xref: /openbmc/linux/fs/binfmt_elf_fdpic.c (revision 174cd4b1)
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/sched/coredump.h>
39 #include <linux/dax.h>
40 
41 #include <linux/uaccess.h>
42 #include <asm/param.h>
43 #include <asm/pgalloc.h>
44 
45 typedef char *elf_caddr_t;
46 
47 #if 0
48 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
49 #else
50 #define kdebug(fmt, ...) do {} while(0)
51 #endif
52 
53 #if 0
54 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
55 #else
56 #define kdcore(fmt, ...) do {} while(0)
57 #endif
58 
59 MODULE_LICENSE("GPL");
60 
61 static int load_elf_fdpic_binary(struct linux_binprm *);
62 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
63 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
64 			      struct mm_struct *, const char *);
65 
66 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
67 				   struct elf_fdpic_params *,
68 				   struct elf_fdpic_params *);
69 
70 #ifndef CONFIG_MMU
71 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
72 						   struct file *,
73 						   struct mm_struct *);
74 #endif
75 
76 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
77 					     struct file *, struct mm_struct *);
78 
79 #ifdef CONFIG_ELF_CORE
80 static int elf_fdpic_core_dump(struct coredump_params *cprm);
81 #endif
82 
83 static struct linux_binfmt elf_fdpic_format = {
84 	.module		= THIS_MODULE,
85 	.load_binary	= load_elf_fdpic_binary,
86 #ifdef CONFIG_ELF_CORE
87 	.core_dump	= elf_fdpic_core_dump,
88 #endif
89 	.min_coredump	= ELF_EXEC_PAGESIZE,
90 };
91 
92 static int __init init_elf_fdpic_binfmt(void)
93 {
94 	register_binfmt(&elf_fdpic_format);
95 	return 0;
96 }
97 
98 static void __exit exit_elf_fdpic_binfmt(void)
99 {
100 	unregister_binfmt(&elf_fdpic_format);
101 }
102 
103 core_initcall(init_elf_fdpic_binfmt);
104 module_exit(exit_elf_fdpic_binfmt);
105 
106 static int is_elf(struct elfhdr *hdr, struct file *file)
107 {
108 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
109 		return 0;
110 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
111 		return 0;
112 	if (!elf_check_arch(hdr))
113 		return 0;
114 	if (!file->f_op->mmap)
115 		return 0;
116 	return 1;
117 }
118 
119 #ifndef elf_check_fdpic
120 #define elf_check_fdpic(x) 0
121 #endif
122 
123 #ifndef elf_check_const_displacement
124 #define elf_check_const_displacement(x) 0
125 #endif
126 
127 static int is_constdisp(struct elfhdr *hdr)
128 {
129 	if (!elf_check_fdpic(hdr))
130 		return 1;
131 	if (elf_check_const_displacement(hdr))
132 		return 1;
133 	return 0;
134 }
135 
136 /*****************************************************************************/
137 /*
138  * read the program headers table into memory
139  */
140 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
141 				 struct file *file)
142 {
143 	struct elf32_phdr *phdr;
144 	unsigned long size;
145 	int retval, loop;
146 
147 	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
148 		return -ENOMEM;
149 	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
150 		return -ENOMEM;
151 
152 	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
153 	params->phdrs = kmalloc(size, GFP_KERNEL);
154 	if (!params->phdrs)
155 		return -ENOMEM;
156 
157 	retval = kernel_read(file, params->hdr.e_phoff,
158 			     (char *) params->phdrs, size);
159 	if (unlikely(retval != size))
160 		return retval < 0 ? retval : -ENOEXEC;
161 
162 	/* determine stack size for this binary */
163 	phdr = params->phdrs;
164 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
165 		if (phdr->p_type != PT_GNU_STACK)
166 			continue;
167 
168 		if (phdr->p_flags & PF_X)
169 			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
170 		else
171 			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
172 
173 		params->stack_size = phdr->p_memsz;
174 		break;
175 	}
176 
177 	return 0;
178 }
179 
180 /*****************************************************************************/
181 /*
182  * load an fdpic binary into various bits of memory
183  */
184 static int load_elf_fdpic_binary(struct linux_binprm *bprm)
185 {
186 	struct elf_fdpic_params exec_params, interp_params;
187 	struct pt_regs *regs = current_pt_regs();
188 	struct elf_phdr *phdr;
189 	unsigned long stack_size, entryaddr;
190 #ifdef ELF_FDPIC_PLAT_INIT
191 	unsigned long dynaddr;
192 #endif
193 #ifndef CONFIG_MMU
194 	unsigned long stack_prot;
195 #endif
196 	struct file *interpreter = NULL; /* to shut gcc up */
197 	char *interpreter_name = NULL;
198 	int executable_stack;
199 	int retval, i;
200 
201 	kdebug("____ LOAD %d ____", current->pid);
202 
203 	memset(&exec_params, 0, sizeof(exec_params));
204 	memset(&interp_params, 0, sizeof(interp_params));
205 
206 	exec_params.hdr = *(struct elfhdr *) bprm->buf;
207 	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
208 
209 	/* check that this is a binary we know how to deal with */
210 	retval = -ENOEXEC;
211 	if (!is_elf(&exec_params.hdr, bprm->file))
212 		goto error;
213 	if (!elf_check_fdpic(&exec_params.hdr)) {
214 #ifdef CONFIG_MMU
215 		/* binfmt_elf handles non-fdpic elf except on nommu */
216 		goto error;
217 #else
218 		/* nommu can only load ET_DYN (PIE) ELF */
219 		if (exec_params.hdr.e_type != ET_DYN)
220 			goto error;
221 #endif
222 	}
223 
224 	/* read the program header table */
225 	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
226 	if (retval < 0)
227 		goto error;
228 
229 	/* scan for a program header that specifies an interpreter */
230 	phdr = exec_params.phdrs;
231 
232 	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
233 		switch (phdr->p_type) {
234 		case PT_INTERP:
235 			retval = -ENOMEM;
236 			if (phdr->p_filesz > PATH_MAX)
237 				goto error;
238 			retval = -ENOENT;
239 			if (phdr->p_filesz < 2)
240 				goto error;
241 
242 			/* read the name of the interpreter into memory */
243 			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
244 			if (!interpreter_name)
245 				goto error;
246 
247 			retval = kernel_read(bprm->file,
248 					     phdr->p_offset,
249 					     interpreter_name,
250 					     phdr->p_filesz);
251 			if (unlikely(retval != phdr->p_filesz)) {
252 				if (retval >= 0)
253 					retval = -ENOEXEC;
254 				goto error;
255 			}
256 
257 			retval = -ENOENT;
258 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
259 				goto error;
260 
261 			kdebug("Using ELF interpreter %s", interpreter_name);
262 
263 			/* replace the program with the interpreter */
264 			interpreter = open_exec(interpreter_name);
265 			retval = PTR_ERR(interpreter);
266 			if (IS_ERR(interpreter)) {
267 				interpreter = NULL;
268 				goto error;
269 			}
270 
271 			/*
272 			 * If the binary is not readable then enforce
273 			 * mm->dumpable = 0 regardless of the interpreter's
274 			 * permissions.
