xref: /linux/fs/binfmt_elf_fdpic.c (revision ce615f5c1f73537c8267035d58b3d0c70e19b8da)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
3  *
4  * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  * Derived from binfmt_elf.c
7  */
8 
9 #include <linux/module.h>
10 
11 #include <linux/fs.h>
12 #include <linux/stat.h>
13 #include <linux/sched.h>
14 #include <linux/sched/coredump.h>
15 #include <linux/sched/task_stack.h>
16 #include <linux/sched/cputime.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/errno.h>
20 #include <linux/signal.h>
21 #include <linux/binfmts.h>
22 #include <linux/string.h>
23 #include <linux/file.h>
24 #include <linux/fcntl.h>
25 #include <linux/slab.h>
26 #include <linux/pagemap.h>
27 #include <linux/security.h>
28 #include <linux/highmem.h>
29 #include <linux/highuid.h>
30 #include <linux/personality.h>
31 #include <linux/ptrace.h>
32 #include <linux/init.h>
33 #include <linux/elf.h>
34 #include <linux/elf-fdpic.h>
35 #include <linux/elfcore.h>
36 #include <linux/coredump.h>
37 #include <linux/dax.h>
38 
39 #include <linux/uaccess.h>
40 #include <asm/param.h>
41 
42 typedef char *elf_caddr_t;
43 
44 #if 0
45 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
46 #else
47 #define kdebug(fmt, ...) do {} while(0)
48 #endif
49 
50 #if 0
51 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
52 #else
53 #define kdcore(fmt, ...) do {} while(0)
54 #endif
55 
56 MODULE_LICENSE("GPL");
57 
58 static int load_elf_fdpic_binary(struct linux_binprm *);
59 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
60 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
61 			      struct mm_struct *, const char *);
62 
63 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
64 				   struct elf_fdpic_params *,
65 				   struct elf_fdpic_params *);
66 
67 #ifndef CONFIG_MMU
68 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
69 						   struct file *,
70 						   struct mm_struct *);
71 #endif
72 
73 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
74 					     struct file *, struct mm_struct *);
75 
76 #ifdef CONFIG_ELF_CORE
77 static int elf_fdpic_core_dump(struct coredump_params *cprm);
78 #endif
79 
80 static struct linux_binfmt elf_fdpic_format = {
81 	.module		= THIS_MODULE,
82 	.load_binary	= load_elf_fdpic_binary,
83 #ifdef CONFIG_ELF_CORE
84 	.core_dump	= elf_fdpic_core_dump,
85 #endif
86 	.min_coredump	= ELF_EXEC_PAGESIZE,
87 };
88 
89 static int __init init_elf_fdpic_binfmt(void)
90 {
91 	register_binfmt(&elf_fdpic_format);
92 	return 0;
93 }
94 
95 static void __exit exit_elf_fdpic_binfmt(void)
96 {
97 	unregister_binfmt(&elf_fdpic_format);
98 }
99 
100 core_initcall(init_elf_fdpic_binfmt);
101 module_exit(exit_elf_fdpic_binfmt);
102 
103 static int is_elf(struct elfhdr *hdr, struct file *file)
104 {
105 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
106 		return 0;
107 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
108 		return 0;
109 	if (!elf_check_arch(hdr))
110 		return 0;
111 	if (!file->f_op->mmap)
112 		return 0;
113 	return 1;
114 }
115 
116 #ifndef elf_check_fdpic
117 #define elf_check_fdpic(x) 0
118 #endif
119 
120 #ifndef elf_check_const_displacement
121 #define elf_check_const_displacement(x) 0
122 #endif
123 
124 static int is_constdisp(struct elfhdr *hdr)
125 {
126 	if (!elf_check_fdpic(hdr))
127 		return 1;
128 	if (elf_check_const_displacement(hdr))
129 		return 1;
130 	return 0;
131 }
132 
133 /*****************************************************************************/
134 /*
135  * read the program headers table into memory
136  */
137 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
138 				 struct file *file)
139 {
140 	struct elf32_phdr *phdr;
141 	unsigned long size;
142 	int retval, loop;
143 	loff_t pos = params->hdr.e_phoff;
144 
145 	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
