xref: /openbmc/linux/fs/binfmt_elf_fdpic.c (revision e3d786a3)
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 	finalize_exec(bprm);
467 	/* everything is now ready... get the userspace context ready to roll */
468 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
469 	start_thread(regs, entryaddr, current->mm->start_stack);
470 
471 	retval = 0;
472 
473 error:
474 	if (interpreter) {
475 		allow_write_access(interpreter);
476 		fput(interpreter);
477 	}
478 	kfree(interpreter_name);
479 	kfree(exec_params.phdrs);
480 	kfree(exec_params.loadmap);
481 	kfree(interp_params.phdrs);
482 	kfree(interp_params.loadmap);
483 	return retval;
484 }
485 
486 /*****************************************************************************/
487 
488 #ifndef ELF_BASE_PLATFORM
489 /*
490  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
491  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
492  * will be copied to the user stack in the same manner as AT_PLATFORM.
493  */
494 #define ELF_BASE_PLATFORM NULL
495 #endif
496 
497 /*
498  * present useful information to the program by shovelling it onto the new
499  * process's stack
500  */
501 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
502 				   struct mm_struct *mm,
503 				   struct elf_fdpic_params *exec_params,
504 				   struct elf_fdpic_params *interp_params)
505 {
506 	const struct cred *cred = current_cred();
507 	unsigned long sp, csp, nitems;
508 	elf_caddr_t __user *argv, *envp;
509 	size_t platform_len = 0, len;
510 	char *k_platform, *k_base_platform;
511 	char __user *u_platform, *u_base_platform, *p;
512 	int loop;
513 	int nr;	/* reset for each csp adjustment */
514 
515 #ifdef CONFIG_MMU
516 	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
517 	 * by the processes running on the same package. One thing we can do is
518 	 * to shuffle the initial stack for them, so we give the architecture
519 	 * an opportunity to do so here.
520 	 */
521 	sp = arch_align_stack(bprm->p);
522 #else
523 	sp = mm->start_stack;
524 
525 	/* stack the program arguments and environment */
526 	if (transfer_args_to_stack(bprm, &sp) < 0)
527 		return -EFAULT;
528 	sp &= ~15;
529 #endif
530 
531 	/*
532 	 * If this architecture has a platform capability string, copy it
533 	 * to userspace.  In some cases (Sparc), this info is impossible
534 	 * for userspace to get any other way, in others (i386) it is
535 	 * merely difficult.
536 	 */
537 	k_platform = ELF_PLATFORM;
538 	u_platform = NULL;
539 
540 	if (k_platform) {
541 		platform_len = strlen(k_platform) + 1;
542 		sp -= platform_len;
543 		u_platform = (char __user *) sp;
544 		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
545 			return -EFAULT;
546 	}
547 
548 	/*
549 	 * If this architecture has a "base" platform capability
550 	 * string, copy it to userspace.
551 	 */
552 	k_base_platform = ELF_BASE_PLATFORM;
553 	u_base_platform = NULL;
554 
555 	if (k_base_platform) {
556 		platform_len = strlen(k_base_platform) + 1;
557 		sp -= platform_len;
558 		u_base_platform = (char __user *) sp;
559 		if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
560 			return -EFAULT;
561 	}
562 
563 	sp &= ~7UL;
564 
565 	/* stack the load map(s) */
566 	len = sizeof(struct elf32_fdpic_loadmap);
567 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
568 	sp = (sp - len) & ~7UL;
569 	exec_params->map_addr = sp;
570 
571 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
572 		return -EFAULT;
573 
574 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
575 
576 	if (interp_params->loadmap) {
577 		len = sizeof(struct elf32_fdpic_loadmap);
578 		len += sizeof(struct elf32_fdpic_loadseg) *
579 			interp_params->loadmap->nsegs;
580 		sp = (sp - len) & ~7UL;
581 		interp_params->map_addr = sp;
582 
583 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
584 				 len) != 0)
585 			return -EFAULT;
586 
587 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
588 	}
589 
590 	/* force 16 byte _final_ alignment here for generality */
591 #define DLINFO_ITEMS 15
592 
593 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
594 		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
595 
596 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
597 		nitems++;
598 
599 	csp = sp;
600 	sp -= nitems * 2 * sizeof(unsigned long);
601 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
602 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
603 	sp -= 1 * sizeof(unsigned long);		/* argc */
604 
605 	csp -= sp & 15UL;
606 	sp -= sp & 15UL;
607 
608 	/* put the ELF interpreter info on the stack */
609 #define NEW_AUX_ENT(id, val)						\
610 	do {								\
611 		struct { unsigned long _id, _val; } __user *ent;	\
612 									\
613 		ent = (void __user *) csp;				\
614 		__put_user((id), &ent[nr]._id);				\
615 		__put_user((val), &ent[nr]._val);			\
616 		nr++;							\
617 	} while (0)
618 
619 	nr = 0;
620 	csp -= 2 * sizeof(unsigned long);
621 	NEW_AUX_ENT(AT_NULL, 0);
622 	if (k_platform) {
623 		nr = 0;
624 		csp -= 2 * sizeof(unsigned long);
625 		NEW_AUX_ENT(AT_PLATFORM,
626 			    (elf_addr_t) (unsigned long) u_platform);
627 	}
628 
629 	if (k_base_platform) {
630 		nr = 0;
631 		csp -= 2 * sizeof(unsigned long);
632 		NEW_AUX_ENT(AT_BASE_PLATFORM,
633 			    (elf_addr_t) (unsigned long) u_base_platform);
634 	}
635 
636 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
637 		nr = 0;
638 		csp -= 2 * sizeof(unsigned long);
