xref: /openbmc/linux/fs/binfmt_elf_fdpic.c (revision f9834f18)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
3  *
4  * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  * Derived from binfmt_elf.c
7  */
8 
9 #include <linux/module.h>
10 
11 #include <linux/fs.h>
12 #include <linux/stat.h>
13 #include <linux/sched.h>
14 #include <linux/sched/coredump.h>
15 #include <linux/sched/task_stack.h>
16 #include <linux/sched/cputime.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/errno.h>
20 #include <linux/signal.h>
21 #include <linux/binfmts.h>
22 #include <linux/string.h>
23 #include <linux/file.h>
24 #include <linux/fcntl.h>
25 #include <linux/slab.h>
26 #include <linux/pagemap.h>
27 #include <linux/security.h>
28 #include <linux/highmem.h>
29 #include <linux/highuid.h>
30 #include <linux/personality.h>
31 #include <linux/ptrace.h>
32 #include <linux/init.h>
33 #include <linux/elf.h>
34 #include <linux/elf-fdpic.h>
35 #include <linux/elfcore.h>
36 #include <linux/coredump.h>
37 #include <linux/dax.h>
38 
39 #include <linux/uaccess.h>
40 #include <asm/param.h>
41 #include <asm/pgalloc.h>
42 
43 typedef char *elf_caddr_t;
44 
45 #if 0
46 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
47 #else
48 #define kdebug(fmt, ...) do {} while(0)
49 #endif
50 
51 #if 0
52 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
53 #else
54 #define kdcore(fmt, ...) do {} while(0)
55 #endif
56 
57 MODULE_LICENSE("GPL");
58 
59 static int load_elf_fdpic_binary(struct linux_binprm *);
60 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
61 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
62 			      struct mm_struct *, const char *);
63 
64 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
65 				   struct elf_fdpic_params *,
66 				   struct elf_fdpic_params *);
67 
68 #ifndef CONFIG_MMU
69 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
70 						   struct file *,
71 						   struct mm_struct *);
72 #endif
73 
74 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
75 					     struct file *, struct mm_struct *);
76 
77 #ifdef CONFIG_ELF_CORE
78 static int elf_fdpic_core_dump(struct coredump_params *cprm);
79 #endif
80 
81 static struct linux_binfmt elf_fdpic_format = {
82 	.module		= THIS_MODULE,
83 	.load_binary	= load_elf_fdpic_binary,
84 #ifdef CONFIG_ELF_CORE
85 	.core_dump	= elf_fdpic_core_dump,
86 #endif
87 	.min_coredump	= ELF_EXEC_PAGESIZE,
88 };
89 
90 static int __init init_elf_fdpic_binfmt(void)
91 {
92 	register_binfmt(&elf_fdpic_format);
93 	return 0;
94 }
95 
96 static void __exit exit_elf_fdpic_binfmt(void)
97 {
98 	unregister_binfmt(&elf_fdpic_format);
99 }
100 
101 core_initcall(init_elf_fdpic_binfmt);
102 module_exit(exit_elf_fdpic_binfmt);
103 
104 static int is_elf(struct elfhdr *hdr, struct file *file)
105 {
106 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
107 		return 0;
108 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
109 		return 0;
110 	if (!elf_check_arch(hdr))
111 		return 0;
112 	if (!file->f_op->mmap)
113 		return 0;
114 	return 1;
115 }
116 
117 #ifndef elf_check_fdpic
118 #define elf_check_fdpic(x) 0
119 #endif
120 
121 #ifndef elf_check_const_displacement
122 #define elf_check_const_displacement(x) 0
123 #endif
124 
125 static int is_constdisp(struct elfhdr *hdr)
126 {
127 	if (!elf_check_fdpic(hdr))
128 		return 1;
129 	if (elf_check_const_displacement(hdr))
130 		return 1;
131 	return 0;
132 }
133 
134 /*****************************************************************************/
135 /*
136  * read the program headers table into memory
137  */
138 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
139 				 struct file *file)
140 {
141 	struct elf32_phdr *phdr;
142 	unsigned long size;
143 	int retval, loop;
144 	loff_t pos = params->hdr.e_phoff;
145 
146 	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
147 		return -ENOMEM;
148 	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
149 		return -ENOMEM;
150 
151 	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
152 	params->phdrs = kmalloc(size, GFP_KERNEL);
153 	if (!params->phdrs)
154 		return -ENOMEM;
155 
156 	retval = kernel_read(file, params->phdrs, size, &pos);
157 	if (unlikely(retval != size))
158 		return retval < 0 ? retval : -ENOEXEC;
159 
160 	/* determine stack size for this binary */
161 	phdr = params->phdrs;
162 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
163 		if (phdr->p_type != PT_GNU_STACK)
164 			continue;
165 
166 		if (phdr->p_flags & PF_X)
167 			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
168 		else
169 			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
170 
171 		params->stack_size = phdr->p_memsz;
172 		break;
173 	}
174 
175 	return 0;
176 }
177 
178 /*****************************************************************************/
179 /*
180  * load an fdpic binary into various bits of memory
181  */
182 static int load_elf_fdpic_binary(struct linux_binprm *bprm)
183 {
184 	struct elf_fdpic_params exec_params, interp_params;
185 	struct pt_regs *regs = current_pt_regs();
186 	struct elf_phdr *phdr;
187 	unsigned long stack_size, entryaddr;
188 #ifdef ELF_FDPIC_PLAT_INIT
189 	unsigned long dynaddr;
190 #endif
191 #ifndef CONFIG_MMU
192 	unsigned long stack_prot;
193 #endif
194 	struct file *interpreter = NULL; /* to shut gcc up */
195 	char *interpreter_name = NULL;
196 	int executable_stack;
197 	int retval, i;
198 	loff_t pos;
199 
200 	kdebug("____ LOAD %d ____", current->pid);
201 
202 	memset(&exec_params, 0, sizeof(exec_params));
203 	memset(&interp_params, 0, sizeof(interp_params));
204 
205 	exec_params.hdr = *(struct elfhdr *) bprm->buf;
206 	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
207 
208 	/* check that this is a binary we know how to deal with */
209 	retval = -ENOEXEC;
210 	if (!is_elf(&exec_params.hdr, bprm->file))
211 		goto error;
212 	if (!elf_check_fdpic(&exec_params.hdr)) {
213 #ifdef CONFIG_MMU
214 		/* binfmt_elf handles non-fdpic elf except on nommu */
215 		goto error;
216 #else
217 		/* nommu can only load ET_DYN (PIE) ELF */
218 		if (exec_params.hdr.e_type != ET_DYN)
219 			goto error;
220 #endif
221 	}
222 
223 	/* read the program header table */
224 	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
225 	if (retval < 0)
226 		goto error;
227 
228 	/* scan for a program header that specifies an interpreter */
229 	phdr = exec_params.phdrs;
230 
231 	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
232 		switch (phdr->p_type) {
233 		case PT_INTERP:
234 			retval = -ENOMEM;
235 			if (phdr->p_filesz > PATH_MAX)
236 				goto error;
237 			retval = -ENOENT;
238 			if (phdr->p_filesz < 2)
239 				goto error;
240 
241 			/* read the name of the interpreter into memory */
242 			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
243 			if (!interpreter_name)
244 				goto error;
245 
246 			pos = phdr->p_offset;
247 			retval = kernel_read(bprm->file, interpreter_name,
248 					     phdr->p_filesz, &pos);
249 			if (unlikely(retval != phdr->p_filesz)) {
250 				if (retval >= 0)
251 					retval = -ENOEXEC;
252 				goto error;
253 			}
254 
255 			retval = -ENOENT;
256 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
257 				goto error;
258 
259 			kdebug("Using ELF interpreter %s", interpreter_name);
260 
261 			/* replace the program with the interpreter */
262 			interpreter = open_exec(interpreter_name);
263 			retval = PTR_ERR(interpreter);
264 			if (IS_ERR(interpreter)) {
265 				interpreter = NULL;
266 				goto error;
267 			}
268 
269 			/*
270 			 * If the binary is not readable then enforce
271 			 * mm->dumpable = 0 regardless of the interpreter's
272 			 * permissions.
