xref: /linux/fs/binfmt_elf_fdpic.c (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
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/mm.h>
19 #include <linux/mman.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/slab.h>
27 #include <linux/pagemap.h>
28 #include <linux/security.h>
29 #include <linux/highmem.h>
30 #include <linux/highuid.h>
31 #include <linux/personality.h>
32 #include <linux/ptrace.h>
33 #include <linux/init.h>
34 #include <linux/elf.h>
35 #include <linux/elf-fdpic.h>
36 #include <linux/elfcore.h>
37 #include <linux/coredump.h>
38 #include <linux/dax.h>
39 
40 #include <asm/uaccess.h>
41 #include <asm/param.h>
42 #include <asm/pgalloc.h>
43 
44 typedef char *elf_caddr_t;
45 
46 #if 0
47 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
48 #else
49 #define kdebug(fmt, ...) do {} while(0)
50 #endif
51 
52 #if 0
53 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
54 #else
55 #define kdcore(fmt, ...) do {} while(0)
56 #endif
57 
58 MODULE_LICENSE("GPL");
59 
60 static int load_elf_fdpic_binary(struct linux_binprm *);
61 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
62 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
63 			      struct mm_struct *, const char *);
64 
65 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
66 				   struct elf_fdpic_params *,
67 				   struct elf_fdpic_params *);
68 
69 #ifndef CONFIG_MMU
70 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
71 						   struct file *,
72 						   struct mm_struct *);
73 #endif
74 
75 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
76 					     struct file *, struct mm_struct *);
77 
78 #ifdef CONFIG_ELF_CORE
79 static int elf_fdpic_core_dump(struct coredump_params *cprm);
80 #endif
81 
82 static struct linux_binfmt elf_fdpic_format = {
83 	.module		= THIS_MODULE,
84 	.load_binary	= load_elf_fdpic_binary,
85 #ifdef CONFIG_ELF_CORE
86 	.core_dump	= elf_fdpic_core_dump,
87 #endif
88 	.min_coredump	= ELF_EXEC_PAGESIZE,
89 };
90 
91 static int __init init_elf_fdpic_binfmt(void)
92 {
93 	register_binfmt(&elf_fdpic_format);
94 	return 0;
95 }
96 
97 static void __exit exit_elf_fdpic_binfmt(void)
98 {
99 	unregister_binfmt(&elf_fdpic_format);
100 }
101 
102 core_initcall(init_elf_fdpic_binfmt);
103 module_exit(exit_elf_fdpic_binfmt);
104 
105 static int is_elf(struct elfhdr *hdr, struct file *file)
106 {
107 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
108 		return 0;
109 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
110 		return 0;
111 	if (!elf_check_arch(hdr))
112 		return 0;
113 	if (!file->f_op->mmap)
114 		return 0;
115 	return 1;
116 }
117 
118 #ifndef elf_check_fdpic
119 #define elf_check_fdpic(x) 0
120 #endif
121 
122 #ifndef elf_check_const_displacement
123 #define elf_check_const_displacement(x) 0
124 #endif
125 
126 static int is_constdisp(struct elfhdr *hdr)
127 {
128 	if (!elf_check_fdpic(hdr))
129 		return 1;
130 	if (elf_check_const_displacement(hdr))
131 		return 1;
132 	return 0;
133 }
134 
135 /*****************************************************************************/
136 /*
137  * read the program headers table into memory
138  */
139 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
140 				 struct file *file)
141 {
142 	struct elf32_phdr *phdr;
143 	unsigned long size;
144 	int retval, loop;
145 
146 	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
147 		return -ENOMEM;
148 	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
149 		return -ENOMEM;
150 
151 	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
152 	params->phdrs = kmalloc(size, GFP_KERNEL);
153 	if (!params->phdrs)
154 		return -ENOMEM;
155 
156 	retval = kernel_read(file, params->hdr.e_phoff,
157 			     (char *) params->phdrs, size);
158 	if (unlikely(retval != size))
159 		return retval < 0 ? retval : -ENOEXEC;
160 
161 	/* determine stack size for this binary */
162 	phdr = params->phdrs;
163 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
164 		if (phdr->p_type != PT_GNU_STACK)
165 			continue;
166 
167 		if (phdr->p_flags & PF_X)
168 			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
169 		else
170 			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
171 
172 		params->stack_size = phdr->p_memsz;
173 		break;
174 	}
175 
176 	return 0;
177 }
178 
179 /*****************************************************************************/
180 /*
181  * load an fdpic binary into various bits of memory
182  */
183 static int load_elf_fdpic_binary(struct linux_binprm *bprm)
184 {
185 	struct elf_fdpic_params exec_params, interp_params;
186 	struct pt_regs *regs = current_pt_regs();
187 	struct elf_phdr *phdr;
188 	unsigned long stack_size, entryaddr;
189 #ifdef ELF_FDPIC_PLAT_INIT
190 	unsigned long dynaddr;
191 #endif
192 #ifndef CONFIG_MMU
193 	unsigned long stack_prot;
194 #endif
195 	struct file *interpreter = NULL; /* to shut gcc up */
196 	char *interpreter_name = NULL;
197 	int executable_stack;
198 	int retval, i;
199 
200 	kdebug("____ LOAD %d ____", current->pid);
201 
202 	memset(&exec_params, 0, sizeof(exec_params));
203 	memset(&interp_params, 0, sizeof(interp_params));
204 
205 	exec_params.hdr = *(struct elfhdr *) bprm->buf;
206 	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
207 
208 	/* check that this is a binary we know how to deal with */
209 	retval = -ENOEXEC;
210 	if (!is_elf(&exec_params.hdr, bprm->file))
211 		goto error;
212 	if (!elf_check_fdpic(&exec_params.hdr)) {
213 #ifdef CONFIG_MMU
214 		/* binfmt_elf handles non-fdpic elf except on nommu */
215 		goto error;
216 #else
217 		/* nommu can only load ET_DYN (PIE) ELF */
218 		if (exec_params.hdr.e_type != ET_DYN)
219 			goto error;
220 #endif
221 	}
222 
223 	/* read the program header table */
224 	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
225 	if (retval < 0)
226 		goto error;
227 
228 	/* scan for a program header that specifies an interpreter */
229 	phdr = exec_params.phdrs;
230 
231 	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
232 		switch (phdr->p_type) {
233 		case PT_INTERP:
234 			retval = -ENOMEM;
235 			if (phdr->p_filesz > PATH_MAX)
236 				goto error;
237 			retval = -ENOENT;
238 			if (phdr->p_filesz < 2)
239 				goto error;
240 
241 			/* read the name of the interpreter into memory */
242 			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
243 			if (!interpreter_name)
244 				goto error;
245 
246 			retval = kernel_read(bprm->file,
247 					     phdr->p_offset,
248 					     interpreter_name,
249 					     phdr->p_filesz);
250 			if (unlikely(retval != phdr->p_filesz)) {
251 				if (retval >= 0)
252 					retval = -ENOEXEC;
253 				goto error;
254 			}
255 
256 			retval = -ENOENT;
257 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
258 				goto error;
259 
260 			kdebug("Using ELF interpreter %s", interpreter_name);
261 
262 			/* replace the program with the interpreter */
263 			interpreter = open_exec(interpreter_name);
264 			retval = PTR_ERR(interpreter);
265 			if (IS_ERR(interpreter)) {
266 				interpreter = NULL;
267 				goto error;
268 			}
269 
270 			/*
271 			 * If the binary is not readable then enforce
272 			 * mm->dumpable = 0 regardless of the interpreter's
273 			 * permissions.