639 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
640 	}
641 
642 	nr = 0;
643 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
644 	NEW_AUX_ENT(AT_HWCAP,	ELF_HWCAP);
645 #ifdef ELF_HWCAP2
646 	NEW_AUX_ENT(AT_HWCAP2,	ELF_HWCAP2);
647 #endif
648 	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
649 	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
650 	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
651 	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
652 	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
653 	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
654 	NEW_AUX_ENT(AT_FLAGS,	0);
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 	__put_user(bprm->argc, (unsigned long __user *) csp);
683 
684 	BUG_ON(csp != sp);
685 
686 	/* fill in the argv[] array */
687 #ifdef CONFIG_MMU
688 	current->mm->arg_start = bprm->p;
689 #else
690 	current->mm->arg_start = current->mm->start_stack -
691 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
692 #endif
693 
694 	p = (char __user *) current->mm->arg_start;
695 	for (loop = bprm->argc; loop > 0; loop--) {
696 		__put_user((elf_caddr_t) p, argv++);
697 		len = strnlen_user(p, MAX_ARG_STRLEN);
698 		if (!len || len > MAX_ARG_STRLEN)
699 			return -EINVAL;
700 		p += len;
701 	}
702 	__put_user(NULL, argv);
703 	current->mm->arg_end = (unsigned long) p;
704 
705 	/* fill in the envv[] array */
706 	current->mm->env_start = (unsigned long) p;
707 	for (loop = bprm->envc; loop > 0; loop--) {
708 		__put_user((elf_caddr_t)(unsigned long) p, envp++);
709 		len = strnlen_user(p, MAX_ARG_STRLEN);
710 		if (!len || len > MAX_ARG_STRLEN)
711 			return -EINVAL;
712 		p += len;
713 	}
714 	__put_user(NULL, envp);
715 	current->mm->env_end = (unsigned long) p;
716 
717 	mm->start_stack = (unsigned long) sp;
718 	return 0;
719 }
720 
721 /*****************************************************************************/
722 /*
723  * load the appropriate binary image (executable or interpreter) into memory
724  * - we assume no MMU is available
725  * - if no other PIC bits are set in params->hdr->e_flags
726  *   - we assume that the LOADable segments in the binary are independently relocatable
727  *   - we assume R/O executable segments are shareable
728  * - else
729  *   - we assume the loadable parts of the image to require fixed displacement
730  *   - the image is not shareable
731  */
732 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
733 			      struct file *file,
734 			      struct mm_struct *mm,
735 			      const char *what)
736 {
737 	struct elf32_fdpic_loadmap *loadmap;
738 #ifdef CONFIG_MMU
739 	struct elf32_fdpic_loadseg *mseg;
740 #endif
741 	struct elf32_fdpic_loadseg *seg;
742 	struct elf32_phdr *phdr;
743 	unsigned long load_addr, stop;
744 	unsigned nloads, tmp;
745 	size_t size;
746 	int loop, ret;
747 
748 	/* allocate a load map table */
749 	nloads = 0;
750 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
751 		if (params->phdrs[loop].p_type == PT_LOAD)
752 			nloads++;
753 
754 	if (nloads == 0)
755 		return -ELIBBAD;
756 
757 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
758 	loadmap = kzalloc(size, GFP_KERNEL);
759 	if (!loadmap)
760 		return -ENOMEM;
761 
762 	params->loadmap = loadmap;
763 
764 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
765 	loadmap->nsegs = nloads;
766 
767 	load_addr = params->load_addr;
768 	seg = loadmap->segs;
769 
770 	/* map the requested LOADs into the memory space */
771 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
772 	case ELF_FDPIC_FLAG_CONSTDISP:
773 	case ELF_FDPIC_FLAG_CONTIGUOUS:
774 #ifndef CONFIG_MMU
775 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
776 		if (ret < 0)
777 			return ret;
778 		break;
779 #endif
780 	default:
781 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
782 		if (ret < 0)
783 			return ret;
784 		break;
785 	}
786 
787 	/* map the entry point */
788 	if (params->hdr.e_entry) {
789 		seg = loadmap->segs;
790 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
791 			if (params->hdr.e_entry >= seg->p_vaddr &&
792 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
793 				params->entry_addr =
794 					(params->hdr.e_entry - seg->p_vaddr) +
795 					seg->addr;
796 				break;
797 			}
798 		}
799 	}
800 
801 	/* determine where the program header table has wound up if mapped */
802 	stop = params->hdr.e_phoff;
803 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
804 	phdr = params->phdrs;
805 
806 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
807 		if (phdr->p_type != PT_LOAD)
808 			continue;
809 
810 		if (phdr->p_offset > params->hdr.e_phoff ||
811 		    phdr->p_offset + phdr->p_filesz < stop)
812 			continue;
813 
814 		seg = loadmap->segs;
815 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
816 			if (phdr->p_vaddr >= seg->p_vaddr &&
817 			    phdr->p_vaddr + phdr->p_filesz <=
818 			    seg->p_vaddr + seg->p_memsz) {
819 				params->ph_addr =
820 					(phdr->p_vaddr - seg->p_vaddr) +
821 					seg->addr +
822 					params->hdr.e_phoff - phdr->p_offset;
823 				break;
824 			}
825 		}
826 		break;
827 	}
828 
829 	/* determine where the dynamic section has wound up if there is one */
830 	phdr = params->phdrs;
831 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
832 		if (phdr->p_type != PT_DYNAMIC)
833 			continue;
834 
835 		seg = loadmap->segs;
836 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
