include <linux/sched.h>
14 #include <linux/sched/coredump.h>
15 #include <linux/sched/task_stack.h>
16 #include <linux/sched/cputime.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/errno.h>
20 #include <linux/signal.h>
21 #include <linux/binfmts.h>
22 #include <linux/string.h>
23 #include <linux/file.h>
24 #include <linux/fcntl.h>
25 #include <linux/slab.h>
26 #include <linux/pagemap.h>
27 #include <linux/security.h>
28 #include <linux/highmem.h>
29 #include <linux/highuid.h>
30 #include <linux/personality.h>
31 #include <linux/ptrace.h>
32 #include <linux/init.h>
33 #include <linux/elf.h>
34 #include <linux/elf-fdpic.h>
35 #include <linux/elfcore.h>
36 #include <linux/coredump.h>
37 #include <linux/dax.h>
38 #include <linux/regset.h>
39 
40 #include <linux/uaccess.h>
41 #include <asm/param.h>
42 
43 typedef char *elf_caddr_t;
44 
45 #if 0
46 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
47 #else
48 #define kdebug(fmt, ...) do {} while(0)
49 #endif
50 
51 #if 0
52 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
53 #else
54 #define kdcore(fmt, ...) do {} while(0)
55 #endif
56 
57 MODULE_LICENSE("GPL");
58 
59 static int load_elf_fdpic_binary(struct linux_binprm *);
60 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
61 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
62 			      struct mm_struct *, const char *);
63 
64 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
65 				   struct elf_fdpic_params *,
66 				   struct elf_fdpic_params *);
67 
68 #ifndef CONFIG_MMU
69 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
70 						   struct file *,
71 						   struct mm_struct *);
72 #endif
73 
74 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
75 					     struct file *, struct mm_struct *);
76 
77 #ifdef CONFIG_ELF_CORE
78 static int elf_fdpic_core_dump(struct coredump_params *cprm);
79 #endif
80 
81 static struct linux_binfmt elf_fdpic_format = {
82 	.module		= THIS_MODULE,
83 	.load_binary	= load_elf_fdpic_binary,
84 #ifdef CONFIG_ELF_CORE
85 	.core_dump	= elf_fdpic_core_dump,
86 #endif
87 	.min_coredump	= ELF_EXEC_PAGESIZE,
88 };
89 
90 static int __init init_elf_fdpic_binfmt(void)
91 {
92 	register_binfmt(&elf_fdpic_format);
93 	return 0;
94 }
95 
96 static void __exit exit_elf_fdpic_binfmt(void)
97 {
98 	unregister_binfmt(&elf_fdpic_format);
99 }
100 
101 core_initcall(init_elf_fdpic_binfmt);
102 module_exit(exit_elf_fdpic_binfmt);
103 
104 static int is_elf(struct elfhdr *hdr, struct file *file)
105 {
106 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
107 		return 0;
108 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
109 		return 0;
110 	if (!elf_check_arch(hdr))
111 		return 0;
112 	if (!file->f_op->mmap)
113 		return 0;
114 	return 1;
115 }
116 
117 #ifndef elf_check_fdpic
118 #define elf_check_fdpic(x) 0
119 #endif
120 
121 #ifndef elf_check_const_displacement
122 #define elf_check_const_displacement(x) 0
123 #endif
124 
125 static int is_constdisp(struct elfhdr *hdr)
126 {
127 	if (!elf_check_fdpic(hdr))
128 		return 1;
129 	if (elf_check_const_displacement(hdr))
130 		return 1;
131 	return 0;
132 }
133 
134 /*****************************************************************************/
135 /*
136  * read the program headers table into memory
137  */
138 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
139 				 struct file *file)
140 {
141 	struct elf32_phdr *phdr;
142 	unsigned long size;
143 	int retval, loop;
144 	loff_t pos = params->hdr.e_phoff;
145 
146 	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
147 		return -ENOMEM;
148 	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
149 		return -ENOMEM;
150 
151 	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
152 	params->phdrs = kmalloc(size, GFP_KERNEL);
153 	if (!params->phdrs)
154 		return -ENOMEM;
155 
156 	retval = kernel_read(file, params->phdrs, size, &pos);
157 	if (unlikely(retval != size))
158 		return retval < 0 ? retval : -ENOEXEC;
159 
160 	/* determine stack size for this binary */
161 	phdr = params->phdrs;
162 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
163 		if (phdr->p_type != PT_GNU_STACK)
164 			continue;
165 
166 		if (phdr->p_flags & PF_X)
167 			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
168 		else
169 			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
170 
171 		params->stack_size = phdr->p_memsz;
172 		break;
173 	}
174 
175 	return 0;
176 }
177 
178 /*****************************************************************************/
179 /*
180  * load an fdpic binary into various bits of memory
181  */
182 static int load_elf_fdpic_binary(struct linux_binprm *bprm)
183 {
184 	struct elf_fdpic_params exec_params, interp_params;
185 	struct pt_regs *regs = current_pt_regs();
186 	struct elf_phdr *phdr;
187 	unsigned long stack_size, entryaddr;
188 #ifdef ELF_FDPIC_PLAT_INIT
189 	unsigned long dynaddr;
190 #endif
191 #ifndef CONFIG_MMU
192 	unsigned long stack_prot;
193 #endif
194 	struct file *interpreter = NULL; /* to shut gcc up */
195 	char *interpreter_name = NULL;
196 	int executable_stack;
197 	int retval, i;
198 	loff_t pos;
199 
200 	kdebug("____ LOAD %d ____", current->pid);
201 
202 	memset(&exec_params, 0, sizeof(exec_params));
203 	memset(&interp_params, 0, sizeof(interp_params));
204 
205 	exec_params.hdr = *(struct elfhdr *) bprm->buf;
206 	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
207 
208 	/* check that this is a binary we know how to deal with */
209 	retval = -ENOEXEC;
210 	if (!is_elf(&exec_params.hdr, bprm->file))
211 		goto error;
212 	if (!elf_check_fdpic(&exec_params.hdr)) {
213 #ifdef CONFIG_MMU
214 		/* binfmt_elf handles non-fdpic elf except on nommu */
215 		goto error;
216 #else
217 		/* nommu can only load ET_DYN (PIE) ELF */
218 		if (exec_params.hdr.e_type != ET_DYN)
219 			goto error;
220 #endif
221 	}
222 
223 	/* read the program header table */
224 	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
225 	if (retval < 0)
226 		goto error;
227 
228 	/* scan for a program header that specifies an interpreter */
229 	phdr = exec_params.phdrs;
230 
231 	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
232 		switch (phdr->p_type) {
233 		case PT_INTERP:
234 			retval = -ENOMEM;
235 			if (phdr->p_filesz > PATH_MAX)
236 				goto error;
237 			retval = -ENOENT;
238 			if (phdr->p_filesz < 2)
239 				goto error;
240 
241 			/* read the name of the interpreter into memory */
242 			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
243 			if (!interpreter_name)
244 				goto error;
245 
246 			pos = phdr->p_offset;
247 			retval = kernel_read(bprm->file, interpreter_name,
248 					     phdr->p_filesz, &pos);
249 			if (unlikely(retval != phdr->p_filesz)) {
250 				if (retval >= 0)
251 					retval = -ENOEXEC;
252 				goto error;
253 			}
254 
255 			retval = -ENOENT;
256 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
257 				goto error;
258 
259 			kdebug("Using ELF interpreter %s", interpreter_name);
260 
261 			/* replace the program with the interpreter */
262 			interpreter = open_exec(interpreter_name);
263 			retval = PTR_ERR(interpreter);
264 			if (IS_ERR(interpreter)) {
265 				interpreter = NULL;
266 				goto error;
267 			}
268 
269 			/*
270 			 * If the binary is not readable then enforce
271 			 * mm->dumpable = 0 regardless of the interpreter's
272 			 * permissions.
