xref: /linux/fs/binfmt_elf_fdpic.c (revision 9a379e77033f02c4a071891afdf0f0a01eff8ccb)
1 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
2  *
3  * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  * Derived from binfmt_elf.c
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License
9  * as published by the Free Software Foundation; either version
10  * 2 of the License, or (at your option) any later version.
11  */
12 
13 #include <linux/module.h>
14 
15 #include <linux/fs.h>
16 #include <linux/stat.h>
17 #include <linux/sched.h>
18 #include <linux/sched/coredump.h>
19 #include <linux/sched/task_stack.h>
20 #include <linux/sched/cputime.h>
21 #include <linux/mm.h>
22 #include <linux/mman.h>
23 #include <linux/errno.h>
24 #include <linux/signal.h>
25 #include <linux/binfmts.h>
26 #include <linux/string.h>
27 #include <linux/file.h>
28 #include <linux/fcntl.h>
29 #include <linux/slab.h>
30 #include <linux/pagemap.h>
31 #include <linux/security.h>
32 #include <linux/highmem.h>
33 #include <linux/highuid.h>
34 #include <linux/personality.h>
35 #include <linux/ptrace.h>
36 #include <linux/init.h>
37 #include <linux/elf.h>
38 #include <linux/elf-fdpic.h>
39 #include <linux/elfcore.h>
40 #include <linux/coredump.h>
41 #include <linux/dax.h>
42 
43 #include <linux/uaccess.h>
44 #include <asm/param.h>
45 #include <asm/pgalloc.h>
46 
47 typedef char *elf_caddr_t;
48 
49 #if 0
50 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
51 #else
52 #define kdebug(fmt, ...) do {} while(0)
53 #endif
54 
55 #if 0
56 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
57 #else
58 #define kdcore(fmt, ...) do {} while(0)
59 #endif
60 
61 MODULE_LICENSE("GPL");
62 
63 static int load_elf_fdpic_binary(struct linux_binprm *);
64 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
65 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
66 			      struct mm_struct *, const char *);
67 
68 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
69 				   struct elf_fdpic_params *,
70 				   struct elf_fdpic_params *);
71 
72 #ifndef CONFIG_MMU
73 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
74 						   struct file *,
75 						   struct mm_struct *);
76 #endif
77 
78 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
79 					     struct file *, struct mm_struct *);
80 
81 #ifdef CONFIG_ELF_CORE
82 static int elf_fdpic_core_dump(struct coredump_params *cprm);
83 #endif
84 
85 static struct linux_binfmt elf_fdpic_format = {
86 	.module		= THIS_MODULE,
87 	.load_binary	= load_elf_fdpic_binary,
88 #ifdef CONFIG_ELF_CORE
89 	.core_dump	= elf_fdpic_core_dump,
90 #endif
91 	.min_coredump	= ELF_EXEC_PAGESIZE,
92 };
93 
94 static int __init init_elf_fdpic_binfmt(void)
95 {
96 	register_binfmt(&elf_fdpic_format);
97 	return 0;
98 }
99 
100 static void __exit exit_elf_fdpic_binfmt(void)
101 {
102 	unregister_binfmt(&elf_fdpic_format);
103 }
104 
105 core_initcall(init_elf_fdpic_binfmt);
106 module_exit(exit_elf_fdpic_binfmt);
107 
108 static int is_elf(struct elfhdr *hdr, struct file *file)
109 {
110 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
111 		return 0;
112 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
113 		return 0;
114 	if (!elf_check_arch(hdr))
115 		return 0;
116 	if (!file->f_op->mmap)
117 		return 0;
118 	return 1;
119 }
120 
121 #ifndef elf_check_fdpic
122 #define elf_check_fdpic(x) 0
123 #endif
124 
125 #ifndef elf_check_const_displacement
126 #define elf_check_const_displacement(x) 0
127 #endif
128 
129 static int is_constdisp(struct elfhdr *hdr)
130 {
131 	if (!elf_check_fdpic(hdr))
132 		return 1;
133 	if (elf_check_const_displacement(hdr))
134 		return 1;
135 	return 0;
136 }
137 
138 /*****************************************************************************/
139 /*
140  * read the program headers table into memory
141  */
142 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
143 				 struct file *file)
144 {
145 	struct elf32_phdr *phdr;
146 	unsigned long size;
147 	int retval, loop;
148 	loff_t pos = params->hdr.e_phoff;
149 
150 	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
151 		return -ENOMEM;
152 	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
153 		return -ENOMEM;
154 
155 	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
156 	params->phdrs = kmalloc(size, GFP_KERNEL);
157 	if (!params->phdrs)
158 		return -ENOMEM;
159 
160 	retval = kernel_read(file, params->phdrs, size, &pos);
161 	if (unlikely(retval != size))
162 		return retval < 0 ? retval : -ENOEXEC;
163 
164 	/* determine stack size for this binary */
165 	phdr = params->phdrs;
166 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
167 		if (phdr->p_type != PT_GNU_STACK)
168 			continue;
169 
170 		if (phdr->p_flags & PF_X)
171 			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
172 		else
173 			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
174 
175 		params->stack_size = phdr->p_memsz;
176 		break;
177 	}
178 
179 	return 0;
180 }
181 
182 /*****************************************************************************/
183 /*
184  * load an fdpic binary into various bits of memory
185  */
186 static int load_elf_fdpic_binary(struct linux_binprm *bprm)
187 {
188 	struct elf_fdpic_params exec_params, interp_params;
189 	struct pt_regs *regs = current_pt_regs();
190 	struct elf_phdr *phdr;
191 	unsigned long stack_size, entryaddr;
192 #ifdef ELF_FDPIC_PLAT_INIT
193 	unsigned long dynaddr;
194 #endif
195 #ifndef CONFIG_MMU
196 	unsigned long stack_prot;
197 #endif
198 	struct file *interpreter = NULL; /* to shut gcc up */
199 	char *interpreter_name = NULL;
200 	int executable_stack;
201 	int retval, i;
202 	loff_t pos;
203 
204 	kdebug("____ LOAD %d ____", current->pid);
205 
206 	memset(&exec_params, 0, sizeof(exec_params));
207 	memset(&interp_params, 0, sizeof(interp_params));
208 
209 	exec_params.hdr = *(struct elfhdr *) bprm->buf;
210 	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
211 
212 	/* check that this is a binary we know how to deal with */
213 	retval = -ENOEXEC;
214 	if (!is_elf(&exec_params.hdr, bprm->file))
215 		goto error;
216 	if (!elf_check_fdpic(&exec_params.hdr)) {
217 #ifdef CONFIG_MMU
218 		/* binfmt_elf handles non-fdpic elf except on nommu */
219 		goto error;
220 #else
221 		/* nommu can only load ET_DYN (PIE) ELF */
222 		if (exec_params.hdr.e_type != ET_DYN)
223 			goto error;
224 #endif
225 	}
226 
227 	/* read the program header table */
228 	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
229 	if (retval < 0)
230 		goto error;
231 
232 	/* scan for a program header that specifies an interpreter */
233 	phdr = exec_params.phdrs;
234 
235 	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
236 		switch (phdr->p_type) {
237 		case PT_INTERP:
238 			retval = -ENOMEM;
239 			if (phdr->p_filesz > PATH_MAX)
240 				goto error;
241 			retval = -ENOENT;
242 			if (phdr->p_filesz < 2)
243 				goto error;
244 
245 			/* read the name of the interpreter into memory */
246 			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
247 			if (!interpreter_name)
248 				goto error;
249 
250 			pos = phdr->p_offset;
251 			retval = kernel_read(bprm->file, interpreter_name,
252 					     phdr->p_filesz, &pos);
253 			if (unlikely(retval != phdr->p_filesz)) {
254 				if (retval >= 0)
255 					retval = -ENOEXEC;
256 				goto error;
257 			}
258 
259 			retval = -ENOENT;
260 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
261 				goto error;
262 
263 			kdebug("Using ELF interpreter %s", interpreter_name);
264 
265 			/* replace the program with the interpreter */
266 			interpreter = open_exec(interpreter_name);
267 			retval = PTR_ERR(interpreter);
268 			if (IS_ERR(interpreter)) {
269 				interpreter = NULL;
270 				goto error;
271 			}
272 
273 			/*
274 			 * If the binary is not readable then enforce
275 			 * mm->dumpable = 0 regardless of the interpreter's
276 			 * permissions.