275 			 */
276 			would_dump(bprm, interpreter);
277 
278 			retval = kernel_read(interpreter, 0, bprm->buf,
279 					     BINPRM_BUF_SIZE);
280 			if (unlikely(retval != BINPRM_BUF_SIZE)) {
281 				if (retval >= 0)
282 					retval = -ENOEXEC;
283 				goto error;
284 			}
285 
286 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
287 			break;
288 
289 		case PT_LOAD:
290 #ifdef CONFIG_MMU
291 			if (exec_params.load_addr == 0)
292 				exec_params.load_addr = phdr->p_vaddr;
293 #endif
294 			break;
295 		}
296 
297 	}
298 
299 	if (is_constdisp(&exec_params.hdr))
300 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
301 
302 	/* perform insanity checks on the interpreter */
303 	if (interpreter_name) {
304 		retval = -ELIBBAD;
305 		if (!is_elf(&interp_params.hdr, interpreter))
306 			goto error;
307 
308 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
309 
310 		/* read the interpreter's program header table */
311 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
312 		if (retval < 0)
313 			goto error;
314 	}
315 
316 	stack_size = exec_params.stack_size;
317 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
318 		executable_stack = EXSTACK_ENABLE_X;
319 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
320 		executable_stack = EXSTACK_DISABLE_X;
321 	else
322 		executable_stack = EXSTACK_DEFAULT;
323 
324 	if (stack_size == 0) {
325 		stack_size = interp_params.stack_size;
326 		if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
327 			executable_stack = EXSTACK_ENABLE_X;
328 		else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
329 			executable_stack = EXSTACK_DISABLE_X;
330 		else
331 			executable_stack = EXSTACK_DEFAULT;
332 	}
333 
334 	retval = -ENOEXEC;
335 	if (stack_size == 0)
336 		stack_size = 131072UL; /* same as exec.c's default commit */
337 
338 	if (is_constdisp(&interp_params.hdr))
339 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
340 
341 	/* flush all traces of the currently running executable */
342 	retval = flush_old_exec(bprm);
343 	if (retval)
344 		goto error;
345 
346 	/* there's now no turning back... the old userspace image is dead,
347 	 * defunct, deceased, etc.
348 	 */
349 	if (elf_check_fdpic(&exec_params.hdr))
350 		set_personality(PER_LINUX_FDPIC);
351 	else
352 		set_personality(PER_LINUX);
353 	if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
354 		current->personality |= READ_IMPLIES_EXEC;
355 
356 	setup_new_exec(bprm);
357 
358 	set_binfmt(&elf_fdpic_format);
359 
360 	current->mm->start_code = 0;
361 	current->mm->end_code = 0;
362 	current->mm->start_stack = 0;
363 	current->mm->start_data = 0;
364 	current->mm->end_data = 0;
365 	current->mm->context.exec_fdpic_loadmap = 0;
366 	current->mm->context.interp_fdpic_loadmap = 0;
367 
368 #ifdef CONFIG_MMU
369 	elf_fdpic_arch_lay_out_mm(&exec_params,
370 				  &interp_params,
371 				  &current->mm->start_stack,
372 				  &current->mm->start_brk);
373 
374 	retval = setup_arg_pages(bprm, current->mm->start_stack,
375 				 executable_stack);
376 	if (retval < 0)
377 		goto error;
378 #endif
379 
380 	/* load the executable and interpreter into memory */
381 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
382 				    "executable");
383 	if (retval < 0)
384 		goto error;
385 
386 	if (interpreter_name) {
387 		retval = elf_fdpic_map_file(&interp_params, interpreter,
388 					    current->mm, "interpreter");
389 		if (retval < 0) {
390 			printk(KERN_ERR "Unable to load interpreter\n");
391 			goto error;
392 		}
393 
394 		allow_write_access(interpreter);
395 		fput(interpreter);
396 		interpreter = NULL;
397 	}
398 
399 #ifdef CONFIG_MMU
400 	if (!current->mm->start_brk)
401 		current->mm->start_brk = current->mm->end_data;
402 
403 	current->mm->brk = current->mm->start_brk =
404 		PAGE_ALIGN(current->mm->start_brk);
405 
406 #else
407 	/* create a stack area and zero-size brk area */
408 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
409 	if (stack_size < PAGE_SIZE * 2)
410 		stack_size = PAGE_SIZE * 2;
411 
412 	stack_prot = PROT_READ | PROT_WRITE;
413 	if (executable_stack == EXSTACK_ENABLE_X ||
414 	    (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
415 		stack_prot |= PROT_EXEC;
416 
417 	current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
418 					 MAP_PRIVATE | MAP_ANONYMOUS |
419 					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
420 					 0);
421 
422 	if (IS_ERR_VALUE(current->mm->start_brk)) {
423 		retval = current->mm->start_brk;
424 		current->mm->start_brk = 0;
425 		goto error;
426 	}
427 
428 	current->mm->brk = current->mm->start_brk;
429 	current->mm->context.end_brk = current->mm->start_brk;
430 	current->mm->start_stack = current->mm->start_brk + stack_size;
431 #endif
432 
433 	install_exec_creds(bprm);
434 	if (create_elf_fdpic_tables(bprm, current->mm,
435 				    &exec_params, &interp_params) < 0)
436 		goto error;
437 
438 	kdebug("- start_code  %lx", current->mm->start_code);
439 	kdebug("- end_code    %lx", current->mm->end_code);
440 	kdebug("- start_data  %lx", current->mm->start_data);
441 	kdebug("- end_data    %lx", current->mm->end_data);
442 	kdebug("- start_brk   %lx", current->mm->start_brk);
443 	kdebug("- brk         %lx", current->mm->brk);
444 	kdebug("- start_stack %lx", current->mm->start_stack);
445 
446 #ifdef ELF_FDPIC_PLAT_INIT
447 	/*
448 	 * The ABI may specify that certain registers be set up in special
449 	 * ways (on i386 %edx is the address of a DT_FINI function, for
450 	 * example.  This macro performs whatever initialization to
451 	 * the regs structure is required.
452 	 */
453 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
454 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
455 			    dynaddr);
456 #endif
457 
458 	/* everything is now ready... get the userspace context ready to roll */
459 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
460 	start_thread(regs, entryaddr, current->mm->start_stack);
461 
462 	retval = 0;
463 
464 error:
465 	if (interpreter) {
466 		allow_write_access(interpreter);
467 		fput(interpreter);
468 	}
469 	kfree(interpreter_name);
470 	kfree(exec_params.phdrs);
471 	kfree(exec_params.loadmap);
472 	kfree(interp_params.phdrs);
473 	kfree(interp_params.loadmap);
474 	return retval;
475 }
476 
477 /*****************************************************************************/
478 
479 #ifndef ELF_BASE_PLATFORM
480 /*
481  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
482  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
483  * will be copied to the user stack in the same manner as AT_PLATFORM.