273 			 */
274 			would_dump(bprm, interpreter);
275 
276 			pos = 0;
277 			retval = kernel_read(interpreter, bprm->buf,
278 					BINPRM_BUF_SIZE, &pos);
279 			if (unlikely(retval != BINPRM_BUF_SIZE)) {
280 				if (retval >= 0)
281 					retval = -ENOEXEC;
282 				goto error;
283 			}
284 
285 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
286 			break;
287 
288 		case PT_LOAD:
289 #ifdef CONFIG_MMU
290 			if (exec_params.load_addr == 0)
291 				exec_params.load_addr = phdr->p_vaddr;
292 #endif
293 			break;
294 		}
295 
296 	}
297 
298 	if (is_constdisp(&exec_params.hdr))
299 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
300 
301 	/* perform insanity checks on the interpreter */
302 	if (interpreter_name) {
303 		retval = -ELIBBAD;
304 		if (!is_elf(&interp_params.hdr, interpreter))
305 			goto error;
306 
307 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
308 
309 		/* read the interpreter's program header table */
310 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
311 		if (retval < 0)
312 			goto error;
313 	}
314 
315 	stack_size = exec_params.stack_size;
316 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
317 		executable_stack = EXSTACK_ENABLE_X;
318 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
319 		executable_stack = EXSTACK_DISABLE_X;
320 	else
321 		executable_stack = EXSTACK_DEFAULT;
322 
323 	if (stack_size == 0) {
324 		stack_size = interp_params.stack_size;
325 		if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
326 			executable_stack = EXSTACK_ENABLE_X;
327 		else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
328 			executable_stack = EXSTACK_DISABLE_X;
329 		else
330 			executable_stack = EXSTACK_DEFAULT;
331 	}
332 
333 	retval = -ENOEXEC;
334 	if (stack_size == 0)
335 		stack_size = 131072UL; /* same as exec.c's default commit */
336 
337 	if (is_constdisp(&interp_params.hdr))
338 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
339 
340 	/* flush all traces of the currently running executable */
341 	retval = 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 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
378 	retval = arch_setup_additional_pages(bprm, !!interpreter_name);
379 	if (retval < 0)
380 		goto error;
381 #endif
382 #endif
383 
384 	/* load the executable and interpreter into memory */
385 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
386 				    "executable");
387 	if (retval < 0)
388 		goto error;
389 
390 	if (interpreter_name) {
391 		retval = elf_fdpic_map_file(&interp_params, interpreter,
392 					    current->mm, "interpreter");
393 		if (retval < 0) {
394 			printk(KERN_ERR "Unable to load interpreter\n");
395 			goto error;
396 		}
397 
398 		allow_write_access(interpreter);
399 		fput(interpreter);
400 		interpreter = NULL;
401 	}
402 
403 #ifdef CONFIG_MMU
404 	if (!current->mm->start_brk)
405 		current->mm->start_brk = current->mm->end_data;
406 
407 	current->mm->brk = current->mm->start_brk =
408 		PAGE_ALIGN(current->mm->start_brk);
409 
410 #else
411 	/* create a stack area and zero-size brk area */
412 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
413 	if (stack_size < PAGE_SIZE * 2)
414 		stack_size = PAGE_SIZE * 2;
415 
416 	stack_prot = PROT_READ | PROT_WRITE;
417 	if (executable_stack == EXSTACK_ENABLE_X ||
418 	    (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
419 		stack_prot |= PROT_EXEC;
420 
421 	current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
422 					 MAP_PRIVATE | MAP_ANONYMOUS |
423 					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
424 					 0);
425 
426 	if (IS_ERR_VALUE(current->mm->start_brk)) {
427 		retval = current->mm->start_brk;
428 		current->mm->start_brk = 0;
429 		goto error;
430 	}
431 
432 	current->mm->brk = current->mm->start_brk;
433 	current->mm->context.end_brk = current->mm->start_brk;
434 	current->mm->start_stack = current->mm->start_brk + stack_size;
435 #endif
436 
437 	install_exec_creds(bprm);
438 	if (create_elf_fdpic_tables(bprm, current->mm,
439 				    &exec_params, &interp_params) < 0)
440 		goto error;
441 
442 	kdebug("- start_code  %lx", current->mm->start_code);
443 	kdebug("- end_code    %lx", current->mm->end_code);
444 	kdebug("- start_data  %lx", current->mm->start_data);
445 	kdebug("- end_data    %lx", current->mm->end_data);
446 	kdebug("- start_brk   %lx", current->mm->start_brk);
447 	kdebug("- brk         %lx", current->mm->brk);
448 	kdebug("- start_stack %lx", current->mm->start_stack);
449 
450 #ifdef ELF_FDPIC_PLAT_INIT
451 	/*
452 	 * The ABI may specify that certain registers be set up in special
453 	 * ways (on i386 %edx is the address of a DT_FINI function, for
454 	 * example.  This macro performs whatever initialization to
455 	 * the regs structure is required.
456 	 */
457 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
458 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
459 			    dynaddr);
460 #endif
461 
462 	finalize_exec(bprm);
463 	/* everything is now ready... get the userspace context ready to roll */
464 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
465 	start_thread(regs, entryaddr, current->mm->start_stack);
466 
467 	retval = 0;
468 
469 error:
470 	if (interpreter) {
471 		allow_write_access(interpreter);
472 		fput(interpreter);
473 	}
474 	kfree(interpreter_name);
475 	kfree(exec_params.phdrs);
476 	kfree(exec_params.loadmap);
477 	kfree(interp_params.phdrs);
478 	kfree(interp_params.loadmap);
479 	return retval;
480 }
481 
482 /*****************************************************************************/
483 
484 #ifndef ELF_BASE_PLATFORM
485 /*
486  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
487  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
488  * will be copied to the user stack in the same manner as AT_PLATFORM.