837 			if (phdr->p_vaddr >= seg->p_vaddr &&
838 			    phdr->p_vaddr + phdr->p_memsz <=
839 			    seg->p_vaddr + seg->p_memsz) {
840 				Elf32_Dyn __user *dyn;
841 				Elf32_Sword d_tag;
842 
843 				params->dynamic_addr =
844 					(phdr->p_vaddr - seg->p_vaddr) +
845 					seg->addr;
846 
847 				/* check the dynamic section contains at least
848 				 * one item, and that the last item is a NULL
849 				 * entry */
850 				if (phdr->p_memsz == 0 ||
851 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
852 					goto dynamic_error;
853 
854 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
855 				dyn = (Elf32_Dyn __user *)params->dynamic_addr;
856 				__get_user(d_tag, &dyn[tmp - 1].d_tag);
857 				if (d_tag != 0)
858 					goto dynamic_error;
859 				break;
860 			}
861 		}
862 		break;
863 	}
864 
865 	/* now elide adjacent segments in the load map on MMU linux
866 	 * - on uClinux the holes between may actually be filled with system
867 	 *   stuff or stuff from other processes
868 	 */
869 #ifdef CONFIG_MMU
870 	nloads = loadmap->nsegs;
871 	mseg = loadmap->segs;
872 	seg = mseg + 1;
873 	for (loop = 1; loop < nloads; loop++) {
874 		/* see if we have a candidate for merging */
875 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
876 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
877 			if (load_addr == (seg->addr & PAGE_MASK)) {
878 				mseg->p_memsz +=
879 					load_addr -
880 					(mseg->addr + mseg->p_memsz);
881 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
882 				mseg->p_memsz += seg->p_memsz;
883 				loadmap->nsegs--;
884 				continue;
885 			}
886 		}
887 
888 		mseg++;
889 		if (mseg != seg)
890 			*mseg = *seg;
891 	}
892 #endif
893 
894 	kdebug("Mapped Object [%s]:", what);
895 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
896 	kdebug("- entry    : %lx", params->entry_addr);
897 	kdebug("- PHDR[]   : %lx", params->ph_addr);
898 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
899 	seg = loadmap->segs;
900 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
901 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
902 		       loop,
903 		       seg->addr, seg->addr + seg->p_memsz - 1,
904 		       seg->p_vaddr, seg->p_memsz);
905 
906 	return 0;
907 
908 dynamic_error:
909 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
910 	       what, file_inode(file)->i_ino);
911 	return -ELIBBAD;
912 }
913 
914 /*****************************************************************************/
915 /*
916  * map a file with constant displacement under uClinux
917  */
918 #ifndef CONFIG_MMU
919 static int elf_fdpic_map_file_constdisp_on_uclinux(
920 	struct elf_fdpic_params *params,
921 	struct file *file,
922 	struct mm_struct *mm)
923 {
924 	struct elf32_fdpic_loadseg *seg;
925 	struct elf32_phdr *phdr;
926 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
927 	int loop, ret;
928 
929 	load_addr = params->load_addr;
930 	seg = params->loadmap->segs;
931 
932 	/* determine the bounds of the contiguous overall allocation we must
933 	 * make */
934 	phdr = params->phdrs;
935 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
936 		if (params->phdrs[loop].p_type != PT_LOAD)
937 			continue;
938 
939 		if (base > phdr->p_vaddr)
940 			base = phdr->p_vaddr;
941 		if (top < phdr->p_vaddr + phdr->p_memsz)
942 			top = phdr->p_vaddr + phdr->p_memsz;
943 	}
944 
945 	/* allocate one big anon block for everything */
946 	mflags = MAP_PRIVATE;
947 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
948 		mflags |= MAP_EXECUTABLE;
949 
950 	maddr = vm_mmap(NULL, load_addr, top - base,
951 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
952 	if (IS_ERR_VALUE(maddr))
953 		return (int) maddr;
954 
955 	if (load_addr != 0)
956 		load_addr += PAGE_ALIGN(top - base);
957 
958 	/* and then load the file segments into it */
959 	phdr = params->phdrs;
960 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
961 		if (params->phdrs[loop].p_type != PT_LOAD)
962 			continue;
963 
964 		seg->addr = maddr + (phdr->p_vaddr - base);
965 		seg->p_vaddr = phdr->p_vaddr;
966 		seg->p_memsz = phdr->p_memsz;
967 
968 		ret = read_code(file, seg->addr, phdr->p_offset,
969 				       phdr->p_filesz);
970 		if (ret < 0)
971 			return ret;
972 
973 		/* map the ELF header address if in this segment */
974 		if (phdr->p_offset == 0)
975 			params->elfhdr_addr = seg->addr;
976 
977 		/* clear any space allocated but not loaded */
978 		if (phdr->p_filesz < phdr->p_memsz) {
979 			if (clear_user((void *) (seg->addr + phdr->p_filesz),
980 				       phdr->p_memsz - phdr->p_filesz))
981 				return -EFAULT;
982 		}
983 
984 		if (mm) {
985 			if (phdr->p_flags & PF_X) {
986 				if (!mm->start_code) {
987 					mm->start_code = seg->addr;
988 					mm->end_code = seg->addr +
989 						phdr->p_memsz;
990 				}
991 			} else if (!mm->start_data) {
992 				mm->start_data = seg->addr;
993 				mm->end_data = seg->addr + phdr->p_memsz;
994 			}
995 		}
996 
997 		seg++;
998 	}
999 
1000 	return 0;
1001 }
1002 #endif
1003 
1004 /*****************************************************************************/
1005 /*
1006  * map a binary by direct mmap() of the individual PT_LOAD segments
1007  */
1008 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1009 					     struct file *file,
1010 					     struct mm_struct *mm)
1011 {
1012 	struct elf32_fdpic_loadseg *seg;
1013 	struct elf32_phdr *phdr;