277 			 */
278 			would_dump(bprm, interpreter);
279 
280 			pos = 0;
281 			retval = kernel_read(interpreter, bprm->buf,
282 					BINPRM_BUF_SIZE, &pos);
283 			if (unlikely(retval != BINPRM_BUF_SIZE)) {
284 				if (retval >= 0)
285 					retval = -ENOEXEC;
286 				goto error;
287 			}
288 
289 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
290 			break;
291 
292 		case PT_LOAD:
293 #ifdef CONFIG_MMU
294 			if (exec_params.load_addr == 0)
295 				exec_params.load_addr = phdr->p_vaddr;
296 #endif
297 			break;
298 		}
299 
300 	}
301 
302 	if (is_constdisp(&exec_params.hdr))
303 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
304 
305 	/* perform insanity checks on the interpreter */
306 	if (interpreter_name) {
307 		retval = -ELIBBAD;
308 		if (!is_elf(&interp_params.hdr, interpreter))
309 			goto error;
310 
311 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
312 
313 		/* read the interpreter's program header table */
314 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
315 		if (retval < 0)
316 			goto error;
317 	}
318 
319 	stack_size = exec_params.stack_size;
320 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
321 		executable_stack = EXSTACK_ENABLE_X;
322 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
323 		executable_stack = EXSTACK_DISABLE_X;
324 	else
325 		executable_stack = EXSTACK_DEFAULT;
326 
327 	if (stack_size == 0) {
328 		stack_size = interp_params.stack_size;
329 		if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
330 			executable_stack = EXSTACK_ENABLE_X;
331 		else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
332 			executable_stack = EXSTACK_DISABLE_X;
333 		else
334 			executable_stack = EXSTACK_DEFAULT;
335 	}
336 
337 	retval = -ENOEXEC;
338 	if (stack_size == 0)
339 		stack_size = 131072UL; /* same as exec.c's default commit */
340 
341 	if (is_constdisp(&interp_params.hdr))
342 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
343 
344 	/* flush all traces of the currently running executable */
345 	retval = flush_old_exec(bprm);
346 	if (retval)
347 		goto error;
348 
349 	/* there's now no turning back... the old userspace image is dead,
350 	 * defunct, deceased, etc.
351 	 */
352 	if (elf_check_fdpic(&exec_params.hdr))
353 		set_personality(PER_LINUX_FDPIC);
354 	else
355 		set_personality(PER_LINUX);
356 	if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
357 		current->personality |= READ_IMPLIES_EXEC;
358 
359 	setup_new_exec(bprm);
360 
361 	set_binfmt(&elf_fdpic_format);
362 
363 	current->mm->start_code = 0;
364 	current->mm->end_code = 0;
365 	current->mm->start_stack = 0;
366 	current->mm->start_data = 0;
367 	current->mm->end_data = 0;
368 	current->mm->context.exec_fdpic_loadmap = 0;
369 	current->mm->context.interp_fdpic_loadmap = 0;
370 
371 #ifdef CONFIG_MMU
372 	elf_fdpic_arch_lay_out_mm(&exec_params,
373 				  &interp_params,
374 				  &current->mm->start_stack,
375 				  &current->mm->start_brk);
376 
377 	retval = setup_arg_pages(bprm, current->mm->start_stack,
378 				 executable_stack);
379 	if (retval < 0)
380 		goto error;
381 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
382 	retval = arch_setup_additional_pages(bprm, !!interpreter_name);
383 	if (retval < 0)
384 		goto error;
385 #endif
386 #endif
387 
388 	/* load the executable and interpreter into memory */
389 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
390 				    "executable");
391 	if (retval < 0)
392 		goto error;
393 
394 	if (interpreter_name) {
395 		retval = elf_fdpic_map_file(&interp_params, interpreter,
396 					    current->mm, "interpreter");
397 		if (retval < 0) {
398 			printk(KERN_ERR "Unable to load interpreter\n");
399 			goto error;
400 		}
401 
402 		allow_write_access(interpreter);
403 		fput(interpreter);
404 		interpreter = NULL;
405 	}
406 
407 #ifdef CONFIG_MMU
408 	if (!current->mm->start_brk)
409 		current->mm->start_brk = current->mm->end_data;
410 
411 	current->mm->brk = current->mm->start_brk =
412 		PAGE_ALIGN(current->mm->start_brk);
413 
414 #else
415 	/* create a stack area and zero-size brk area */
416 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
417 	if (stack_size < PAGE_SIZE * 2)
418 		stack_size = PAGE_SIZE * 2;
419 
420 	stack_prot = PROT_READ | PROT_WRITE;
421 	if (executable_stack == EXSTACK_ENABLE_X ||
422 	    (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
423 		stack_prot |= PROT_EXEC;
424 
425 	current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
426 					 MAP_PRIVATE | MAP_ANONYMOUS |
427 					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
428 					 0);
429 
430 	if (IS_ERR_VALUE(current->mm->start_brk)) {
431 		retval = current->mm->start_brk;
432 		current->mm->start_brk = 0;
433 		goto error;
434 	}
435 
436 	current->mm->brk = current->mm->start_brk;
437 	current->mm->context.end_brk = current->mm->start_brk;
438 	current->mm->start_stack = current->mm->start_brk + stack_size;
439 #endif
440 
441 	install_exec_creds(bprm);
442 	if (create_elf_fdpic_tables(bprm, current->mm,
443 				    &exec_params, &interp_params) < 0)
444 		goto error;
445 
446 	kdebug("- start_code  %lx", current->mm->start_code);
447 	kdebug("- end_code    %lx", current->mm->end_code);
448 	kdebug("- start_data  %lx", current->mm->start_data);
449 	kdebug("- end_data    %lx", current->mm->end_data);
450 	kdebug("- start_brk   %lx", current->mm->start_brk);
451 	kdebug("- brk         %lx", current->mm->brk);
452 	kdebug("- start_stack %lx", current->mm->start_stack);
453 
454 #ifdef ELF_FDPIC_PLAT_INIT
455 	/*
456 	 * The ABI may specify that certain registers be set up in special
457 	 * ways (on i386 %edx is the address of a DT_FINI function, for
458 	 * example.  This macro performs whatever initialization to
459 	 * the regs structure is required.