484  */
485 #define ELF_BASE_PLATFORM NULL
486 #endif
487 
488 /*
489  * present useful information to the program by shovelling it onto the new
490  * process's stack
491  */
492 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
493 				   struct mm_struct *mm,
494 				   struct elf_fdpic_params *exec_params,
495 				   struct elf_fdpic_params *interp_params)
496 {
497 	const struct cred *cred = current_cred();
498 	unsigned long sp, csp, nitems;
499 	elf_caddr_t __user *argv, *envp;
500 	size_t platform_len = 0, len;
501 	char *k_platform, *k_base_platform;
502 	char __user *u_platform, *u_base_platform, *p;
503 	int loop;
504 	int nr;	/* reset for each csp adjustment */
505 
506 #ifdef CONFIG_MMU
507 	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
508 	 * by the processes running on the same package. One thing we can do is
509 	 * to shuffle the initial stack for them, so we give the architecture
510 	 * an opportunity to do so here.
511 	 */
512 	sp = arch_align_stack(bprm->p);
513 #else
514 	sp = mm->start_stack;
515 
516 	/* stack the program arguments and environment */
517 	if (transfer_args_to_stack(bprm, &sp) < 0)
518 		return -EFAULT;
519 	sp &= ~15;
520 #endif
521 
522 	/*
523 	 * If this architecture has a platform capability string, copy it
524 	 * to userspace.  In some cases (Sparc), this info is impossible
525 	 * for userspace to get any other way, in others (i386) it is
526 	 * merely difficult.
527 	 */
528 	k_platform = ELF_PLATFORM;
529 	u_platform = NULL;
530 
531 	if (k_platform) {
532 		platform_len = strlen(k_platform) + 1;
533 		sp -= platform_len;
534 		u_platform = (char __user *) sp;
535 		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
536 			return -EFAULT;
537 	}
538 
539 	/*
540 	 * If this architecture has a "base" platform capability
541 	 * string, copy it to userspace.
542 	 */
543 	k_base_platform = ELF_BASE_PLATFORM;
544 	u_base_platform = NULL;
545 
546 	if (k_base_platform) {
547 		platform_len = strlen(k_base_platform) + 1;
548 		sp -= platform_len;
549 		u_base_platform = (char __user *) sp;
550 		if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
551 			return -EFAULT;
552 	}
553 
554 	sp &= ~7UL;
555 
556 	/* stack the load map(s) */
557 	len = sizeof(struct elf32_fdpic_loadmap);
558 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
559 	sp = (sp - len) & ~7UL;
560 	exec_params->map_addr = sp;
561 
562 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
563 		return -EFAULT;
564 
565 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
566 
567 	if (interp_params->loadmap) {
568 		len = sizeof(struct elf32_fdpic_loadmap);
569 		len += sizeof(struct elf32_fdpic_loadseg) *
570 			interp_params->loadmap->nsegs;
571 		sp = (sp - len) & ~7UL;
572 		interp_params->map_addr = sp;
573 
574 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
575 				 len) != 0)
576 			return -EFAULT;
577 
578 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
579 	}
580 
581 	/* force 16 byte _final_ alignment here for generality */
582 #define DLINFO_ITEMS 15
583 
584 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
585 		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
586 
587 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
588 		nitems++;
589 
590 	csp = sp;
591 	sp -= nitems * 2 * sizeof(unsigned long);
592 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
593 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
594 	sp -= 1 * sizeof(unsigned long);		/* argc */
595 
596 	csp -= sp & 15UL;
597 	sp -= sp & 15UL;
598 
599 	/* put the ELF interpreter info on the stack */
600 #define NEW_AUX_ENT(id, val)						\
601 	do {								\
602 		struct { unsigned long _id, _val; } __user *ent;	\
603 									\
604 		ent = (void __user *) csp;				\
605 		__put_user((id), &ent[nr]._id);				\
606 		__put_user((val), &ent[nr]._val);			\
607 		nr++;							\
608 	} while (0)
609 
610 	nr = 0;
611 	csp -= 2 * sizeof(unsigned long);
612 	NEW_AUX_ENT(AT_NULL, 0);
613 	if (k_platform) {
614 		nr = 0;
615 		csp -= 2 * sizeof(unsigned long);
616 		NEW_AUX_ENT(AT_PLATFORM,
617 			    (elf_addr_t) (unsigned long) u_platform);
618 	}
619 
620 	if (k_base_platform) {
621 		nr = 0;
622 		csp -= 2 * sizeof(unsigned long);
623 		NEW_AUX_ENT(AT_BASE_PLATFORM,
624 			    (elf_addr_t) (unsigned long) u_base_platform);
625 	}
626 
627 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
628 		nr = 0;
629 		csp -= 2 * sizeof(unsigned long);
630 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
631 	}
632 
633 	nr = 0;
634 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
635 	NEW_AUX_ENT(AT_HWCAP,	ELF_HWCAP);
636 #ifdef ELF_HWCAP2
637 	NEW_AUX_ENT(AT_HWCAP2,	ELF_HWCAP2);
638 #endif
639 	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
640 	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
641 	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
642 	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
643 	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
644 	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
645 	NEW_AUX_ENT(AT_FLAGS,	0);
646 	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
647 	NEW_AUX_ENT(AT_UID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
648 	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
649 	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
650 	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
651 	NEW_AUX_ENT(AT_SECURE,	security_bprm_secureexec(bprm));
652 	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
653 
654 #ifdef ARCH_DLINFO
655 	nr = 0;
656 	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
657 
658 	/* ARCH_DLINFO must come last so platform specific code can enforce
659 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
660 	 */
661 	ARCH_DLINFO;
662 #endif
663 #undef NEW_AUX_ENT
664 
665 	/* allocate room for argv[] and envv[] */
666 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
667 	envp = (elf_caddr_t __user *) csp;
668 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
669 	argv = (elf_caddr_t __user *) csp;
670 
671 	/* stack argc */
672 	csp -= sizeof(unsigned long);
673 	__put_user(bprm->argc, (unsigned long __user *) csp);
674 
675 	BUG_ON(csp != sp);
676 
677 	/* fill in the argv[] array */
678 #ifdef CONFIG_MMU
679 	current->mm->arg_start = bprm->p;
680 #else
681 	current->mm->arg_start = current->mm->start_stack -
682 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
683 #endif
684 
685 	p = (char __user *) current->mm->arg_start;
686 	for (loop = bprm->argc; loop > 0; loop--) {
687 		__put_user((elf_caddr_t) p, argv++);
688 		len = strnlen_user(p, MAX_ARG_STRLEN);
689 		if (!len || len > MAX_ARG_STRLEN)
690 			return -EINVAL;
691 		p += len;
692 	}
693 	__put_user(NULL, argv);
694 	current->mm->arg_end = (unsigned long) p;
695 
696 	/* fill in the envv[] array */
697 	current->mm->env_start = (unsigned long) p;
698 	for (loop = bprm->envc; loop > 0; loop--) {
699 		__put_user((elf_caddr_t)(unsigned long) p, envp++);
700 		len = strnlen_user(p, MAX_ARG_STRLEN);
701 		if (!len || len > MAX_ARG_STRLEN)
702 			return -EINVAL;
703 		p += len;
704 	}
705 	__put_user(NULL, envp);
706 	current->mm->env_end = (unsigned long) p;