273 			 */
274 			would_dump(bprm, interpreter);
275 
276 			pos = 0;
277 			retval = kernel_read(interpreter, bprm->buf,
278 					BINPRM_BUF_SIZE, &pos);
279 			if (unlikely(retval != BINPRM_BUF_SIZE)) {
280 				if (retval >= 0)
281 					retval = -ENOEXEC;
282 				goto error;
283 			}
284 
285 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
286 			break;
287 
288 		case PT_LOAD:
289 #ifdef CONFIG_MMU
290 			if (exec_params.load_addr == 0)
291 				exec_params.load_addr = phdr->p_vaddr;
292 #endif
293 			break;
294 		}
295 
296 	}
297 
298 	if (is_constdisp(&exec_params.hdr))
299 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
300 
301 	/* perform insanity checks on the interpreter */
302 	if (interpreter_name) {
303 		retval = -ELIBBAD;
304 		if (!is_elf(&interp_params.hdr, interpreter))
305 			goto error;
306 
307 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
308 
309 		/* read the interpreter's program header table */
310 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
311 		if (retval < 0)
312 			goto error;
313 	}
314 
315 	stack_size = exec_params.stack_size;
316 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
317 		executable_stack = EXSTACK_ENABLE_X;
318 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
319 		executable_stack = EXSTACK_DISABLE_X;
320 	else
321 		executable_stack = EXSTACK_DEFAULT;
322 
323 	if (stack_size == 0) {
324 		stack_size = interp_params.stack_size;
325 		if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
326 			executable_stack = EXSTACK_ENABLE_X;
327 		else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
328 			executable_stack = EXSTACK_DISABLE_X;
329 		else
330 			executable_stack = EXSTACK_DEFAULT;
331 	}
332 
333 	retval = -ENOEXEC;
334 	if (stack_size == 0)
335 		stack_size = 131072UL; /* same as exec.c's default commit */
336 
337 	if (is_constdisp(&interp_params.hdr))
338 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
339 
340 	/* flush all traces of the currently running executable */
341 	retval = begin_new_exec(bprm);
342 	if (retval)
343 		goto error;
344 
345 	/* there's now no turning back... the old userspace image is dead,
346 	 * defunct, deceased, etc.
347 	 */
348 	if (elf_check_fdpic(&exec_params.hdr))
349 		set_personality(PER_LINUX_FDPIC);
350 	else
351 		set_personality(PER_LINUX);
352 	if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
353 		current->personality |= READ_IMPLIES_EXEC;
354 
355 	setup_new_exec(bprm);
356 
357 	set_binfmt(&elf_fdpic_format);
358 
359 	current->mm->start_code = 0;
360 	current->mm->end_code = 0;
361 	current->mm->start_stack = 0;
362 	current->mm->start_data = 0;
363 	current->mm->end_data = 0;
364 	current->mm->context.exec_fdpic_loadmap = 0;
365 	current->mm->context.interp_fdpic_loadmap = 0;
366 
367 #ifdef CONFIG_MMU
368 	elf_fdpic_arch_lay_out_mm(&exec_params,
369 				  &interp_params,
370 				  &current->mm->start_stack,
371 				  &current->mm->start_brk);
372 
373 	retval = setup_arg_pages(bprm, current->mm->start_stack,
374 				 executable_stack);
375 	if (retval < 0)
376 		goto error;
377 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
378 	retval = arch_setup_additional_pages(bprm, !!interpreter_name);
379 	if (retval < 0)
380 		goto error;
381 #endif
382 #endif
383 
384 	/* load the executable and interpreter into memory */
385 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
386 				    "executable");
387 	if (retval < 0)
388 		goto error;
389 
390 	if (interpreter_name) {
391 		retval = elf_fdpic_map_file(&interp_params, interpreter,
392 					    current->mm, "interpreter");
393 		if (retval < 0) {
394 			printk(KERN_ERR "Unable to load interpreter\n");
395 			goto error;
396 		}
397 
398 		allow_write_access(interpreter);
399 		fput(interpreter);
400 		interpreter = NULL;
401 	}
402 
403 #ifdef CONFIG_MMU
404 	if (!current->mm->start_brk)
405 		current->mm->start_brk = current->mm->end_data;
406 
407 	current->mm->brk = current->mm->start_brk =
408 		PAGE_ALIGN(current->mm->start_brk);
409 
410 #else
411 	/* create a stack area and zero-size brk area */
412 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
413 	if (stack_size < PAGE_SIZE * 2)
414 		stack_size = PAGE_SIZE * 2;
415 
416 	stack_prot = PROT_READ | PROT_WRITE;
417 	if (executable_stack == EXSTACK_ENABLE_X ||
418 	    (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
419 		stack_prot |= PROT_EXEC;
420 
421 	current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
422 					 MAP_PRIVATE | MAP_ANONYMOUS |
423 					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
424 					 0);
425 
426 	if (IS_ERR_VALUE(current->mm->start_brk)) {
427 		retval = current->mm->start_brk;
428 		current->mm->start_brk = 0;
429 		goto error;
430 	}
431 
432 	current->mm->brk = current->mm->start_brk;
433 	current->mm->context.end_brk = current->mm->start_brk;
434 	current->mm->start_stack = current->mm->start_brk + stack_size;
435 #endif
436 
437 	if (create_elf_fdpic_tables(bprm, current->mm,
438 				    &exec_params, &interp_params) < 0)
439 		goto error;
440 
441 	kdebug("- start_code  %lx", current->mm->start_code);
442 	kdebug("- end_code    %lx", current->mm->end_code);
443 	kdebug("- start_data  %lx", current->mm->start_data);
444 	kdebug("- end_data    %lx", current->mm->end_data);
445 	kdebug("- start_brk   %lx", current->mm->start_brk);
446 	kdebug("- brk         %lx", current->mm->brk);
447 	kdebug("- start_stack %lx", current->mm->start_stack);
448 
449 #ifdef ELF_FDPIC_PLAT_INIT
450 	/*
451 	 * The ABI may specify that certain registers be set up in special
452 	 * ways (on i386 %edx is the address of a DT_FINI function, for
453 	 * example.  This macro performs whatever initialization to
454 	 * the regs structure is required.