460 	 */
461 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
462 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
463 			    dynaddr);
464 #endif
465 
466 	/* everything is now ready... get the userspace context ready to roll */
467 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
468 	start_thread(regs, entryaddr, current->mm->start_stack);
469 
470 	retval = 0;
471 
472 error:
473 	if (interpreter) {
474 		allow_write_access(interpreter);
475 		fput(interpreter);
476 	}
477 	kfree(interpreter_name);
478 	kfree(exec_params.phdrs);
479 	kfree(exec_params.loadmap);
480 	kfree(interp_params.phdrs);
481 	kfree(interp_params.loadmap);
482 	return retval;
483 }
484 
485 /*****************************************************************************/
486 
487 #ifndef ELF_BASE_PLATFORM
488 /*
489  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
490  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
491  * will be copied to the user stack in the same manner as AT_PLATFORM.
492  */
493 #define ELF_BASE_PLATFORM NULL
494 #endif
495 
496 /*
497  * present useful information to the program by shovelling it onto the new
498  * process's stack
499  */
500 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
501 				   struct mm_struct *mm,
502 				   struct elf_fdpic_params *exec_params,
503 				   struct elf_fdpic_params *interp_params)
504 {
505 	const struct cred *cred = current_cred();
506 	unsigned long sp, csp, nitems;
507 	elf_caddr_t __user *argv, *envp;
508 	size_t platform_len = 0, len;
509 	char *k_platform, *k_base_platform;
510 	char __user *u_platform, *u_base_platform, *p;
511 	int loop;
512 	int nr;	/* reset for each csp adjustment */
513 
514 #ifdef CONFIG_MMU
515 	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
516 	 * by the processes running on the same package. One thing we can do is
517 	 * to shuffle the initial stack for them, so we give the architecture
518 	 * an opportunity to do so here.
519 	 */
520 	sp = arch_align_stack(bprm->p);
521 #else
522 	sp = mm->start_stack;
523 
524 	/* stack the program arguments and environment */
525 	if (transfer_args_to_stack(bprm, &sp) < 0)
526 		return -EFAULT;
527 	sp &= ~15;
528 #endif
529 
530 	/*
531 	 * If this architecture has a platform capability string, copy it
532 	 * to userspace.  In some cases (Sparc), this info is impossible
533 	 * for userspace to get any other way, in others (i386) it is
534 	 * merely difficult.
535 	 */
536 	k_platform = ELF_PLATFORM;
537 	u_platform = NULL;
538 
539 	if (k_platform) {
540 		platform_len = strlen(k_platform) + 1;
541 		sp -= platform_len;
542 		u_platform = (char __user *) sp;
543 		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
544 			return -EFAULT;
545 	}
546 
547 	/*
548 	 * If this architecture has a "base" platform capability
549 	 * string, copy it to userspace.
550 	 */
551 	k_base_platform = ELF_BASE_PLATFORM;
552 	u_base_platform = NULL;
553 
554 	if (k_base_platform) {
555 		platform_len = strlen(k_base_platform) + 1;
556 		sp -= platform_len;
557 		u_base_platform = (char __user *) sp;
558 		if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
559 			return -EFAULT;
560 	}
561 
562 	sp &= ~7UL;
563 
564 	/* stack the load map(s) */
565 	len = sizeof(struct elf32_fdpic_loadmap);
566 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
567 	sp = (sp - len) & ~7UL;
568 	exec_params->map_addr = sp;
569 
570 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
571 		return -EFAULT;
572 
573 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
574 
575 	if (interp_params->loadmap) {
576 		len = sizeof(struct elf32_fdpic_loadmap);
577 		len += sizeof(struct elf32_fdpic_loadseg) *
578 			interp_params->loadmap->nsegs;
579 		sp = (sp - len) & ~7UL;
580 		interp_params->map_addr = sp;
581 
582 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
583 				 len) != 0)
584 			return -EFAULT;
585 
586 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
587 	}
588 
589 	/* force 16 byte _final_ alignment here for generality */
590 #define DLINFO_ITEMS 15
591 
592 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
593 		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
594 
595 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
596 		nitems++;
597 
598 	csp = sp;
599 	sp -= nitems * 2 * sizeof(unsigned long);
600 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
601 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
602 	sp -= 1 * sizeof(unsigned long);		/* argc */
603 
604 	csp -= sp & 15UL;
605 	sp -= sp & 15UL;
606 
607 	/* put the ELF interpreter info on the stack */
608 #define NEW_AUX_ENT(id, val)						\
609 	do {								\
610 		struct { unsigned long _id, _val; } __user *ent;	\
611 									\
612 		ent = (void __user *) csp;				\
613 		__put_user((id), &ent[nr]._id);				\
614 		__put_user((val), &ent[nr]._val);			\
615 		nr++;							\
616 	} while (0)
617 
618 	nr = 0;
619 	csp -= 2 * sizeof(unsigned long);
620 	NEW_AUX_ENT(AT_NULL, 0);
621 	if (k_platform) {
622 		nr = 0;
623 		csp -= 2 * sizeof(unsigned long);
624 		NEW_AUX_ENT(AT_PLATFORM,
625 			    (elf_addr_t) (unsigned long) u_platform);
626 	}
627 
628 	if (k_base_platform) {
629 		nr = 0;
630 		csp -= 2 * sizeof(unsigned long);
631 		NEW_AUX_ENT(AT_BASE_PLATFORM,
632 			    (elf_addr_t) (unsigned long) u_base_platform);
633 	}
634 
635 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
636 		nr = 0;
637 		csp -= 2 * sizeof(unsigned long);
638 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
639 	}
640 
641 	nr = 0;
642 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
643 	NEW_AUX_ENT(AT_HWCAP,	ELF_HWCAP);
644 #ifdef ELF_HWCAP2
645 	NEW_AUX_ENT(AT_HWCAP2,	ELF_HWCAP2);
646 #endif
647 	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
648 	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
649 	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
650 	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
651 	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
652 	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
653 	NEW_AUX_ENT(AT_FLAGS,	0);
654 	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
655 	NEW_AUX_ENT(AT_UID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
656 	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
657 	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
658 	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
659 	NEW_AUX_ENT(AT_SECURE,	bprm->secureexec);
660 	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
661 
662 #ifdef ARCH_DLINFO
663 	nr = 0;
664 	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
665 
666 	/* ARCH_DLINFO must come last so platform specific code can enforce
667 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
668 	 */
669 	ARCH_DLINFO;
670 #endif
671 #undef NEW_AUX_ENT
672 
673 	/* allocate room for argv[] and envv[] */
674 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
675 	envp = (elf_caddr_t __user *) csp;
676 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
677 	argv = (elf_caddr_t __user *) csp;
678 
679 	/* stack argc */
680 	csp -= sizeof(unsigned long);
681 	__put_user(bprm->argc, (unsigned long __user *) csp);
682 
683 	BUG_ON(csp != sp);
684 
685 	/* fill in the argv[] array */
686 #ifdef CONFIG_MMU
687 	current->mm->arg_start = bprm->p;
688 #else
689 	current->mm->arg_start = current->mm->start_stack -
690 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
691 #endif
692 
693 	p = (char __user *) current->mm->arg_start;
694 	for (loop = bprm->argc; loop > 0; loop--) {
695 		__put_user((elf_caddr_t) p, argv++);
