xref: /linux/fs/binfmt_elf_fdpic.c (revision 4b132aacb0768ac1e652cf517097ea6f237214b9)
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 #include <linux/regset.h>
39 
40 #include <linux/uaccess.h>
41 #include <asm/param.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 	.min_coredump	= ELF_EXEC_PAGESIZE,
87 #endif
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 elf_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 && interp_params.flags & ELF_FDPIC_FLAG_PRESENT) {
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 = begin_new_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 	SET_PERSONALITY(exec_params.hdr);
349 	if (elf_check_fdpic(&exec_params.hdr))
350 		current->personality |= PER_LINUX_FDPIC;
351 	if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
352 		current->personality |= READ_IMPLIES_EXEC;
353 
354 	setup_new_exec(bprm);
355 
356 	set_binfmt(&elf_fdpic_format);
357 
358 	current->mm->start_code = 0;
359 	current->mm->end_code = 0;
360 	current->mm->start_stack = 0;
361 	current->mm->start_data = 0;
362 	current->mm->end_data = 0;
363 	current->mm->context.exec_fdpic_loadmap = 0;
364 	current->mm->context.interp_fdpic_loadmap = 0;
365 
366 #ifdef CONFIG_MMU
367 	elf_fdpic_arch_lay_out_mm(&exec_params,
368 				  &interp_params,
369 				  &current->mm->start_stack,
370 				  &current->mm->start_brk);
371 
372 	retval = setup_arg_pages(bprm, current->mm->start_stack,
373 				 executable_stack);
374 	if (retval < 0)
375 		goto error;
376 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
377 	retval = arch_setup_additional_pages(bprm, !!interpreter_name);
378 	if (retval < 0)
379 		goto error;
380 #endif
381 #endif
382 
383 	/* load the executable and interpreter into memory */
384 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
385 				    "executable");
386 	if (retval < 0)
387 		goto error;
388 
389 	if (interpreter_name) {
390 		retval = elf_fdpic_map_file(&interp_params, interpreter,
391 					    current->mm, "interpreter");
392 		if (retval < 0) {
393 			printk(KERN_ERR "Unable to load interpreter\n");
394 			goto error;
395 		}
396 
397 		fput(interpreter);
398 		interpreter = NULL;
399 	}
400 
401 #ifdef CONFIG_MMU
402 	if (!current->mm->start_brk)
403 		current->mm->start_brk = current->mm->end_data;
404 
405 	current->mm->brk = current->mm->start_brk =
406 		PAGE_ALIGN(current->mm->start_brk);
407 
408 #else
409 	/* create a stack area and zero-size brk area */
410 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
411 	if (stack_size < PAGE_SIZE * 2)
412 		stack_size = PAGE_SIZE * 2;
413 
414 	stack_prot = PROT_READ | PROT_WRITE;
415 	if (executable_stack == EXSTACK_ENABLE_X ||
416 	    (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
417 		stack_prot |= PROT_EXEC;
418 
419 	current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
420 					 MAP_PRIVATE | MAP_ANONYMOUS |
421 					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
422 					 0);
423 
424 	if (IS_ERR_VALUE(current->mm->start_brk)) {
425 		retval = current->mm->start_brk;
426 		current->mm->start_brk = 0;
427 		goto error;
428 	}
429 
430 	current->mm->brk = current->mm->start_brk;
431 	current->mm->context.end_brk = current->mm->start_brk;
432 	current->mm->start_stack = current->mm->start_brk + stack_size;
433 #endif
434 
435 	retval = create_elf_fdpic_tables(bprm, current->mm, &exec_params,
436 					 &interp_params);
437 	if (retval < 0)
438 		goto error;
439 
440 	kdebug("- start_code  %lx", current->mm->start_code);
441 	kdebug("- end_code    %lx", current->mm->end_code);
442 	kdebug("- start_data  %lx", current->mm->start_data);
443 	kdebug("- end_data    %lx", current->mm->end_data);
444 	kdebug("- start_brk   %lx", current->mm->start_brk);
445 	kdebug("- brk         %lx", current->mm->brk);
446 	kdebug("- start_stack %lx", current->mm->start_stack);
447 
448 #ifdef ELF_FDPIC_PLAT_INIT
449 	/*
450 	 * The ABI may specify that certain registers be set up in special
451 	 * ways (on i386 %edx is the address of a DT_FINI function, for
452 	 * example.  This macro performs whatever initialization to
453 	 * the regs structure is required.
454 	 */
455 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
456 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
457 			    dynaddr);
458 #endif
459 
460 	finalize_exec(bprm);
461 	/* everything is now ready... get the userspace context ready to roll */
462 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
463 	start_thread(regs, entryaddr, current->mm->start_stack);
464 
465 	retval = 0;
466 
467 error:
468 	if (interpreter)
469 		fput(interpreter);
470 	kfree(interpreter_name);
471 	kfree(exec_params.phdrs);
472 	kfree(exec_params.loadmap);
473 	kfree(interp_params.phdrs);
474 	kfree(interp_params.loadmap);
475 	return retval;
476 }
477 
478 /*****************************************************************************/
479 
480 #ifndef ELF_BASE_PLATFORM
481 /*
482  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
483  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
484  * will be copied to the user stack in the same manner as AT_PLATFORM.