455 	 */
456 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
457 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
458 			    dynaddr);
459 #endif
460 
461 	finalize_exec(bprm);
462 	/* everything is now ready... get the userspace context ready to roll */
463 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
464 	start_thread(regs, entryaddr, current->mm->start_stack);
465 
466 	retval = 0;
467 
468 error:
469 	if (interpreter) {
470 		allow_write_access(interpreter);
471 		fput(interpreter);
472 	}
473 	kfree(interpreter_name);
474 	kfree(exec_params.phdrs);
475 	kfree(exec_params.loadmap);
476 	kfree(interp_params.phdrs);
477 	kfree(interp_params.loadmap);
478 	return retval;
479 }
480 
481 /*****************************************************************************/
482 
483 #ifndef ELF_BASE_PLATFORM
484 /*
485  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
486  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
487  * will be copied to the user stack in the same manner as AT_PLATFORM.
488  */
489 #define ELF_BASE_PLATFORM NULL
490 #endif
491 
492 /*
493  * present useful information to the program by shovelling it onto the new
494  * process's stack
495  */
496 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
497 				   struct mm_struct *mm,
498 				   struct elf_fdpic_params *exec_params,
499 				   struct elf_fdpic_params *interp_params)
500 {
501 	const struct cred *cred = current_cred();
502 	unsigned long sp, csp, nitems;
503 	elf_caddr_t __user *argv, *envp;
504 	size_t platform_len = 0, len;
505 	char *k_platform, *k_base_platform;
506 	char __user *u_platform, *u_base_platform, *p;
507 	int loop;
508 	int nr;	/* reset for each csp adjustment */
509 	unsigned long flags = 0;
510 
511 #ifdef CONFIG_MMU
512 	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
513 	 * by the processes running on the same package. One thing we can do is
514 	 * to shuffle the initial stack for them, so we give the architecture
515 	 * an opportunity to do so here.
516 	 */
517 	sp = arch_align_stack(bprm->p);
518 #else
519 	sp = mm->start_stack;
520 
521 	/* stack the program arguments and environment */
522 	if (transfer_args_to_stack(bprm, &sp) < 0)
523 		return -EFAULT;
524 	sp &= ~15;
525 #endif
526 
527 	/*
528 	 * If this architecture has a platform capability string, copy it
529 	 * to userspace.  In some cases (Sparc), this info is impossible
530 	 * for userspace to get any other way, in others (i386) it is
531 	 * merely difficult.
532 	 */
533 	k_platform = ELF_PLATFORM;
534 	u_platform = NULL;
535 
536 	if (k_platform) {
537 		platform_len = strlen(k_platform) + 1;
538 		sp -= platform_len;
539 		u_platform = (char __user *) sp;
540 		if (copy_to_user(u_platform, k_platform, platform_len) != 0)
541 			return -EFAULT;
542 	}
543 
544 	/*
545 	 * If this architecture has a "base" platform capability
546 	 * string, copy it to userspace.
547 	 */
548 	k_base_platform = ELF_BASE_PLATFORM;
549 	u_base_platform = NULL;
550 
551 	if (k_base_platform) {
552 		platform_len = strlen(k_base_platform) + 1;
553 		sp -= platform_len;
554 		u_base_platform = (char __user *) sp;
555 		if (copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
556 			return -EFAULT;
557 	}
558 
559 	sp &= ~7UL;
560 
561 	/* stack the load map(s) */
562 	len = sizeof(struct elf32_fdpic_loadmap);
563 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
564 	sp = (sp - len) & ~7UL;
565 	exec_params->map_addr = sp;
566 
567 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
568 		return -EFAULT;
569 
570 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
571 
572 	if (interp_params->loadmap) {
573 		len = sizeof(struct elf32_fdpic_loadmap);
574 		len += sizeof(struct elf32_fdpic_loadseg) *
575 			interp_params->loadmap->nsegs;
576 		sp = (sp - len) & ~7UL;
577 		interp_params->map_addr = sp;
578 
579 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
580 				 len) != 0)
581 			return -EFAULT;
582 
583 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
584 	}
585 
586 	/* force 16 byte _final_ alignment here for generality */
587 #define DLINFO_ITEMS 15
588 
589 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
590 		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
591 
592 	if (bprm->have_execfd)
593 		nitems++;
594 
595 	csp = sp;
596 	sp -= nitems * 2 * sizeof(unsigned long);
597 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
598 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
599 	sp -= 1 * sizeof(unsigned long);		/* argc */
600 
601 	csp -= sp & 15UL;
602 	sp -= sp & 15UL;
603 
604 	/* put the ELF interpreter info on the stack */
605 #define NEW_AUX_ENT(id, val)						\
606 	do {								\
607 		struct { unsigned long _id, _val; } __user *ent, v;	\
608 									\
609 		ent = (void __user *) csp;				\
610 		v._id = (id);						\
611 		v._val = (val);						\
612 		if (copy_to_user(ent + nr, &v, sizeof(v)))		\
613 			return -EFAULT;					\
614 		nr++;							\
615 	} while (0)
616 
617 	nr = 0;
618 	csp -= 2 * sizeof(unsigned long);
619 	NEW_AUX_ENT(AT_NULL, 0);
620 	if (k_platform) {
621 		nr = 0;
622 		csp -= 2 * sizeof(unsigned long);
623 		NEW_AUX_ENT(AT_PLATFORM,
624 			    (elf_addr_t) (unsigned long) u_platform);
625 	}
626 
627 	if (k_base_platform) {
628 		nr = 0;
629 		csp -= 2 * sizeof(unsigned long);
630 		NEW_AUX_ENT(AT_BASE_PLATFORM,
631 			    (elf_addr_t) (unsigned long) u_base_platform);
632 	}
633 
634 	if (bprm->have_execfd) {
635 		nr = 0;
636 		csp -= 2 * sizeof(unsigned long);
637 		NEW_AUX_ENT(AT_EXECFD, bprm->execfd);
638 	}
639 
640 	nr = 0;
641 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
642 	NEW_AUX_ENT(AT_HWCAP,	ELF_HWCAP);
643 #ifdef ELF_HWCAP2
644 	NEW_AUX_ENT(AT_HWCAP2,	ELF_HWCAP2);
645 #endif
646 	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
647 	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
648 	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
649 	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
650 	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
651 	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
652 	if (bprm->interp_flags & BINPRM_FLAGS_PRESERVE_ARGV0)
653 		flags |= AT_FLAGS_PRESERVE_ARGV0;
654 	NEW_AUX_ENT(AT_FLAGS,	flags);
655 	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
656 	NEW_AUX_ENT(AT_UID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
657 	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