696 		len = strnlen_user(p, MAX_ARG_STRLEN);
697 		if (!len || len > MAX_ARG_STRLEN)
698 			return -EINVAL;
699 		p += len;
700 	}
701 	__put_user(NULL, argv);
702 	current->mm->arg_end = (unsigned long) p;
703 
704 	/* fill in the envv[] array */
705 	current->mm->env_start = (unsigned long) p;
706 	for (loop = bprm->envc; loop > 0; loop--) {
707 		__put_user((elf_caddr_t)(unsigned long) p, envp++);
708 		len = strnlen_user(p, MAX_ARG_STRLEN);
709 		if (!len || len > MAX_ARG_STRLEN)
710 			return -EINVAL;
711 		p += len;
712 	}
713 	__put_user(NULL, envp);
714 	current->mm->env_end = (unsigned long) p;
715 
716 	mm->start_stack = (unsigned long) sp;
717 	return 0;
718 }
719 
720 /*****************************************************************************/
721 /*
722  * load the appropriate binary image (executable or interpreter) into memory
723  * - we assume no MMU is available
724  * - if no other PIC bits are set in params->hdr->e_flags
725  *   - we assume that the LOADable segments in the binary are independently relocatable
726  *   - we assume R/O executable segments are shareable
727  * - else
728  *   - we assume the loadable parts of the image to require fixed displacement
729  *   - the image is not shareable
730  */
731 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
732 			      struct file *file,
733 			      struct mm_struct *mm,
734 			      const char *what)
735 {
736 	struct elf32_fdpic_loadmap *loadmap;
737 #ifdef CONFIG_MMU
738 	struct elf32_fdpic_loadseg *mseg;
739 #endif
740 	struct elf32_fdpic_loadseg *seg;
741 	struct elf32_phdr *phdr;
742 	unsigned long load_addr, stop;
743 	unsigned nloads, tmp;
744 	size_t size;
745 	int loop, ret;
746 
747 	/* allocate a load map table */
748 	nloads = 0;
749 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
750 		if (params->phdrs[loop].p_type == PT_LOAD)
751 			nloads++;
752 
753 	if (nloads == 0)
754 		return -ELIBBAD;
755 
756 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
757 	loadmap = kzalloc(size, GFP_KERNEL);
758 	if (!loadmap)
759 		return -ENOMEM;
760 
761 	params->loadmap = loadmap;
762 
763 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
764 	loadmap->nsegs = nloads;
765 
766 	load_addr = params->load_addr;
767 	seg = loadmap->segs;
768 
769 	/* map the requested LOADs into the memory space */
770 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
771 	case ELF_FDPIC_FLAG_CONSTDISP:
772 	case ELF_FDPIC_FLAG_CONTIGUOUS:
773 #ifndef CONFIG_MMU
774 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
775 		if (ret < 0)
776 			return ret;
777 		break;
778 #endif
779 	default:
780 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
781 		if (ret < 0)
782 			return ret;
783 		break;
784 	}
785 
786 	/* map the entry point */
787 	if (params->hdr.e_entry) {
788 		seg = loadmap->segs;
789 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
790 			if (params->hdr.e_entry >= seg->p_vaddr &&
791 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
792 				params->entry_addr =
793 					(params->hdr.e_entry - seg->p_vaddr) +
794 					seg->addr;
795 				break;
796 			}
797 		}
798 	}
799 
800 	/* determine where the program header table has wound up if mapped */
801 	stop = params->hdr.e_phoff;
802 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
803 	phdr = params->phdrs;
804 
805 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
806 		if (phdr->p_type != PT_LOAD)
807 			continue;
808 
809 		if (phdr->p_offset > params->hdr.e_phoff ||
810 		    phdr->p_offset + phdr->p_filesz < stop)
811 			continue;
812 
813 		seg = loadmap->segs;
814 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
815 			if (phdr->p_vaddr >= seg->p_vaddr &&
816 			    phdr->p_vaddr + phdr->p_filesz <=
817 			    seg->p_vaddr + seg->p_memsz) {
818 				params->ph_addr =
819 					(phdr->p_vaddr - seg->p_vaddr) +
820 					seg->addr +
821 					params->hdr.e_phoff - phdr->p_offset;
822 				break;
823 			}
824 		}
825 		break;
826 	}
827 
828 	/* determine where the dynamic section has wound up if there is one */
829 	phdr = params->phdrs;
830 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
831 		if (phdr->p_type != PT_DYNAMIC)
832 			continue;
833 
834 		seg = loadmap->segs;
835 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
836 			if (phdr->p_vaddr >= seg->p_vaddr &&
837 			    phdr->p_vaddr + phdr->p_memsz <=
838 			    seg->p_vaddr + seg->p_memsz) {
839 				Elf32_Dyn __user *dyn;
840 				Elf32_Sword d_tag;
841 
842 				params->dynamic_addr =
843 					(phdr->p_vaddr - seg->p_vaddr) +
844 					seg->addr;
845 
846 				/* check the dynamic section contains at least
847 				 * one item, and that the last item is a NULL
848 				 * entry */
849 				if (phdr->p_memsz == 0 ||
850 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
851 					goto dynamic_error;
852 
853 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
854 				dyn = (Elf32_Dyn __user *)params->dynamic_addr;
855 				__get_user(d_tag, &dyn[tmp - 1].d_tag);
856 				if (d_tag != 0)
857 					goto dynamic_error;
858 				break;
859 			}
860 		}
861 		break;
862 	}
863 
864 	/* now elide adjacent segments in the load map on MMU linux
865 	 * - on uClinux the holes between may actually be filled with system
866 	 *   stuff or stuff from other processes
867 	 */
868 #ifdef CONFIG_MMU
869 	nloads = loadmap->nsegs;
870 	mseg = loadmap->segs;
871 	seg = mseg + 1;
872 	for (loop = 1; loop < nloads; loop++) {
873 		/* see if we have a candidate for merging */
874 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
875 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
876 			if (load_addr == (seg->addr & PAGE_MASK)) {
877 				mseg->p_memsz +=
878 					load_addr -
879 					(mseg->addr + mseg->p_memsz);
880 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
881 				mseg->p_memsz += seg->p_memsz;
882 				loadmap->nsegs--;
883 				continue;
884 			}
885 		}
886 
887 		mseg++;
888 		if (mseg != seg)
889 			*mseg = *seg;
890 	}
891 #endif
892 
893 	kdebug("Mapped Object [%s]:", what);
894 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
895 	kdebug("- entry    : %lx", params->entry_addr);
896 	kdebug("- PHDR[]   : %lx", params->ph_addr);
897 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
898 	seg = loadmap->segs;
899 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
900 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
901 		       loop,
902 		       seg->addr, seg->addr + seg->p_memsz - 1,
903 		       seg->p_vaddr, seg->p_memsz);
904 
905 	return 0;
906 
907 dynamic_error:
908 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
909 	       what, file_inode(file)->i_ino);
910 	return -ELIBBAD;
911 }
912 
913 /*****************************************************************************/
914 /*
915  * map a file with constant displacement under uClinux
916  */
917 #ifndef CONFIG_MMU
918 static int elf_fdpic_map_file_constdisp_on_uclinux(
919 	struct elf_fdpic_params *params,
920 	struct file *file,
921 	struct mm_struct *mm)
922 {
923 	struct elf32_fdpic_loadseg *seg;
924 	struct elf32_phdr *phdr;
925 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
926 	int loop, ret;
927 
928 	load_addr = params->load_addr;
929 	seg = params->loadmap->segs;
930 
931 	/* determine the bounds of the contiguous overall allocation we must
932 	 * make */
933 	phdr = params->phdrs;
934 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
935 		if (params->phdrs[loop].p_type != PT_LOAD)
936 			continue;
937 
938 		if (base > phdr->p_vaddr)
939 			base = phdr->p_vaddr;