707 
708 	mm->start_stack = (unsigned long) sp;
709 	return 0;
710 }
711 
712 /*****************************************************************************/
713 /*
714  * load the appropriate binary image (executable or interpreter) into memory
715  * - we assume no MMU is available
716  * - if no other PIC bits are set in params->hdr->e_flags
717  *   - we assume that the LOADable segments in the binary are independently relocatable
718  *   - we assume R/O executable segments are shareable
719  * - else
720  *   - we assume the loadable parts of the image to require fixed displacement
721  *   - the image is not shareable
722  */
723 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
724 			      struct file *file,
725 			      struct mm_struct *mm,
726 			      const char *what)
727 {
728 	struct elf32_fdpic_loadmap *loadmap;
729 #ifdef CONFIG_MMU
730 	struct elf32_fdpic_loadseg *mseg;
731 #endif
732 	struct elf32_fdpic_loadseg *seg;
733 	struct elf32_phdr *phdr;
734 	unsigned long load_addr, stop;
735 	unsigned nloads, tmp;
736 	size_t size;
737 	int loop, ret;
738 
739 	/* allocate a load map table */
740 	nloads = 0;
741 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
742 		if (params->phdrs[loop].p_type == PT_LOAD)
743 			nloads++;
744 
745 	if (nloads == 0)
746 		return -ELIBBAD;
747 
748 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
749 	loadmap = kzalloc(size, GFP_KERNEL);
750 	if (!loadmap)
751 		return -ENOMEM;
752 
753 	params->loadmap = loadmap;
754 
755 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
756 	loadmap->nsegs = nloads;
757 
758 	load_addr = params->load_addr;
759 	seg = loadmap->segs;
760 
761 	/* map the requested LOADs into the memory space */
762 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
763 	case ELF_FDPIC_FLAG_CONSTDISP:
764 	case ELF_FDPIC_FLAG_CONTIGUOUS:
765 #ifndef CONFIG_MMU
766 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
767 		if (ret < 0)
768 			return ret;
769 		break;
770 #endif
771 	default:
772 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
773 		if (ret < 0)
774 			return ret;
775 		break;
776 	}
777 
778 	/* map the entry point */
779 	if (params->hdr.e_entry) {
780 		seg = loadmap->segs;
781 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
782 			if (params->hdr.e_entry >= seg->p_vaddr &&
783 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
784 				params->entry_addr =
785 					(params->hdr.e_entry - seg->p_vaddr) +
786 					seg->addr;
787 				break;
788 			}
789 		}
790 	}
791 
792 	/* determine where the program header table has wound up if mapped */
793 	stop = params->hdr.e_phoff;
794 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
795 	phdr = params->phdrs;
796 
797 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
798 		if (phdr->p_type != PT_LOAD)
799 			continue;
800 
801 		if (phdr->p_offset > params->hdr.e_phoff ||
802 		    phdr->p_offset + phdr->p_filesz < stop)
803 			continue;
804 
805 		seg = loadmap->segs;
806 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
807 			if (phdr->p_vaddr >= seg->p_vaddr &&
808 			    phdr->p_vaddr + phdr->p_filesz <=
809 			    seg->p_vaddr + seg->p_memsz) {
810 				params->ph_addr =
811 					(phdr->p_vaddr - seg->p_vaddr) +
812 					seg->addr +
813 					params->hdr.e_phoff - phdr->p_offset;
814 				break;
815 			}
816 		}
817 		break;
818 	}
819 
820 	/* determine where the dynamic section has wound up if there is one */
821 	phdr = params->phdrs;
822 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
823 		if (phdr->p_type != PT_DYNAMIC)
824 			continue;
825 
826 		seg = loadmap->segs;
827 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
828 			if (phdr->p_vaddr >= seg->p_vaddr &&
829 			    phdr->p_vaddr + phdr->p_memsz <=
830 			    seg->p_vaddr + seg->p_memsz) {
831 				params->dynamic_addr =
832 					(phdr->p_vaddr - seg->p_vaddr) +
833 					seg->addr;
834 
835 				/* check the dynamic section contains at least
836 				 * one item, and that the last item is a NULL
837 				 * entry */
838 				if (phdr->p_memsz == 0 ||
839 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
840 					goto dynamic_error;
841 
842 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
843 				if (((Elf32_Dyn *)
844 				     params->dynamic_addr)[tmp - 1].d_tag != 0)
845 					goto dynamic_error;
846 				break;
847 			}
848 		}
849 		break;
850 	}
851 
852 	/* now elide adjacent segments in the load map on MMU linux
853 	 * - on uClinux the holes between may actually be filled with system
854 	 *   stuff or stuff from other processes
855 	 */
856 #ifdef CONFIG_MMU
857 	nloads = loadmap->nsegs;
858 	mseg = loadmap->segs;
859 	seg = mseg + 1;
860 	for (loop = 1; loop < nloads; loop++) {
861 		/* see if we have a candidate for merging */
862 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
863 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
864 			if (load_addr == (seg->addr & PAGE_MASK)) {
865 				mseg->p_memsz +=
866 					load_addr -
867 					(mseg->addr + mseg->p_memsz);
868 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
869 				mseg->p_memsz += seg->p_memsz;
870 				loadmap->nsegs--;
871 				continue;
872 			}
873 		}
874 
875 		mseg++;
876 		if (mseg != seg)
877 			*mseg = *seg;
878 	}
879 #endif
880 
881 	kdebug("Mapped Object [%s]:", what);
882 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
883 	kdebug("- entry    : %lx", params->entry_addr);
884 	kdebug("- PHDR[]   : %lx", params->ph_addr);
885 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
886 	seg = loadmap->segs;
887 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
888 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
889 		       loop,
890 		       seg->addr, seg->addr + seg->p_memsz - 1,
891 		       seg->p_vaddr, seg->p_memsz);
892 
893 	return 0;
894 
895 dynamic_error:
896 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
897 	       what, file_inode(file)->i_ino);
898 	return -ELIBBAD;
899 }
900 
901 /*****************************************************************************/
902 /*
903  * map a file with constant displacement under uClinux
904  */
905 #ifndef CONFIG_MMU
906 static int elf_fdpic_map_file_constdisp_on_uclinux(
907 	struct elf_fdpic_params *params,
908 	struct file *file,
909 	struct mm_struct *mm)
910 {
911 	struct elf32_fdpic_loadseg *seg;
912 	struct elf32_phdr *phdr;
913 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
914 	int loop, ret;
915 
916 	load_addr = params->load_addr;
917 	seg = params->loadmap->segs;
918 
919 	/* determine the bounds of the contiguous overall allocation we must
920 	 * make */
921 	phdr = params->phdrs;
922 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
923 		if (params->phdrs[loop].p_type != PT_LOAD)
924 			continue;
925 
926 		if (base > phdr->p_vaddr)
927 			base = phdr->p_vaddr;
928 		if (top < phdr->p_vaddr + phdr->p_memsz)
929 			top = phdr->p_vaddr + phdr->p_memsz;
930 	}
931 
932 	/* allocate one big anon block for everything */
933 	mflags = MAP_PRIVATE;
934 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
935 		mflags |= MAP_EXECUTABLE;
936 
937 	maddr = vm_mmap(NULL, load_addr, top - base,
938 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