274 			 */
275 			would_dump(bprm, interpreter);
276 
277 			retval = kernel_read(interpreter, 0, bprm->buf,
278 					     BINPRM_BUF_SIZE);
279 			if (unlikely(retval != BINPRM_BUF_SIZE)) {
280 				if (retval >= 0)
281 					retval = -ENOEXEC;
282 				goto error;
283 			}
284 
285 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
286 			break;
287 
288 		case PT_LOAD:
289 #ifdef CONFIG_MMU
290 			if (exec_params.load_addr == 0)
291 				exec_params.load_addr = phdr->p_vaddr;
292 #endif
293 			break;
294 		}
295 
296 	}
297 
298 	if (is_constdisp(&exec_params.hdr))
299 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
300 
301 	/* perform insanity checks on the interpreter */
302 	if (interpreter_name) {
303 		retval = -ELIBBAD;
304 		if (!is_elf(&interp_params.hdr, interpreter))
305 			goto error;
306 
307 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
308 
309 		/* read the interpreter's program header table */
310 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
311 		if (retval < 0)
312 			goto error;
313 	}
314 
315 	stack_size = exec_params.stack_size;
316 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
317 		executable_stack = EXSTACK_ENABLE_X;
318 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
319 		executable_stack = EXSTACK_DISABLE_X;
320 	else
321 		executable_stack = EXSTACK_DEFAULT;
322 
323 	if (stack_size == 0) {
324 		stack_size = interp_params.stack_size;
325 		if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
326 			executable_stack = EXSTACK_ENABLE_X;
327 		else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
328 			executable_stack = EXSTACK_DISABLE_X;
329 		else
330 			executable_stack = EXSTACK_DEFAULT;
331 	}
332 
333 	retval = -ENOEXEC;
334 	if (stack_size == 0)
335 		stack_size = 131072UL; /* same as exec.c's default commit */
336 
337 	if (is_constdisp(&interp_params.hdr))
338 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
339 
340 	/* flush all traces of the currently running executable */
341 	retval = flush_old_exec(bprm);
342 	if (retval)
343 		goto error;
344 
345 	/* there's now no turning back... the old userspace image is dead,
346 	 * defunct, deceased, etc.
347 	 */
348 	if (elf_check_fdpic(&exec_params.hdr))
349 		set_personality(PER_LINUX_FDPIC);
350 	else
351 		set_personality(PER_LINUX);
352 	if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
353 		current->personality |= READ_IMPLIES_EXEC;
354 
355 	setup_new_exec(bprm);
356 
357 	set_binfmt(&elf_fdpic_format);
358 
359 	current->mm->start_code = 0;
360 	current->mm->end_code = 0;
361 	current->mm->start_stack = 0;
362 	current->mm->start_data = 0;
363 	current->mm->end_data = 0;
364 	current->mm->context.exec_fdpic_loadmap = 0;
365 	current->mm->context.interp_fdpic_loadmap = 0;
366 
367 #ifdef CONFIG_MMU
368 	elf_fdpic_arch_lay_out_mm(&exec_params,
369 				  &interp_params,
370 				  &current->mm->start_stack,
371 				  &current->mm->start_brk);
372 
373 	retval = setup_arg_pages(bprm, current->mm->start_stack,
374 				 executable_stack);
375 	if (retval < 0)
376 		goto error;
377 #endif
378 
379 	/* load the executable and interpreter into memory */
380 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
381 				    "executable");
382 	if (retval < 0)
383 		goto error;
384 
385 	if (interpreter_name) {
386 		retval = elf_fdpic_map_file(&interp_params, interpreter,
387 					    current->mm, "interpreter");
388 		if (retval < 0) {
389 			printk(KERN_ERR "Unable to load interpreter\n");
390 			goto error;
391 		}
392 
393 		allow_write_access(interpreter);
394 		fput(interpreter);
395 		interpreter = NULL;
396 	}
397 
398 #ifdef CONFIG_MMU
399 	if (!current->mm->start_brk)
400 		current->mm->start_brk = current->mm->end_data;
401 
402 	current->mm->brk = current->mm->start_brk =
403 		PAGE_ALIGN(current->mm->start_brk);
404 
405 #else
406 	/* create a stack area and zero-size brk area */
407 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
408 	if (stack_size < PAGE_SIZE * 2)
409 		stack_size = PAGE_SIZE * 2;
410 
411 	stack_prot = PROT_READ | PROT_WRITE;
412 	if (executable_stack == EXSTACK_ENABLE_X ||
413 	    (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
414 		stack_prot |= PROT_EXEC;
415 
416 	current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
417 					 MAP_PRIVATE | MAP_ANONYMOUS |
418 					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
419 					 0);
420 
421 	if (IS_ERR_VALUE(current->mm->start_brk)) {
422 		retval = current->mm->start_brk;
423 		current->mm->start_brk = 0;
424 		goto error;
425 	}
426 
427 	current->mm->brk = current->mm->start_brk;
428 	current->mm->context.end_brk = current->mm->start_brk;
429 	current->mm->start_stack = current->mm->start_brk + stack_size;
430 #endif
431 
432 	install_exec_creds(bprm);
433 	if (create_elf_fdpic_tables(bprm, current->mm,
434 				    &exec_params, &interp_params) < 0)
435 		goto error;
436 
437 	kdebug("- start_code  %lx", current->mm->start_code);
438 	kdebug("- end_code    %lx", current->mm->end_code);
439 	kdebug("- start_data  %lx", current->mm->start_data);
440 	kdebug("- end_data    %lx", current->mm->end_data);
441 	kdebug("- start_brk   %lx", current->mm->start_brk);
442 	kdebug("- brk         %lx", current->mm->brk);
443 	kdebug("- start_stack %lx", current->mm->start_stack);
444 
445 #ifdef ELF_FDPIC_PLAT_INIT
446 	/*
447 	 * The ABI may specify that certain registers be set up in special
448 	 * ways (on i386 %edx is the address of a DT_FINI function, for
449 	 * example.  This macro performs whatever initialization to
450 	 * the regs structure is required.
451 	 */
452 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
453 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
454 			    dynaddr);
455 #endif
456 
457 	/* everything is now ready... get the userspace context ready to roll */
458 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
459 	start_thread(regs, entryaddr, current->mm->start_stack);
460 
461 	retval = 0;
462 
463 error:
464 	if (interpreter) {
465 		allow_write_access(interpreter);
466 		fput(interpreter);
467 	}
468 	kfree(interpreter_name);
469 	kfree(exec_params.phdrs);
470 	kfree(exec_params.loadmap);
471 	kfree(interp_params.phdrs);
472 	kfree(interp_params.loadmap);
473 	return retval;
474 }
475 
476 /*****************************************************************************/
477 
478 #ifndef ELF_BASE_PLATFORM
479 /*
480  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
481  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
482  * will be copied to the user stack in the same manner as AT_PLATFORM.