940 		if (top < phdr->p_vaddr + phdr->p_memsz)
941 			top = phdr->p_vaddr + phdr->p_memsz;
942 	}
943 
944 	/* allocate one big anon block for everything */
945 	mflags = MAP_PRIVATE;
946 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
947 		mflags |= MAP_EXECUTABLE;
948 
949 	maddr = vm_mmap(NULL, load_addr, top - base,
950 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
951 	if (IS_ERR_VALUE(maddr))
952 		return (int) maddr;
953 
954 	if (load_addr != 0)
955 		load_addr += PAGE_ALIGN(top - base);
956 
957 	/* and then load the file segments into it */
958 	phdr = params->phdrs;
959 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
960 		if (params->phdrs[loop].p_type != PT_LOAD)
961 			continue;
962 
963 		seg->addr = maddr + (phdr->p_vaddr - base);
964 		seg->p_vaddr = phdr->p_vaddr;
965 		seg->p_memsz = phdr->p_memsz;
966 
967 		ret = read_code(file, seg->addr, phdr->p_offset,
968 				       phdr->p_filesz);
969 		if (ret < 0)
970 			return ret;
971 
972 		/* map the ELF header address if in this segment */
973 		if (phdr->p_offset == 0)
974 			params->elfhdr_addr = seg->addr;
975 
976 		/* clear any space allocated but not loaded */
977 		if (phdr->p_filesz < phdr->p_memsz) {
978 			if (clear_user((void *) (seg->addr + phdr->p_filesz),
979 				       phdr->p_memsz - phdr->p_filesz))
980 				return -EFAULT;
981 		}
982 
983 		if (mm) {
984 			if (phdr->p_flags & PF_X) {
985 				if (!mm->start_code) {
986 					mm->start_code = seg->addr;
987 					mm->end_code = seg->addr +
988 						phdr->p_memsz;
989 				}
990 			} else if (!mm->start_data) {
991 				mm->start_data = seg->addr;
992 				mm->end_data = seg->addr + phdr->p_memsz;
993 			}
994 		}
995 
996 		seg++;
997 	}
998 
999 	return 0;
1000 }
1001 #endif
1002 
1003 /*****************************************************************************/
1004 /*
1005  * map a binary by direct mmap() of the individual PT_LOAD segments
1006  */
1007 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1008 					     struct file *file,
1009 					     struct mm_struct *mm)
1010 {
1011 	struct elf32_fdpic_loadseg *seg;
1012 	struct elf32_phdr *phdr;
1013 	unsigned long load_addr, delta_vaddr;
1014 	int loop, dvset;
1015 
1016 	load_addr = params->load_addr;
1017 	delta_vaddr = 0;
1018 	dvset = 0;
1019 
1020 	seg = params->loadmap->segs;
1021 
1022 	/* deal with each load segment separately */
1023 	phdr = params->phdrs;
1024 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1025 		unsigned long maddr, disp, excess, excess1;
1026 		int prot = 0, flags;
1027 
1028 		if (phdr->p_type != PT_LOAD)
1029 			continue;
1030 
1031 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1032 		       (unsigned long) phdr->p_vaddr,
1033 		       (unsigned long) phdr->p_offset,
1034 		       (unsigned long) phdr->p_filesz,
1035 		       (unsigned long) phdr->p_memsz);
1036 
1037 		/* determine the mapping parameters */
1038 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1039 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1040 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1041 
1042 		flags = MAP_PRIVATE | MAP_DENYWRITE;
1043 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1044 			flags |= MAP_EXECUTABLE;
1045 
1046 		maddr = 0;
1047 
1048 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1049 		case ELF_FDPIC_FLAG_INDEPENDENT:
1050 			/* PT_LOADs are independently locatable */
1051 			break;
1052 
1053 		case ELF_FDPIC_FLAG_HONOURVADDR:
1054 			/* the specified virtual address must be honoured */
1055 			maddr = phdr->p_vaddr;
1056 			flags |= MAP_FIXED;
1057 			break;
1058 
1059 		case ELF_FDPIC_FLAG_CONSTDISP:
1060 			/* constant displacement
1061 			 * - can be mapped anywhere, but must be mapped as a
1062 			 *   unit
1063 			 */
1064 			if (!dvset) {
1065 				maddr = load_addr;
1066 				delta_vaddr = phdr->p_vaddr;
1067 				dvset = 1;
1068 			} else {
1069 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1070 				flags |= MAP_FIXED;
1071 			}
1072 			break;
1073 
1074 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1075 			/* contiguity handled later */
1076 			break;
1077 
1078 		default:
1079 			BUG();
1080 		}
1081 
1082 		maddr &= PAGE_MASK;
1083 
1084 		/* create the mapping */
1085 		disp = phdr->p_vaddr & ~PAGE_MASK;
1086 		maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1087 				phdr->p_offset - disp);
1088 
1089 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1090 		       loop, phdr->p_memsz + disp, prot, flags,
1091 		       phdr->p_offset - disp, maddr);
1092 
1093 		if (IS_ERR_VALUE(maddr))
1094 			return (int) maddr;
1095 
1096 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1097 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1098 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1099 
1100 		seg->addr = maddr + disp;
1101 		seg->p_vaddr = phdr->p_vaddr;
1102 		seg->p_memsz = phdr->p_memsz;
1103 
1104 		/* map the ELF header address if in this segment */
1105 		if (phdr->p_offset == 0)
1106 			params->elfhdr_addr = seg->addr;
1107 
1108 		/* clear the bit between beginning of mapping and beginning of
1109 		 * PT_LOAD */
1110 		if (prot & PROT_WRITE && disp > 0) {
1111 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1112 			if (clear_user((void __user *) maddr, disp))
1113 				return -EFAULT;
1114 			maddr += disp;
1115 		}
1116 
1117 		/* clear any space allocated but not loaded
1118 		 * - on uClinux we can just clear the lot
1119 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1120 		 *   extant in the file
1121 		 */
1122 		excess = phdr->p_memsz - phdr->p_filesz;
1123 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1124 
1125 #ifdef CONFIG_MMU
1126 		if (excess > excess1) {
1127 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1128 			unsigned long xmaddr;
1129 
1130 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1131 			xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1132 					 prot, flags, 0);
1133 
1134 			kdebug("mmap[%d] <anon>"
1135 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1136 			       loop, xaddr, excess - excess1, prot, flags,
1137 			       xmaddr);
1138 
1139 			if (xmaddr != xaddr)
1140 				return -ENOMEM;
1141 		}
1142 
1143 		if (prot & PROT_WRITE && excess1 > 0) {
1144 			kdebug("clear[%d] ad=%lx sz=%lx",
1145 			       loop, maddr + phdr->p_filesz, excess1);
1146 			if (clear_user((void __user *) maddr + phdr->p_filesz,
1147 				       excess1))
1148 				return -EFAULT;
1149 		}
1150 
1151 #else
1152 		if (excess > 0) {
1153 			kdebug("clear[%d] ad=%lx sz=%lx",
1154 			       loop, maddr + phdr->p_filesz, excess);
1155 			if (clear_user((void *) maddr + phdr->p_filesz, excess))
1156 				return -EFAULT;
1157 		}
1158 #endif
1159 
1160 		if (mm) {
1161 			if (phdr->p_flags & PF_X) {
1162 				if (!mm->start_code) {
1163 					mm->start_code = maddr;
1164 					mm->end_code = maddr + phdr->p_memsz;
1165 				}
1166 			} else if (!mm->start_data) {
1167 				mm->start_data = maddr;
1168 				mm->end_data = maddr + phdr->p_memsz;
1169 			}
1170 		}
1171 
1172 		seg++;
1173 	}
1174 
1175 	return 0;
1176 }
1177 
1178 /*****************************************************************************/
1179 /*
1180  * ELF-FDPIC core dumper
1181  *
1182  * Modelled on fs/exec.c:aout_core_dump()
1183  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1184  *
1185  * Modelled on fs/binfmt_elf.c core dumper
1186  */
1187 #ifdef CONFIG_ELF_CORE
1188 
1189 /*
1190  * Decide whether a segment is worth dumping; default is yes to be
1191  * sure (missing info is worse than too much; etc).