939 	if (IS_ERR_VALUE(maddr))
940 		return (int) maddr;
941 
942 	if (load_addr != 0)
943 		load_addr += PAGE_ALIGN(top - base);
944 
945 	/* and then load the file segments into it */
946 	phdr = params->phdrs;
947 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
948 		if (params->phdrs[loop].p_type != PT_LOAD)
949 			continue;
950 
951 		seg->addr = maddr + (phdr->p_vaddr - base);
952 		seg->p_vaddr = phdr->p_vaddr;
953 		seg->p_memsz = phdr->p_memsz;
954 
955 		ret = read_code(file, seg->addr, phdr->p_offset,
956 				       phdr->p_filesz);
957 		if (ret < 0)
958 			return ret;
959 
960 		/* map the ELF header address if in this segment */
961 		if (phdr->p_offset == 0)
962 			params->elfhdr_addr = seg->addr;
963 
964 		/* clear any space allocated but not loaded */
965 		if (phdr->p_filesz < phdr->p_memsz) {
966 			if (clear_user((void *) (seg->addr + phdr->p_filesz),
967 				       phdr->p_memsz - phdr->p_filesz))
968 				return -EFAULT;
969 		}
970 
971 		if (mm) {
972 			if (phdr->p_flags & PF_X) {
973 				if (!mm->start_code) {
974 					mm->start_code = seg->addr;
975 					mm->end_code = seg->addr +
976 						phdr->p_memsz;
977 				}
978 			} else if (!mm->start_data) {
979 				mm->start_data = seg->addr;
980 				mm->end_data = seg->addr + phdr->p_memsz;
981 			}
982 		}
983 
984 		seg++;
985 	}
986 
987 	return 0;
988 }
989 #endif
990 
991 /*****************************************************************************/
992 /*
993  * map a binary by direct mmap() of the individual PT_LOAD segments
994  */
995 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
996 					     struct file *file,
997 					     struct mm_struct *mm)
998 {
999 	struct elf32_fdpic_loadseg *seg;
1000 	struct elf32_phdr *phdr;
1001 	unsigned long load_addr, delta_vaddr;
1002 	int loop, dvset;
1003 
1004 	load_addr = params->load_addr;
1005 	delta_vaddr = 0;
1006 	dvset = 0;
1007 
1008 	seg = params->loadmap->segs;
1009 
1010 	/* deal with each load segment separately */
1011 	phdr = params->phdrs;
1012 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1013 		unsigned long maddr, disp, excess, excess1;
1014 		int prot = 0, flags;
1015 
1016 		if (phdr->p_type != PT_LOAD)
1017 			continue;
1018 
1019 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1020 		       (unsigned long) phdr->p_vaddr,
1021 		       (unsigned long) phdr->p_offset,
1022 		       (unsigned long) phdr->p_filesz,
1023 		       (unsigned long) phdr->p_memsz);
1024 
1025 		/* determine the mapping parameters */
1026 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1027 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1028 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1029 
1030 		flags = MAP_PRIVATE | MAP_DENYWRITE;
1031 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1032 			flags |= MAP_EXECUTABLE;
1033 
1034 		maddr = 0;
1035 
1036 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1037 		case ELF_FDPIC_FLAG_INDEPENDENT:
1038 			/* PT_LOADs are independently locatable */
1039 			break;
1040 
1041 		case ELF_FDPIC_FLAG_HONOURVADDR:
1042 			/* the specified virtual address must be honoured */
1043 			maddr = phdr->p_vaddr;
1044 			flags |= MAP_FIXED;
1045 			break;
1046 
1047 		case ELF_FDPIC_FLAG_CONSTDISP:
1048 			/* constant displacement
1049 			 * - can be mapped anywhere, but must be mapped as a
1050 			 *   unit
1051 			 */
1052 			if (!dvset) {
1053 				maddr = load_addr;
1054 				delta_vaddr = phdr->p_vaddr;
1055 				dvset = 1;
1056 			} else {
1057 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1058 				flags |= MAP_FIXED;
1059 			}
1060 			break;
1061 
1062 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1063 			/* contiguity handled later */
1064 			break;
1065 
1066 		default:
1067 			BUG();
1068 		}
1069 
1070 		maddr &= PAGE_MASK;
1071 
1072 		/* create the mapping */
1073 		disp = phdr->p_vaddr & ~PAGE_MASK;
1074 		maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1075 				phdr->p_offset - disp);
1076 
1077 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1078 		       loop, phdr->p_memsz + disp, prot, flags,
1079 		       phdr->p_offset - disp, maddr);
1080 
1081 		if (IS_ERR_VALUE(maddr))
1082 			return (int) maddr;
1083 
1084 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1085 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1086 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1087 
1088 		seg->addr = maddr + disp;
1089 		seg->p_vaddr = phdr->p_vaddr;
1090 		seg->p_memsz = phdr->p_memsz;
1091 
1092 		/* map the ELF header address if in this segment */
1093 		if (phdr->p_offset == 0)
1094 			params->elfhdr_addr = seg->addr;
1095 
1096 		/* clear the bit between beginning of mapping and beginning of
1097 		 * PT_LOAD */
1098 		if (prot & PROT_WRITE && disp > 0) {
1099 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1100 			if (clear_user((void __user *) maddr, disp))
1101 				return -EFAULT;
1102 			maddr += disp;
1103 		}
1104 
1105 		/* clear any space allocated but not loaded
1106 		 * - on uClinux we can just clear the lot
1107 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1108 		 *   extant in the file
1109 		 */
1110 		excess = phdr->p_memsz - phdr->p_filesz;
1111 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1112 
1113 #ifdef CONFIG_MMU
1114 		if (excess > excess1) {
1115 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1116 			unsigned long xmaddr;
1117 
1118 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1119 			xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1120 					 prot, flags, 0);
1121 
1122 			kdebug("mmap[%d] <anon>"
1123 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1124 			       loop, xaddr, excess - excess1, prot, flags,
1125 			       xmaddr);
1126 
1127 			if (xmaddr != xaddr)
1128 				return -ENOMEM;
1129 		}
1130 
1131 		if (prot & PROT_WRITE && excess1 > 0) {
1132 			kdebug("clear[%d] ad=%lx sz=%lx",
1133 			       loop, maddr + phdr->p_filesz, excess1);
1134 			if (clear_user((void __user *) maddr + phdr->p_filesz,
1135 				       excess1))
1136 				return -EFAULT;
1137 		}
1138 
1139 #else
1140 		if (excess > 0) {
1141 			kdebug("clear[%d] ad=%lx sz=%lx",
1142 			       loop, maddr + phdr->p_filesz, excess);
1143 			if (clear_user((void *) maddr + phdr->p_filesz, excess))
1144 				return -EFAULT;
1145 		}
1146 #endif
1147 
1148 		if (mm) {
1149 			if (phdr->p_flags & PF_X) {
1150 				if (!mm->start_code) {
1151 					mm->start_code = maddr;
1152 					mm->end_code = maddr + phdr->p_memsz;
1153 				}
1154 			} else if (!mm->start_data) {
1155 				mm->start_data = maddr;
1156 				mm->end_data = maddr + phdr->p_memsz;
1157 			}
1158 		}
1159 
1160 		seg++;
1161 	}
1162 
1163 	return 0;
1164 }
1165 
1166 /*****************************************************************************/
1167 /*
1168  * ELF-FDPIC core dumper
1169  *
1170  * Modelled on fs/exec.c:aout_core_dump()
1171  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1172  *
1173  * Modelled on fs/binfmt_elf.c core dumper
1174  */
1175 #ifdef CONFIG_ELF_CORE
1176 
1177 /*
1178  * Decide whether a segment is worth dumping; default is yes to be
1179  * sure (missing info is worse than too much; etc).