483  */
484 #define ELF_BASE_PLATFORM NULL
485 #endif
486 
487 /*
488  * present useful information to the program by shovelling it onto the new
489  * process's stack
490  */
491 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
492 				   struct mm_struct *mm,
493 				   struct elf_fdpic_params *exec_params,
494 				   struct elf_fdpic_params *interp_params)
495 {
496 	const struct cred *cred = current_cred();
497 	unsigned long sp, csp, nitems;
498 	elf_caddr_t __user *argv, *envp;
499 	size_t platform_len = 0, len;
500 	char *k_platform, *k_base_platform;
501 	char __user *u_platform, *u_base_platform, *p;
502 	int loop;
503 	int nr;	/* reset for each csp adjustment */
504 
505 #ifdef CONFIG_MMU
506 	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
507 	 * by the processes running on the same package. One thing we can do is
508 	 * to shuffle the initial stack for them, so we give the architecture
509 	 * an opportunity to do so here.
510 	 */
511 	sp = arch_align_stack(bprm->p);
512 #else
513 	sp = mm->start_stack;
514 
515 	/* stack the program arguments and environment */
516 	if (transfer_args_to_stack(bprm, &sp) < 0)
517 		return -EFAULT;
518 	sp &= ~15;
519 #endif
520 
521 	/*
522 	 * If this architecture has a platform capability string, copy it
523 	 * to userspace.  In some cases (Sparc), this info is impossible
524 	 * for userspace to get any other way, in others (i386) it is
525 	 * merely difficult.
526 	 */
527 	k_platform = ELF_PLATFORM;
528 	u_platform = NULL;
529 
530 	if (k_platform) {
531 		platform_len = strlen(k_platform) + 1;
532 		sp -= platform_len;
533 		u_platform = (char __user *) sp;
534 		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
535 			return -EFAULT;
536 	}
537 
538 	/*
539 	 * If this architecture has a "base" platform capability
540 	 * string, copy it to userspace.
541 	 */
542 	k_base_platform = ELF_BASE_PLATFORM;
543 	u_base_platform = NULL;
544 
545 	if (k_base_platform) {
546 		platform_len = strlen(k_base_platform) + 1;
547 		sp -= platform_len;
548 		u_base_platform = (char __user *) sp;
549 		if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
550 			return -EFAULT;
551 	}
552 
553 	sp &= ~7UL;
554 
555 	/* stack the load map(s) */
556 	len = sizeof(struct elf32_fdpic_loadmap);
557 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
558 	sp = (sp - len) & ~7UL;
559 	exec_params->map_addr = sp;
560 
561 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
562 		return -EFAULT;
563 
564 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
565 
566 	if (interp_params->loadmap) {
567 		len = sizeof(struct elf32_fdpic_loadmap);
568 		len += sizeof(struct elf32_fdpic_loadseg) *
569 			interp_params->loadmap->nsegs;
570 		sp = (sp - len) & ~7UL;
571 		interp_params->map_addr = sp;
572 
573 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
574 				 len) != 0)
575 			return -EFAULT;
576 
577 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
578 	}
579 
580 	/* force 16 byte _final_ alignment here for generality */
581 #define DLINFO_ITEMS 15
582 
583 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
584 		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
585 
586 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
587 		nitems++;
588 
589 	csp = sp;
590 	sp -= nitems * 2 * sizeof(unsigned long);
591 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
592 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
593 	sp -= 1 * sizeof(unsigned long);		/* argc */
594 
595 	csp -= sp & 15UL;
596 	sp -= sp & 15UL;
597 
598 	/* put the ELF interpreter info on the stack */
599 #define NEW_AUX_ENT(id, val)						\
600 	do {								\
601 		struct { unsigned long _id, _val; } __user *ent;	\
602 									\
603 		ent = (void __user *) csp;				\
604 		__put_user((id), &ent[nr]._id);				\
605 		__put_user((val), &ent[nr]._val);			\
606 		nr++;							\
607 	} while (0)
608 
609 	nr = 0;
610 	csp -= 2 * sizeof(unsigned long);
611 	NEW_AUX_ENT(AT_NULL, 0);
612 	if (k_platform) {
613 		nr = 0;
614 		csp -= 2 * sizeof(unsigned long);
615 		NEW_AUX_ENT(AT_PLATFORM,
616 			    (elf_addr_t) (unsigned long) u_platform);
617 	}
618 
619 	if (k_base_platform) {
620 		nr = 0;
621 		csp -= 2 * sizeof(unsigned long);
622 		NEW_AUX_ENT(AT_BASE_PLATFORM,
623 			    (elf_addr_t) (unsigned long) u_base_platform);
624 	}
625 
626 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
627 		nr = 0;
628 		csp -= 2 * sizeof(unsigned long);
629 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
630 	}
631 
632 	nr = 0;
633 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
634 	NEW_AUX_ENT(AT_HWCAP,	ELF_HWCAP);
635 #ifdef ELF_HWCAP2
636 	NEW_AUX_ENT(AT_HWCAP2,	ELF_HWCAP2);
637 #endif
638 	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
639 	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
640 	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
641 	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
642 	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
643 	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
644 	NEW_AUX_ENT(AT_FLAGS,	0);
645 	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
646 	NEW_AUX_ENT(AT_UID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
647 	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
648 	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
649 	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
650 	NEW_AUX_ENT(AT_SECURE,	security_bprm_secureexec(bprm));
651 	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
652 
653 #ifdef ARCH_DLINFO
654 	nr = 0;
655 	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
656 
657 	/* ARCH_DLINFO must come last so platform specific code can enforce
658 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
659 	 */
660 	ARCH_DLINFO;
661 #endif
662 #undef NEW_AUX_ENT
663 
664 	/* allocate room for argv[] and envv[] */
665 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
666 	envp = (elf_caddr_t __user *) csp;
667 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
668 	argv = (elf_caddr_t __user *) csp;
669 
670 	/* stack argc */
671 	csp -= sizeof(unsigned long);
672 	__put_user(bprm->argc, (unsigned long __user *) csp);
673 
674 	BUG_ON(csp != sp);
675 
676 	/* fill in the argv[] array */
677 #ifdef CONFIG_MMU
678 	current->mm->arg_start = bprm->p;
679 #else
680 	current->mm->arg_start = current->mm->start_stack -
681 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
682 #endif
683 
684 	p = (char __user *) current->mm->arg_start;
685 	for (loop = bprm->argc; loop > 0; loop--) {
686 		__put_user((elf_caddr_t) p, argv++);
687 		len = strnlen_user(p, MAX_ARG_STRLEN);
688 		if (!len || len > MAX_ARG_STRLEN)
689 			return -EINVAL;
690 		p += len;
691 	}
692 	__put_user(NULL, argv);
693 	current->mm->arg_end = (unsigned long) p;
694 
695 	/* fill in the envv[] array */
696 	current->mm->env_start = (unsigned long) p;
697 	for (loop = bprm->envc; loop > 0; loop--) {
698 		__put_user((elf_caddr_t)(unsigned long) p, envp++);
699 		len = strnlen_user(p, MAX_ARG_STRLEN);
700 		if (!len || len > MAX_ARG_STRLEN)
701 			return -EINVAL;
702 		p += len;
703 	}
704 	__put_user(NULL, envp);
705 	current->mm->env_end = (unsigned long) p;