489  */
490 #define ELF_BASE_PLATFORM NULL
491 #endif
492 
493 /*
494  * present useful information to the program by shovelling it onto the new
495  * process's stack
496  */
497 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
498 				   struct mm_struct *mm,
499 				   struct elf_fdpic_params *exec_params,
500 				   struct elf_fdpic_params *interp_params)
501 {
502 	const struct cred *cred = current_cred();
503 	unsigned long sp, csp, nitems;
504 	elf_caddr_t __user *argv, *envp;
505 	size_t platform_len = 0, len;
506 	char *k_platform, *k_base_platform;
507 	char __user *u_platform, *u_base_platform, *p;
508 	int loop;
509 	int nr;	/* reset for each csp adjustment */
510 
511 #ifdef CONFIG_MMU
512 	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
513 	 * by the processes running on the same package. One thing we can do is
514 	 * to shuffle the initial stack for them, so we give the architecture
515 	 * an opportunity to do so here.
516 	 */
517 	sp = arch_align_stack(bprm->p);
518 #else
519 	sp = mm->start_stack;
520 
521 	/* stack the program arguments and environment */
522 	if (transfer_args_to_stack(bprm, &sp) < 0)
523 		return -EFAULT;
524 	sp &= ~15;
525 #endif
526 
527 	/*
528 	 * If this architecture has a platform capability string, copy it
529 	 * to userspace.  In some cases (Sparc), this info is impossible
530 	 * for userspace to get any other way, in others (i386) it is
531 	 * merely difficult.
532 	 */
533 	k_platform = ELF_PLATFORM;
534 	u_platform = NULL;
535 
536 	if (k_platform) {
537 		platform_len = strlen(k_platform) + 1;
538 		sp -= platform_len;
539 		u_platform = (char __user *) sp;
540 		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
541 			return -EFAULT;
542 	}
543 
544 	/*
545 	 * If this architecture has a "base" platform capability
546 	 * string, copy it to userspace.
547 	 */
548 	k_base_platform = ELF_BASE_PLATFORM;
549 	u_base_platform = NULL;
550 
551 	if (k_base_platform) {
552 		platform_len = strlen(k_base_platform) + 1;
553 		sp -= platform_len;
554 		u_base_platform = (char __user *) sp;
555 		if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
556 			return -EFAULT;
557 	}
558 
559 	sp &= ~7UL;
560 
561 	/* stack the load map(s) */
562 	len = sizeof(struct elf32_fdpic_loadmap);
563 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
564 	sp = (sp - len) & ~7UL;
565 	exec_params->map_addr = sp;
566 
567 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
568 		return -EFAULT;
569 
570 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
571 
572 	if (interp_params->loadmap) {
573 		len = sizeof(struct elf32_fdpic_loadmap);
574 		len += sizeof(struct elf32_fdpic_loadseg) *
575 			interp_params->loadmap->nsegs;
576 		sp = (sp - len) & ~7UL;
577 		interp_params->map_addr = sp;
578 
579 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
580 				 len) != 0)
581 			return -EFAULT;
582 
583 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
584 	}
585 
586 	/* force 16 byte _final_ alignment here for generality */
587 #define DLINFO_ITEMS 15
588 
589 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
590 		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
591 
592 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
593 		nitems++;
594 
595 	csp = sp;
596 	sp -= nitems * 2 * sizeof(unsigned long);
597 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
598 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
599 	sp -= 1 * sizeof(unsigned long);		/* argc */
600 
601 	csp -= sp & 15UL;
602 	sp -= sp & 15UL;
603 
604 	/* put the ELF interpreter info on the stack */
605 #define NEW_AUX_ENT(id, val)						\
606 	do {								\
607 		struct { unsigned long _id, _val; } __user *ent;	\
608 									\
609 		ent = (void __user *) csp;				\
610 		__put_user((id), &ent[nr]._id);				\
611 		__put_user((val), &ent[nr]._val);			\
612 		nr++;							\
613 	} while (0)
614 
615 	nr = 0;
616 	csp -= 2 * sizeof(unsigned long);
617 	NEW_AUX_ENT(AT_NULL, 0);
618 	if (k_platform) {
619 		nr = 0;
620 		csp -= 2 * sizeof(unsigned long);
621 		NEW_AUX_ENT(AT_PLATFORM,
622 			    (elf_addr_t) (unsigned long) u_platform);
623 	}
624 
625 	if (k_base_platform) {
626 		nr = 0;
627 		csp -= 2 * sizeof(unsigned long);
628 		NEW_AUX_ENT(AT_BASE_PLATFORM,
629 			    (elf_addr_t) (unsigned long) u_base_platform);
630 	}
631 
632 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
633 		nr = 0;
634 		csp -= 2 * sizeof(unsigned long);
635 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
636 	}
637 
638 	nr = 0;
639 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
640 	NEW_AUX_ENT(AT_HWCAP,	ELF_HWCAP);
641 #ifdef ELF_HWCAP2
642 	NEW_AUX_ENT(AT_HWCAP2,	ELF_HWCAP2);
643 #endif
644 	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
645 	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
646 	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
647 	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
648 	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
649 	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
650 	NEW_AUX_ENT(AT_FLAGS,	0);
651 	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
652 	NEW_AUX_ENT(AT_UID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
653 	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
654 	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
655 	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
656 	NEW_AUX_ENT(AT_SECURE,	bprm->secureexec);
657 	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
658 
659 #ifdef ARCH_DLINFO
660 	nr = 0;
661 	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
662 
663 	/* ARCH_DLINFO must come last so platform specific code can enforce
664 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
665 	 */
666 	ARCH_DLINFO;
667 #endif
668 #undef NEW_AUX_ENT
669 
670 	/* allocate room for argv[] and envv[] */
671 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
672 	envp = (elf_caddr_t __user *) csp;
673 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
674 	argv = (elf_caddr_t __user *) csp;
675 
676 	/* stack argc */
677 	csp -= sizeof(unsigned long);
678 	__put_user(bprm->argc, (unsigned long __user *) csp);
679 
680 	BUG_ON(csp != sp);
681 
682 	/* fill in the argv[] array */
683 #ifdef CONFIG_MMU
684 	current->mm->arg_start = bprm->p;
685 #else
686 	current->mm->arg_start = current->mm->start_stack -
687 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
688 #endif
689 
690 	p = (char __user *) current->mm->arg_start;
691 	for (loop = bprm->argc; loop > 0; loop--) {
692 		__put_user((elf_caddr_t) p, argv++);
693 		len = strnlen_user(p, MAX_ARG_STRLEN);
694 		if (!len || len > MAX_ARG_STRLEN)
695 			return -EINVAL;
696 		p += len;
697 	}
698 	__put_user(NULL, argv);
699 	current->mm->arg_end = (unsigned long) p;
700 
701 	/* fill in the envv[] array */
702 	current->mm->env_start = (unsigned long) p;
703 	for (loop = bprm->envc; loop > 0; loop--) {
704 		__put_user((elf_caddr_t)(unsigned long) p, envp++);
705 		len = strnlen_user(p, MAX_ARG_STRLEN);
706 		if (!len || len > MAX_ARG_STRLEN)
707 			return -EINVAL;
708 		p += len;
709 	}
710 	__put_user(NULL, envp);
711 	current->mm->env_end = (unsigned long) p;
712 
713 	mm->start_stack = (unsigned long) sp;