1014 	unsigned long load_addr, delta_vaddr;
1015 	int loop, dvset;
1016 
1017 	load_addr = params->load_addr;
1018 	delta_vaddr = 0;
1019 	dvset = 0;
1020 
1021 	seg = params->loadmap->segs;
1022 
1023 	/* deal with each load segment separately */
1024 	phdr = params->phdrs;
1025 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1026 		unsigned long maddr, disp, excess, excess1;
1027 		int prot = 0, flags;
1028 
1029 		if (phdr->p_type != PT_LOAD)
1030 			continue;
1031 
1032 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1033 		       (unsigned long) phdr->p_vaddr,
1034 		       (unsigned long) phdr->p_offset,
1035 		       (unsigned long) phdr->p_filesz,
1036 		       (unsigned long) phdr->p_memsz);
1037 
1038 		/* determine the mapping parameters */
1039 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1040 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1041 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1042 
1043 		flags = MAP_PRIVATE | MAP_DENYWRITE;
1044 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1045 			flags |= MAP_EXECUTABLE;
1046 
1047 		maddr = 0;
1048 
1049 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1050 		case ELF_FDPIC_FLAG_INDEPENDENT:
1051 			/* PT_LOADs are independently locatable */
1052 			break;
1053 
1054 		case ELF_FDPIC_FLAG_HONOURVADDR:
1055 			/* the specified virtual address must be honoured */
1056 			maddr = phdr->p_vaddr;
1057 			flags |= MAP_FIXED;
1058 			break;
1059 
1060 		case ELF_FDPIC_FLAG_CONSTDISP:
1061 			/* constant displacement
1062 			 * - can be mapped anywhere, but must be mapped as a
1063 			 *   unit
1064 			 */
1065 			if (!dvset) {
1066 				maddr = load_addr;
1067 				delta_vaddr = phdr->p_vaddr;
1068 				dvset = 1;
1069 			} else {
1070 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1071 				flags |= MAP_FIXED;
1072 			}
1073 			break;
1074 
1075 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1076 			/* contiguity handled later */
1077 			break;
1078 
1079 		default:
1080 			BUG();
1081 		}
1082 
1083 		maddr &= PAGE_MASK;
1084 
1085 		/* create the mapping */
1086 		disp = phdr->p_vaddr & ~PAGE_MASK;
1087 		maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1088 				phdr->p_offset - disp);
1089 
1090 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1091 		       loop, phdr->p_memsz + disp, prot, flags,
1092 		       phdr->p_offset - disp, maddr);
1093 
1094 		if (IS_ERR_VALUE(maddr))
1095 			return (int) maddr;
1096 
1097 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1098 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1099 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1100 
1101 		seg->addr = maddr + disp;
1102 		seg->p_vaddr = phdr->p_vaddr;
1103 		seg->p_memsz = phdr->p_memsz;
1104 
1105 		/* map the ELF header address if in this segment */
1106 		if (phdr->p_offset == 0)
1107 			params->elfhdr_addr = seg->addr;
1108 
1109 		/* clear the bit between beginning of mapping and beginning of
1110 		 * PT_LOAD */
1111 		if (prot & PROT_WRITE && disp > 0) {
1112 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1113 			if (clear_user((void __user *) maddr, disp))
1114 				return -EFAULT;
1115 			maddr += disp;
1116 		}
1117 
1118 		/* clear any space allocated but not loaded
1119 		 * - on uClinux we can just clear the lot
1120 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1121 		 *   extant in the file
1122 		 */
1123 		excess = phdr->p_memsz - phdr->p_filesz;
1124 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1125 
1126 #ifdef CONFIG_MMU
1127 		if (excess > excess1) {
1128 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1129 			unsigned long xmaddr;
1130 
1131 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1132 			xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1133 					 prot, flags, 0);
1134 
1135 			kdebug("mmap[%d] <anon>"
1136 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1137 			       loop, xaddr, excess - excess1, prot, flags,
1138 			       xmaddr);
1139 
1140 			if (xmaddr != xaddr)
1141 				return -ENOMEM;
1142 		}
1143 
1144 		if (prot & PROT_WRITE && excess1 > 0) {
1145 			kdebug("clear[%d] ad=%lx sz=%lx",
1146 			       loop, maddr + phdr->p_filesz, excess1);
1147 			if (clear_user((void __user *) maddr + phdr->p_filesz,
1148 				       excess1))
1149 				return -EFAULT;
1150 		}
1151 
1152 #else
1153 		if (excess > 0) {
1154 			kdebug("clear[%d] ad=%lx sz=%lx",
1155 			       loop, maddr + phdr->p_filesz, excess);
1156 			if (clear_user((void *) maddr + phdr->p_filesz, excess))
1157 				return -EFAULT;
1158 		}
1159 #endif
1160 
1161 		if (mm) {
1162 			if (phdr->p_flags & PF_X) {
1163 				if (!mm->start_code) {
1164 					mm->start_code = maddr;
1165 					mm->end_code = maddr + phdr->p_memsz;
1166 				}
1167 			} else if (!mm->start_data) {
1168 				mm->start_data = maddr;
1169 				mm->end_data = maddr + phdr->p_memsz;
1170 			}
1171 		}
1172 
1173 		seg++;
1174 	}
1175 
1176 	return 0;
1177 }
1178 
1179 /*****************************************************************************/
1180 /*
1181  * ELF-FDPIC core dumper
1182  *
1183  * Modelled on fs/exec.c:aout_core_dump()
1184  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1185  *
1186  * Modelled on fs/binfmt_elf.c core dumper
1187  */
1188 #ifdef CONFIG_ELF_CORE
1189 
1190 /*
1191  * Decide whether a segment is worth dumping; default is yes to be
1192  * sure (missing info is worse than too much; etc).