485  */
486 #define ELF_BASE_PLATFORM NULL
487 #endif
488 
489 /*
490  * present useful information to the program by shovelling it onto the new
491  * process's stack
492  */
493 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
494 				   struct mm_struct *mm,
495 				   struct elf_fdpic_params *exec_params,
496 				   struct elf_fdpic_params *interp_params)
497 {
498 	const struct cred *cred = current_cred();
499 	unsigned long sp, csp, nitems;
500 	elf_caddr_t __user *argv, *envp;
501 	size_t platform_len = 0, len;
502 	char *k_platform, *k_base_platform;
503 	char __user *u_platform, *u_base_platform, *p;
504 	int loop;
505 	unsigned long flags = 0;
506 	int ei_index;
507 	elf_addr_t *elf_info;
508 
509 #ifdef CONFIG_MMU
510 	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
511 	 * by the processes running on the same package. One thing we can do is
512 	 * to shuffle the initial stack for them, so we give the architecture
513 	 * an opportunity to do so here.
514 	 */
515 	sp = arch_align_stack(bprm->p);
516 #else
517 	sp = mm->start_stack;
518 
519 	/* stack the program arguments and environment */
520 	if (transfer_args_to_stack(bprm, &sp) < 0)
521 		return -EFAULT;
522 	sp &= ~15;
523 #endif
524 
525 	/*
526 	 * If this architecture has a platform capability string, copy it
527 	 * to userspace.  In some cases (Sparc), this info is impossible
528 	 * for userspace to get any other way, in others (i386) it is
529 	 * merely difficult.
530 	 */
531 	k_platform = ELF_PLATFORM;
532 	u_platform = NULL;
533 
534 	if (k_platform) {
535 		platform_len = strlen(k_platform) + 1;
536 		sp -= platform_len;
537 		u_platform = (char __user *) sp;
538 		if (copy_to_user(u_platform, k_platform, platform_len) != 0)
539 			return -EFAULT;
540 	}
541 
542 	/*
543 	 * If this architecture has a "base" platform capability
544 	 * string, copy it to userspace.
545 	 */
546 	k_base_platform = ELF_BASE_PLATFORM;
547 	u_base_platform = NULL;
548 
549 	if (k_base_platform) {
550 		platform_len = strlen(k_base_platform) + 1;
551 		sp -= platform_len;
552 		u_base_platform = (char __user *) sp;
553 		if (copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
554 			return -EFAULT;
555 	}
556 
557 	sp &= ~7UL;
558 
559 	/* stack the load map(s) */
560 	len = sizeof(struct elf_fdpic_loadmap);
561 	len += sizeof(struct elf_fdpic_loadseg) * exec_params->loadmap->nsegs;
562 	sp = (sp - len) & ~7UL;
563 	exec_params->map_addr = sp;
564 
565 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
566 		return -EFAULT;
567 
568 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
569 
570 	if (interp_params->loadmap) {
571 		len = sizeof(struct elf_fdpic_loadmap);
572 		len += sizeof(struct elf_fdpic_loadseg) *
573 			interp_params->loadmap->nsegs;
574 		sp = (sp - len) & ~7UL;
575 		interp_params->map_addr = sp;
576 
577 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
578 				 len) != 0)
579 			return -EFAULT;
580 
581 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
582 	}
583 
584 	/* force 16 byte _final_ alignment here for generality */
585 #define DLINFO_ITEMS 15
586 
587 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
588 		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
589 
590 	if (bprm->have_execfd)
591 		nitems++;
592 #ifdef ELF_HWCAP2
593 	nitems++;
594 #endif
595 
596 	csp = sp;
597 	sp -= nitems * 2 * sizeof(unsigned long);
598 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
599 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
600 	sp -= 1 * sizeof(unsigned long);		/* argc */
601 
602 	csp -= sp & 15UL;
603 	sp -= sp & 15UL;
604 
605 	/* Create the ELF interpreter info */
606 	elf_info = (elf_addr_t *)mm->saved_auxv;
607 	/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
608 #define NEW_AUX_ENT(id, val) \
609 	do { \
610 		*elf_info++ = id; \
611 		*elf_info++ = val; \
612 	} while (0)
613 
614 #ifdef ARCH_DLINFO
615 	/*
616 	 * ARCH_DLINFO must come first so PPC can do its special alignment of
617 	 * AUXV.