714 	return 0;
715 }
716 
717 /*****************************************************************************/
718 /*
719  * load the appropriate binary image (executable or interpreter) into memory
720  * - we assume no MMU is available
721  * - if no other PIC bits are set in params->hdr->e_flags
722  *   - we assume that the LOADable segments in the binary are independently relocatable
723  *   - we assume R/O executable segments are shareable
724  * - else
725  *   - we assume the loadable parts of the image to require fixed displacement
726  *   - the image is not shareable
727  */
728 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
729 			      struct file *file,
730 			      struct mm_struct *mm,
731 			      const char *what)
732 {
733 	struct elf32_fdpic_loadmap *loadmap;
734 #ifdef CONFIG_MMU
735 	struct elf32_fdpic_loadseg *mseg;
736 #endif
737 	struct elf32_fdpic_loadseg *seg;
738 	struct elf32_phdr *phdr;
739 	unsigned long load_addr, stop;
740 	unsigned nloads, tmp;
741 	size_t size;
742 	int loop, ret;
743 
744 	/* allocate a load map table */
745 	nloads = 0;
746 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
747 		if (params->phdrs[loop].p_type == PT_LOAD)
748 			nloads++;
749 
750 	if (nloads == 0)
751 		return -ELIBBAD;
752 
753 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
754 	loadmap = kzalloc(size, GFP_KERNEL);
755 	if (!loadmap)
756 		return -ENOMEM;
757 
758 	params->loadmap = loadmap;
759 
760 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
761 	loadmap->nsegs = nloads;
762 
763 	load_addr = params->load_addr;
764 	seg = loadmap->segs;
765 
766 	/* map the requested LOADs into the memory space */
767 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
768 	case ELF_FDPIC_FLAG_CONSTDISP:
769 	case ELF_FDPIC_FLAG_CONTIGUOUS:
770 #ifndef CONFIG_MMU
771 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
772 		if (ret < 0)
773 			return ret;
774 		break;
775 #endif
776 	default:
777 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
778 		if (ret < 0)
779 			return ret;
780 		break;
781 	}
782 
783 	/* map the entry point */
784 	if (params->hdr.e_entry) {
785 		seg = loadmap->segs;
786 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
787 			if (params->hdr.e_entry >= seg->p_vaddr &&
788 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
789 				params->entry_addr =
790 					(params->hdr.e_entry - seg->p_vaddr) +
791 					seg->addr;
792 				break;
793 			}
794 		}
795 	}
796 
797 	/* determine where the program header table has wound up if mapped */
798 	stop = params->hdr.e_phoff;
799 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
800 	phdr = params->phdrs;
801 
802 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
803 		if (phdr->p_type != PT_LOAD)
804 			continue;
805 
806 		if (phdr->p_offset > params->hdr.e_phoff ||
807 		    phdr->p_offset + phdr->p_filesz < stop)
808 			continue;
809 
810 		seg = loadmap->segs;
811 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
812 			if (phdr->p_vaddr >= seg->p_vaddr &&
813 			    phdr->p_vaddr + phdr->p_filesz <=
814 			    seg->p_vaddr + seg->p_memsz) {
815 				params->ph_addr =
816 					(phdr->p_vaddr - seg->p_vaddr) +
817 					seg->addr +
818 					params->hdr.e_phoff - phdr->p_offset;
819 				break;
820 			}
821 		}
822 		break;
823 	}
824 
825 	/* determine where the dynamic section has wound up if there is one */
826 	phdr = params->phdrs;
827 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
828 		if (phdr->p_type != PT_DYNAMIC)
829 			continue;
830 
831 		seg = loadmap->segs;
832 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
833 			if (phdr->p_vaddr >= seg->p_vaddr &&
834 			    phdr->p_vaddr + phdr->p_memsz <=
835 			    seg->p_vaddr + seg->p_memsz) {
836 				Elf32_Dyn __user *dyn;
837 				Elf32_Sword d_tag;
838 
839 				params->dynamic_addr =
840 					(phdr->p_vaddr - seg->p_vaddr) +
841 					seg->addr;
842 
843 				/* check the dynamic section contains at least
844 				 * one item, and that the last item is a NULL
845 				 * entry */
846 				if (phdr->p_memsz == 0 ||
847 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
848 					goto dynamic_error;
849 
850 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
851 				dyn = (Elf32_Dyn __user *)params->dynamic_addr;
852 				__get_user(d_tag, &dyn[tmp - 1].d_tag);
853 				if (d_tag != 0)
854 					goto dynamic_error;
855 				break;
856 			}
857 		}
858 		break;
859 	}
860 
861 	/* now elide adjacent segments in the load map on MMU linux
862 	 * - on uClinux the holes between may actually be filled with system
863 	 *   stuff or stuff from other processes
864 	 */
865 #ifdef CONFIG_MMU
866 	nloads = loadmap->nsegs;
867 	mseg = loadmap->segs;
868 	seg = mseg + 1;
869 	for (loop = 1; loop < nloads; loop++) {
870 		/* see if we have a candidate for merging */
871 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
872 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
873 			if (load_addr == (seg->addr & PAGE_MASK)) {
874 				mseg->p_memsz +=
875 					load_addr -
876 					(mseg->addr + mseg->p_memsz);
877 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
878 				mseg->p_memsz += seg->p_memsz;
879 				loadmap->nsegs--;
880 				continue;
881 			}
882 		}
883 
884 		mseg++;
885 		if (mseg != seg)
886 			*mseg = *seg;
887 	}
888 #endif
889 
890 	kdebug("Mapped Object [%s]:", what);
891 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
892 	kdebug("- entry    : %lx", params->entry_addr);
893 	kdebug("- PHDR[]   : %lx", params->ph_addr);
894 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
895 	seg = loadmap->segs;
896 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
897 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
898 		       loop,
899 		       seg->addr, seg->addr + seg->p_memsz - 1,
900 		       seg->p_vaddr, seg->p_memsz);
901 
902 	return 0;
903 
904 dynamic_error:
905 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
906 	       what, file_inode(file)->i_ino);
907 	return -ELIBBAD;
908 }
909 
910 /*****************************************************************************/
911 /*
912  * map a file with constant displacement under uClinux
913  */
914 #ifndef CONFIG_MMU
915 static int elf_fdpic_map_file_constdisp_on_uclinux(
916 	struct elf_fdpic_params *params,
917 	struct file *file,
918 	struct mm_struct *mm)
919 {
920 	struct elf32_fdpic_loadseg *seg;
921 	struct elf32_phdr *phdr;
922 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
923 	int loop, ret;
924 
925 	load_addr = params->load_addr;
926 	seg = params->loadmap->segs;
927 
928 	/* determine the bounds of the contiguous overall allocation we must
929 	 * make */
930 	phdr = params->phdrs;
931 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
932 		if (params->phdrs[loop].p_type != PT_LOAD)
933 			continue;
934 
935 		if (base > phdr->p_vaddr)
936 			base = phdr->p_vaddr;
937 		if (top < phdr->p_vaddr + phdr->p_memsz)
938 			top = phdr->p_vaddr + phdr->p_memsz;
939 	}
940 
941 	/* allocate one big anon block for everything */
942 	mflags = MAP_PRIVATE;
943 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
944 		mflags |= MAP_EXECUTABLE;
945 
946 	maddr = vm_mmap(NULL, load_addr, top - base,