1193  * Personally I'd include everything, and use the coredump limit...
1194  *
1195  * I think we should skip something. But I am not sure how. H.J.
1196  */
1197 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1198 {
1199 	int dump_ok;
1200 
1201 	/* Do not dump I/O mapped devices or special mappings */
1202 	if (vma->vm_flags & VM_IO) {
1203 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1204 		return 0;
1205 	}
1206 
1207 	/* If we may not read the contents, don't allow us to dump
1208 	 * them either. "dump_write()" can't handle it anyway.
1209 	 */
1210 	if (!(vma->vm_flags & VM_READ)) {
1211 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1212 		return 0;
1213 	}
1214 
1215 	/* support for DAX */
1216 	if (vma_is_dax(vma)) {
1217 		if (vma->vm_flags & VM_SHARED) {
1218 			dump_ok = test_bit(MMF_DUMP_DAX_SHARED, &mm_flags);
1219 			kdcore("%08lx: %08lx: %s (DAX shared)", vma->vm_start,
1220 			       vma->vm_flags, dump_ok ? "yes" : "no");
1221 		} else {
1222 			dump_ok = test_bit(MMF_DUMP_DAX_PRIVATE, &mm_flags);
1223 			kdcore("%08lx: %08lx: %s (DAX private)", vma->vm_start,
1224 			       vma->vm_flags, dump_ok ? "yes" : "no");
1225 		}
1226 		return dump_ok;
1227 	}
1228 
1229 	/* By default, dump shared memory if mapped from an anonymous file. */
1230 	if (vma->vm_flags & VM_SHARED) {
1231 		if (file_inode(vma->vm_file)->i_nlink == 0) {
1232 			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1233 			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1234 			       vma->vm_flags, dump_ok ? "yes" : "no");
1235 			return dump_ok;
1236 		}
1237 
1238 		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1239 		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1240 		       vma->vm_flags, dump_ok ? "yes" : "no");
1241 		return dump_ok;
1242 	}
1243 
1244 #ifdef CONFIG_MMU
1245 	/* By default, if it hasn't been written to, don't write it out */
1246 	if (!vma->anon_vma) {
1247 		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1248 		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1249 		       vma->vm_flags, dump_ok ? "yes" : "no");
1250 		return dump_ok;
1251 	}
1252 #endif
1253 
1254 	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1255 	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1256 	       dump_ok ? "yes" : "no");
1257 	return dump_ok;
1258 }
1259 
1260 /* An ELF note in memory */
1261 struct memelfnote
1262 {
1263 	const char *name;
1264 	int type;
1265 	unsigned int datasz;
1266 	void *data;
1267 };
1268 
1269 static int notesize(struct memelfnote *en)
1270 {
1271 	int sz;
1272 
1273 	sz = sizeof(struct elf_note);
1274 	sz += roundup(strlen(en->name) + 1, 4);
1275 	sz += roundup(en->datasz, 4);
1276 
1277 	return sz;
1278 }
1279 
1280 /* #define DEBUG */
1281 
1282 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1283 {
1284 	struct elf_note en;
1285 	en.n_namesz = strlen(men->name) + 1;
1286 	en.n_descsz = men->datasz;
1287 	en.n_type = men->type;
1288 
1289 	return dump_emit(cprm, &en, sizeof(en)) &&
1290 		dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1291 		dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1292 }
1293 
1294 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1295 {
1296 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1297 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1298 	elf->e_ident[EI_DATA] = ELF_DATA;
1299 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1300 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1301 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1302 
1303 	elf->e_type = ET_CORE;
1304 	elf->e_machine = ELF_ARCH;
1305 	elf->e_version = EV_CURRENT;
1306 	elf->e_entry = 0;
1307 	elf->e_phoff = sizeof(struct elfhdr);
1308 	elf->e_shoff = 0;
1309 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1310 	elf->e_ehsize = sizeof(struct elfhdr);
1311 	elf->e_phentsize = sizeof(struct elf_phdr);
1312 	elf->e_phnum = segs;
1313 	elf->e_shentsize = 0;
1314 	elf->e_shnum = 0;
1315 	elf->e_shstrndx = 0;
1316 	return;
1317 }
1318 
1319 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1320 {
1321 	phdr->p_type = PT_NOTE;
1322 	phdr->p_offset = offset;
1323 	phdr->p_vaddr = 0;
1324 	phdr->p_paddr = 0;
1325 	phdr->p_filesz = sz;
1326 	phdr->p_memsz = 0;
1327 	phdr->p_flags = 0;
1328 	phdr->p_align = 0;
1329 	return;
1330 }
1331 
1332 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1333 		unsigned int sz, void *data)
1334 {
1335 	note->name = name;
1336 	note->type = type;
1337 	note->datasz = sz;
1338 	note->data = data;
1339 	return;
1340 }
1341 
1342 /*
1343  * fill up all the fields in prstatus from the given task struct, except
1344  * registers which need to be filled up separately.