618 	 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
619 	 * ARCH_DLINFO changes
620 	 */
621 	ARCH_DLINFO;
622 #endif
623 	NEW_AUX_ENT(AT_HWCAP,	ELF_HWCAP);
624 #ifdef ELF_HWCAP2
625 	NEW_AUX_ENT(AT_HWCAP2,	ELF_HWCAP2);
626 #endif
627 	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
628 	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
629 	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
630 	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
631 	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
632 	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
633 	if (bprm->interp_flags & BINPRM_FLAGS_PRESERVE_ARGV0)
634 		flags |= AT_FLAGS_PRESERVE_ARGV0;
635 	NEW_AUX_ENT(AT_FLAGS,	flags);
636 	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
637 	NEW_AUX_ENT(AT_UID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
638 	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
639 	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
640 	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
641 	NEW_AUX_ENT(AT_SECURE,	bprm->secureexec);
642 	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
643 	if (k_platform)
644 		NEW_AUX_ENT(AT_PLATFORM,
645 			    (elf_addr_t)(unsigned long)u_platform);
646 	if (k_base_platform)
647 		NEW_AUX_ENT(AT_BASE_PLATFORM,
648 			    (elf_addr_t)(unsigned long)u_base_platform);
649 	if (bprm->have_execfd)
650 		NEW_AUX_ENT(AT_EXECFD, bprm->execfd);
651 #undef NEW_AUX_ENT
652 	/* AT_NULL is zero; clear the rest too */
653 	memset(elf_info, 0, (char *)mm->saved_auxv +
654 	       sizeof(mm->saved_auxv) - (char *)elf_info);
655 
656 	/* And advance past the AT_NULL entry.  */
657 	elf_info += 2;
658 
659 	ei_index = elf_info - (elf_addr_t *)mm->saved_auxv;
660 	csp -= ei_index * sizeof(elf_addr_t);
661 
662 	/* Put the elf_info on the stack in the right place.  */
663 	if (copy_to_user((void __user *)csp, mm->saved_auxv,
664 			 ei_index * sizeof(elf_addr_t)))
665 		return -EFAULT;
666 
667 	/* allocate room for argv[] and envv[] */
668 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
669 	envp = (elf_caddr_t __user *) csp;
670 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
671 	argv = (elf_caddr_t __user *) csp;
672 
673 	/* stack argc */
674 	csp -= sizeof(unsigned long);
675 	if (put_user(bprm->argc, (unsigned long __user *) csp))
676 		return -EFAULT;
677 
678 	BUG_ON(csp != sp);
679 
680 	/* fill in the argv[] array */
681 #ifdef CONFIG_MMU
682 	current->mm->arg_start = bprm->p;
683 #else
684 	current->mm->arg_start = current->mm->start_stack -
685 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
686 #endif
687 
688 	p = (char __user *) current->mm->arg_start;
689 	for (loop = bprm->argc; loop > 0; loop--) {
690 		if (put_user((elf_caddr_t) p, argv++))
691 			return -EFAULT;
692 		len = strnlen_user(p, MAX_ARG_STRLEN);
693 		if (!len || len > MAX_ARG_STRLEN)
694 			return -EINVAL;
695 		p += len;
696 	}
697 	if (put_user(NULL, argv))
698 		return -EFAULT;
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 		if (put_user((elf_caddr_t)(unsigned long) p, envp++))
705 			return -EFAULT;
706 		len = strnlen_user(p, MAX_ARG_STRLEN);
707 		if (!len || len > MAX_ARG_STRLEN)
708 			return -EINVAL;
709 		p += len;
710 	}
711 	if (put_user(NULL, envp))
712 		return -EFAULT;
713 	current->mm->env_end = (unsigned long) p;
714 
715 	mm->start_stack = (unsigned long) sp;
716 	return 0;
717 }
718 
719 /*****************************************************************************/
720 /*
721  * load the appropriate binary image (executable or interpreter) into memory
722  * - we assume no MMU is available
723  * - if no other PIC bits are set in params->hdr->e_flags
724  *   - we assume that the LOADable segments in the binary are independently relocatable
725  *   - we assume R/O executable segments are shareable
726  * - else
727  *   - we assume the loadable parts of the image to require fixed displacement
728  *   - the image is not shareable
729  */
730 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
731 			      struct file *file,
732 			      struct mm_struct *mm,
733 			      const char *what)
734 {
735 	struct elf_fdpic_loadmap *loadmap;
736 #ifdef CONFIG_MMU
737 	struct elf_fdpic_loadseg *mseg;
738 	unsigned long load_addr;
739 #endif
740 	struct elf_fdpic_loadseg *seg;
741 	struct elf_phdr *phdr;
742 	unsigned nloads, tmp;
743 	unsigned long stop;
744 	int loop, ret;
745 
746 	/* allocate a load map table */
747 	nloads = 0;
748 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
749 		if (params->phdrs[loop].p_type == PT_LOAD)
750 			nloads++;
751 
752 	if (nloads == 0)
753 		return -ELIBBAD;
754 
755 	loadmap = kzalloc(struct_size(loadmap, segs, nloads), GFP_KERNEL);
756 	if (!loadmap)
757 		return -ENOMEM;
758 
759 	params->loadmap = loadmap;
760 
761 	loadmap->version = ELF_FDPIC_LOADMAP_VERSION;
762 	loadmap->nsegs = nloads;
763 
764 	/* map the requested LOADs into the memory space */
765 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
766 	case ELF_FDPIC_FLAG_CONSTDISP:
767 	case ELF_FDPIC_FLAG_CONTIGUOUS:
768 #ifndef CONFIG_MMU
769 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
770 		if (ret < 0)
771 			return ret;
772 		break;
773 #endif
774 	default:
775 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
776 		if (ret < 0)
777 			return ret;
778 		break;
779 	}
780 
781 	/* map the entry point */
782 	if (params->hdr.e_entry) {
783 		seg = loadmap->segs;
784 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
785 			if (params->hdr.e_entry >= seg->p_vaddr &&
786 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
787 				params->entry_addr =
788 					(params->hdr.e_entry - seg->p_vaddr) +