658 	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
659 	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
660 	NEW_AUX_ENT(AT_SECURE,	bprm->secureexec);
661 	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
662 
663 #ifdef ARCH_DLINFO
664 	nr = 0;
665 	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
666 
667 	/* ARCH_DLINFO must come last so platform specific code can enforce
668 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
669 	 */
670 	ARCH_DLINFO;
671 #endif
672 #undef NEW_AUX_ENT
673 
674 	/* allocate room for argv[] and envv[] */
675 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
676 	envp = (elf_caddr_t __user *) csp;
677 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
678 	argv = (elf_caddr_t __user *) csp;
679 
680 	/* stack argc */
681 	csp -= sizeof(unsigned long);
682 	if (put_user(bprm->argc, (unsigned long __user *) csp))
683 		return -EFAULT;
684 
685 	BUG_ON(csp != sp);
686 
687 	/* fill in the argv[] array */
688 #ifdef CONFIG_MMU
689 	current->mm->arg_start = bprm->p;
690 #else
691 	current->mm->arg_start = current->mm->start_stack -
692 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
693 #endif
694 
695 	p = (char __user *) current->mm->arg_start;
696 	for (loop = bprm->argc; loop > 0; loop--) {
697 		if (put_user((elf_caddr_t) p, argv++))
698 			return -EFAULT;
699 		len = strnlen_user(p, MAX_ARG_STRLEN);
700 		if (!len || len > MAX_ARG_STRLEN)
701 			return -EINVAL;
702 		p += len;
703 	}
704 	if (put_user(NULL, argv))
705 		return -EFAULT;
706 	current->mm->arg_end = (unsigned long) p;
707 
708 	/* fill in the envv[] array */
709 	current->mm->env_start = (unsigned long) p;
710 	for (loop = bprm->envc; loop > 0; loop--) {
711 		if (put_user((elf_caddr_t)(unsigned long) p, envp++))
712 			return -EFAULT;
713 		len = strnlen_user(p, MAX_ARG_STRLEN);
714 		if (!len || len > MAX_ARG_STRLEN)
715 			return -EINVAL;
716 		p += len;
717 	}
718 	if (put_user(NULL, envp))
719 		return -EFAULT;
720 	current->mm->env_end = (unsigned long) p;
721 
722 	mm->start_stack = (unsigned long) sp;
723 	return 0;
724 }
725 
726 /*****************************************************************************/
727 /*
728  * load the appropriate binary image (executable or interpreter) into memory
729  * - we assume no MMU is available
730  * - if no other PIC bits are set in params->hdr->e_flags
731  *   - we assume that the LOADable segments in the binary are independently relocatable
732  *   - we assume R/O executable segments are shareable
733  * - else
734  *   - we assume the loadable parts of the image to require fixed displacement
735  *   - the image is not shareable
736  */
737 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
738 			      struct file *file,
739 			      struct mm_struct *mm,
740 			      const char *what)
741 {
742 	struct elf32_fdpic_loadmap *loadmap;
743 #ifdef CONFIG_MMU
744 	struct elf32_fdpic_loadseg *mseg;
745 #endif
746 	struct elf32_fdpic_loadseg *seg;
747 	struct elf32_phdr *phdr;
748 	unsigned long load_addr, stop;
749 	unsigned nloads, tmp;
750 	size_t size;
751 	int loop, ret;
752 
753 	/* allocate a load map table */
754 	nloads = 0;
755 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
756 		if (params->phdrs[loop].p_type == PT_LOAD)
757 			nloads++;
758 
759 	if (nloads == 0)
760 		return -ELIBBAD;
761 
762 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
763 	loadmap = kzalloc(size, GFP_KERNEL);
764 	if (!loadmap)
765 		return -ENOMEM;
766 
767 	params->loadmap = loadmap;
768 
769 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
770 	loadmap->nsegs = nloads;
771 
772 	load_addr = params->load_addr;
773 	seg = loadmap->segs;
774 
775 	/* map the requested LOADs into the memory space */
776 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
777 	case ELF_FDPIC_FLAG_CONSTDISP:
778 	case ELF_FDPIC_FLAG_CONTIGUOUS:
779 #ifndef CONFIG_MMU
780 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
781 		if (ret < 0)
782 			return ret;
783 		break;
784 #endif
785 	default:
786 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
787 		if (ret < 0)
788 			return ret;
789 		break;
790 	}
791 
792 	/* map the entry point */
793 	if (params->hdr.e_entry) {
794 		seg = loadmap->segs;
795 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
796 			if (params->hdr.e_entry >= seg->p_vaddr &&
797 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
798 				params->entry_addr =
799 					(params->hdr.e_entry - seg->p_vaddr) +
800 					seg->addr;
801 				break;
802 			}
803 		}
804 	}
805 
806 	/* determine where the program header table has wound up if mapped */
807 	stop = params->hdr.e_phoff;
808 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
809 	phdr = params->phdrs;
810 
811 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
812 		if (phdr->p_type != PT_LOAD)
813 			continue;
814 
815 		if (phdr->p_offset > params->hdr.e_phoff ||
816 		    phdr->p_offset + phdr->p_filesz < stop)
817 			continue;
818 
819 		seg = loadmap->segs;
820 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
821 			if (phdr->p_vaddr >= seg->p_vaddr &&
822 			    phdr->p_vaddr + phdr->p_filesz <=
823 			    seg->p_vaddr + seg->p_memsz) {
824 				params->ph_addr =
825 					(phdr->p_vaddr - seg->p_vaddr) +
826 					seg->addr +
827 					params->hdr.e_phoff - phdr->p_offset;
828 				break;
829 			}
830 		}
831 		break;
832 	}
833 
834 	/* determine where the dynamic section has wound up if there is one */
835 	phdr = params->phdrs;
836 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
837 		if (phdr->p_type != PT_DYNAMIC)
838 			continue;
839 
840 		seg = loadmap->segs;
841 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
842 			if (phdr->p_vaddr >= seg->p_vaddr &&
843 			    phdr->p_vaddr + phdr->p_memsz <=
844 			    seg->p_vaddr + seg->p_memsz) {
845 				Elf32_Dyn __user *dyn;
846 				Elf32_Sword d_tag;
847 
848 				params->dynamic_addr =
849 					(phdr->p_vaddr - seg->p_vaddr) +
850 					seg->addr;
851 
852 				/* check the dynamic section contains at least
853 				 * one item, and that the last item is a NULL
854 				 * entry */
855 				if (phdr->p_memsz == 0 ||
856 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
857 					goto dynamic_error;
858 
859 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
860 				dyn = (Elf32_Dyn __user *)params->dynamic_addr;
861 				if (get_user(d_tag, &dyn[tmp - 1].d_tag) ||
862 				    d_tag != 0)
863 					goto dynamic_error;
864 				break;
865 			}