1180  * Personally I'd include everything, and use the coredump limit...
1181  *
1182  * I think we should skip something. But I am not sure how. H.J.
1183  */
1184 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1185 {
1186 	int dump_ok;
1187 
1188 	/* Do not dump I/O mapped devices or special mappings */
1189 	if (vma->vm_flags & VM_IO) {
1190 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1191 		return 0;
1192 	}
1193 
1194 	/* If we may not read the contents, don't allow us to dump
1195 	 * them either. "dump_write()" can't handle it anyway.
1196 	 */
1197 	if (!(vma->vm_flags & VM_READ)) {
1198 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1199 		return 0;
1200 	}
1201 
1202 	/* support for DAX */
1203 	if (vma_is_dax(vma)) {
1204 		if (vma->vm_flags & VM_SHARED) {
1205 			dump_ok = test_bit(MMF_DUMP_DAX_SHARED, &mm_flags);
1206 			kdcore("%08lx: %08lx: %s (DAX shared)", vma->vm_start,
1207 			       vma->vm_flags, dump_ok ? "yes" : "no");
1208 		} else {
1209 			dump_ok = test_bit(MMF_DUMP_DAX_PRIVATE, &mm_flags);
1210 			kdcore("%08lx: %08lx: %s (DAX private)", vma->vm_start,
1211 			       vma->vm_flags, dump_ok ? "yes" : "no");
1212 		}
1213 		return dump_ok;
1214 	}
1215 
1216 	/* By default, dump shared memory if mapped from an anonymous file. */
1217 	if (vma->vm_flags & VM_SHARED) {
1218 		if (file_inode(vma->vm_file)->i_nlink == 0) {
1219 			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1220 			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1221 			       vma->vm_flags, dump_ok ? "yes" : "no");
1222 			return dump_ok;
1223 		}
1224 
1225 		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1226 		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1227 		       vma->vm_flags, dump_ok ? "yes" : "no");
1228 		return dump_ok;
1229 	}
1230 
1231 #ifdef CONFIG_MMU
1232 	/* By default, if it hasn't been written to, don't write it out */
1233 	if (!vma->anon_vma) {
1234 		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1235 		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1236 		       vma->vm_flags, dump_ok ? "yes" : "no");
1237 		return dump_ok;
1238 	}
1239 #endif
1240 
1241 	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1242 	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1243 	       dump_ok ? "yes" : "no");
1244 	return dump_ok;
1245 }
1246 
1247 /* An ELF note in memory */
1248 struct memelfnote
1249 {
1250 	const char *name;
1251 	int type;
1252 	unsigned int datasz;
1253 	void *data;
1254 };
1255 
1256 static int notesize(struct memelfnote *en)
1257 {
1258 	int sz;
1259 
1260 	sz = sizeof(struct elf_note);
1261 	sz += roundup(strlen(en->name) + 1, 4);
1262 	sz += roundup(en->datasz, 4);
1263 
1264 	return sz;
1265 }
1266 
1267 /* #define DEBUG */
1268 
1269 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1270 {
1271 	struct elf_note en;
1272 	en.n_namesz = strlen(men->name) + 1;
1273 	en.n_descsz = men->datasz;
1274 	en.n_type = men->type;
1275 
1276 	return dump_emit(cprm, &en, sizeof(en)) &&
1277 		dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1278 		dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1279 }
1280 
1281 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1282 {
1283 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1284 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1285 	elf->e_ident[EI_DATA] = ELF_DATA;
1286 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1287 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1288 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1289 
1290 	elf->e_type = ET_CORE;
1291 	elf->e_machine = ELF_ARCH;
1292 	elf->e_version = EV_CURRENT;
1293 	elf->e_entry = 0;
1294 	elf->e_phoff = sizeof(struct elfhdr);
1295 	elf->e_shoff = 0;
1296 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1297 	elf->e_ehsize = sizeof(struct elfhdr);
1298 	elf->e_phentsize = sizeof(struct elf_phdr);
1299 	elf->e_phnum = segs;
1300 	elf->e_shentsize = 0;
1301 	elf->e_shnum = 0;
1302 	elf->e_shstrndx = 0;
1303 	return;
1304 }
1305 
1306 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1307 {
1308 	phdr->p_type = PT_NOTE;
1309 	phdr->p_offset = offset;
1310 	phdr->p_vaddr = 0;
1311 	phdr->p_paddr = 0;
1312 	phdr->p_filesz = sz;
1313 	phdr->p_memsz = 0;
1314 	phdr->p_flags = 0;
1315 	phdr->p_align = 0;
1316 	return;
1317 }
1318 
1319 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1320 		unsigned int sz, void *data)
1321 {
1322 	note->name = name;
1323 	note->type = type;
1324 	note->datasz = sz;
1325 	note->data = data;
1326 	return;
1327 }
1328 
1329 /*
1330  * fill up all the fields in prstatus from the given task struct, except
1331  * registers which need to be filled up separately.
1332  */
1333 static void fill_prstatus(struct elf_prstatus *prstatus,
1334 			  struct task_struct *p, long signr)
1335 {
1336 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1337 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1338 	prstatus->pr_sighold = p->blocked.sig[0];
1339 	rcu_read_lock();
1340 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1341 	rcu_read_unlock();
1342 	prstatus->pr_pid = task_pid_vnr(p);
1343 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1344 	prstatus->pr_sid = task_session_vnr(p);
1345 	if (thread_group_leader(p)) {
1346 		struct task_cputime cputime;
1347 
1348 		/*
1349 		 * This is the record for the group leader.  It shows the
1350 		 * group-wide total, not its individual thread total.