789 					seg->addr;
790 				break;
791 			}
792 		}
793 	}
794 
795 	/* determine where the program header table has wound up if mapped */
796 	stop = params->hdr.e_phoff;
797 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
798 	phdr = params->phdrs;
799 
800 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
801 		if (phdr->p_type != PT_LOAD)
802 			continue;
803 
804 		if (phdr->p_offset > params->hdr.e_phoff ||
805 		    phdr->p_offset + phdr->p_filesz < stop)
806 			continue;
807 
808 		seg = loadmap->segs;
809 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
810 			if (phdr->p_vaddr >= seg->p_vaddr &&
811 			    phdr->p_vaddr + phdr->p_filesz <=
812 			    seg->p_vaddr + seg->p_memsz) {
813 				params->ph_addr =
814 					(phdr->p_vaddr - seg->p_vaddr) +
815 					seg->addr +
816 					params->hdr.e_phoff - phdr->p_offset;
817 				break;
818 			}
819 		}
820 		break;
821 	}
822 
823 	/* determine where the dynamic section has wound up if there is one */
824 	phdr = params->phdrs;
825 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
826 		if (phdr->p_type != PT_DYNAMIC)
827 			continue;
828 
829 		seg = loadmap->segs;
830 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
831 			if (phdr->p_vaddr >= seg->p_vaddr &&
832 			    phdr->p_vaddr + phdr->p_memsz <=
833 			    seg->p_vaddr + seg->p_memsz) {
834 				Elf_Dyn __user *dyn;
835 				Elf_Sword d_tag;
836 
837 				params->dynamic_addr =
838 					(phdr->p_vaddr - seg->p_vaddr) +
839 					seg->addr;
840 
841 				/* check the dynamic section contains at least
842 				 * one item, and that the last item is a NULL
843 				 * entry */
844 				if (phdr->p_memsz == 0 ||
845 				    phdr->p_memsz % sizeof(Elf_Dyn) != 0)
846 					goto dynamic_error;
847 
848 				tmp = phdr->p_memsz / sizeof(Elf_Dyn);
849 				dyn = (Elf_Dyn __user *)params->dynamic_addr;
850 				if (get_user(d_tag, &dyn[tmp - 1].d_tag) ||
851 				    d_tag != 0)
852 					goto dynamic_error;
853 				break;
854 			}
855 		}
856 		break;
857 	}
858 
859 	/* now elide adjacent segments in the load map on MMU linux
860 	 * - on uClinux the holes between may actually be filled with system
861 	 *   stuff or stuff from other processes
862 	 */
863 #ifdef CONFIG_MMU
864 	nloads = loadmap->nsegs;
865 	mseg = loadmap->segs;
866 	seg = mseg + 1;
867 	for (loop = 1; loop < nloads; loop++) {
868 		/* see if we have a candidate for merging */
869 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
870 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
871 			if (load_addr == (seg->addr & PAGE_MASK)) {
872 				mseg->p_memsz +=
873 					load_addr -
874 					(mseg->addr + mseg->p_memsz);
875 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
876 				mseg->p_memsz += seg->p_memsz;
877 				loadmap->nsegs--;
878 				continue;
879 			}
880 		}
881 
882 		mseg++;
883 		if (mseg != seg)
884 			*mseg = *seg;
885 	}
886 #endif
887 
888 	kdebug("Mapped Object [%s]:", what);
889 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
890 	kdebug("- entry    : %lx", params->entry_addr);
891 	kdebug("- PHDR[]   : %lx", params->ph_addr);
892 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
893 	seg = loadmap->segs;
894 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
895 		kdebug("- LOAD[%d] : %08llx-%08llx [va=%llx ms=%llx]",
896 		       loop,
897 		       (unsigned long long) seg->addr,
898 		       (unsigned long long) seg->addr + seg->p_memsz - 1,
899 		       (unsigned long long) seg->p_vaddr,
900 		       (unsigned long long) 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 elf_fdpic_loadseg *seg;
921 	struct elf_phdr *phdr;
922 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0;
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 	maddr = vm_mmap(NULL, load_addr, top - base,
943 			PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE, 0);
944 	if (IS_ERR_VALUE(maddr))
945 		return (int) maddr;
946 
947 	if (load_addr != 0)
948 		load_addr += PAGE_ALIGN(top - base);
949 
950 	/* and then load the file segments into it */
951 	phdr = params->phdrs;
952 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
953 		if (params->phdrs[loop].p_type != PT_LOAD)
954 			continue;
955 
956 		seg->addr = maddr + (phdr->p_vaddr - base);
957 		seg->p_vaddr = phdr->p_vaddr;
958 		seg->p_memsz = phdr->p_memsz;
959 
960 		ret = read_code(file, seg->addr, phdr->p_offset,
961 				       phdr->p_filesz);
962 		if (ret < 0)
963 			return ret;
964 
965 		/* map the ELF header address if in this segment */
966 		if (phdr->p_offset == 0)
967 			params->elfhdr_addr = seg->addr;
968 
969 		/* clear any space allocated but not loaded */
970 		if (phdr->p_filesz < phdr->p_memsz) {
971 			if (clear_user((void *) (seg->addr + phdr->p_filesz),
972 				       phdr->p_memsz - phdr->p_filesz))
973 				return -EFAULT;
974 		}
975 
976 		if (mm) {
977 			if (phdr->p_flags & PF_X) {
978 				if (!mm->start_code) {
979 					mm->start_code = seg->addr;
980 					mm->end_code = seg->addr +
981 						phdr->p_memsz;
982 				}
983 			} else if (!mm->start_data) {
984 				mm->start_data = seg->addr;
985 				mm->end_data = seg->addr + phdr->p_memsz;
986 			}
987 		}
988 
989 		seg++;
990 	}
991 
992 	return 0;
993 }
994 #endif
995 
996 /*****************************************************************************/
997 /*
998  * map a binary by direct mmap() of the individual PT_LOAD segments
999  */
1000 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1001 					     struct file *file,
1002 					     struct mm_struct *mm)
1003 {
1004 	struct elf_fdpic_loadseg *seg;
1005 	struct elf_phdr *phdr;
1006 	unsigned long load_addr, delta_vaddr;
1007 	int loop, dvset;
1008 