947 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
948 	if (IS_ERR_VALUE(maddr))
949 		return (int) maddr;
950 
951 	if (load_addr != 0)
952 		load_addr += PAGE_ALIGN(top - base);
953 
954 	/* and then load the file segments into it */
955 	phdr = params->phdrs;
956 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
957 		if (params->phdrs[loop].p_type != PT_LOAD)
958 			continue;
959 
960 		seg->addr = maddr + (phdr->p_vaddr - base);
961 		seg->p_vaddr = phdr->p_vaddr;
962 		seg->p_memsz = phdr->p_memsz;
963 
964 		ret = read_code(file, seg->addr, phdr->p_offset,
965 				       phdr->p_filesz);
966 		if (ret < 0)
967 			return ret;
968 
969 		/* map the ELF header address if in this segment */
970 		if (phdr->p_offset == 0)
971 			params->elfhdr_addr = seg->addr;
972 
973 		/* clear any space allocated but not loaded */
974 		if (phdr->p_filesz < phdr->p_memsz) {
975 			if (clear_user((void *) (seg->addr + phdr->p_filesz),
976 				       phdr->p_memsz - phdr->p_filesz))
977 				return -EFAULT;
978 		}
979 
980 		if (mm) {
981 			if (phdr->p_flags & PF_X) {
982 				if (!mm->start_code) {
983 					mm->start_code = seg->addr;
984 					mm->end_code = seg->addr +
985 						phdr->p_memsz;
986 				}
987 			} else if (!mm->start_data) {
988 				mm->start_data = seg->addr;
989 				mm->end_data = seg->addr + phdr->p_memsz;
990 			}
991 		}
992 
993 		seg++;
994 	}
995 
996 	return 0;
997 }
998 #endif
999 
1000 /*****************************************************************************/
1001 /*
1002  * map a binary by direct mmap() of the individual PT_LOAD segments
1003  */
1004 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1005 					     struct file *file,
1006 					     struct mm_struct *mm)
1007 {
1008 	struct elf32_fdpic_loadseg *seg;
1009 	struct elf32_phdr *phdr;
1010 	unsigned long load_addr, delta_vaddr;
1011 	int loop, dvset;
1012 
1013 	load_addr = params->load_addr;
1014 	delta_vaddr = 0;
1015 	dvset = 0;
1016 
1017 	seg = params->loadmap->segs;
1018 
1019 	/* deal with each load segment separately */
1020 	phdr = params->phdrs;
1021 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1022 		unsigned long maddr, disp, excess, excess1;
1023 		int prot = 0, flags;
1024 
1025 		if (phdr->p_type != PT_LOAD)
1026 			continue;
1027 
1028 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1029 		       (unsigned long) phdr->p_vaddr,
1030 		       (unsigned long) phdr->p_offset,
1031 		       (unsigned long) phdr->p_filesz,
1032 		       (unsigned long) phdr->p_memsz);
1033 
1034 		/* determine the mapping parameters */
1035 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1036 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1037 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1038 
1039 		flags = MAP_PRIVATE | MAP_DENYWRITE;
1040 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1041 			flags |= MAP_EXECUTABLE;
1042 
1043 		maddr = 0;
1044 
1045 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1046 		case ELF_FDPIC_FLAG_INDEPENDENT:
1047 			/* PT_LOADs are independently locatable */
1048 			break;
1049 
1050 		case ELF_FDPIC_FLAG_HONOURVADDR:
1051 			/* the specified virtual address must be honoured */
1052 			maddr = phdr->p_vaddr;
1053 			flags |= MAP_FIXED;
1054 			break;
1055 
1056 		case ELF_FDPIC_FLAG_CONSTDISP:
1057 			/* constant displacement
1058 			 * - can be mapped anywhere, but must be mapped as a
1059 			 *   unit
1060 			 */
1061 			if (!dvset) {
1062 				maddr = load_addr;
1063 				delta_vaddr = phdr->p_vaddr;
1064 				dvset = 1;
1065 			} else {
1066 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1067 				flags |= MAP_FIXED;
1068 			}
1069 			break;
1070 
1071 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1072 			/* contiguity handled later */
1073 			break;
1074 
1075 		default:
1076 			BUG();
1077 		}
1078 
1079 		maddr &= PAGE_MASK;
1080 
1081 		/* create the mapping */
1082 		disp = phdr->p_vaddr & ~PAGE_MASK;
1083 		maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1084 				phdr->p_offset - disp);
1085 
1086 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1087 		       loop, phdr->p_memsz + disp, prot, flags,
1088 		       phdr->p_offset - disp, maddr);
1089 
1090 		if (IS_ERR_VALUE(maddr))
1091 			return (int) maddr;
1092 
1093 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1094 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1095 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1096 
1097 		seg->addr = maddr + disp;
1098 		seg->p_vaddr = phdr->p_vaddr;
1099 		seg->p_memsz = phdr->p_memsz;
1100 
1101 		/* map the ELF header address if in this segment */
1102 		if (phdr->p_offset == 0)
1103 			params->elfhdr_addr = seg->addr;
1104 
1105 		/* clear the bit between beginning of mapping and beginning of
1106 		 * PT_LOAD */
1107 		if (prot & PROT_WRITE && disp > 0) {
1108 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1109 			if (clear_user((void __user *) maddr, disp))
1110 				return -EFAULT;
1111 			maddr += disp;
1112 		}
1113 
1114 		/* clear any space allocated but not loaded
1115 		 * - on uClinux we can just clear the lot
1116 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1117 		 *   extant in the file
1118 		 */
1119 		excess = phdr->p_memsz - phdr->p_filesz;
1120 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1121 
1122 #ifdef CONFIG_MMU
1123 		if (excess > excess1) {
1124 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1125 			unsigned long xmaddr;
1126 
1127 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1128 			xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1129 					 prot, flags, 0);
1130 
1131 			kdebug("mmap[%d] <anon>"
1132 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1133 			       loop, xaddr, excess - excess1, prot, flags,
1134 			       xmaddr);
1135 
1136 			if (xmaddr != xaddr)
1137 				return -ENOMEM;
1138 		}
1139 
1140 		if (prot & PROT_WRITE && excess1 > 0) {
1141 			kdebug("clear[%d] ad=%lx sz=%lx",
1142 			       loop, maddr + phdr->p_filesz, excess1);
1143 			if (clear_user((void __user *) maddr + phdr->p_filesz,
1144 				       excess1))
1145 				return -EFAULT;
1146 		}
1147 
1148 #else
1149 		if (excess > 0) {
1150 			kdebug("clear[%d] ad=%lx sz=%lx",
1151 			       loop, maddr + phdr->p_filesz, excess);
1152 			if (clear_user((void *) maddr + phdr->p_filesz, excess))
1153 				return -EFAULT;
1154 		}
1155 #endif
1156 
1157 		if (mm) {
1158 			if (phdr->p_flags & PF_X) {
1159 				if (!mm->start_code) {
1160 					mm->start_code = maddr;
1161 					mm->end_code = maddr + phdr->p_memsz;
1162 				}
1163 			} else if (!mm->start_data) {
1164 				mm->start_data = maddr;
1165 				mm->end_data = maddr + phdr->p_memsz;
1166 			}
1167 		}
1168 
1169 		seg++;
1170 	}
1171 
1172 	return 0;
1173 }
1174 
1175 /*****************************************************************************/
1176 /*
1177  * ELF-FDPIC core dumper
1178  *
1179  * Modelled on fs/exec.c:aout_core_dump()
1180  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1181  *
1182  * Modelled on fs/binfmt_elf.c core dumper
1183  */
1184 #ifdef CONFIG_ELF_CORE
1185 
1186 /*
1187  * Decide whether a segment is worth dumping; default is yes to be
1188  * sure (missing info is worse than too much; etc).