1351 		 */
1352 		thread_group_cputime(p, &cputime);
1353 		prstatus->pr_utime = ns_to_timeval(cputime.utime);
1354 		prstatus->pr_stime = ns_to_timeval(cputime.stime);
1355 	} else {
1356 		u64 utime, stime;
1357 
1358 		task_cputime(p, &utime, &stime);
1359 		prstatus->pr_utime = ns_to_timeval(utime);
1360 		prstatus->pr_stime = ns_to_timeval(stime);
1361 	}
1362 	prstatus->pr_cutime = ns_to_timeval(p->signal->cutime);
1363 	prstatus->pr_cstime = ns_to_timeval(p->signal->cstime);
1364 
1365 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1366 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1367 }
1368 
1369 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1370 		       struct mm_struct *mm)
1371 {
1372 	const struct cred *cred;
1373 	unsigned int i, len;
1374 
1375 	/* first copy the parameters from user space */
1376 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1377 
1378 	len = mm->arg_end - mm->arg_start;
1379 	if (len >= ELF_PRARGSZ)
1380 		len = ELF_PRARGSZ - 1;
1381 	if (copy_from_user(&psinfo->pr_psargs,
1382 		           (const char __user *) mm->arg_start, len))
1383 		return -EFAULT;
1384 	for (i = 0; i < len; i++)
1385 		if (psinfo->pr_psargs[i] == 0)
1386 			psinfo->pr_psargs[i] = ' ';
1387 	psinfo->pr_psargs[len] = 0;
1388 
1389 	rcu_read_lock();
1390 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1391 	rcu_read_unlock();
1392 	psinfo->pr_pid = task_pid_vnr(p);
1393 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1394 	psinfo->pr_sid = task_session_vnr(p);
1395 
1396 	i = p->state ? ffz(~p->state) + 1 : 0;
1397 	psinfo->pr_state = i;
1398 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1399 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1400 	psinfo->pr_nice = task_nice(p);
1401 	psinfo->pr_flag = p->flags;
1402 	rcu_read_lock();
1403 	cred = __task_cred(p);
1404 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1405 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1406 	rcu_read_unlock();
1407 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1408 
1409 	return 0;
1410 }
1411 
1412 /* Here is the structure in which status of each thread is captured. */
1413 struct elf_thread_status
1414 {
1415 	struct list_head list;
1416 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1417 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1418 	struct task_struct *thread;
1419 #ifdef ELF_CORE_COPY_XFPREGS
1420 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1421 #endif
1422 	struct memelfnote notes[3];
1423 	int num_notes;
1424 };
1425 
1426 /*
1427  * In order to add the specific thread information for the elf file format,
1428  * we need to keep a linked list of every thread's pr_status and then create
1429  * a single section for them in the final core file.
1430  */
1431 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1432 {
1433 	struct task_struct *p = t->thread;
1434 	int sz = 0;
1435 
1436 	t->num_notes = 0;
1437 
1438 	fill_prstatus(&t->prstatus, p, signr);
1439 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1440 
1441 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1442 		  &t->prstatus);
1443 	t->num_notes++;
1444 	sz += notesize(&t->notes[0]);
1445 
1446 	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1447 	if (t->prstatus.pr_fpvalid) {
1448 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1449 			  &t->fpu);
1450 		t->num_notes++;
1451 		sz += notesize(&t->notes[1]);
1452 	}
1453 
1454 #ifdef ELF_CORE_COPY_XFPREGS
1455 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1456 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1457 			  sizeof(t->xfpu), &t->xfpu);
1458 		t->num_notes++;
1459 		sz += notesize(&t->notes[2]);
1460 	}
1461 #endif
1462 	return sz;
1463 }
1464 
1465 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1466 			     elf_addr_t e_shoff, int segs)
1467 {
1468 	elf->e_shoff = e_shoff;
1469 	elf->e_shentsize = sizeof(*shdr4extnum);
1470 	elf->e_shnum = 1;
1471 	elf->e_shstrndx = SHN_UNDEF;
1472 
1473 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1474 
1475 	shdr4extnum->sh_type = SHT_NULL;
1476 	shdr4extnum->sh_size = elf->e_shnum;
1477 	shdr4extnum->sh_link = elf->e_shstrndx;
1478 	shdr4extnum->sh_info = segs;
1479 }
1480 
1481 /*
1482  * dump the segments for an MMU process
1483  */
1484 static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1485 {
1486 	struct vm_area_struct *vma;
1487 
1488 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1489 		unsigned long addr;
1490 
1491 		if (!maydump(vma, cprm->mm_flags))
1492 			continue;
1493 
1494 #ifdef CONFIG_MMU
1495 		for (addr = vma->vm_start; addr < vma->vm_end;
1496 							addr += PAGE_SIZE) {
1497 			bool res;
1498 			struct page *page = get_dump_page(addr);
1499 			if (page) {
1500 				void *kaddr = kmap(page);
1501 				res = dump_emit(cprm, kaddr, PAGE_SIZE);
1502 				kunmap(page);
1503 				put_page(page);
1504 			} else {
1505 				res = dump_skip(cprm, PAGE_SIZE);
1506 			}
1507 			if (!res)
1508 				return false;
1509 		}
1510 #else
1511 		if (!dump_emit(cprm, (void *) vma->vm_start,
1512 				vma->vm_end - vma->vm_start))
1513 			return false;
1514 #endif
1515 	}
1516 	return true;
1517 }
1518 
1519 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1520 {
1521 	struct vm_area_struct *vma;
1522 	size_t size = 0;
1523 
1524 	for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1525 		if (maydump(vma, mm_flags))
1526 			size += vma->vm_end - vma->vm_start;
1527 	return size;
1528 }
1529 
1530 /*
1531  * Actual dumper
1532  *
1533  * This is a two-pass process; first we find the offsets of the bits,
1534  * and then they are actually written out.  If we run out of core limit
1535  * we just truncate.
1536  */
1537 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1538 {
1539 #define	NUM_NOTES	6
1540 	int has_dumped = 0;
1541 	mm_segment_t fs;
1542 	int segs;
1543 	int i;
1544 	struct vm_area_struct *vma;
1545 	struct elfhdr *elf = NULL;
1546 	loff_t offset = 0, dataoff;
1547 	int numnote;
1548 	struct memelfnote *notes = NULL;
1549 	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1550 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1551  	LIST_HEAD(thread_list);
1552  	struct list_head *t;
1553 	elf_fpregset_t *fpu = NULL;
1554 #ifdef ELF_CORE_COPY_XFPREGS
1555 	elf_fpxregset_t *xfpu = NULL;
1556 #endif
1557 	int thread_status_size = 0;
1558 	elf_addr_t *auxv;
1559 	struct elf_phdr *phdr4note = NULL;
1560 	struct elf_shdr *shdr4extnum = NULL;
1561 	Elf_Half e_phnum;
1562 	elf_addr_t e_shoff;
1563 	struct core_thread *ct;
1564 	struct elf_thread_status *tmp;
1565 
1566 	/*
1567 	 * We no longer stop all VM operations.
1568 	 *
1569 	 * This is because those proceses that could possibly change map_count
1570 	 * or the mmap / vma pages are now blocked in do_exit on current
1571 	 * finishing this core dump.
1572 	 *
1573 	 * Only ptrace can touch these memory addresses, but it doesn't change
1574 	 * the map_count or the pages allocated. So no possibility of crashing
1575 	 * exists while dumping the mm->vm_next areas to the core file.