1009 	load_addr = params->load_addr;
1010 	delta_vaddr = 0;
1011 	dvset = 0;
1012 
1013 	seg = params->loadmap->segs;
1014 
1015 	/* deal with each load segment separately */
1016 	phdr = params->phdrs;
1017 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1018 		unsigned long maddr, disp, excess, excess1;
1019 		int prot = 0, flags;
1020 
1021 		if (phdr->p_type != PT_LOAD)
1022 			continue;
1023 
1024 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1025 		       (unsigned long) phdr->p_vaddr,
1026 		       (unsigned long) phdr->p_offset,
1027 		       (unsigned long) phdr->p_filesz,
1028 		       (unsigned long) phdr->p_memsz);
1029 
1030 		/* determine the mapping parameters */
1031 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1032 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1033 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1034 
1035 		flags = MAP_PRIVATE;
1036 		maddr = 0;
1037 
1038 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1039 		case ELF_FDPIC_FLAG_INDEPENDENT:
1040 			/* PT_LOADs are independently locatable */
1041 			break;
1042 
1043 		case ELF_FDPIC_FLAG_HONOURVADDR:
1044 			/* the specified virtual address must be honoured */
1045 			maddr = phdr->p_vaddr;
1046 			flags |= MAP_FIXED;
1047 			break;
1048 
1049 		case ELF_FDPIC_FLAG_CONSTDISP:
1050 			/* constant displacement
1051 			 * - can be mapped anywhere, but must be mapped as a
1052 			 *   unit
1053 			 */
1054 			if (!dvset) {
1055 				maddr = load_addr;
1056 				delta_vaddr = phdr->p_vaddr;
1057 				dvset = 1;
1058 			} else {
1059 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1060 				flags |= MAP_FIXED;
1061 			}
1062 			break;
1063 
1064 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1065 			/* contiguity handled later */
1066 			break;
1067 
1068 		default:
1069 			BUG();
1070 		}
1071 
1072 		maddr &= PAGE_MASK;
1073 
1074 		/* create the mapping */
1075 		disp = phdr->p_vaddr & ~PAGE_MASK;
1076 		maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1077 				phdr->p_offset - disp);
1078 
1079 		kdebug("mmap[%d] <file> sz=%llx pr=%x fl=%x of=%llx --> %08lx",
1080 		       loop, (unsigned long long) phdr->p_memsz + disp,
1081 		       prot, flags, (unsigned long long) phdr->p_offset - disp,
1082 		       maddr);
1083 
1084 		if (IS_ERR_VALUE(maddr))
1085 			return (int) maddr;
1086 
1087 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1088 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1089 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1090 
1091 		seg->addr = maddr + disp;
1092 		seg->p_vaddr = phdr->p_vaddr;
1093 		seg->p_memsz = phdr->p_memsz;
1094 
1095 		/* map the ELF header address if in this segment */
1096 		if (phdr->p_offset == 0)
1097 			params->elfhdr_addr = seg->addr;
1098 
1099 		/* clear the bit between beginning of mapping and beginning of
1100 		 * PT_LOAD */
1101 		if (prot & PROT_WRITE && disp > 0) {
1102 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1103 			if (clear_user((void __user *) maddr, disp))
1104 				return -EFAULT;
1105 			maddr += disp;
1106 		}
1107 
1108 		/* clear any space allocated but not loaded
1109 		 * - on uClinux we can just clear the lot
1110 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1111 		 *   extant in the file
1112 		 */
1113 		excess = phdr->p_memsz - phdr->p_filesz;
1114 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1115 
1116 #ifdef CONFIG_MMU
1117 		if (excess > excess1) {
1118 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1119 			unsigned long xmaddr;
1120 
1121 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1122 			xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1123 					 prot, flags, 0);
1124 
1125 			kdebug("mmap[%d] <anon>"
1126 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1127 			       loop, xaddr, excess - excess1, prot, flags,
1128 			       xmaddr);
1129 
1130 			if (xmaddr != xaddr)
1131 				return -ENOMEM;
1132 		}
1133 
1134 		if (prot & PROT_WRITE && excess1 > 0) {
1135 			kdebug("clear[%d] ad=%lx sz=%lx",
1136 			       loop, maddr + phdr->p_filesz, excess1);
1137 			if (clear_user((void __user *) maddr + phdr->p_filesz,
1138 				       excess1))
1139 				return -EFAULT;
1140 		}
1141 
1142 #else
1143 		if (excess > 0) {
1144 			kdebug("clear[%d] ad=%llx sz=%lx", loop,
1145 			       (unsigned long long) maddr + phdr->p_filesz,
1146 			       excess);
1147 			if (clear_user((void *) maddr + phdr->p_filesz, excess))
1148 				return -EFAULT;
1149 		}
1150 #endif
1151 
1152 		if (mm) {
1153 			if (phdr->p_flags & PF_X) {
1154 				if (!mm->start_code) {
1155 					mm->start_code = maddr;
1156 					mm->end_code = maddr + phdr->p_memsz;
1157 				}
1158 			} else if (!mm->start_data) {
1159 				mm->start_data = maddr;
1160 				mm->end_data = maddr + phdr->p_memsz;
1161 			}
1162 		}
1163 
1164 		seg++;
1165 	}
1166 
1167 	return 0;
1168 }
1169 
1170 /*****************************************************************************/
1171 /*
1172  * ELF-FDPIC core dumper
1173  *
1174  * Modelled on fs/exec.c:aout_core_dump()
1175  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1176  *
1177  * Modelled on fs/binfmt_elf.c core dumper
1178  */
1179 #ifdef CONFIG_ELF_CORE
1180 
1181 struct elf_prstatus_fdpic
1182 {
1183 	struct elf_prstatus_common	common;
1184 	elf_gregset_t pr_reg;	/* GP registers */
1185 	/* When using FDPIC, the loadmap addresses need to be communicated
1186 	 * to GDB in order for GDB to do the necessary relocations.  The
1187 	 * fields (below) used to communicate this information are placed
1188 	 * immediately after ``pr_reg'', so that the loadmap addresses may
1189 	 * be viewed as part of the register set if so desired.