1189  * Personally I'd include everything, and use the coredump limit...
1190  *
1191  * I think we should skip something. But I am not sure how. H.J.
1192  */
1193 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1194 {
1195 	int dump_ok;
1196 
1197 	/* Do not dump I/O mapped devices or special mappings */
1198 	if (vma->vm_flags & VM_IO) {
1199 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1200 		return 0;
1201 	}
1202 
1203 	/* If we may not read the contents, don't allow us to dump
1204 	 * them either. "dump_write()" can't handle it anyway.
1205 	 */
1206 	if (!(vma->vm_flags & VM_READ)) {
1207 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1208 		return 0;
1209 	}
1210 
1211 	/* support for DAX */
1212 	if (vma_is_dax(vma)) {
1213 		if (vma->vm_flags & VM_SHARED) {
1214 			dump_ok = test_bit(MMF_DUMP_DAX_SHARED, &mm_flags);
1215 			kdcore("%08lx: %08lx: %s (DAX shared)", vma->vm_start,
1216 			       vma->vm_flags, dump_ok ? "yes" : "no");
1217 		} else {
1218 			dump_ok = test_bit(MMF_DUMP_DAX_PRIVATE, &mm_flags);
1219 			kdcore("%08lx: %08lx: %s (DAX private)", vma->vm_start,
1220 			       vma->vm_flags, dump_ok ? "yes" : "no");
1221 		}
1222 		return dump_ok;
1223 	}
1224 
1225 	/* By default, dump shared memory if mapped from an anonymous file. */
1226 	if (vma->vm_flags & VM_SHARED) {
1227 		if (file_inode(vma->vm_file)->i_nlink == 0) {
1228 			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1229 			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1230 			       vma->vm_flags, dump_ok ? "yes" : "no");
1231 			return dump_ok;
1232 		}
1233 
1234 		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1235 		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1236 		       vma->vm_flags, dump_ok ? "yes" : "no");
1237 		return dump_ok;
1238 	}
1239 
1240 #ifdef CONFIG_MMU
1241 	/* By default, if it hasn't been written to, don't write it out */
1242 	if (!vma->anon_vma) {
1243 		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1244 		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1245 		       vma->vm_flags, dump_ok ? "yes" : "no");
1246 		return dump_ok;
1247 	}
1248 #endif
1249 
1250 	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1251 	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1252 	       dump_ok ? "yes" : "no");
1253 	return dump_ok;
1254 }
1255 
1256 /* An ELF note in memory */
1257 struct memelfnote
1258 {
1259 	const char *name;
1260 	int type;
1261 	unsigned int datasz;
1262 	void *data;
1263 };
1264 
1265 static int notesize(struct memelfnote *en)
1266 {
1267 	int sz;
1268 
1269 	sz = sizeof(struct elf_note);
1270 	sz += roundup(strlen(en->name) + 1, 4);
1271 	sz += roundup(en->datasz, 4);
1272 
1273 	return sz;
1274 }
1275 
1276 /* #define DEBUG */
1277 
1278 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1279 {
1280 	struct elf_note en;
1281 	en.n_namesz = strlen(men->name) + 1;
1282 	en.n_descsz = men->datasz;
1283 	en.n_type = men->type;
1284 
1285 	return dump_emit(cprm, &en, sizeof(en)) &&
1286 		dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1287 		dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1288 }
1289 
1290 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1291 {
1292 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1293 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1294 	elf->e_ident[EI_DATA] = ELF_DATA;
1295 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1296 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1297 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1298 
1299 	elf->e_type = ET_CORE;
1300 	elf->e_machine = ELF_ARCH;
1301 	elf->e_version = EV_CURRENT;
1302 	elf->e_entry = 0;
1303 	elf->e_phoff = sizeof(struct elfhdr);
1304 	elf->e_shoff = 0;
1305 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1306 	elf->e_ehsize = sizeof(struct elfhdr);
1307 	elf->e_phentsize = sizeof(struct elf_phdr);
1308 	elf->e_phnum = segs;
1309 	elf->e_shentsize = 0;
1310 	elf->e_shnum = 0;
1311 	elf->e_shstrndx = 0;
1312 	return;
1313 }
1314 
1315 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1316 {
1317 	phdr->p_type = PT_NOTE;
1318 	phdr->p_offset = offset;
1319 	phdr->p_vaddr = 0;
1320 	phdr->p_paddr = 0;
1321 	phdr->p_filesz = sz;
1322 	phdr->p_memsz = 0;
1323 	phdr->p_flags = 0;
1324 	phdr->p_align = 0;
1325 	return;
1326 }
1327 
1328 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1329 		unsigned int sz, void *data)
1330 {
1331 	note->name = name;
1332 	note->type = type;
1333 	note->datasz = sz;
1334 	note->data = data;
1335 	return;
1336 }
1337 
1338 /*
1339  * fill up all the fields in prstatus from the given task struct, except
1340  * registers which need to be filled up separately.
1341  */
1342 static void fill_prstatus(struct elf_prstatus *prstatus,
1343 			  struct task_struct *p, long signr)
1344 {
1345 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1346 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1347 	prstatus->pr_sighold = p->blocked.sig[0];
1348 	rcu_read_lock();
1349 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1350 	rcu_read_unlock();
1351 	prstatus->pr_pid = task_pid_vnr(p);
1352 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1353 	prstatus->pr_sid = task_session_vnr(p);
1354 	if (thread_group_leader(p)) {
1355 		struct task_cputime cputime;
1356 
1357 		/*
1358 		 * This is the record for the group leader.  It shows the
1359 		 * group-wide total, not its individual thread total.