1576 	 */
1577 
1578 	/* alloc memory for large data structures: too large to be on stack */
1579 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1580 	if (!elf)
1581 		goto cleanup;
1582 	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1583 	if (!prstatus)
1584 		goto cleanup;
1585 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1586 	if (!psinfo)
1587 		goto cleanup;
1588 	notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1589 	if (!notes)
1590 		goto cleanup;
1591 	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1592 	if (!fpu)
1593 		goto cleanup;
1594 #ifdef ELF_CORE_COPY_XFPREGS
1595 	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1596 	if (!xfpu)
1597 		goto cleanup;
1598 #endif
1599 
1600 	for (ct = current->mm->core_state->dumper.next;
1601 					ct; ct = ct->next) {
1602 		tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1603 		if (!tmp)
1604 			goto cleanup;
1605 
1606 		tmp->thread = ct->task;
1607 		list_add(&tmp->list, &thread_list);
1608 	}
1609 
1610 	list_for_each(t, &thread_list) {
1611 		struct elf_thread_status *tmp;
1612 		int sz;
1613 
1614 		tmp = list_entry(t, struct elf_thread_status, list);
1615 		sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1616 		thread_status_size += sz;
1617 	}
1618 
1619 	/* now collect the dump for the current */
1620 	fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1621 	elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1622 
1623 	segs = current->mm->map_count;
1624 	segs += elf_core_extra_phdrs();
1625 
1626 	/* for notes section */
1627 	segs++;
1628 
1629 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1630 	 * this, kernel supports extended numbering. Have a look at
1631 	 * include/linux/elf.h for further information. */
1632 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1633 
1634 	/* Set up header */
1635 	fill_elf_fdpic_header(elf, e_phnum);
1636 
1637 	has_dumped = 1;
1638 	/*
1639 	 * Set up the notes in similar form to SVR4 core dumps made
1640 	 * with info from their /proc.
1641 	 */
1642 
1643 	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1644 	fill_psinfo(psinfo, current->group_leader, current->mm);
1645 	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1646 
1647 	numnote = 2;
1648 
1649 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1650 
1651 	i = 0;
1652 	do
1653 		i += 2;
1654 	while (auxv[i - 2] != AT_NULL);
1655 	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1656 		  i * sizeof(elf_addr_t), auxv);
1657 
1658   	/* Try to dump the FPU. */
1659 	if ((prstatus->pr_fpvalid =
1660 	     elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1661 		fill_note(notes + numnote++,
1662 			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1663 #ifdef ELF_CORE_COPY_XFPREGS
1664 	if (elf_core_copy_task_xfpregs(current, xfpu))
1665 		fill_note(notes + numnote++,
1666 			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1667 #endif
1668 
1669 	fs = get_fs();
1670 	set_fs(KERNEL_DS);
1671 
1672 	offset += sizeof(*elf);				/* Elf header */
1673 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1674 
1675 	/* Write notes phdr entry */
1676 	{
1677 		int sz = 0;
1678 
1679 		for (i = 0; i < numnote; i++)
1680 			sz += notesize(notes + i);
1681 
1682 		sz += thread_status_size;
1683 
1684 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1685 		if (!phdr4note)
1686 			goto end_coredump;
1687 
1688 		fill_elf_note_phdr(phdr4note, sz, offset);
1689 		offset += sz;
1690 	}
1691 
1692 	/* Page-align dumped data */
1693 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1694 
1695 	offset += elf_core_vma_data_size(cprm->mm_flags);
1696 	offset += elf_core_extra_data_size();
1697 	e_shoff = offset;
1698 
1699 	if (e_phnum == PN_XNUM) {
1700 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1701 		if (!shdr4extnum)
1702 			goto end_coredump;
1703 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1704 	}
1705 
1706 	offset = dataoff;
1707 
1708 	if (!dump_emit(cprm, elf, sizeof(*elf)))
1709 		goto end_coredump;
1710 
1711 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1712 		goto end_coredump;
1713 
1714 	/* write program headers for segments dump */
1715 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1716 		struct elf_phdr phdr;
1717 		size_t sz;
1718 
1719 		sz = vma->vm_end - vma->vm_start;
1720 
1721 		phdr.p_type = PT_LOAD;
1722 		phdr.p_offset = offset;
1723 		phdr.p_vaddr = vma->vm_start;
1724 		phdr.p_paddr = 0;
1725 		phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1726 		phdr.p_memsz = sz;
1727 		offset += phdr.p_filesz;
1728 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1729 		if (vma->vm_flags & VM_WRITE)
1730 			phdr.p_flags |= PF_W;
1731 		if (vma->vm_flags & VM_EXEC)
1732 			phdr.p_flags |= PF_X;
1733 		phdr.p_align = ELF_EXEC_PAGESIZE;
1734 
1735 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1736 			goto end_coredump;
1737 	}
1738 
1739 	if (!elf_core_write_extra_phdrs(cprm, offset))
1740 		goto end_coredump;
1741 
1742  	/* write out the notes section */
1743 	for (i = 0; i < numnote; i++)
1744 		if (!writenote(notes + i, cprm))
1745 			goto end_coredump;
1746 
1747 	/* write out the thread status notes section */
1748 	list_for_each(t, &thread_list) {
1749 		struct elf_thread_status *tmp =
1750 				list_entry(t, struct elf_thread_status, list);
1751 
1752 		for (i = 0; i < tmp->num_notes; i++)
1753 			if (!writenote(&tmp->notes[i], cprm))
1754 				goto end_coredump;
1755 	}
1756 
1757 	if (!dump_skip(cprm, dataoff - cprm->pos))
1758 		goto end_coredump;
1759 
1760 	if (!elf_fdpic_dump_segments(cprm))
1761 		goto end_coredump;
1762 
1763 	if (!elf_core_write_extra_data(cprm))
1764 		goto end_coredump;
1765 
1766 	if (e_phnum == PN_XNUM) {
1767 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1768 			goto end_coredump;
1769 	}
1770 
1771 	if (cprm->file->f_pos != offset) {
1772 		/* Sanity check */
1773 		printk(KERN_WARNING
1774 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1775 		       cprm->file->f_pos, offset);
1776 	}
1777 
1778 end_coredump:
1779 	set_fs(fs);
1780 
1781 cleanup:
1782 	while (!list_empty(&thread_list)) {
1783 		struct list_head *tmp = thread_list.next;
1784 		list_del(tmp);
1785 		kfree(list_entry(tmp, struct elf_thread_status, list));
1786 	}
1787 	kfree(phdr4note);
1788 	kfree(elf);
1789 	kfree(prstatus);
1790 	kfree(psinfo);
1791 	kfree(notes);
1792 	kfree(fpu);
1793 	kfree(shdr4extnum);
1794 #ifdef ELF_CORE_COPY_XFPREGS
1795 	kfree(xfpu);
1796 #endif
1797 	return has_dumped;
1798 #undef NUM_NOTES
1799 }
1800 
1801 #endif		/* CONFIG_ELF_CORE */
1802