706 
707 	mm->start_stack = (unsigned long) sp;
708 	return 0;
709 }
710 
711 /*****************************************************************************/
712 /*
713  * load the appropriate binary image (executable or interpreter) into memory
714  * - we assume no MMU is available
715  * - if no other PIC bits are set in params->hdr->e_flags
716  *   - we assume that the LOADable segments in the binary are independently relocatable
717  *   - we assume R/O executable segments are shareable
718  * - else
719  *   - we assume the loadable parts of the image to require fixed displacement
720  *   - the image is not shareable
721  */
722 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
723 			      struct file *file,
724 			      struct mm_struct *mm,
725 			      const char *what)
726 {
727 	struct elf32_fdpic_loadmap *loadmap;
728 #ifdef CONFIG_MMU
729 	struct elf32_fdpic_loadseg *mseg;
730 #endif
731 	struct elf32_fdpic_loadseg *seg;
732 	struct elf32_phdr *phdr;
733 	unsigned long load_addr, stop;
734 	unsigned nloads, tmp;
735 	size_t size;
736 	int loop, ret;
737 
738 	/* allocate a load map table */
739 	nloads = 0;
740 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
741 		if (params->phdrs[loop].p_type == PT_LOAD)
742 			nloads++;
743 
744 	if (nloads == 0)
745 		return -ELIBBAD;
746 
747 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
748 	loadmap = kzalloc(size, GFP_KERNEL);
749 	if (!loadmap)
750 		return -ENOMEM;
751 
752 	params->loadmap = loadmap;
753 
754 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
755 	loadmap->nsegs = nloads;
756 
757 	load_addr = params->load_addr;
758 	seg = loadmap->segs;
759 
760 	/* map the requested LOADs into the memory space */
761 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
762 	case ELF_FDPIC_FLAG_CONSTDISP:
763 	case ELF_FDPIC_FLAG_CONTIGUOUS:
764 #ifndef CONFIG_MMU
765 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
766 		if (ret < 0)
767 			return ret;
768 		break;
769 #endif
770 	default:
771 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
772 		if (ret < 0)
773 			return ret;
774 		break;
775 	}
776 
777 	/* map the entry point */
778 	if (params->hdr.e_entry) {
779 		seg = loadmap->segs;
780 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
781 			if (params->hdr.e_entry >= seg->p_vaddr &&
782 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
783 				params->entry_addr =
784 					(params->hdr.e_entry - seg->p_vaddr) +
785 					seg->addr;
786 				break;
787 			}
788 		}
789 	}
790 
791 	/* determine where the program header table has wound up if mapped */
792 	stop = params->hdr.e_phoff;
793 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
794 	phdr = params->phdrs;
795 
796 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
797 		if (phdr->p_type != PT_LOAD)
798 			continue;
799 
800 		if (phdr->p_offset > params->hdr.e_phoff ||
801 		    phdr->p_offset + phdr->p_filesz < stop)
802 			continue;
803 
804 		seg = loadmap->segs;
805 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
806 			if (phdr->p_vaddr >= seg->p_vaddr &&
807 			    phdr->p_vaddr + phdr->p_filesz <=
808 			    seg->p_vaddr + seg->p_memsz) {
809 				params->ph_addr =
810 					(phdr->p_vaddr - seg->p_vaddr) +
811 					seg->addr +
812 					params->hdr.e_phoff - phdr->p_offset;
813 				break;
814 			}
815 		}
816 		break;
817 	}
818 
819 	/* determine where the dynamic section has wound up if there is one */
820 	phdr = params->phdrs;
821 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
822 		if (phdr->p_type != PT_DYNAMIC)
823 			continue;
824 
825 		seg = loadmap->segs;
826 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
827 			if (phdr->p_vaddr >= seg->p_vaddr &&
828 			    phdr->p_vaddr + phdr->p_memsz <=
829 			    seg->p_vaddr + seg->p_memsz) {
830 				params->dynamic_addr =
831 					(phdr->p_vaddr - seg->p_vaddr) +
832 					seg->addr;
833 
834 				/* check the dynamic section contains at least
835 				 * one item, and that the last item is a NULL
836 				 * entry */
837 				if (phdr->p_memsz == 0 ||
838 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
839 					goto dynamic_error;
840 
841 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
842 				if (((Elf32_Dyn *)
843 				     params->dynamic_addr)[tmp - 1].d_tag != 0)
844 					goto dynamic_error;
845 				break;
846 			}
847 		}
848 		break;
849 	}
850 
851 	/* now elide adjacent segments in the load map on MMU linux
852 	 * - on uClinux the holes between may actually be filled with system
853 	 *   stuff or stuff from other processes
854 	 */
855 #ifdef CONFIG_MMU
856 	nloads = loadmap->nsegs;
857 	mseg = loadmap->segs;
858 	seg = mseg + 1;
859 	for (loop = 1; loop < nloads; loop++) {
860 		/* see if we have a candidate for merging */
861 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
862 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
863 			if (load_addr == (seg->addr & PAGE_MASK)) {
864 				mseg->p_memsz +=
865 					load_addr -
866 					(mseg->addr + mseg->p_memsz);
867 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
868 				mseg->p_memsz += seg->p_memsz;
869 				loadmap->nsegs--;
870 				continue;
871 			}
872 		}
873 
874 		mseg++;
875 		if (mseg != seg)
876 			*mseg = *seg;
877 	}
878 #endif
879 
880 	kdebug("Mapped Object [%s]:", what);
881 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
882 	kdebug("- entry    : %lx", params->entry_addr);
883 	kdebug("- PHDR[]   : %lx", params->ph_addr);
884 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
885 	seg = loadmap->segs;
886 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
887 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
888 		       loop,
889 		       seg->addr, seg->addr + seg->p_memsz - 1,
890 		       seg->p_vaddr, seg->p_memsz);
891 
892 	return 0;
893 
894 dynamic_error:
895 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
896 	       what, file_inode(file)->i_ino);
897 	return -ELIBBAD;
898 }
899 
900 /*****************************************************************************/
901 /*
902  * map a file with constant displacement under uClinux
903  */
904 #ifndef CONFIG_MMU
905 static int elf_fdpic_map_file_constdisp_on_uclinux(
906 	struct elf_fdpic_params *params,
907 	struct file *file,
908 	struct mm_struct *mm)
909 {
910 	struct elf32_fdpic_loadseg *seg;
911 	struct elf32_phdr *phdr;
912 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
913 	int loop, ret;
914 
915 	load_addr = params->load_addr;
916 	seg = params->loadmap->segs;
917 
918 	/* determine the bounds of the contiguous overall allocation we must
919 	 * make */
920 	phdr = params->phdrs;
921 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
922 		if (params->phdrs[loop].p_type != PT_LOAD)
923 			continue;
924 
925 		if (base > phdr->p_vaddr)
926 			base = phdr->p_vaddr;
927 		if (top < phdr->p_vaddr + phdr->p_memsz)
928 			top = phdr->p_vaddr + phdr->p_memsz;
929 	}
930 
931 	/* allocate one big anon block for everything */
932 	mflags = MAP_PRIVATE;
933 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
934 		mflags |= MAP_EXECUTABLE;
935 
936 	maddr = vm_mmap(NULL, load_addr, top - base,
937 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