1360 		 */
1361 		thread_group_cputime(p, &cputime);
1362 		prstatus->pr_utime = ns_to_kernel_old_timeval(cputime.utime);
1363 		prstatus->pr_stime = ns_to_kernel_old_timeval(cputime.stime);
1364 	} else {
1365 		u64 utime, stime;
1366 
1367 		task_cputime(p, &utime, &stime);
1368 		prstatus->pr_utime = ns_to_kernel_old_timeval(utime);
1369 		prstatus->pr_stime = ns_to_kernel_old_timeval(stime);
1370 	}
1371 	prstatus->pr_cutime = ns_to_kernel_old_timeval(p->signal->cutime);
1372 	prstatus->pr_cstime = ns_to_kernel_old_timeval(p->signal->cstime);
1373 
1374 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1375 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1376 }
1377 
1378 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1379 		       struct mm_struct *mm)
1380 {
1381 	const struct cred *cred;
1382 	unsigned int i, len;
1383 
1384 	/* first copy the parameters from user space */
1385 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1386 
1387 	len = mm->arg_end - mm->arg_start;
1388 	if (len >= ELF_PRARGSZ)
1389 		len = ELF_PRARGSZ - 1;
1390 	if (copy_from_user(&psinfo->pr_psargs,
1391 		           (const char __user *) mm->arg_start, len))
1392 		return -EFAULT;
1393 	for (i = 0; i < len; i++)
1394 		if (psinfo->pr_psargs[i] == 0)
1395 			psinfo->pr_psargs[i] = ' ';
1396 	psinfo->pr_psargs[len] = 0;
1397 
1398 	rcu_read_lock();
1399 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1400 	rcu_read_unlock();
1401 	psinfo->pr_pid = task_pid_vnr(p);
1402 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1403 	psinfo->pr_sid = task_session_vnr(p);
1404 
1405 	i = p->state ? ffz(~p->state) + 1 : 0;
1406 	psinfo->pr_state = i;
1407 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1408 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1409 	psinfo->pr_nice = task_nice(p);
1410 	psinfo->pr_flag = p->flags;
1411 	rcu_read_lock();
1412 	cred = __task_cred(p);
1413 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1414 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1415 	rcu_read_unlock();
1416 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1417 
1418 	return 0;
1419 }
1420 
1421 /* Here is the structure in which status of each thread is captured. */
1422 struct elf_thread_status
1423 {
1424 	struct list_head list;
1425 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1426 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1427 	struct task_struct *thread;
1428 #ifdef ELF_CORE_COPY_XFPREGS
1429 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1430 #endif
1431 	struct memelfnote notes[3];
1432 	int num_notes;
1433 };
1434 
1435 /*
1436  * In order to add the specific thread information for the elf file format,
1437  * we need to keep a linked list of every thread's pr_status and then create
1438  * a single section for them in the final core file.
1439  */
1440 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1441 {
1442 	struct task_struct *p = t->thread;
1443 	int sz = 0;
1444 
1445 	t->num_notes = 0;
1446 
1447 	fill_prstatus(&t->prstatus, p, signr);
1448 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1449 
1450 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1451 		  &t->prstatus);
1452 	t->num_notes++;
1453 	sz += notesize(&t->notes[0]);
1454 
1455 	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1456 	if (t->prstatus.pr_fpvalid) {
1457 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1458 			  &t->fpu);
1459 		t->num_notes++;
1460 		sz += notesize(&t->notes[1]);
1461 	}
1462 
1463 #ifdef ELF_CORE_COPY_XFPREGS
1464 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1465 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1466 			  sizeof(t->xfpu), &t->xfpu);
1467 		t->num_notes++;
1468 		sz += notesize(&t->notes[2]);
1469 	}
1470 #endif
1471 	return sz;
1472 }
1473 
1474 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1475 			     elf_addr_t e_shoff, int segs)
1476 {
1477 	elf->e_shoff = e_shoff;
1478 	elf->e_shentsize = sizeof(*shdr4extnum);
1479 	elf->e_shnum = 1;
1480 	elf->e_shstrndx = SHN_UNDEF;
1481 
1482 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1483 
1484 	shdr4extnum->sh_type = SHT_NULL;
1485 	shdr4extnum->sh_size = elf->e_shnum;
1486 	shdr4extnum->sh_link = elf->e_shstrndx;
1487 	shdr4extnum->sh_info = segs;
1488 }
1489 
1490 /*
1491  * dump the segments for an MMU process
1492  */
1493 static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1494 {
1495 	struct vm_area_struct *vma;
1496 
1497 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1498 #ifdef CONFIG_MMU
1499 		unsigned long addr;
1500 #endif
1501 
1502 		if (!maydump(vma, cprm->mm_flags))
1503 			continue;
1504 
1505 #ifdef CONFIG_MMU
1506 		for (addr = vma->vm_start; addr < vma->vm_end;
1507 							addr += PAGE_SIZE) {
1508 			bool res;
1509 			struct page *page = get_dump_page(addr);
1510 			if (page) {
1511 				void *kaddr = kmap(page);
1512 				res = dump_emit(cprm, kaddr, PAGE_SIZE);
1513 				kunmap(page);
1514 				put_page(page);
1515 			} else {
1516 				res = dump_skip(cprm, PAGE_SIZE);
1517 			}
1518 			if (!res)
1519 				return false;
1520 		}
1521 #else
1522 		if (!dump_emit(cprm, (void *) vma->vm_start,
1523 				vma->vm_end - vma->vm_start))
1524 			return false;
1525 #endif
1526 	}
1527 	return true;
1528 }
1529 
1530 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1531 {
1532 	struct vm_area_struct *vma;
1533 	size_t size = 0;
1534 
1535 	for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1536 		if (maydump(vma, mm_flags))
1537 			size += vma->vm_end - vma->vm_start;
1538 	return size;
1539 }
1540 
1541 /*
1542  * Actual dumper
1543  *
1544  * This is a two-pass process; first we find the offsets of the bits,
1545  * and then they are actually written out.  If we run out of core limit
1546  * we just truncate.
1547  */
1548 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1549 {
1550 #define	NUM_NOTES	6
1551 	int has_dumped = 0;
1552 	mm_segment_t fs;
1553 	int segs;
1554 	int i;
1555 	struct vm_area_struct *vma;
1556 	struct elfhdr *elf = NULL;
1557 	loff_t offset = 0, dataoff;
1558 	int numnote;
1559 	struct memelfnote *notes = NULL;
1560 	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1561 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1562  	LIST_HEAD(thread_list);
1563  	struct list_head *t;
1564 	elf_fpregset_t *fpu = NULL;
1565 #ifdef ELF_CORE_COPY_XFPREGS
1566 	elf_fpxregset_t *xfpu = NULL;
1567 #endif
1568 	int thread_status_size = 0;
1569 	elf_addr_t *auxv;
1570 	struct elf_phdr *phdr4note = NULL;
1571 	struct elf_shdr *shdr4extnum = NULL;
1572 	Elf_Half e_phnum;
1573 	elf_addr_t e_shoff;
1574 	struct core_thread *ct;
1575 	struct elf_thread_status *tmp;
1576 
1577 	/*
1578 	 * We no longer stop all VM operations.
1579 	 *
1580 	 * This is because those proceses that could possibly change map_count
1581 	 * or the mmap / vma pages are now blocked in do_exit on current
1582 	 * finishing this core dump.
1583 	 *
1584 	 * Only ptrace can touch these memory addresses, but it doesn't change
1585 	 * the map_count or the pages allocated. So no possibility of crashing
1586 	 * exists while dumping the mm->vm_next areas to the core file.