1192  * Personally I'd include everything, and use the coredump limit...
1193  *
1194  * I think we should skip something. But I am not sure how. H.J.
1195  */
1196 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1197 {
1198 	int dump_ok;
1199 
1200 	/* Do not dump I/O mapped devices or special mappings */
1201 	if (vma->vm_flags & VM_IO) {
1202 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1203 		return 0;
1204 	}
1205 
1206 	/* If we may not read the contents, don't allow us to dump
1207 	 * them either. "dump_write()" can't handle it anyway.
1208 	 */
1209 	if (!(vma->vm_flags & VM_READ)) {
1210 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1211 		return 0;
1212 	}
1213 
1214 	/* support for DAX */
1215 	if (vma_is_dax(vma)) {
1216 		if (vma->vm_flags & VM_SHARED) {
1217 			dump_ok = test_bit(MMF_DUMP_DAX_SHARED, &mm_flags);
1218 			kdcore("%08lx: %08lx: %s (DAX shared)", vma->vm_start,
1219 			       vma->vm_flags, dump_ok ? "yes" : "no");
1220 		} else {
1221 			dump_ok = test_bit(MMF_DUMP_DAX_PRIVATE, &mm_flags);
1222 			kdcore("%08lx: %08lx: %s (DAX private)", vma->vm_start,
1223 			       vma->vm_flags, dump_ok ? "yes" : "no");
1224 		}
1225 		return dump_ok;
1226 	}
1227 
1228 	/* By default, dump shared memory if mapped from an anonymous file. */
1229 	if (vma->vm_flags & VM_SHARED) {
1230 		if (file_inode(vma->vm_file)->i_nlink == 0) {
1231 			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1232 			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1233 			       vma->vm_flags, dump_ok ? "yes" : "no");
1234 			return dump_ok;
1235 		}
1236 
1237 		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1238 		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1239 		       vma->vm_flags, dump_ok ? "yes" : "no");
1240 		return dump_ok;
1241 	}
1242 
1243 #ifdef CONFIG_MMU
1244 	/* By default, if it hasn't been written to, don't write it out */
1245 	if (!vma->anon_vma) {
1246 		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1247 		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1248 		       vma->vm_flags, dump_ok ? "yes" : "no");
1249 		return dump_ok;
1250 	}
1251 #endif
1252 
1253 	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1254 	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1255 	       dump_ok ? "yes" : "no");
1256 	return dump_ok;
1257 }
1258 
1259 /* An ELF note in memory */
1260 struct memelfnote
1261 {
1262 	const char *name;
1263 	int type;
1264 	unsigned int datasz;
1265 	void *data;
1266 };
1267 
1268 static int notesize(struct memelfnote *en)
1269 {
1270 	int sz;
1271 
1272 	sz = sizeof(struct elf_note);
1273 	sz += roundup(strlen(en->name) + 1, 4);
1274 	sz += roundup(en->datasz, 4);
1275 
1276 	return sz;
1277 }
1278 
1279 /* #define DEBUG */
1280 
1281 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1282 {
1283 	struct elf_note en;
1284 	en.n_namesz = strlen(men->name) + 1;
1285 	en.n_descsz = men->datasz;
1286 	en.n_type = men->type;
1287 
1288 	return dump_emit(cprm, &en, sizeof(en)) &&
1289 		dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1290 		dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1291 }
1292 
1293 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1294 {
1295 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1296 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1297 	elf->e_ident[EI_DATA] = ELF_DATA;
1298 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1299 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1300 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1301 
1302 	elf->e_type = ET_CORE;
1303 	elf->e_machine = ELF_ARCH;
1304 	elf->e_version = EV_CURRENT;
1305 	elf->e_entry = 0;
1306 	elf->e_phoff = sizeof(struct elfhdr);
1307 	elf->e_shoff = 0;
1308 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1309 	elf->e_ehsize = sizeof(struct elfhdr);
1310 	elf->e_phentsize = sizeof(struct elf_phdr);
1311 	elf->e_phnum = segs;
1312 	elf->e_shentsize = 0;
1313 	elf->e_shnum = 0;
1314 	elf->e_shstrndx = 0;
1315 	return;
1316 }
1317 
1318 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1319 {
1320 	phdr->p_type = PT_NOTE;
1321 	phdr->p_offset = offset;
1322 	phdr->p_vaddr = 0;
1323 	phdr->p_paddr = 0;
1324 	phdr->p_filesz = sz;
1325 	phdr->p_memsz = 0;
1326 	phdr->p_flags = 0;
1327 	phdr->p_align = 0;
1328 	return;
1329 }
1330 
1331 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1332 		unsigned int sz, void *data)
1333 {
1334 	note->name = name;
1335 	note->type = type;
1336 	note->datasz = sz;
1337 	note->data = data;
1338 	return;
1339 }
1340 
1341 /*
1342  * fill up all the fields in prstatus from the given task struct, except
1343  * registers which need to be filled up separately.
1344  */
1345 static void fill_prstatus(struct elf_prstatus *prstatus,
1346 			  struct task_struct *p, long signr)
1347 {
1348 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1349 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1350 	prstatus->pr_sighold = p->blocked.sig[0];
1351 	rcu_read_lock();
1352 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1353 	rcu_read_unlock();
1354 	prstatus->pr_pid = task_pid_vnr(p);
1355 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1356 	prstatus->pr_sid = task_session_vnr(p);
1357 	if (thread_group_leader(p)) {
1358 		struct task_cputime cputime;
1359 
1360 		/*
1361 		 * This is the record for the group leader.  It shows the
1362 		 * group-wide total, not its individual thread total.