1190 	 */
1191 	unsigned long pr_exec_fdpic_loadmap;
1192 	unsigned long pr_interp_fdpic_loadmap;
1193 	int pr_fpvalid;		/* True if math co-processor being used.  */
1194 };
1195 
1196 /* An ELF note in memory */
1197 struct memelfnote
1198 {
1199 	const char *name;
1200 	int type;
1201 	unsigned int datasz;
1202 	void *data;
1203 };
1204 
1205 static int notesize(struct memelfnote *en)
1206 {
1207 	int sz;
1208 
1209 	sz = sizeof(struct elf_note);
1210 	sz += roundup(strlen(en->name) + 1, 4);
1211 	sz += roundup(en->datasz, 4);
1212 
1213 	return sz;
1214 }
1215 
1216 /* #define DEBUG */
1217 
1218 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1219 {
1220 	struct elf_note en;
1221 	en.n_namesz = strlen(men->name) + 1;
1222 	en.n_descsz = men->datasz;
1223 	en.n_type = men->type;
1224 
1225 	return dump_emit(cprm, &en, sizeof(en)) &&
1226 		dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1227 		dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1228 }
1229 
1230 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1231 {
1232 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1233 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1234 	elf->e_ident[EI_DATA] = ELF_DATA;
1235 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1236 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1237 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1238 
1239 	elf->e_type = ET_CORE;
1240 	elf->e_machine = ELF_ARCH;
1241 	elf->e_version = EV_CURRENT;
1242 	elf->e_entry = 0;
1243 	elf->e_phoff = sizeof(struct elfhdr);
1244 	elf->e_shoff = 0;
1245 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1246 	elf->e_ehsize = sizeof(struct elfhdr);
1247 	elf->e_phentsize = sizeof(struct elf_phdr);
1248 	elf->e_phnum = segs;
1249 	elf->e_shentsize = 0;
1250 	elf->e_shnum = 0;
1251 	elf->e_shstrndx = 0;
1252 	return;
1253 }
1254 
1255 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1256 {
1257 	phdr->p_type = PT_NOTE;
1258 	phdr->p_offset = offset;
1259 	phdr->p_vaddr = 0;
1260 	phdr->p_paddr = 0;
1261 	phdr->p_filesz = sz;
1262 	phdr->p_memsz = 0;
1263 	phdr->p_flags = 0;
1264 	phdr->p_align = 4;
1265 	return;
1266 }
1267 
1268 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1269 		unsigned int sz, void *data)
1270 {
1271 	note->name = name;
1272 	note->type = type;
1273 	note->datasz = sz;
1274 	note->data = data;
1275 	return;
1276 }
1277 
1278 /*
1279  * fill up all the fields in prstatus from the given task struct, except
1280  * registers which need to be filled up separately.
1281  */
1282 static void fill_prstatus(struct elf_prstatus_common *prstatus,
1283 			  struct task_struct *p, long signr)
1284 {
1285 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1286 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1287 	prstatus->pr_sighold = p->blocked.sig[0];
1288 	rcu_read_lock();
1289 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1290 	rcu_read_unlock();
1291 	prstatus->pr_pid = task_pid_vnr(p);
1292 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1293 	prstatus->pr_sid = task_session_vnr(p);
1294 	if (thread_group_leader(p)) {
1295 		struct task_cputime cputime;
1296 
1297 		/*
1298 		 * This is the record for the group leader.  It shows the
1299 		 * group-wide total, not its individual thread total.
1300 		 */
1301 		thread_group_cputime(p, &cputime);
1302 		prstatus->pr_utime = ns_to_kernel_old_timeval(cputime.utime);
1303 		prstatus->pr_stime = ns_to_kernel_old_timeval(cputime.stime);
1304 	} else {
1305 		u64 utime, stime;
1306 
1307 		task_cputime(p, &utime, &stime);
1308 		prstatus->pr_utime = ns_to_kernel_old_timeval(utime);
1309 		prstatus->pr_stime = ns_to_kernel_old_timeval(stime);
1310 	}
1311 	prstatus->pr_cutime = ns_to_kernel_old_timeval(p->signal->cutime);
1312 	prstatus->pr_cstime = ns_to_kernel_old_timeval(p->signal->cstime);
1313 }
1314 
1315 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1316 		       struct mm_struct *mm)
1317 {
1318 	const struct cred *cred;
1319 	unsigned int i, len;
1320 	unsigned int state;
1321 
1322 	/* first copy the parameters from user space */
1323 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1324 
1325 	len = mm->arg_end - mm->arg_start;
1326 	if (len >= ELF_PRARGSZ)
1327 		len = ELF_PRARGSZ - 1;
1328 	if (copy_from_user(&psinfo->pr_psargs,
1329 		           (const char __user *) mm->arg_start, len))
1330 		return -EFAULT;
1331 	for (i = 0; i < len; i++)
1332 		if (psinfo->pr_psargs[i] == 0)
1333 			psinfo->pr_psargs[i] = ' ';
1334 	psinfo->pr_psargs[len] = 0;
1335 
1336 	rcu_read_lock();
1337 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1338 	rcu_read_unlock();
1339 	psinfo->pr_pid = task_pid_vnr(p);
1340 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1341 	psinfo->pr_sid = task_session_vnr(p);
1342 
1343 	state = READ_ONCE(p->__state);
1344 	i = state ? ffz(~state) + 1 : 0;
1345 	psinfo->pr_state = i;
1346 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1347 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1348 	psinfo->pr_nice = task_nice(p);
1349 	psinfo->pr_flag = p->flags;
1350 	rcu_read_lock();
1351 	cred = __task_cred(p);
1352 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1353 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1354 	rcu_read_unlock();
1355 	get_task_comm(psinfo->pr_fname, p);
1356 
1357 	return 0;
1358 }
1359 
1360 /* Here is the structure in which status of each thread is captured. */
1361 struct elf_thread_status
1362 {
1363 	struct elf_thread_status *next;
1364 	struct elf_prstatus_fdpic prstatus;	/* NT_PRSTATUS */
1365 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1366 	struct memelfnote notes[2];
1367 	int num_notes;
1368 };
1369 
1370 /*
1371  * In order to add the specific thread information for the elf file format,
1372  * we need to keep a linked list of every thread's pr_status and then create
1373  * a single section for them in the final core file.