1587 	 */
1588 
1589 	/* alloc memory for large data structures: too large to be on stack */
1590 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1591 	if (!elf)
1592 		goto cleanup;
1593 	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1594 	if (!prstatus)
1595 		goto cleanup;
1596 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1597 	if (!psinfo)
1598 		goto cleanup;
1599 	notes = kmalloc_array(NUM_NOTES, sizeof(struct memelfnote),
1600 			      GFP_KERNEL);
1601 	if (!notes)
1602 		goto cleanup;
1603 	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1604 	if (!fpu)
1605 		goto cleanup;
1606 #ifdef ELF_CORE_COPY_XFPREGS
1607 	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1608 	if (!xfpu)
1609 		goto cleanup;
1610 #endif
1611 
1612 	for (ct = current->mm->core_state->dumper.next;
1613 					ct; ct = ct->next) {
1614 		tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1615 		if (!tmp)
1616 			goto cleanup;
1617 
1618 		tmp->thread = ct->task;
1619 		list_add(&tmp->list, &thread_list);
1620 	}
1621 
1622 	list_for_each(t, &thread_list) {
1623 		struct elf_thread_status *tmp;
1624 		int sz;
1625 
1626 		tmp = list_entry(t, struct elf_thread_status, list);
1627 		sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1628 		thread_status_size += sz;
1629 	}
1630 
1631 	/* now collect the dump for the current */
1632 	fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1633 	elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1634 
1635 	segs = current->mm->map_count;
1636 	segs += elf_core_extra_phdrs();
1637 
1638 	/* for notes section */
1639 	segs++;
1640 
1641 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1642 	 * this, kernel supports extended numbering. Have a look at
1643 	 * include/linux/elf.h for further information. */
1644 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1645 
1646 	/* Set up header */
1647 	fill_elf_fdpic_header(elf, e_phnum);
1648 
1649 	has_dumped = 1;
1650 	/*
1651 	 * Set up the notes in similar form to SVR4 core dumps made
1652 	 * with info from their /proc.
1653 	 */
1654 
1655 	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1656 	fill_psinfo(psinfo, current->group_leader, current->mm);
1657 	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1658 
1659 	numnote = 2;
1660 
1661 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1662 
1663 	i = 0;
1664 	do
1665 		i += 2;
1666 	while (auxv[i - 2] != AT_NULL);
1667 	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1668 		  i * sizeof(elf_addr_t), auxv);
1669 
1670   	/* Try to dump the FPU. */
1671 	if ((prstatus->pr_fpvalid =
1672 	     elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1673 		fill_note(notes + numnote++,
1674 			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1675 #ifdef ELF_CORE_COPY_XFPREGS
1676 	if (elf_core_copy_task_xfpregs(current, xfpu))
1677 		fill_note(notes + numnote++,
1678 			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1679 #endif
1680 
1681 	fs = get_fs();
1682 	set_fs(KERNEL_DS);
1683 
1684 	offset += sizeof(*elf);				/* Elf header */
1685 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1686 
1687 	/* Write notes phdr entry */
1688 	{
1689 		int sz = 0;
1690 
1691 		for (i = 0; i < numnote; i++)
1692 			sz += notesize(notes + i);
1693 
1694 		sz += thread_status_size;
1695 
1696 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1697 		if (!phdr4note)
1698 			goto end_coredump;
1699 
1700 		fill_elf_note_phdr(phdr4note, sz, offset);
1701 		offset += sz;
1702 	}
1703 
1704 	/* Page-align dumped data */
1705 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1706 
1707 	offset += elf_core_vma_data_size(cprm->mm_flags);
1708 	offset += elf_core_extra_data_size();
1709 	e_shoff = offset;
1710 
1711 	if (e_phnum == PN_XNUM) {
1712 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1713 		if (!shdr4extnum)
1714 			goto end_coredump;
1715 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1716 	}
1717 
1718 	offset = dataoff;
1719 
1720 	if (!dump_emit(cprm, elf, sizeof(*elf)))
1721 		goto end_coredump;
1722 
1723 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1724 		goto end_coredump;
1725 
1726 	/* write program headers for segments dump */
1727 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1728 		struct elf_phdr phdr;
1729 		size_t sz;
1730 
1731 		sz = vma->vm_end - vma->vm_start;
1732 
1733 		phdr.p_type = PT_LOAD;
1734 		phdr.p_offset = offset;
1735 		phdr.p_vaddr = vma->vm_start;
1736 		phdr.p_paddr = 0;
1737 		phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1738 		phdr.p_memsz = sz;
1739 		offset += phdr.p_filesz;
1740 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1741 		if (vma->vm_flags & VM_WRITE)
1742 			phdr.p_flags |= PF_W;
1743 		if (vma->vm_flags & VM_EXEC)
1744 			phdr.p_flags |= PF_X;
1745 		phdr.p_align = ELF_EXEC_PAGESIZE;
1746 
1747 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1748 			goto end_coredump;
1749 	}
1750 
1751 	if (!elf_core_write_extra_phdrs(cprm, offset))
1752 		goto end_coredump;
1753 
1754  	/* write out the notes section */
1755 	for (i = 0; i < numnote; i++)
1756 		if (!writenote(notes + i, cprm))
1757 			goto end_coredump;
1758 
1759 	/* write out the thread status notes section */
1760 	list_for_each(t, &thread_list) {
1761 		struct elf_thread_status *tmp =
1762 				list_entry(t, struct elf_thread_status, list);
1763 
1764 		for (i = 0; i < tmp->num_notes; i++)
1765 			if (!writenote(&tmp->notes[i], cprm))
1766 				goto end_coredump;
1767 	}
1768 
1769 	if (!dump_skip(cprm, dataoff - cprm->pos))
1770 		goto end_coredump;
1771 
1772 	if (!elf_fdpic_dump_segments(cprm))
1773 		goto end_coredump;
1774 
1775 	if (!elf_core_write_extra_data(cprm))
1776 		goto end_coredump;
1777 
1778 	if (e_phnum == PN_XNUM) {
1779 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1780 			goto end_coredump;
1781 	}
1782 
1783 	if (cprm->file->f_pos != offset) {
1784 		/* Sanity check */
1785 		printk(KERN_WARNING
1786 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1787 		       cprm->file->f_pos, offset);
1788 	}
1789 
1790 end_coredump:
1791 	set_fs(fs);
1792 
1793 cleanup:
1794 	while (!list_empty(&thread_list)) {
1795 		struct list_head *tmp = thread_list.next;
1796 		list_del(tmp);
1797 		kfree(list_entry(tmp, struct elf_thread_status, list));
1798 	}
1799 	kfree(phdr4note);
1800 	kfree(elf);
1801 	kfree(prstatus);
1802 	kfree(psinfo);
1803 	kfree(notes);
1804 	kfree(fpu);
1805 	kfree(shdr4extnum);
1806 #ifdef ELF_CORE_COPY_XFPREGS
1807 	kfree(xfpu);
1808 #endif
1809 	return has_dumped;
1810 #undef NUM_NOTES
1811 }
1812 
1813 #endif		/* CONFIG_ELF_CORE */
1814