1363 		 */
1364 		thread_group_cputime(p, &cputime);
1365 		prstatus->pr_utime = ns_to_timeval(cputime.utime);
1366 		prstatus->pr_stime = ns_to_timeval(cputime.stime);
1367 	} else {
1368 		u64 utime, stime;
1369 
1370 		task_cputime(p, &utime, &stime);
1371 		prstatus->pr_utime = ns_to_timeval(utime);
1372 		prstatus->pr_stime = ns_to_timeval(stime);
1373 	}
1374 	prstatus->pr_cutime = ns_to_timeval(p->signal->cutime);
1375 	prstatus->pr_cstime = ns_to_timeval(p->signal->cstime);
1376 
1377 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1378 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1379 }
1380 
1381 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1382 		       struct mm_struct *mm)
1383 {
1384 	const struct cred *cred;
1385 	unsigned int i, len;
1386 
1387 	/* first copy the parameters from user space */
1388 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1389 
1390 	len = mm->arg_end - mm->arg_start;
1391 	if (len >= ELF_PRARGSZ)
1392 		len = ELF_PRARGSZ - 1;
1393 	if (copy_from_user(&psinfo->pr_psargs,
1394 		           (const char __user *) mm->arg_start, len))
1395 		return -EFAULT;
1396 	for (i = 0; i < len; i++)
1397 		if (psinfo->pr_psargs[i] == 0)
1398 			psinfo->pr_psargs[i] = ' ';
1399 	psinfo->pr_psargs[len] = 0;
1400 
1401 	rcu_read_lock();
1402 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1403 	rcu_read_unlock();
1404 	psinfo->pr_pid = task_pid_vnr(p);
1405 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1406 	psinfo->pr_sid = task_session_vnr(p);
1407 
1408 	i = p->state ? ffz(~p->state) + 1 : 0;
1409 	psinfo->pr_state = i;
1410 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1411 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1412 	psinfo->pr_nice = task_nice(p);
1413 	psinfo->pr_flag = p->flags;
1414 	rcu_read_lock();
1415 	cred = __task_cred(p);
1416 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1417 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1418 	rcu_read_unlock();
1419 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1420 
1421 	return 0;
1422 }
1423 
1424 /* Here is the structure in which status of each thread is captured. */
1425 struct elf_thread_status
1426 {
1427 	struct list_head list;
1428 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1429 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1430 	struct task_struct *thread;
1431 #ifdef ELF_CORE_COPY_XFPREGS
1432 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1433 #endif
1434 	struct memelfnote notes[3];
1435 	int num_notes;
1436 };
1437 
1438 /*
1439  * In order to add the specific thread information for the elf file format,
1440  * we need to keep a linked list of every thread's pr_status and then create
1441  * a single section for them in the final core file.
1442  */
1443 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1444 {
1445 	struct task_struct *p = t->thread;
1446 	int sz = 0;
1447 
1448 	t->num_notes = 0;
1449 
1450 	fill_prstatus(&t->prstatus, p, signr);
1451 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1452 
1453 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1454 		  &t->prstatus);
1455 	t->num_notes++;
1456 	sz += notesize(&t->notes[0]);
1457 
1458 	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1459 	if (t->prstatus.pr_fpvalid) {
1460 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1461 			  &t->fpu);
1462 		t->num_notes++;
1463 		sz += notesize(&t->notes[1]);
1464 	}
1465 
1466 #ifdef ELF_CORE_COPY_XFPREGS
1467 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1468 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1469 			  sizeof(t->xfpu), &t->xfpu);
1470 		t->num_notes++;
1471 		sz += notesize(&t->notes[2]);
1472 	}
1473 #endif
1474 	return sz;
1475 }
1476 
1477 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1478 			     elf_addr_t e_shoff, int segs)
1479 {
1480 	elf->e_shoff = e_shoff;
1481 	elf->e_shentsize = sizeof(*shdr4extnum);
1482 	elf->e_shnum = 1;
1483 	elf->e_shstrndx = SHN_UNDEF;
1484 
1485 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1486 
1487 	shdr4extnum->sh_type = SHT_NULL;
1488 	shdr4extnum->sh_size = elf->e_shnum;
1489 	shdr4extnum->sh_link = elf->e_shstrndx;
1490 	shdr4extnum->sh_info = segs;
1491 }
1492 
1493 /*
1494  * dump the segments for an MMU process
1495  */
1496 static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1497 {
1498 	struct vm_area_struct *vma;
1499 
1500 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1501 #ifdef CONFIG_MMU
1502 		unsigned long addr;
1503 #endif
1504 
1505 		if (!maydump(vma, cprm->mm_flags))
1506 			continue;
1507 
1508 #ifdef CONFIG_MMU
1509 		for (addr = vma->vm_start; addr < vma->vm_end;
1510 							addr += PAGE_SIZE) {
1511 			bool res;
1512 			struct page *page = get_dump_page(addr);
1513 			if (page) {
1514 				void *kaddr = kmap(page);
1515 				res = dump_emit(cprm, kaddr, PAGE_SIZE);
1516 				kunmap(page);
1517 				put_page(page);
1518 			} else {
1519 				res = dump_skip(cprm, PAGE_SIZE);
1520 			}
1521 			if (!res)
1522 				return false;
1523 		}
1524 #else
1525 		if (!dump_emit(cprm, (void *) vma->vm_start,
1526 				vma->vm_end - vma->vm_start))
1527 			return false;
1528 #endif
1529 	}
1530 	return true;
1531 }
1532 
1533 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1534 {
1535 	struct vm_area_struct *vma;
1536 	size_t size = 0;
1537 
1538 	for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1539 		if (maydump(vma, mm_flags))
1540 			size += vma->vm_end - vma->vm_start;
1541 	return size;
1542 }
1543 
1544 /*
1545  * Actual dumper
1546  *
1547  * This is a two-pass process; first we find the offsets of the bits,
1548  * and then they are actually written out.  If we run out of core limit
1549  * we just truncate.
1550  */
1551 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1552 {
1553 #define	NUM_NOTES	6
1554 	int has_dumped = 0;
1555 	mm_segment_t fs;
1556 	int segs;
1557 	int i;
1558 	struct vm_area_struct *vma;
1559 	struct elfhdr *elf = NULL;
1560 	loff_t offset = 0, dataoff;
1561 	int numnote;
1562 	struct memelfnote *notes = NULL;
1563 	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1564 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1565  	LIST_HEAD(thread_list);
1566  	struct list_head *t;
1567 	elf_fpregset_t *fpu = NULL;
1568 #ifdef ELF_CORE_COPY_XFPREGS
1569 	elf_fpxregset_t *xfpu = NULL;
1570 #endif
1571 	int thread_status_size = 0;
1572 	elf_addr_t *auxv;
1573 	struct elf_phdr *phdr4note = NULL;
1574 	struct elf_shdr *shdr4extnum = NULL;
1575 	Elf_Half e_phnum;
1576 	elf_addr_t e_shoff;
1577 	struct core_thread *ct;
1578 	struct elf_thread_status *tmp;
1579 
1580 	/*
1581 	 * We no longer stop all VM operations.
1582 	 *
1583 	 * This is because those proceses that could possibly change map_count
1584 	 * or the mmap / vma pages are now blocked in do_exit on current
1585 	 * finishing this core dump.