866 		}
867 		break;
868 	}
869 
870 	/* now elide adjacent segments in the load map on MMU linux
871 	 * - on uClinux the holes between may actually be filled with system
872 	 *   stuff or stuff from other processes
873 	 */
874 #ifdef CONFIG_MMU
875 	nloads = loadmap->nsegs;
876 	mseg = loadmap->segs;
877 	seg = mseg + 1;
878 	for (loop = 1; loop < nloads; loop++) {
879 		/* see if we have a candidate for merging */
880 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
881 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
882 			if (load_addr == (seg->addr & PAGE_MASK)) {
883 				mseg->p_memsz +=
884 					load_addr -
885 					(mseg->addr + mseg->p_memsz);
886 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
887 				mseg->p_memsz += seg->p_memsz;
888 				loadmap->nsegs--;
889 				continue;
890 			}
891 		}
892 
893 		mseg++;
894 		if (mseg != seg)
895 			*mseg = *seg;
896 	}
897 #endif
898 
899 	kdebug("Mapped Object [%s]:", what);
900 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
901 	kdebug("- entry    : %lx", params->entry_addr);
902 	kdebug("- PHDR[]   : %lx", params->ph_addr);
903 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
904 	seg = loadmap->segs;
905 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
906 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
907 		       loop,
908 		       seg->addr, seg->addr + seg->p_memsz - 1,
909 		       seg->p_vaddr, seg->p_memsz);
910 
911 	return 0;
912 
913 dynamic_error:
914 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
915 	       what, file_inode(file)->i_ino);
916 	return -ELIBBAD;
917 }
918 
919 /*****************************************************************************/
920 /*
921  * map a file with constant displacement under uClinux
922  */
923 #ifndef CONFIG_MMU
924 static int elf_fdpic_map_file_constdisp_on_uclinux(
925 	struct elf_fdpic_params *params,
926 	struct file *file,
927 	struct mm_struct *mm)
928 {
929 	struct elf32_fdpic_loadseg *seg;
930 	struct elf32_phdr *phdr;
931 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
932 	int loop, ret;
933 
934 	load_addr = params->load_addr;
935 	seg = params->loadmap->segs;
936 
937 	/* determine the bounds of the contiguous overall allocation we must
938 	 * make */
939 	phdr = params->phdrs;
940 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
941 		if (params->phdrs[loop].p_type != PT_LOAD)
942 			continue;
943 
944 		if (base > phdr->p_vaddr)
945 			base = phdr->p_vaddr;
946 		if (top < phdr->p_vaddr + phdr->p_memsz)
947 			top = phdr->p_vaddr + phdr->p_memsz;
948 	}
949 
950 	/* allocate one big anon block for everything */
951 	mflags = MAP_PRIVATE;
952 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
953 		mflags |= MAP_EXECUTABLE;
954 
955 	maddr = vm_mmap(NULL, load_addr, top - base,
956 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
957 	if (IS_ERR_VALUE(maddr))
958 		return (int) maddr;
959 
960 	if (load_addr != 0)
961 		load_addr += PAGE_ALIGN(top - base);
962 
963 	/* and then load the file segments into it */
964 	phdr = params->phdrs;
965 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
966 		if (params->phdrs[loop].p_type != PT_LOAD)
967 			continue;
968 
969 		seg->addr = maddr + (phdr->p_vaddr - base);
970 		seg->p_vaddr = phdr->p_vaddr;
971 		seg->p_memsz = phdr->p_memsz;
972 
973 		ret = read_code(file, seg->addr, phdr->p_offset,
974 				       phdr->p_filesz);
975 		if (ret < 0)
976 			return ret;
977 
978 		/* map the ELF header address if in this segment */
979 		if (phdr->p_offset == 0)
980 			params->elfhdr_addr = seg->addr;
981 
982 		/* clear any space allocated but not loaded */
983 		if (phdr->p_filesz < phdr->p_memsz) {
984 			if (clear_user((void *) (seg->addr + phdr->p_filesz),
985 				       phdr->p_memsz - phdr->p_filesz))
986 				return -EFAULT;
987 		}
988 
989 		if (mm) {
990 			if (phdr->p_flags & PF_X) {
991 				if (!mm->start_code) {
992 					mm->start_code = seg->addr;
993 					mm->end_code = seg->addr +
994 						phdr->p_memsz;
995 				}
996 			} else if (!mm->start_data) {
997 				mm->start_data = seg->addr;
998 				mm->end_data = seg->addr + phdr->p_memsz;
999 			}
1000 		}
1001 
1002 		seg++;
1003 	}
1004 
1005 	return 0;
1006 }
1007 #endif
1008 
1009 /*****************************************************************************/
1010 /*
1011  * map a binary by direct mmap() of the individual PT_LOAD segments
1012  */
1013 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1014 					     struct file *file,
1015 					     struct mm_struct *mm)
1016 {
1017 	struct elf32_fdpic_loadseg *seg;
1018 	struct elf32_phdr *phdr;
1019 	unsigned long load_addr, delta_vaddr;
1020 	int loop, dvset;
1021 
1022 	load_addr = params->load_addr;
1023 	delta_vaddr = 0;
1024 	dvset = 0;
1025 
1026 	seg = params->loadmap->segs;
1027 
1028 	/* deal with each load segment separately */
1029 	phdr = params->phdrs;
1030 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1031 		unsigned long maddr, disp, excess, excess1;
1032 		int prot = 0, flags;
1033 
1034 		if (phdr->p_type != PT_LOAD)
1035 			continue;
1036 
1037 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1038 		       (unsigned long) phdr->p_vaddr,
1039 		       (unsigned long) phdr->p_offset,
1040 		       (unsigned long) phdr->p_filesz,
1041 		       (unsigned long) phdr->p_memsz);
1042 
1043 		/* determine the mapping parameters */
1044 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1045 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1046 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1047 
1048 		flags = MAP_PRIVATE | MAP_DENYWRITE;
1049 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1050 			flags |= MAP_EXECUTABLE;
1051 
1052 		maddr = 0;
1053 
1054 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1055 		case ELF_FDPIC_FLAG_INDEPENDENT:
1056 			/* PT_LOADs are independently locatable */
1057 			break;
1058 
1059 		case ELF_FDPIC_FLAG_HONOURVADDR:
1060 			/* the specified virtual address must be honoured */
1061 			maddr = phdr->p_vaddr;
1062 			flags |= MAP_FIXED;
1063 			break;
1064 
1065 		case ELF_FDPIC_FLAG_CONSTDISP:
1066 			/* constant displacement
1067 			 * - can be mapped anywhere, but must be mapped as a
1068 			 *   unit
1069 			 */
1070 			if (!dvset) {
1071 				maddr = load_addr;
1072 				delta_vaddr = phdr->p_vaddr;
1073 				dvset = 1;
1074 			} else {
1075 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1076 				flags |= MAP_FIXED;
1077 			}