938 	if (IS_ERR_VALUE(maddr))
939 		return (int) maddr;
940 
941 	if (load_addr != 0)
942 		load_addr += PAGE_ALIGN(top - base);
943 
944 	/* and then load the file segments into it */
945 	phdr = params->phdrs;
946 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
947 		if (params->phdrs[loop].p_type != PT_LOAD)
948 			continue;
949 
950 		seg->addr = maddr + (phdr->p_vaddr - base);
951 		seg->p_vaddr = phdr->p_vaddr;
952 		seg->p_memsz = phdr->p_memsz;
953 
954 		ret = read_code(file, seg->addr, phdr->p_offset,
955 				       phdr->p_filesz);
956 		if (ret < 0)
957 			return ret;
958 
959 		/* map the ELF header address if in this segment */
960 		if (phdr->p_offset == 0)
961 			params->elfhdr_addr = seg->addr;
962 
963 		/* clear any space allocated but not loaded */
964 		if (phdr->p_filesz < phdr->p_memsz) {
965 			if (clear_user((void *) (seg->addr + phdr->p_filesz),
966 				       phdr->p_memsz - phdr->p_filesz))
967 				return -EFAULT;
968 		}
969 
970 		if (mm) {
971 			if (phdr->p_flags & PF_X) {
972 				if (!mm->start_code) {
973 					mm->start_code = seg->addr;
974 					mm->end_code = seg->addr +
975 						phdr->p_memsz;
976 				}
977 			} else if (!mm->start_data) {
978 				mm->start_data = seg->addr;
979 				mm->end_data = seg->addr + phdr->p_memsz;
980 			}
981 		}
982 
983 		seg++;
984 	}
985 
986 	return 0;
987 }
988 #endif
989 
990 /*****************************************************************************/
991 /*
992  * map a binary by direct mmap() of the individual PT_LOAD segments
993  */
994 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
995 					     struct file *file,
996 					     struct mm_struct *mm)
997 {
998 	struct elf32_fdpic_loadseg *seg;
999 	struct elf32_phdr *phdr;
1000 	unsigned long load_addr, delta_vaddr;
1001 	int loop, dvset;
1002 
1003 	load_addr = params->load_addr;
1004 	delta_vaddr = 0;
1005 	dvset = 0;
1006 
1007 	seg = params->loadmap->segs;
1008 
1009 	/* deal with each load segment separately */
1010 	phdr = params->phdrs;
1011 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1012 		unsigned long maddr, disp, excess, excess1;
1013 		int prot = 0, flags;
1014 
1015 		if (phdr->p_type != PT_LOAD)
1016 			continue;
1017 
1018 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1019 		       (unsigned long) phdr->p_vaddr,
1020 		       (unsigned long) phdr->p_offset,
1021 		       (unsigned long) phdr->p_filesz,
1022 		       (unsigned long) phdr->p_memsz);
1023 
1024 		/* determine the mapping parameters */
1025 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1026 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1027 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1028 
1029 		flags = MAP_PRIVATE | MAP_DENYWRITE;
1030 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1031 			flags |= MAP_EXECUTABLE;
1032 
1033 		maddr = 0;
1034 
1035 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1036 		case ELF_FDPIC_FLAG_INDEPENDENT:
1037 			/* PT_LOADs are independently locatable */
1038 			break;
1039 
1040 		case ELF_FDPIC_FLAG_HONOURVADDR:
1041 			/* the specified virtual address must be honoured */
1042 			maddr = phdr->p_vaddr;
1043 			flags |= MAP_FIXED;
1044 			break;
1045 
1046 		case ELF_FDPIC_FLAG_CONSTDISP:
1047 			/* constant displacement
1048 			 * - can be mapped anywhere, but must be mapped as a
1049 			 *   unit
1050 			 */
1051 			if (!dvset) {
1052 				maddr = load_addr;
1053 				delta_vaddr = phdr->p_vaddr;
1054 				dvset = 1;
1055 			} else {
1056 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1057 				flags |= MAP_FIXED;
1058 			}
1059 			break;
1060 
1061 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1062 			/* contiguity handled later */
1063 			break;
1064 
1065 		default:
1066 			BUG();
1067 		}
1068 
1069 		maddr &= PAGE_MASK;
1070 
1071 		/* create the mapping */
1072 		disp = phdr->p_vaddr & ~PAGE_MASK;
1073 		maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1074 				phdr->p_offset - disp);
1075 
1076 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1077 		       loop, phdr->p_memsz + disp, prot, flags,
1078 		       phdr->p_offset - disp, maddr);
1079 
1080 		if (IS_ERR_VALUE(maddr))
1081 			return (int) maddr;
1082 
1083 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1084 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1085 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1086 
1087 		seg->addr = maddr + disp;
1088 		seg->p_vaddr = phdr->p_vaddr;
1089 		seg->p_memsz = phdr->p_memsz;
1090 
1091 		/* map the ELF header address if in this segment */
1092 		if (phdr->p_offset == 0)
1093 			params->elfhdr_addr = seg->addr;
1094 
1095 		/* clear the bit between beginning of mapping and beginning of
1096 		 * PT_LOAD */
1097 		if (prot & PROT_WRITE && disp > 0) {
1098 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1099 			if (clear_user((void __user *) maddr, disp))
1100 				return -EFAULT;
1101 			maddr += disp;
1102 		}
1103 
1104 		/* clear any space allocated but not loaded
1105 		 * - on uClinux we can just clear the lot
1106 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1107 		 *   extant in the file
1108 		 */
1109 		excess = phdr->p_memsz - phdr->p_filesz;
1110 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1111 
1112 #ifdef CONFIG_MMU
1113 		if (excess > excess1) {
1114 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1115 			unsigned long xmaddr;
1116 
1117 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1118 			xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1119 					 prot, flags, 0);
1120 
1121 			kdebug("mmap[%d] <anon>"
1122 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1123 			       loop, xaddr, excess - excess1, prot, flags,
1124 			       xmaddr);
1125 
1126 			if (xmaddr != xaddr)
1127 				return -ENOMEM;
1128 		}
1129 
1130 		if (prot & PROT_WRITE && excess1 > 0) {
1131 			kdebug("clear[%d] ad=%lx sz=%lx",
1132 			       loop, maddr + phdr->p_filesz, excess1);
1133 			if (clear_user((void __user *) maddr + phdr->p_filesz,
1134 				       excess1))
1135 				return -EFAULT;
1136 		}
1137 
1138 #else
1139 		if (excess > 0) {
1140 			kdebug("clear[%d] ad=%lx sz=%lx",
1141 			       loop, maddr + phdr->p_filesz, excess);
1142 			if (clear_user((void *) maddr + phdr->p_filesz, excess))
1143 				return -EFAULT;
1144 		}
1145 #endif
1146 
1147 		if (mm) {
1148 			if (phdr->p_flags & PF_X) {
1149 				if (!mm->start_code) {
1150 					mm->start_code = maddr;
1151 					mm->end_code = maddr + phdr->p_memsz;
1152 				}
1153 			} else if (!mm->start_data) {
1154 				mm->start_data = maddr;
1155 				mm->end_data = maddr + phdr->p_memsz;
1156 			}
1157 		}
1158 
1159 		seg++;
1160 	}
1161 
1162 	return 0;
1163 }
1164 
1165 /*****************************************************************************/
1166 /*
1167  * ELF-FDPIC core dumper
1168  *
1169  * Modelled on fs/exec.c:aout_core_dump()
1170  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1171  *
1172  * Modelled on fs/binfmt_elf.c core dumper
1173  */
1174 #ifdef CONFIG_ELF_CORE
1175 
1176 /*
1177  * Decide whether a segment is worth dumping; default is yes to be
1178  * sure (missing info is worse than too much; etc).