1586 	 *
1587 	 * Only ptrace can touch these memory addresses, but it doesn't change
1588 	 * the map_count or the pages allocated. So no possibility of crashing
1589 	 * exists while dumping the mm->vm_next areas to the core file.
1590 	 */
1591 
1592 	/* alloc memory for large data structures: too large to be on stack */
1593 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1594 	if (!elf)
1595 		goto cleanup;
1596 	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1597 	if (!prstatus)
1598 		goto cleanup;
1599 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1600 	if (!psinfo)
1601 		goto cleanup;
1602 	notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1603 	if (!notes)
1604 		goto cleanup;
1605 	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1606 	if (!fpu)
1607 		goto cleanup;
1608 #ifdef ELF_CORE_COPY_XFPREGS
1609 	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1610 	if (!xfpu)
1611 		goto cleanup;
1612 #endif
1613 
1614 	for (ct = current->mm->core_state->dumper.next;
1615 					ct; ct = ct->next) {
1616 		tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1617 		if (!tmp)
1618 			goto cleanup;
1619 
1620 		tmp->thread = ct->task;
1621 		list_add(&tmp->list, &thread_list);
1622 	}
1623 
1624 	list_for_each(t, &thread_list) {
1625 		struct elf_thread_status *tmp;
1626 		int sz;
1627 
1628 		tmp = list_entry(t, struct elf_thread_status, list);
1629 		sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1630 		thread_status_size += sz;
1631 	}
1632 
1633 	/* now collect the dump for the current */
1634 	fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1635 	elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1636 
1637 	segs = current->mm->map_count;
1638 	segs += elf_core_extra_phdrs();
1639 
1640 	/* for notes section */
1641 	segs++;
1642 
1643 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1644 	 * this, kernel supports extended numbering. Have a look at
1645 	 * include/linux/elf.h for further information. */
1646 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1647 
1648 	/* Set up header */
1649 	fill_elf_fdpic_header(elf, e_phnum);
1650 
1651 	has_dumped = 1;
1652 	/*
1653 	 * Set up the notes in similar form to SVR4 core dumps made
1654 	 * with info from their /proc.
1655 	 */
1656 
1657 	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1658 	fill_psinfo(psinfo, current->group_leader, current->mm);
1659 	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1660 
1661 	numnote = 2;
1662 
1663 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1664 
1665 	i = 0;
1666 	do
1667 		i += 2;
1668 	while (auxv[i - 2] != AT_NULL);
1669 	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1670 		  i * sizeof(elf_addr_t), auxv);
1671 
1672   	/* Try to dump the FPU. */
1673 	if ((prstatus->pr_fpvalid =
1674 	     elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1675 		fill_note(notes + numnote++,
1676 			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1677 #ifdef ELF_CORE_COPY_XFPREGS
1678 	if (elf_core_copy_task_xfpregs(current, xfpu))
1679 		fill_note(notes + numnote++,
1680 			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1681 #endif
1682 
1683 	fs = get_fs();
1684 	set_fs(KERNEL_DS);
1685 
1686 	offset += sizeof(*elf);				/* Elf header */
1687 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1688 
1689 	/* Write notes phdr entry */
1690 	{
1691 		int sz = 0;
1692 
1693 		for (i = 0; i < numnote; i++)
1694 			sz += notesize(notes + i);
1695 
1696 		sz += thread_status_size;
1697 
1698 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1699 		if (!phdr4note)
1700 			goto end_coredump;
1701 
1702 		fill_elf_note_phdr(phdr4note, sz, offset);
1703 		offset += sz;
1704 	}
1705 
1706 	/* Page-align dumped data */
1707 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1708 
1709 	offset += elf_core_vma_data_size(cprm->mm_flags);
1710 	offset += elf_core_extra_data_size();
1711 	e_shoff = offset;
1712 
1713 	if (e_phnum == PN_XNUM) {
1714 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1715 		if (!shdr4extnum)
1716 			goto end_coredump;
1717 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1718 	}
1719 
1720 	offset = dataoff;
1721 
1722 	if (!dump_emit(cprm, elf, sizeof(*elf)))
1723 		goto end_coredump;
1724 
1725 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1726 		goto end_coredump;
1727 
1728 	/* write program headers for segments dump */
1729 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1730 		struct elf_phdr phdr;
1731 		size_t sz;
1732 
1733 		sz = vma->vm_end - vma->vm_start;
1734 
1735 		phdr.p_type = PT_LOAD;
1736 		phdr.p_offset = offset;
1737 		phdr.p_vaddr = vma->vm_start;
1738 		phdr.p_paddr = 0;
1739 		phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1740 		phdr.p_memsz = sz;
1741 		offset += phdr.p_filesz;
1742 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1743 		if (vma->vm_flags & VM_WRITE)
1744 			phdr.p_flags |= PF_W;
1745 		if (vma->vm_flags & VM_EXEC)
1746 			phdr.p_flags |= PF_X;
1747 		phdr.p_align = ELF_EXEC_PAGESIZE;
1748 
1749 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1750 			goto end_coredump;
1751 	}
1752 
1753 	if (!elf_core_write_extra_phdrs(cprm, offset))
1754 		goto end_coredump;
1755 
1756  	/* write out the notes section */
1757 	for (i = 0; i < numnote; i++)
1758 		if (!writenote(notes + i, cprm))
1759 			goto end_coredump;
1760 
1761 	/* write out the thread status notes section */
1762 	list_for_each(t, &thread_list) {
1763 		struct elf_thread_status *tmp =
1764 				list_entry(t, struct elf_thread_status, list);
1765 
1766 		for (i = 0; i < tmp->num_notes; i++)
1767 			if (!writenote(&tmp->notes[i], cprm))
1768 				goto end_coredump;
1769 	}
1770 
1771 	if (!dump_skip(cprm, dataoff - cprm->pos))
1772 		goto end_coredump;
1773 
1774 	if (!elf_fdpic_dump_segments(cprm))
1775 		goto end_coredump;
1776 
1777 	if (!elf_core_write_extra_data(cprm))
1778 		goto end_coredump;
1779 
1780 	if (e_phnum == PN_XNUM) {
1781 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1782 			goto end_coredump;
1783 	}
1784 
1785 	if (cprm->file->f_pos != offset) {
1786 		/* Sanity check */
1787 		printk(KERN_WARNING
1788 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1789 		       cprm->file->f_pos, offset);
1790 	}
1791 
1792 end_coredump:
1793 	set_fs(fs);
1794 
1795 cleanup:
1796 	while (!list_empty(&thread_list)) {
1797 		struct list_head *tmp = thread_list.next;
1798 		list_del(tmp);
1799 		kfree(list_entry(tmp, struct elf_thread_status, list));
1800 	}
1801 	kfree(phdr4note);
1802 	kfree(elf);
1803 	kfree(prstatus);
1804 	kfree(psinfo);
1805 	kfree(notes);
1806 	kfree(fpu);
1807 	kfree(shdr4extnum);
1808 #ifdef ELF_CORE_COPY_XFPREGS
1809 	kfree(xfpu);
1810 #endif
1811 	return has_dumped;
1812 #undef NUM_NOTES
1813 }
1814 
1815 #endif		/* CONFIG_ELF_CORE */
1816