1374  */
1375 static struct elf_thread_status *elf_dump_thread_status(long signr, struct task_struct *p, int *sz)
1376 {
1377 	const struct user_regset_view *view = task_user_regset_view(p);
1378 	struct elf_thread_status *t;
1379 	int i, ret;
1380 
1381 	t = kzalloc(sizeof(struct elf_thread_status), GFP_KERNEL);
1382 	if (!t)
1383 		return t;
1384 
1385 	fill_prstatus(&t->prstatus.common, p, signr);
1386 	t->prstatus.pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1387 	t->prstatus.pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1388 	regset_get(p, &view->regsets[0],
1389 		   sizeof(t->prstatus.pr_reg), &t->prstatus.pr_reg);
1390 
1391 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1392 		  &t->prstatus);
1393 	t->num_notes++;
1394 	*sz += notesize(&t->notes[0]);
1395 
1396 	for (i = 1; i < view->n; ++i) {
1397 		const struct user_regset *regset = &view->regsets[i];
1398 		if (regset->core_note_type != NT_PRFPREG)
1399 			continue;
1400 		if (regset->active && regset->active(p, regset) <= 0)
1401 			continue;
1402 		ret = regset_get(p, regset, sizeof(t->fpu), &t->fpu);
1403 		if (ret >= 0)
1404 			t->prstatus.pr_fpvalid = 1;
1405 		break;
1406 	}
1407 
1408 	if (t->prstatus.pr_fpvalid) {
1409 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1410 			  &t->fpu);
1411 		t->num_notes++;
1412 		*sz += notesize(&t->notes[1]);
1413 	}
1414 	return t;
1415 }
1416 
1417 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1418 			     elf_addr_t e_shoff, int segs)
1419 {
1420 	elf->e_shoff = e_shoff;
1421 	elf->e_shentsize = sizeof(*shdr4extnum);
1422 	elf->e_shnum = 1;
1423 	elf->e_shstrndx = SHN_UNDEF;
1424 
1425 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1426 
1427 	shdr4extnum->sh_type = SHT_NULL;
1428 	shdr4extnum->sh_size = elf->e_shnum;
1429 	shdr4extnum->sh_link = elf->e_shstrndx;
1430 	shdr4extnum->sh_info = segs;
1431 }
1432 
1433 /*
1434  * dump the segments for an MMU process
1435  */
1436 static bool elf_fdpic_dump_segments(struct coredump_params *cprm,
1437 				    struct core_vma_metadata *vma_meta,
1438 				    int vma_count)
1439 {
1440 	int i;
1441 
1442 	for (i = 0; i < vma_count; i++) {
1443 		struct core_vma_metadata *meta = vma_meta + i;
1444 
1445 		if (!dump_user_range(cprm, meta->start, meta->dump_size))
1446 			return false;
1447 	}
1448 	return true;
1449 }
1450 
1451 /*
1452  * Actual dumper
1453  *
1454  * This is a two-pass process; first we find the offsets of the bits,
1455  * and then they are actually written out.  If we run out of core limit
1456  * we just truncate.
1457  */
1458 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1459 {
1460 	int has_dumped = 0;
1461 	int segs;
1462 	int i;
1463 	struct elfhdr *elf = NULL;
1464 	loff_t offset = 0, dataoff;
1465 	struct memelfnote psinfo_note, auxv_note;
1466 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1467 	struct elf_thread_status *thread_list = NULL;
1468 	int thread_status_size = 0;
1469 	elf_addr_t *auxv;
1470 	struct elf_phdr *phdr4note = NULL;
1471 	struct elf_shdr *shdr4extnum = NULL;
1472 	Elf_Half e_phnum;
1473 	elf_addr_t e_shoff;
1474 	struct core_thread *ct;
1475 	struct elf_thread_status *tmp;
1476 
1477 	/* alloc memory for large data structures: too large to be on stack */
1478 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1479 	if (!elf)
1480 		goto end_coredump;
1481 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1482 	if (!psinfo)
1483 		goto end_coredump;
1484 
1485 	for (ct = current->signal->core_state->dumper.next;
1486 					ct; ct = ct->next) {
1487 		tmp = elf_dump_thread_status(cprm->siginfo->si_signo,
1488 					     ct->task, &thread_status_size);
1489 		if (!tmp)
1490 			goto end_coredump;
1491 
1492 		tmp->next = thread_list;
1493 		thread_list = tmp;
1494 	}
1495 
1496 	/* now collect the dump for the current */
1497 	tmp = elf_dump_thread_status(cprm->siginfo->si_signo,
1498 				     current, &thread_status_size);
1499 	if (!tmp)
1500 		goto end_coredump;
1501 	tmp->next = thread_list;
1502 	thread_list = tmp;
1503 
1504 	segs = cprm->vma_count + elf_core_extra_phdrs(cprm);
1505 
1506 	/* for notes section */
1507 	segs++;
1508 
1509 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1510 	 * this, kernel supports extended numbering. Have a look at
1511 	 * include/linux/elf.h for further information. */
1512 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1513 
1514 	/* Set up header */
1515 	fill_elf_fdpic_header(elf, e_phnum);
1516 
1517 	has_dumped = 1;
1518 	/*
1519 	 * Set up the notes in similar form to SVR4 core dumps made
1520 	 * with info from their /proc.