1179  * Personally I'd include everything, and use the coredump limit...
1180  *
1181  * I think we should skip something. But I am not sure how. H.J.
1182  */
1183 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1184 {
1185 	int dump_ok;
1186 
1187 	/* Do not dump I/O mapped devices or special mappings */
1188 	if (vma->vm_flags & VM_IO) {
1189 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1190 		return 0;
1191 	}
1192 
1193 	/* If we may not read the contents, don't allow us to dump
1194 	 * them either. "dump_write()" can't handle it anyway.
1195 	 */
1196 	if (!(vma->vm_flags & VM_READ)) {
1197 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1198 		return 0;
1199 	}
1200 
1201 	/* support for DAX */
1202 	if (vma_is_dax(vma)) {
1203 		if (vma->vm_flags & VM_SHARED) {
1204 			dump_ok = test_bit(MMF_DUMP_DAX_SHARED, &mm_flags);
1205 			kdcore("%08lx: %08lx: %s (DAX shared)", vma->vm_start,
1206 			       vma->vm_flags, dump_ok ? "yes" : "no");
1207 		} else {
1208 			dump_ok = test_bit(MMF_DUMP_DAX_PRIVATE, &mm_flags);
1209 			kdcore("%08lx: %08lx: %s (DAX private)", vma->vm_start,
1210 			       vma->vm_flags, dump_ok ? "yes" : "no");
1211 		}
1212 		return dump_ok;
1213 	}
1214 
1215 	/* By default, dump shared memory if mapped from an anonymous file. */
1216 	if (vma->vm_flags & VM_SHARED) {
1217 		if (file_inode(vma->vm_file)->i_nlink == 0) {
1218 			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1219 			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1220 			       vma->vm_flags, dump_ok ? "yes" : "no");
1221 			return dump_ok;
1222 		}
1223 
1224 		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1225 		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1226 		       vma->vm_flags, dump_ok ? "yes" : "no");
1227 		return dump_ok;
1228 	}
1229 
1230 #ifdef CONFIG_MMU
1231 	/* By default, if it hasn't been written to, don't write it out */
1232 	if (!vma->anon_vma) {
1233 		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1234 		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1235 		       vma->vm_flags, dump_ok ? "yes" : "no");
1236 		return dump_ok;
1237 	}
1238 #endif
1239 
1240 	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1241 	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1242 	       dump_ok ? "yes" : "no");
1243 	return dump_ok;
1244 }
1245 
1246 /* An ELF note in memory */
1247 struct memelfnote
1248 {
1249 	const char *name;
1250 	int type;
1251 	unsigned int datasz;
1252 	void *data;
1253 };
1254 
1255 static int notesize(struct memelfnote *en)
1256 {
1257 	int sz;
1258 
1259 	sz = sizeof(struct elf_note);
1260 	sz += roundup(strlen(en->name) + 1, 4);
1261 	sz += roundup(en->datasz, 4);
1262 
1263 	return sz;
1264 }
1265 
1266 /* #define DEBUG */
1267 
1268 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1269 {
1270 	struct elf_note en;
1271 	en.n_namesz = strlen(men->name) + 1;
1272 	en.n_descsz = men->datasz;
1273 	en.n_type = men->type;
1274 
1275 	return dump_emit(cprm, &en, sizeof(en)) &&
1276 		dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1277 		dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1278 }
1279 
1280 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1281 {
1282 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1283 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1284 	elf->e_ident[EI_DATA] = ELF_DATA;
1285 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1286 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1287 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1288 
1289 	elf->e_type = ET_CORE;
1290 	elf->e_machine = ELF_ARCH;
1291 	elf->e_version = EV_CURRENT;
1292 	elf->e_entry = 0;
1293 	elf->e_phoff = sizeof(struct elfhdr);
1294 	elf->e_shoff = 0;
1295 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1296 	elf->e_ehsize = sizeof(struct elfhdr);
1297 	elf->e_phentsize = sizeof(struct elf_phdr);
1298 	elf->e_phnum = segs;
1299 	elf->e_shentsize = 0;
1300 	elf->e_shnum = 0;
1301 	elf->e_shstrndx = 0;
1302 	return;
1303 }
1304 
1305 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1306 {
1307 	phdr->p_type = PT_NOTE;
1308 	phdr->p_offset = offset;
1309 	phdr->p_vaddr = 0;
1310 	phdr->p_paddr = 0;
1311 	phdr->p_filesz = sz;
1312 	phdr->p_memsz = 0;
1313 	phdr->p_flags = 0;
1314 	phdr->p_align = 0;
1315 	return;
1316 }
1317 
1318 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1319 		unsigned int sz, void *data)
1320 {
1321 	note->name = name;
1322 	note->type = type;
1323 	note->datasz = sz;
1324 	note->data = data;
1325 	return;
1326 }
1327 
1328 /*
1329  * fill up all the fields in prstatus from the given task struct, except
1330  * registers which need to be filled up separately.
1331  */
1332 static void fill_prstatus(struct elf_prstatus *prstatus,
1333 			  struct task_struct *p, long signr)
1334 {
1335 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1336 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1337 	prstatus->pr_sighold = p->blocked.sig[0];
1338 	rcu_read_lock();
1339 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1340 	rcu_read_unlock();
1341 	prstatus->pr_pid = task_pid_vnr(p);
1342 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1343 	prstatus->pr_sid = task_session_vnr(p);
1344 	if (thread_group_leader(p)) {
1345 		struct task_cputime cputime;
1346 
1347 		/*
1348 		 * This is the record for the group leader.  It shows the
1349 		 * group-wide total, not its individual thread total.