1078 			break;
1079 
1080 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1081 			/* contiguity handled later */
1082 			break;
1083 
1084 		default:
1085 			BUG();
1086 		}
1087 
1088 		maddr &= PAGE_MASK;
1089 
1090 		/* create the mapping */
1091 		disp = phdr->p_vaddr & ~PAGE_MASK;
1092 		maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1093 				phdr->p_offset - disp);
1094 
1095 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1096 		       loop, phdr->p_memsz + disp, prot, flags,
1097 		       phdr->p_offset - disp, maddr);
1098 
1099 		if (IS_ERR_VALUE(maddr))
1100 			return (int) maddr;
1101 
1102 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1103 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1104 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1105 
1106 		seg->addr = maddr + disp;
1107 		seg->p_vaddr = phdr->p_vaddr;
1108 		seg->p_memsz = phdr->p_memsz;
1109 
1110 		/* map the ELF header address if in this segment */
1111 		if (phdr->p_offset == 0)
1112 			params->elfhdr_addr = seg->addr;
1113 
1114 		/* clear the bit between beginning of mapping and beginning of
1115 		 * PT_LOAD */
1116 		if (prot & PROT_WRITE && disp > 0) {
1117 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1118 			if (clear_user((void __user *) maddr, disp))
1119 				return -EFAULT;
1120 			maddr += disp;
1121 		}
1122 
1123 		/* clear any space allocated but not loaded
1124 		 * - on uClinux we can just clear the lot
1125 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1126 		 *   extant in the file
1127 		 */
1128 		excess = phdr->p_memsz - phdr->p_filesz;
1129 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1130 
1131 #ifdef CONFIG_MMU
1132 		if (excess > excess1) {
1133 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1134 			unsigned long xmaddr;
1135 
1136 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1137 			xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1138 					 prot, flags, 0);
1139 
1140 			kdebug("mmap[%d] <anon>"
1141 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1142 			       loop, xaddr, excess - excess1, prot, flags,
1143 			       xmaddr);
1144 
1145 			if (xmaddr != xaddr)
1146 				return -ENOMEM;
1147 		}
1148 
1149 		if (prot & PROT_WRITE && excess1 > 0) {
1150 			kdebug("clear[%d] ad=%lx sz=%lx",
1151 			       loop, maddr + phdr->p_filesz, excess1);
1152 			if (clear_user((void __user *) maddr + phdr->p_filesz,
1153 				       excess1))
1154 				return -EFAULT;
1155 		}
1156 
1157 #else
1158 		if (excess > 0) {
1159 			kdebug("clear[%d] ad=%lx sz=%lx",
1160 			       loop, maddr + phdr->p_filesz, excess);
1161 			if (clear_user((void *) maddr + phdr->p_filesz, excess))
1162 				return -EFAULT;
1163 		}
1164 #endif
1165 
1166 		if (mm) {
1167 			if (phdr->p_flags & PF_X) {
1168 				if (!mm->start_code) {
1169 					mm->start_code = maddr;
1170 					mm->end_code = maddr + phdr->p_memsz;
1171 				}
1172 			} else if (!mm->start_data) {
1173 				mm->start_data = maddr;
1174 				mm->end_data = maddr + phdr->p_memsz;
1175 			}
1176 		}
1177 
1178 		seg++;
1179 	}
1180 
1181 	return 0;
1182 }
1183 
1184 /*****************************************************************************/
1185 /*
1186  * ELF-FDPIC core dumper
1187  *
1188  * Modelled on fs/exec.c:aout_core_dump()
1189  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1190  *
1191  * Modelled on fs/binfmt_elf.c core dumper
1192  */
1193 #ifdef CONFIG_ELF_CORE
1194 
1195 struct elf_prstatus_fdpic
1196 {
1197 	struct elf_prstatus_common	common;
1198 	elf_gregset_t pr_reg;	/* GP registers */
1199 	/* When using FDPIC, the loadmap addresses need to be communicated
1200 	 * to GDB in order for GDB to do the necessary relocations.  The
1201 	 * fields (below) used to communicate this information are placed
1202 	 * immediately after ``pr_reg'', so that the loadmap addresses may
1203 	 * be viewed as part of the register set if so desired.
1204 	 */
1205 	unsigned long pr_exec_fdpic_loadmap;
1206 	unsigned long pr_interp_fdpic_loadmap;
1207 	int pr_fpvalid;		/* True if math co-processor being used.  */
1208 };
1209 
1210 /* An ELF note in memory */
1211 struct memelfnote
1212 {
1213 	const char *name;
1214 	int type;
1215 	unsigned int datasz;
1216 	void *data;
1217 };
1218 
1219 static int notesize(struct memelfnote *en)
1220 {
1221 	int sz;
1222 
1223 	sz = sizeof(struct elf_note);
1224 	sz += roundup(strlen(en->name) + 1, 4);
1225 	sz += roundup(en->datasz, 4);
1226 
1227 	return sz;
1228 }
1229 
1230 /* #define DEBUG */
1231 
1232 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1233 {
1234 	struct elf_note en;
1235 	en.n_namesz = strlen(men->name) + 1;
1236 	en.n_descsz = men->datasz;
1237 	en.n_type = men->type;
1238 
1239 	return dump_emit(cprm, &en, sizeof(en)) &&
1240 		dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1241 		dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1242 }
1243 
1244 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1245 {
1246 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1247 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1248 	elf->e_ident[EI_DATA] = ELF_DATA;
1249 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1250 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1251 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1252 
1253 	elf->e_type = ET_CORE;
1254 	elf->e_machine = ELF_ARCH;
1255 	elf->e_version = EV_CURRENT;
1256 	elf->e_entry = 0;
1257 	elf->e_phoff = sizeof(struct elfhdr);
1258 	elf->e_shoff = 0;
1259 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1260 	elf->e_ehsize = sizeof(struct elfhdr);
1261 	elf->e_phentsize = sizeof(struct elf_phdr);
1262 	elf->e_phnum = segs;
1263 	elf->e_shentsize = 0;
1264 	elf->e_shnum = 0;
1265 	elf->e_shstrndx = 0;
1266 	return;
1267 }
1268 
1269 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1270 {
1271 	phdr->p_type = PT_NOTE;
1272 	phdr->p_offset = offset;
1273 	phdr->p_vaddr = 0;
1274 	phdr->p_paddr = 0;
1275 	phdr->p_filesz = sz;
1276 	phdr->p_memsz = 0;
1277 	phdr->p_flags = 0;
1278 	phdr->p_align = 0;
1279 	return;
1280 }
1281 
1282 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1283 		unsigned int sz, void *data)
1284 {
1285 	note->name = name;
1286 	note->type = type;
1287 	note->datasz = sz;
1288 	note->data = data;
1289 	return;
1290 }
1291 
1292 /*
1293  * fill up all the fields in prstatus from the given task struct, except
1294  * registers which need to be filled up separately.