1345  */
1346 static void fill_prstatus(struct elf_prstatus *prstatus,
1347 			  struct task_struct *p, long signr)
1348 {
1349 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1350 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1351 	prstatus->pr_sighold = p->blocked.sig[0];
1352 	rcu_read_lock();
1353 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1354 	rcu_read_unlock();
1355 	prstatus->pr_pid = task_pid_vnr(p);
1356 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1357 	prstatus->pr_sid = task_session_vnr(p);
1358 	if (thread_group_leader(p)) {
1359 		struct task_cputime cputime;
1360 
1361 		/*
1362 		 * This is the record for the group leader.  It shows the
1363 		 * group-wide total, not its individual thread total.
1364 		 */
1365 		thread_group_cputime(p, &cputime);
1366 		prstatus->pr_utime = ns_to_timeval(cputime.utime);
1367 		prstatus->pr_stime = ns_to_timeval(cputime.stime);
1368 	} else {
1369 		u64 utime, stime;
1370 
1371 		task_cputime(p, &utime, &stime);
1372 		prstatus->pr_utime = ns_to_timeval(utime);
1373 		prstatus->pr_stime = ns_to_timeval(stime);
1374 	}
1375 	prstatus->pr_cutime = ns_to_timeval(p->signal->cutime);
1376 	prstatus->pr_cstime = ns_to_timeval(p->signal->cstime);
1377 
1378 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1379 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1380 }
1381 
1382 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1383 		       struct mm_struct *mm)
1384 {
1385 	const struct cred *cred;
1386 	unsigned int i, len;
1387 
1388 	/* first copy the parameters from user space */
1389 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1390 
1391 	len = mm->arg_end - mm->arg_start;
1392 	if (len >= ELF_PRARGSZ)
1393 		len = ELF_PRARGSZ - 1;
1394 	if (copy_from_user(&psinfo->pr_psargs,
1395 		           (const char __user *) mm->arg_start, len))
1396 		return -EFAULT;
1397 	for (i = 0; i < len; i++)
1398 		if (psinfo->pr_psargs[i] == 0)
1399 			psinfo->pr_psargs[i] = ' ';
1400 	psinfo->pr_psargs[len] = 0;
1401 
1402 	rcu_read_lock();
1403 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1404 	rcu_read_unlock();
1405 	psinfo->pr_pid = task_pid_vnr(p);
1406 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1407 	psinfo->pr_sid = task_session_vnr(p);
1408 
1409 	i = p->state ? ffz(~p->state) + 1 : 0;
1410 	psinfo->pr_state = i;
1411 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1412 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1413 	psinfo->pr_nice = task_nice(p);
1414 	psinfo->pr_flag = p->flags;
1415 	rcu_read_lock();
1416 	cred = __task_cred(p);
1417 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1418 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1419 	rcu_read_unlock();
1420 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1421 
1422 	return 0;
1423 }
1424 
1425 /* Here is the structure in which status of each thread is captured. */
1426 struct elf_thread_status
1427 {
1428 	struct list_head list;
1429 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1430 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1431 	struct task_struct *thread;
1432 #ifdef ELF_CORE_COPY_XFPREGS
1433 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1434 #endif
1435 	struct memelfnote notes[3];
1436 	int num_notes;
1437 };
1438 
1439 /*
1440  * In order to add the specific thread information for the elf file format,
1441  * we need to keep a linked list of every thread's pr_status and then create
1442  * a single section for them in the final core file.