1350 		 */
1351 		thread_group_cputime(p, &cputime);
1352 		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1353 		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1354 	} else {
1355 		cputime_t utime, stime;
1356 
1357 		task_cputime(p, &utime, &stime);
1358 		cputime_to_timeval(utime, &prstatus->pr_utime);
1359 		cputime_to_timeval(stime, &prstatus->pr_stime);
1360 	}
1361 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1362 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1363 
1364 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1365 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1366 }
1367 
1368 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1369 		       struct mm_struct *mm)
1370 {
1371 	const struct cred *cred;
1372 	unsigned int i, len;
1373 
1374 	/* first copy the parameters from user space */
1375 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1376 
1377 	len = mm->arg_end - mm->arg_start;
1378 	if (len >= ELF_PRARGSZ)
1379 		len = ELF_PRARGSZ - 1;
1380 	if (copy_from_user(&psinfo->pr_psargs,
1381 		           (const char __user *) mm->arg_start, len))
1382 		return -EFAULT;
1383 	for (i = 0; i < len; i++)
1384 		if (psinfo->pr_psargs[i] == 0)
1385 			psinfo->pr_psargs[i] = ' ';
1386 	psinfo->pr_psargs[len] = 0;
1387 
1388 	rcu_read_lock();
1389 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1390 	rcu_read_unlock();
1391 	psinfo->pr_pid = task_pid_vnr(p);
1392 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1393 	psinfo->pr_sid = task_session_vnr(p);
1394 
1395 	i = p->state ? ffz(~p->state) + 1 : 0;
1396 	psinfo->pr_state = i;
1397 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1398 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1399 	psinfo->pr_nice = task_nice(p);
1400 	psinfo->pr_flag = p->flags;
1401 	rcu_read_lock();
1402 	cred = __task_cred(p);
1403 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1404 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1405 	rcu_read_unlock();
1406 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1407 
1408 	return 0;
1409 }
1410 
1411 /* Here is the structure in which status of each thread is captured. */
1412 struct elf_thread_status
1413 {
1414 	struct list_head list;
1415 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1416 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1417 	struct task_struct *thread;
1418 #ifdef ELF_CORE_COPY_XFPREGS
1419 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1420 #endif
1421 	struct memelfnote notes[3];
1422 	int num_notes;
1423 };
1424 
1425 /*
1426  * In order to add the specific thread information for the elf file format,
1427  * we need to keep a linked list of every thread's pr_status and then create
1428  * a single section for them in the final core file.
1429  */
1430 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1431 {
1432 	struct task_struct *p = t->thread;
1433 	int sz = 0;
1434 
1435 	t->num_notes = 0;
1436 
1437 	fill_prstatus(&t->prstatus, p, signr);
1438 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1439 
1440 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1441 		  &t->prstatus);
1442 	t->num_notes++;
1443 	sz += notesize(&t->notes[0]);
1444 
1445 	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1446 	if (t->prstatus.pr_fpvalid) {
1447 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1448 			  &t->fpu);
1449 		t->num_notes++;
1450 		sz += notesize(&t->notes[1]);
1451 	}
1452 
1453 #ifdef ELF_CORE_COPY_XFPREGS
1454 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1455 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1456 			  sizeof(t->xfpu), &t->xfpu);
1457 		t->num_notes++;
1458 		sz += notesize(&t->notes[2]);
1459 	}
1460 #endif
1461 	return sz;
1462 }
1463 
1464 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1465 			     elf_addr_t e_shoff, int segs)
1466 {
1467 	elf->e_shoff = e_shoff;
1468 	elf->e_shentsize = sizeof(*shdr4extnum);
1469 	elf->e_shnum = 1;
1470 	elf->e_shstrndx = SHN_UNDEF;
1471 
1472 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1473 
1474 	shdr4extnum->sh_type = SHT_NULL;
1475 	shdr4extnum->sh_size = elf->e_shnum;
1476 	shdr4extnum->sh_link = elf->e_shstrndx;
1477 	shdr4extnum->sh_info = segs;
1478 }
1479 
1480 /*
1481  * dump the segments for an MMU process
1482  */
1483 static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1484 {
1485 	struct vm_area_struct *vma;
1486 
1487 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1488 		unsigned long addr;
1489 
1490 		if (!maydump(vma, cprm->mm_flags))
1491 			continue;
1492 
1493 #ifdef CONFIG_MMU
1494 		for (addr = vma->vm_start; addr < vma->vm_end;
1495 							addr += PAGE_SIZE) {
1496 			bool res;
1497 			struct page *page = get_dump_page(addr);
1498 			if (page) {
1499 				void *kaddr = kmap(page);
1500 				res = dump_emit(cprm, kaddr, PAGE_SIZE);
1501 				kunmap(page);
1502 				put_page(page);
1503 			} else {
1504 				res = dump_skip(cprm, PAGE_SIZE);
1505 			}
1506 			if (!res)
1507 				return false;
1508 		}
1509 #else
1510 		if (!dump_emit(cprm, (void *) vma->vm_start,
1511 				vma->vm_end - vma->vm_start))
1512 			return false;
1513 #endif
1514 	}
1515 	return true;
1516 }
1517 
1518 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1519 {
1520 	struct vm_area_struct *vma;
1521 	size_t size = 0;
1522 
1523 	for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1524 		if (maydump(vma, mm_flags))
1525 			size += vma->vm_end - vma->vm_start;
1526 	return size;
1527 }
1528 
1529 /*
1530  * Actual dumper
1531  *
1532  * This is a two-pass process; first we find the offsets of the bits,
1533  * and then they are actually written out.  If we run out of core limit
1534  * we just truncate.
1535  */
1536 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1537 {
1538 #define	NUM_NOTES	6
1539 	int has_dumped = 0;
1540 	mm_segment_t fs;
1541 	int segs;
1542 	int i;
1543 	struct vm_area_struct *vma;
1544 	struct elfhdr *elf = NULL;
1545 	loff_t offset = 0, dataoff;
1546 	int numnote;
1547 	struct memelfnote *notes = NULL;
1548 	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1549 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1550  	LIST_HEAD(thread_list);
1551  	struct list_head *t;
1552 	elf_fpregset_t *fpu = NULL;
1553 #ifdef ELF_CORE_COPY_XFPREGS
1554 	elf_fpxregset_t *xfpu = NULL;
1555 #endif
1556 	int thread_status_size = 0;
1557 	elf_addr_t *auxv;
1558 	struct elf_phdr *phdr4note = NULL;
1559 	struct elf_shdr *shdr4extnum = NULL;
1560 	Elf_Half e_phnum;
1561 	elf_addr_t e_shoff;
1562 	struct core_thread *ct;
1563 	struct elf_thread_status *tmp;
1564 
1565 	/*
1566 	 * We no longer stop all VM operations.
1567 	 *
1568 	 * This is because those proceses that could possibly change map_count
1569 	 * or the mmap / vma pages are now blocked in do_exit on current
1570 	 * finishing this core dump.
1571 	 *
1572 	 * Only ptrace can touch these memory addresses, but it doesn't change
1573 	 * the map_count or the pages allocated. So no possibility of crashing
1574 	 * exists while dumping the mm->vm_next areas to the core file.