1295  */
1296 static void fill_prstatus(struct elf_prstatus_common *prstatus,
1297 			  struct task_struct *p, long signr)
1298 {
1299 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1300 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1301 	prstatus->pr_sighold = p->blocked.sig[0];
1302 	rcu_read_lock();
1303 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1304 	rcu_read_unlock();
1305 	prstatus->pr_pid = task_pid_vnr(p);
1306 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1307 	prstatus->pr_sid = task_session_vnr(p);
1308 	if (thread_group_leader(p)) {
1309 		struct task_cputime cputime;
1310 
1311 		/*
1312 		 * This is the record for the group leader.  It shows the
1313 		 * group-wide total, not its individual thread total.
1314 		 */
1315 		thread_group_cputime(p, &cputime);
1316 		prstatus->pr_utime = ns_to_kernel_old_timeval(cputime.utime);
1317 		prstatus->pr_stime = ns_to_kernel_old_timeval(cputime.stime);
1318 	} else {
1319 		u64 utime, stime;
1320 
1321 		task_cputime(p, &utime, &stime);
1322 		prstatus->pr_utime = ns_to_kernel_old_timeval(utime);
1323 		prstatus->pr_stime = ns_to_kernel_old_timeval(stime);
1324 	}
1325 	prstatus->pr_cutime = ns_to_kernel_old_timeval(p->signal->cutime);
1326 	prstatus->pr_cstime = ns_to_kernel_old_timeval(p->signal->cstime);
1327 }
1328 
1329 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1330 		       struct mm_struct *mm)
1331 {
1332 	const struct cred *cred;
1333 	unsigned int i, len;
1334 
1335 	/* first copy the parameters from user space */
1336 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1337 
1338 	len = mm->arg_end - mm->arg_start;
1339 	if (len >= ELF_PRARGSZ)
1340 		len = ELF_PRARGSZ - 1;
1341 	if (copy_from_user(&psinfo->pr_psargs,
1342 		           (const char __user *) mm->arg_start, len))
1343 		return -EFAULT;
1344 	for (i = 0; i < len; i++)
1345 		if (psinfo->pr_psargs[i] == 0)
1346 			psinfo->pr_psargs[i] = ' ';
1347 	psinfo->pr_psargs[len] = 0;
1348 
1349 	rcu_read_lock();
1350 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1351 	rcu_read_unlock();
1352 	psinfo->pr_pid = task_pid_vnr(p);
1353 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1354 	psinfo->pr_sid = task_session_vnr(p);
1355 
1356 	i = p->state ? ffz(~p->state) + 1 : 0;
1357 	psinfo->pr_state = i;
1358 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1359 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1360 	psinfo->pr_nice = task_nice(p);
1361 	psinfo->pr_flag = p->flags;
1362 	rcu_read_lock();
1363 	cred = __task_cred(p);
1364 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1365 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1366 	rcu_read_unlock();
1367 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1368 
1369 	return 0;
1370 }
1371 
1372 /* Here is the structure in which status of each thread is captured. */
1373 struct elf_thread_status
1374 {
1375 	struct elf_thread_status *next;
1376 	struct elf_prstatus_fdpic prstatus;	/* NT_PRSTATUS */
1377 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1378 	struct memelfnote notes[2];
1379 	int num_notes;
1380 };
1381 
1382 /*
1383  * In order to add the specific thread information for the elf file format,
1384  * we need to keep a linked list of every thread's pr_status and then create
1385  * a single section for them in the final core file.
1386  */
1387 static struct elf_thread_status *elf_dump_thread_status(long signr, struct task_struct *p, int *sz)
1388 {
1389 	const struct user_regset_view *view = task_user_regset_view(p);
1390 	struct elf_thread_status *t;
1391 	int i, ret;
1392 
1393 	t = kzalloc(sizeof(struct elf_thread_status), GFP_KERNEL);
1394 	if (!t)
1395 		return t;
1396 
1397 	fill_prstatus(&t->prstatus.common, p, signr);
1398 	t->prstatus.pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1399 	t->prstatus.pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1400 	regset_get(p, &view->regsets[0],
1401 		   sizeof(t->prstatus.pr_reg), &t->prstatus.pr_reg);
1402 
1403 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1404 		  &t->prstatus);
1405 	t->num_notes++;
1406 	*sz += notesize(&t->notes[0]);
1407 
1408 	for (i = 1; i < view->n; ++i) {
1409 		const struct user_regset *regset = &view->regsets[i];
1410 		if (regset->core_note_type != NT_PRFPREG)
1411 			continue;
1412 		if (regset->active && regset->active(p, regset) <= 0)
1413 			continue;
1414 		ret = regset_get(p, regset, sizeof(t->fpu), &t->fpu);
1415 		if (ret >= 0)
1416 			t->prstatus.pr_fpvalid = 1;
1417 		break;
1418 	}
1419 
1420 	if (t->prstatus.pr_fpvalid) {
1421 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1422 			  &t->fpu);
1423 		t->num_notes++;
1424 		*sz += notesize(&t->notes[1]);
1425 	}
1426 	return t;
1427 }
1428 
1429 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1430 			     elf_addr_t e_shoff, int segs)
1431 {
1432 	elf->e_shoff = e_shoff;
1433 	elf->e_shentsize = sizeof(*shdr4extnum);
1434 	elf->e_shnum = 1;
1435 	elf->e_shstrndx = SHN_UNDEF;
1436 
1437 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1438 
1439 	shdr4extnum->sh_type = SHT_NULL;
1440 	shdr4extnum->sh_size = elf->e_shnum;
1441 	shdr4extnum->sh_link = elf->e_shstrndx;
1442 	shdr4extnum->sh_info = segs;
1443 }
1444 
1445 /*
1446  * dump the segments for an MMU process
1447  */
1448 static bool elf_fdpic_dump_segments(struct coredump_params *cprm,
1449 				    struct core_vma_metadata *vma_meta,
1450 				    int vma_count)
1451 {
1452 	int i;
1453 
1454 	for (i = 0; i < vma_count; i++) {
1455 		struct core_vma_metadata *meta = vma_meta + i;
1456 
1457 		if (!dump_user_range(cprm, meta->start, meta->dump_size))
1458 			return false;
1459 	}
1460 	return true;
1461 }
1462 
1463 /*
1464  * Actual dumper
1465  *
1466  * This is a two-pass process; first we find the offsets of the bits,
1467  * and then they are actually written out.  If we run out of core limit
1468  * we just truncate.