1443  */
1444 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1445 {
1446 	struct task_struct *p = t->thread;
1447 	int sz = 0;
1448 
1449 	t->num_notes = 0;
1450 
1451 	fill_prstatus(&t->prstatus, p, signr);
1452 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1453 
1454 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1455 		  &t->prstatus);
1456 	t->num_notes++;
1457 	sz += notesize(&t->notes[0]);
1458 
1459 	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1460 	if (t->prstatus.pr_fpvalid) {
1461 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1462 			  &t->fpu);
1463 		t->num_notes++;
1464 		sz += notesize(&t->notes[1]);
1465 	}
1466 
1467 #ifdef ELF_CORE_COPY_XFPREGS
1468 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1469 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1470 			  sizeof(t->xfpu), &t->xfpu);
1471 		t->num_notes++;
1472 		sz += notesize(&t->notes[2]);
1473 	}
1474 #endif
1475 	return sz;
1476 }
1477 
1478 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1479 			     elf_addr_t e_shoff, int segs)
1480 {
1481 	elf->e_shoff = e_shoff;
1482 	elf->e_shentsize = sizeof(*shdr4extnum);
1483 	elf->e_shnum = 1;
1484 	elf->e_shstrndx = SHN_UNDEF;
1485 
1486 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1487 
1488 	shdr4extnum->sh_type = SHT_NULL;
1489 	shdr4extnum->sh_size = elf->e_shnum;
1490 	shdr4extnum->sh_link = elf->e_shstrndx;
1491 	shdr4extnum->sh_info = segs;
1492 }
1493 
1494 /*
1495  * dump the segments for an MMU process
1496  */
1497 static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1498 {
1499 	struct vm_area_struct *vma;
1500 
1501 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1502 #ifdef CONFIG_MMU
1503 		unsigned long addr;
1504 #endif
1505 
1506 		if (!maydump(vma, cprm->mm_flags))
1507 			continue;
1508 
1509 #ifdef CONFIG_MMU
1510 		for (addr = vma->vm_start; addr < vma->vm_end;
1511 							addr += PAGE_SIZE) {
1512 			bool res;
1513 			struct page *page = get_dump_page(addr);
1514 			if (page) {
1515 				void *kaddr = kmap(page);
1516 				res = dump_emit(cprm, kaddr, PAGE_SIZE);
1517 				kunmap(page);
1518 				put_page(page);
1519 			} else {
1520 				res = dump_skip(cprm, PAGE_SIZE);
1521 			}
1522 			if (!res)
1523 				return false;
1524 		}
1525 #else
1526 		if (!dump_emit(cprm, (void *) vma->vm_start,
1527 				vma->vm_end - vma->vm_start))
1528 			return false;
1529 #endif
1530 	}
1531 	return true;
1532 }
1533 
1534 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1535 {
1536 	struct vm_area_struct *vma;
1537 	size_t size = 0;
1538 
1539 	for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1540 		if (maydump(vma, mm_flags))
1541 			size += vma->vm_end - vma->vm_start;
1542 	return size;
1543 }
1544 
1545 /*
1546  * Actual dumper
1547  *
1548  * This is a two-pass process; first we find the offsets of the bits,
1549  * and then they are actually written out.  If we run out of core limit
1550  * we just truncate.
1551  */
1552 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1553 {
1554 #define	NUM_NOTES	6
1555 	int has_dumped = 0;
1556 	mm_segment_t fs;
1557 	int segs;
1558 	int i;
1559 	struct vm_area_struct *vma;
1560 	struct elfhdr *elf = NULL;
1561 	loff_t offset = 0, dataoff;
1562 	int numnote;
1563 	struct memelfnote *notes = NULL;
1564 	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1565 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1566  	LIST_HEAD(thread_list);
1567  	struct list_head *t;
1568 	elf_fpregset_t *fpu = NULL;
1569 #ifdef ELF_CORE_COPY_XFPREGS
1570 	elf_fpxregset_t *xfpu = NULL;
1571 #endif
1572 	int thread_status_size = 0;
1573 	elf_addr_t *auxv;
1574 	struct elf_phdr *phdr4note = NULL;
1575 	struct elf_shdr *shdr4extnum = NULL;
1576 	Elf_Half e_phnum;
1577 	elf_addr_t e_shoff;
1578 	struct core_thread *ct;
1579 	struct elf_thread_status *tmp;
1580 
1581 	/*
1582 	 * We no longer stop all VM operations.
1583 	 *
1584 	 * This is because those proceses that could possibly change map_count
1585 	 * or the mmap / vma pages are now blocked in do_exit on current
1586 	 * finishing this core dump.
1587 	 *
1588 	 * Only ptrace can touch these memory addresses, but it doesn't change
1589 	 * the map_count or the pages allocated. So no possibility of crashing
1590 	 * exists while dumping the mm->vm_next areas to the core file.
1591 	 */
1592 
1593 	/* alloc memory for large data structures: too large to be on stack */
1594 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1595 	if (!elf)
1596 		goto cleanup;
1597 	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1598 	if (!prstatus)
1599 		goto cleanup;
1600 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1601 	if (!psinfo)
1602 		goto cleanup;
1603 	notes = kmalloc_array(NUM_NOTES, sizeof(struct memelfnote),
1604 			      GFP_KERNEL);
1605 	if (!notes)
1606 		goto cleanup;
1607 	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1608 	if (!fpu)
1609 		goto cleanup;
1610 #ifdef ELF_CORE_COPY_XFPREGS
1611 	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1612 	if (!xfpu)
1613 		goto cleanup;
1614 #endif
1615 
1616 	for (ct = current->mm->core_state->dumper.next;
1617 					ct; ct = ct->next) {
1618 		tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1619 		if (!tmp)
1620 			goto cleanup;
1621 
1622 		tmp->thread = ct->task;
1623 		list_add(&tmp->list, &thread_list);
1624 	}
1625 
1626 	list_for_each(t, &thread_list) {
1627 		struct elf_thread_status *tmp;
1628 		int sz;
1629 
1630 		tmp = list_entry(t, struct elf_thread_status, list);
1631 		sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1632 		thread_status_size += sz;
1633 	}
1634 
1635 	/* now collect the dump for the current */
1636 	fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1637 	elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1638 
1639 	segs = current->mm->map_count;
1640 	segs += elf_core_extra_phdrs();
1641 
1642 	/* for notes section */
1643 	segs++;
1644 
1645 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1646 	 * this, kernel supports extended numbering. Have a look at
1647 	 * include/linux/elf.h for further information. */
1648 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1649 
1650 	/* Set up header */
1651 	fill_elf_fdpic_header(elf, e_phnum);
1652 
1653 	has_dumped = 1;
1654 	/*
1655 	 * Set up the notes in similar form to SVR4 core dumps made
1656 	 * with info from their /proc.