1575 	 */
1576 
1577 	/* alloc memory for large data structures: too large to be on stack */
1578 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1579 	if (!elf)
1580 		goto cleanup;
1581 	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1582 	if (!prstatus)
1583 		goto cleanup;
1584 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1585 	if (!psinfo)
1586 		goto cleanup;
1587 	notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1588 	if (!notes)
1589 		goto cleanup;
1590 	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1591 	if (!fpu)
1592 		goto cleanup;
1593 #ifdef ELF_CORE_COPY_XFPREGS
1594 	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1595 	if (!xfpu)
1596 		goto cleanup;
1597 #endif
1598 
1599 	for (ct = current->mm->core_state->dumper.next;
1600 					ct; ct = ct->next) {
1601 		tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1602 		if (!tmp)
1603 			goto cleanup;
1604 
1605 		tmp->thread = ct->task;
1606 		list_add(&tmp->list, &thread_list);
1607 	}
1608 
1609 	list_for_each(t, &thread_list) {
1610 		struct elf_thread_status *tmp;
1611 		int sz;
1612 
1613 		tmp = list_entry(t, struct elf_thread_status, list);
1614 		sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1615 		thread_status_size += sz;
1616 	}
1617 
1618 	/* now collect the dump for the current */
1619 	fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1620 	elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1621 
1622 	segs = current->mm->map_count;
1623 	segs += elf_core_extra_phdrs();
1624 
1625 	/* for notes section */
1626 	segs++;
1627 
1628 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1629 	 * this, kernel supports extended numbering. Have a look at
1630 	 * include/linux/elf.h for further information. */
1631 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1632 
1633 	/* Set up header */
1634 	fill_elf_fdpic_header(elf, e_phnum);
1635 
1636 	has_dumped = 1;
1637 	/*
1638 	 * Set up the notes in similar form to SVR4 core dumps made
1639 	 * with info from their /proc.
1640 	 */
1641 
1642 	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1643 	fill_psinfo(psinfo, current->group_leader, current->mm);
1644 	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1645 
1646 	numnote = 2;
1647 
1648 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1649 
1650 	i = 0;
1651 	do
1652 		i += 2;
1653 	while (auxv[i - 2] != AT_NULL);
1654 	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1655 		  i * sizeof(elf_addr_t), auxv);
1656 
1657   	/* Try to dump the FPU. */
1658 	if ((prstatus->pr_fpvalid =
1659 	     elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1660 		fill_note(notes + numnote++,
1661 			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1662 #ifdef ELF_CORE_COPY_XFPREGS
1663 	if (elf_core_copy_task_xfpregs(current, xfpu))
1664 		fill_note(notes + numnote++,
1665 			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1666 #endif
1667 
1668 	fs = get_fs();
1669 	set_fs(KERNEL_DS);
1670 
1671 	offset += sizeof(*elf);				/* Elf header */
1672 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1673 
1674 	/* Write notes phdr entry */
1675 	{
1676 		int sz = 0;
1677 
1678 		for (i = 0; i < numnote; i++)
1679 			sz += notesize(notes + i);
1680 
1681 		sz += thread_status_size;
1682 
1683 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1684 		if (!phdr4note)
1685 			goto end_coredump;
1686 
1687 		fill_elf_note_phdr(phdr4note, sz, offset);
1688 		offset += sz;
1689 	}
1690 
1691 	/* Page-align dumped data */
1692 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1693 
1694 	offset += elf_core_vma_data_size(cprm->mm_flags);
1695 	offset += elf_core_extra_data_size();
1696 	e_shoff = offset;
1697 
1698 	if (e_phnum == PN_XNUM) {
1699 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1700 		if (!shdr4extnum)
1701 			goto end_coredump;
1702 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1703 	}
1704 
1705 	offset = dataoff;
1706 
1707 	if (!dump_emit(cprm, elf, sizeof(*elf)))
1708 		goto end_coredump;
1709 
1710 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1711 		goto end_coredump;
1712 
1713 	/* write program headers for segments dump */
1714 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1715 		struct elf_phdr phdr;
1716 		size_t sz;
1717 
1718 		sz = vma->vm_end - vma->vm_start;
1719 
1720 		phdr.p_type = PT_LOAD;
1721 		phdr.p_offset = offset;
1722 		phdr.p_vaddr = vma->vm_start;
1723 		phdr.p_paddr = 0;
1724 		phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1725 		phdr.p_memsz = sz;
1726 		offset += phdr.p_filesz;
1727 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1728 		if (vma->vm_flags & VM_WRITE)
1729 			phdr.p_flags |= PF_W;
1730 		if (vma->vm_flags & VM_EXEC)
1731 			phdr.p_flags |= PF_X;
1732 		phdr.p_align = ELF_EXEC_PAGESIZE;
1733 
1734 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1735 			goto end_coredump;
1736 	}
1737 
1738 	if (!elf_core_write_extra_phdrs(cprm, offset))
1739 		goto end_coredump;
1740 
1741  	/* write out the notes section */
1742 	for (i = 0; i < numnote; i++)
1743 		if (!writenote(notes + i, cprm))
1744 			goto end_coredump;
1745 
1746 	/* write out the thread status notes section */
1747 	list_for_each(t, &thread_list) {
1748 		struct elf_thread_status *tmp =
1749 				list_entry(t, struct elf_thread_status, list);
1750 
1751 		for (i = 0; i < tmp->num_notes; i++)
1752 			if (!writenote(&tmp->notes[i], cprm))
1753 				goto end_coredump;
1754 	}
1755 
1756 	if (!dump_skip(cprm, dataoff - cprm->pos))
1757 		goto end_coredump;
1758 
1759 	if (!elf_fdpic_dump_segments(cprm))
1760 		goto end_coredump;
1761 
1762 	if (!elf_core_write_extra_data(cprm))
1763 		goto end_coredump;
1764 
1765 	if (e_phnum == PN_XNUM) {
1766 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1767 			goto end_coredump;
1768 	}
1769 
1770 	if (cprm->file->f_pos != offset) {
1771 		/* Sanity check */
1772 		printk(KERN_WARNING
1773 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1774 		       cprm->file->f_pos, offset);
1775 	}
1776 
1777 end_coredump:
1778 	set_fs(fs);
1779 
1780 cleanup:
1781 	while (!list_empty(&thread_list)) {
1782 		struct list_head *tmp = thread_list.next;
1783 		list_del(tmp);
1784 		kfree(list_entry(tmp, struct elf_thread_status, list));
1785 	}
1786 	kfree(phdr4note);
1787 	kfree(elf);
1788 	kfree(prstatus);
1789 	kfree(psinfo);
1790 	kfree(notes);
1791 	kfree(fpu);
1792 	kfree(shdr4extnum);
1793 #ifdef ELF_CORE_COPY_XFPREGS
1794 	kfree(xfpu);
1795 #endif
1796 	return has_dumped;
1797 #undef NUM_NOTES
1798 }
1799 
1800 #endif		/* CONFIG_ELF_CORE */
1801