1469  */
1470 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1471 {
1472 	int has_dumped = 0;
1473 	int vma_count, segs;
1474 	int i;
1475 	struct elfhdr *elf = NULL;
1476 	loff_t offset = 0, dataoff;
1477 	struct memelfnote psinfo_note, auxv_note;
1478 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1479 	struct elf_thread_status *thread_list = NULL;
1480 	int thread_status_size = 0;
1481 	elf_addr_t *auxv;
1482 	struct elf_phdr *phdr4note = NULL;
1483 	struct elf_shdr *shdr4extnum = NULL;
1484 	Elf_Half e_phnum;
1485 	elf_addr_t e_shoff;
1486 	struct core_thread *ct;
1487 	struct elf_thread_status *tmp;
1488 	struct core_vma_metadata *vma_meta = NULL;
1489 	size_t vma_data_size;
1490 
1491 	/* alloc memory for large data structures: too large to be on stack */
1492 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1493 	if (!elf)
1494 		goto end_coredump;
1495 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1496 	if (!psinfo)
1497 		goto end_coredump;
1498 
1499 	if (dump_vma_snapshot(cprm, &vma_count, &vma_meta, &vma_data_size))
1500 		goto end_coredump;
1501 
1502 	for (ct = current->mm->core_state->dumper.next;
1503 					ct; ct = ct->next) {
1504 		tmp = elf_dump_thread_status(cprm->siginfo->si_signo,
1505 					     ct->task, &thread_status_size);
1506 		if (!tmp)
1507 			goto end_coredump;
1508 
1509 		tmp->next = thread_list;
1510 		thread_list = tmp;
1511 	}
1512 
1513 	/* now collect the dump for the current */
1514 	tmp = elf_dump_thread_status(cprm->siginfo->si_signo,
1515 				     current, &thread_status_size);
1516 	if (!tmp)
1517 		goto end_coredump;
1518 	tmp->next = thread_list;
1519 	thread_list = tmp;
1520 
1521 	segs = vma_count + elf_core_extra_phdrs();
1522 
1523 	/* for notes section */
1524 	segs++;
1525 
1526 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1527 	 * this, kernel supports extended numbering. Have a look at
1528 	 * include/linux/elf.h for further information. */
1529 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1530 
1531 	/* Set up header */
1532 	fill_elf_fdpic_header(elf, e_phnum);
1533 
1534 	has_dumped = 1;
1535 	/*
1536 	 * Set up the notes in similar form to SVR4 core dumps made
1537 	 * with info from their /proc.
1538 	 */
1539 
1540 	fill_psinfo(psinfo, current->group_leader, current->mm);
1541 	fill_note(&psinfo_note, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1542 	thread_status_size += notesize(&psinfo_note);
1543 
1544 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1545 	i = 0;
1546 	do
1547 		i += 2;
1548 	while (auxv[i - 2] != AT_NULL);
1549 	fill_note(&auxv_note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1550 	thread_status_size += notesize(&auxv_note);
1551 
1552 	offset = sizeof(*elf);				/* Elf header */
1553 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1554 
1555 	/* Write notes phdr entry */
1556 	phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1557 	if (!phdr4note)
1558 		goto end_coredump;
1559 
1560 	fill_elf_note_phdr(phdr4note, thread_status_size, offset);
1561 	offset += thread_status_size;
1562 
1563 	/* Page-align dumped data */
1564 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1565 
1566 	offset += vma_data_size;
1567 	offset += elf_core_extra_data_size();
1568 	e_shoff = offset;
1569 
1570 	if (e_phnum == PN_XNUM) {
1571 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1572 		if (!shdr4extnum)
1573 			goto end_coredump;
1574 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1575 	}
1576 
1577 	offset = dataoff;
1578 
1579 	if (!dump_emit(cprm, elf, sizeof(*elf)))
1580 		goto end_coredump;
1581 
1582 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1583 		goto end_coredump;
1584 
1585 	/* write program headers for segments dump */
1586 	for (i = 0; i < vma_count; i++) {
1587 		struct core_vma_metadata *meta = vma_meta + i;
1588 		struct elf_phdr phdr;
1589 		size_t sz;
1590 
1591 		sz = meta->end - meta->start;
1592 
1593 		phdr.p_type = PT_LOAD;
1594 		phdr.p_offset = offset;
1595 		phdr.p_vaddr = meta->start;
1596 		phdr.p_paddr = 0;
1597 		phdr.p_filesz = meta->dump_size;
1598 		phdr.p_memsz = sz;
1599 		offset += phdr.p_filesz;
1600 		phdr.p_flags = 0;
1601 		if (meta->flags & VM_READ)
1602 			phdr.p_flags |= PF_R;
1603 		if (meta->flags & VM_WRITE)
1604 			phdr.p_flags |= PF_W;
1605 		if (meta->flags & VM_EXEC)
1606 			phdr.p_flags |= PF_X;
1607 		phdr.p_align = ELF_EXEC_PAGESIZE;
1608 
1609 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1610 			goto end_coredump;
1611 	}
1612 
1613 	if (!elf_core_write_extra_phdrs(cprm, offset))
1614 		goto end_coredump;
1615 
1616  	/* write out the notes section */
1617 	if (!writenote(thread_list->notes, cprm))
1618 		goto end_coredump;
1619 	if (!writenote(&psinfo_note, cprm))
1620 		goto end_coredump;
1621 	if (!writenote(&auxv_note, cprm))
1622 		goto end_coredump;
1623 	for (i = 1; i < thread_list->num_notes; i++)
1624 		if (!writenote(thread_list->notes + i, cprm))
1625 			goto end_coredump;
1626 
1627 	/* write out the thread status notes section */
1628 	for (tmp = thread_list->next; tmp; tmp = tmp->next) {
1629 		for (i = 0; i < tmp->num_notes; i++)
1630 			if (!writenote(&tmp->notes[i], cprm))
1631 				goto end_coredump;
1632 	}
1633 
1634 	dump_skip_to(cprm, dataoff);
1635 
1636 	if (!elf_fdpic_dump_segments(cprm, vma_meta, vma_count))
1637 		goto end_coredump;
1638 
1639 	if (!elf_core_write_extra_data(cprm))
1640 		goto end_coredump;
1641 
1642 	if (e_phnum == PN_XNUM) {
1643 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1644 			goto end_coredump;
1645 	}
1646 
1647 	if (cprm->file->f_pos != offset) {
1648 		/* Sanity check */
1649 		printk(KERN_WARNING
1650 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1651 		       cprm->file->f_pos, offset);
1652 	}
1653 
1654 end_coredump:
1655 	while (thread_list) {
1656 		tmp = thread_list;
1657 		thread_list = thread_list->next;
1658 		kfree(tmp);
1659 	}
1660 	kvfree(vma_meta);
1661 	kfree(phdr4note);
1662 	kfree(elf);
1663 	kfree(psinfo);
1664 	kfree(shdr4extnum);
1665 	return has_dumped;
1666 }
1667 
1668 #endif		/* CONFIG_ELF_CORE */
1669