1521 	 */
1522 
1523 	fill_psinfo(psinfo, current->group_leader, current->mm);
1524 	fill_note(&psinfo_note, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1525 	thread_status_size += notesize(&psinfo_note);
1526 
1527 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1528 	i = 0;
1529 	do
1530 		i += 2;
1531 	while (auxv[i - 2] != AT_NULL);
1532 	fill_note(&auxv_note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1533 	thread_status_size += notesize(&auxv_note);
1534 
1535 	offset = sizeof(*elf);				/* ELF header */
1536 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1537 
1538 	/* Write notes phdr entry */
1539 	phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1540 	if (!phdr4note)
1541 		goto end_coredump;
1542 
1543 	fill_elf_note_phdr(phdr4note, thread_status_size, offset);
1544 	offset += thread_status_size;
1545 
1546 	/* Page-align dumped data */
1547 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1548 
1549 	offset += cprm->vma_data_size;
1550 	offset += elf_core_extra_data_size(cprm);
1551 	e_shoff = offset;
1552 
1553 	if (e_phnum == PN_XNUM) {
1554 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1555 		if (!shdr4extnum)
1556 			goto end_coredump;
1557 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1558 	}
1559 
1560 	offset = dataoff;
1561 
1562 	if (!dump_emit(cprm, elf, sizeof(*elf)))
1563 		goto end_coredump;
1564 
1565 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1566 		goto end_coredump;
1567 
1568 	/* write program headers for segments dump */
1569 	for (i = 0; i < cprm->vma_count; i++) {
1570 		struct core_vma_metadata *meta = cprm->vma_meta + i;
1571 		struct elf_phdr phdr;
1572 		size_t sz;
1573 
1574 		sz = meta->end - meta->start;
1575 
1576 		phdr.p_type = PT_LOAD;
1577 		phdr.p_offset = offset;
1578 		phdr.p_vaddr = meta->start;
1579 		phdr.p_paddr = 0;
1580 		phdr.p_filesz = meta->dump_size;
1581 		phdr.p_memsz = sz;
1582 		offset += phdr.p_filesz;
1583 		phdr.p_flags = 0;
1584 		if (meta->flags & VM_READ)
1585 			phdr.p_flags |= PF_R;
1586 		if (meta->flags & VM_WRITE)
1587 			phdr.p_flags |= PF_W;
1588 		if (meta->flags & VM_EXEC)
1589 			phdr.p_flags |= PF_X;
1590 		phdr.p_align = ELF_EXEC_PAGESIZE;
1591 
1592 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1593 			goto end_coredump;
1594 	}
1595 
1596 	if (!elf_core_write_extra_phdrs(cprm, offset))
1597 		goto end_coredump;
1598 
1599 	/* write out the notes section */
1600 	if (!writenote(thread_list->notes, cprm))
1601 		goto end_coredump;
1602 	if (!writenote(&psinfo_note, cprm))
1603 		goto end_coredump;
1604 	if (!writenote(&auxv_note, cprm))
1605 		goto end_coredump;
1606 	for (i = 1; i < thread_list->num_notes; i++)
1607 		if (!writenote(thread_list->notes + i, cprm))
1608 			goto end_coredump;
1609 
1610 	/* write out the thread status notes section */
1611 	for (tmp = thread_list->next; tmp; tmp = tmp->next) {
1612 		for (i = 0; i < tmp->num_notes; i++)
1613 			if (!writenote(&tmp->notes[i], cprm))
1614 				goto end_coredump;
1615 	}
1616 
1617 	dump_skip_to(cprm, dataoff);
1618 
1619 	if (!elf_fdpic_dump_segments(cprm, cprm->vma_meta, cprm->vma_count))
1620 		goto end_coredump;
1621 
1622 	if (!elf_core_write_extra_data(cprm))
1623 		goto end_coredump;
1624 
1625 	if (e_phnum == PN_XNUM) {
1626 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1627 			goto end_coredump;
1628 	}
1629 
1630 	if (cprm->file->f_pos != offset) {
1631 		/* Sanity check */
1632 		printk(KERN_WARNING
1633 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1634 		       cprm->file->f_pos, offset);
1635 	}
1636 
1637 end_coredump:
1638 	while (thread_list) {
1639 		tmp = thread_list;
1640 		thread_list = thread_list->next;
1641 		kfree(tmp);
1642 	}
1643 	kfree(phdr4note);
1644 	kfree(elf);
1645 	kfree(psinfo);
1646 	kfree(shdr4extnum);
1647 	return has_dumped;
1648 }
1649 
1650 #endif		/* CONFIG_ELF_CORE */
1651