1657 	 */
1658 
1659 	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1660 	fill_psinfo(psinfo, current->group_leader, current->mm);
1661 	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1662 
1663 	numnote = 2;
1664 
1665 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1666 
1667 	i = 0;
1668 	do
1669 		i += 2;
1670 	while (auxv[i - 2] != AT_NULL);
1671 	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1672 		  i * sizeof(elf_addr_t), auxv);
1673 
1674   	/* Try to dump the FPU. */
1675 	if ((prstatus->pr_fpvalid =
1676 	     elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1677 		fill_note(notes + numnote++,
1678 			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1679 #ifdef ELF_CORE_COPY_XFPREGS
1680 	if (elf_core_copy_task_xfpregs(current, xfpu))
1681 		fill_note(notes + numnote++,
1682 			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1683 #endif
1684 
1685 	fs = get_fs();
1686 	set_fs(KERNEL_DS);
1687 
1688 	offset += sizeof(*elf);				/* Elf header */
1689 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1690 
1691 	/* Write notes phdr entry */
1692 	{
1693 		int sz = 0;
1694 
1695 		for (i = 0; i < numnote; i++)
1696 			sz += notesize(notes + i);
1697 
1698 		sz += thread_status_size;
1699 
1700 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1701 		if (!phdr4note)
1702 			goto end_coredump;
1703 
1704 		fill_elf_note_phdr(phdr4note, sz, offset);
1705 		offset += sz;
1706 	}
1707 
1708 	/* Page-align dumped data */
1709 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1710 
1711 	offset += elf_core_vma_data_size(cprm->mm_flags);
1712 	offset += elf_core_extra_data_size();
1713 	e_shoff = offset;
1714 
1715 	if (e_phnum == PN_XNUM) {
1716 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1717 		if (!shdr4extnum)
1718 			goto end_coredump;
1719 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1720 	}
1721 
1722 	offset = dataoff;
1723 
1724 	if (!dump_emit(cprm, elf, sizeof(*elf)))
1725 		goto end_coredump;
1726 
1727 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1728 		goto end_coredump;
1729 
1730 	/* write program headers for segments dump */
1731 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1732 		struct elf_phdr phdr;
1733 		size_t sz;
1734 
1735 		sz = vma->vm_end - vma->vm_start;
1736 
1737 		phdr.p_type = PT_LOAD;
1738 		phdr.p_offset = offset;
1739 		phdr.p_vaddr = vma->vm_start;
1740 		phdr.p_paddr = 0;
1741 		phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1742 		phdr.p_memsz = sz;
1743 		offset += phdr.p_filesz;
1744 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1745 		if (vma->vm_flags & VM_WRITE)
1746 			phdr.p_flags |= PF_W;
1747 		if (vma->vm_flags & VM_EXEC)
1748 			phdr.p_flags |= PF_X;
1749 		phdr.p_align = ELF_EXEC_PAGESIZE;
1750 
1751 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1752 			goto end_coredump;
1753 	}
1754 
1755 	if (!elf_core_write_extra_phdrs(cprm, offset))
1756 		goto end_coredump;
1757 
1758  	/* write out the notes section */
1759 	for (i = 0; i < numnote; i++)
1760 		if (!writenote(notes + i, cprm))
1761 			goto end_coredump;
1762 
1763 	/* write out the thread status notes section */
1764 	list_for_each(t, &thread_list) {
1765 		struct elf_thread_status *tmp =
1766 				list_entry(t, struct elf_thread_status, list);
1767 
1768 		for (i = 0; i < tmp->num_notes; i++)
1769 			if (!writenote(&tmp->notes[i], cprm))
1770 				goto end_coredump;
1771 	}
1772 
1773 	if (!dump_skip(cprm, dataoff - cprm->pos))
1774 		goto end_coredump;
1775 
1776 	if (!elf_fdpic_dump_segments(cprm))
1777 		goto end_coredump;
1778 
1779 	if (!elf_core_write_extra_data(cprm))
1780 		goto end_coredump;
1781 
1782 	if (e_phnum == PN_XNUM) {
1783 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1784 			goto end_coredump;
1785 	}
1786 
1787 	if (cprm->file->f_pos != offset) {
1788 		/* Sanity check */
1789 		printk(KERN_WARNING
1790 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1791 		       cprm->file->f_pos, offset);
1792 	}
1793 
1794 end_coredump:
1795 	set_fs(fs);
1796 
1797 cleanup:
1798 	while (!list_empty(&thread_list)) {
1799 		struct list_head *tmp = thread_list.next;
1800 		list_del(tmp);
1801 		kfree(list_entry(tmp, struct elf_thread_status, list));
1802 	}
1803 	kfree(phdr4note);
1804 	kfree(elf);
1805 	kfree(prstatus);
1806 	kfree(psinfo);
1807 	kfree(notes);
1808 	kfree(fpu);
1809 	kfree(shdr4extnum);
1810 #ifdef ELF_CORE_COPY_XFPREGS
1811 	kfree(xfpu);
1812 #endif
1813 	return has_dumped;
1814 #undef NUM_NOTES
1815 }
1816 
1817 #endif		/* CONFIG_ELF_CORE */
1818