146 		return -ENOMEM;
147 	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
148 		return -ENOMEM;
149 
150 	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
151 	params->phdrs = kmalloc(size, GFP_KERNEL);
152 	if (!params->phdrs)
153 		return -ENOMEM;
154 
155 	retval = kernel_read(file, params->phdrs, size, &pos);
156 	if (unlikely(retval != size))
157 		return retval < 0 ? retval : -ENOEXEC;
158 
159 	/* determine stack size for this binary */
160 	phdr = params->phdrs;
161 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
162 		if (phdr->p_type != PT_GNU_STACK)
163 			continue;
164 
165 		if (phdr->p_flags & PF_X)
166 			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
167 		else
168 			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
169 
170 		params->stack_size = phdr->p_memsz;
171 		break;
172 	}
173 
174 	return 0;
175 }
176 
177 /*****************************************************************************/
178 /*
179  * load an fdpic binary into various bits of memory
180  */
181 static int load_elf_fdpic_binary(struct linux_binprm *bprm)
182 {
183 	struct elf_fdpic_params exec_params, interp_params;
184 	struct pt_regs *regs = current_pt_regs();
185 	struct elf_phdr *phdr;
186 	unsigned long stack_size, entryaddr;
187 #ifdef ELF_FDPIC_PLAT_INIT
188 	unsigned long dynaddr;
189 #endif
190 #ifndef CONFIG_MMU
191 	unsigned long stack_prot;
192 #endif
193 	struct file *interpreter = NULL; /* to shut gcc up */
194 	char *interpreter_name = NULL;
195 	int executable_stack;
196 	int retval, i;
197 	loff_t pos;
198 
199 	kdebug("____ LOAD %d ____", current->pid);
200 
201 	memset(&exec_params, 0, sizeof(exec_params));
202 	memset(&interp_params, 0, sizeof(interp_params));
203 
204 	exec_params.hdr = *(struct elfhdr *) bprm->buf;
205 	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
206 
207 	/* check that this is a binary we know how to deal with */
208 	retval = -ENOEXEC;
209 	if (!is_elf(&exec_params.hdr, bprm->file))
210 		goto error;
211 	if (!elf_check_fdpic(&exec_params.hdr)) {
212 #ifdef CONFIG_MMU
213 		/* binfmt_elf handles non-fdpic elf except on nommu */
214 		goto error;
215 #else
216 		/* nommu can only load ET_DYN (PIE) ELF */
217 		if (exec_params.hdr.e_type != ET_DYN)
218 			goto error;
219 #endif
220 	}
221 
222 	/* read the program header table */
223 	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
224 	if (retval < 0)
225 		goto error;
226 
227 	/* scan for a program header that specifies an interpreter */
228 	phdr = exec_params.phdrs;
229 
230 	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
231 		switch (phdr->p_type) {
232 		case PT_INTERP:
233 			retval = -ENOMEM;
234 			if (phdr->p_filesz > PATH_MAX)
235 				goto error;
236 			retval = -ENOENT;
237 			if (phdr->p_filesz < 2)
238 				goto error;
239 
240 			/* read the name of the interpreter into memory */
241 			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
242 			if (!interpreter_name)
243 				goto error;
244 
245 			pos = phdr->p_offset;
246 			retval = kernel_read(bprm->file, interpreter_name,
247 					     phdr->p_filesz, &pos);
248 			if (unlikely(retval != phdr->p_filesz)) {
249 				if (retval >= 0)
250 					retval = -ENOEXEC;
251 				goto error;
252 			}
253 
254 			retval = -ENOENT;
255 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
256 				goto error;
257 
258 			kdebug("Using ELF interpreter %s", interpreter_name);
259 
260 			/* replace the program with the interpreter */
261 			interpreter = open_exec(interpreter_name);
262 			retval = PTR_ERR(interpreter);
263 			if (IS_ERR(interpreter)) {
264 				interpreter = NULL;
265 				goto error;
266 			}
267 
268 			/*
269 			 * If the binary is not readable then enforce
270 			 * mm->dumpable = 0 regardless of the interpreter's
271 			 * permissions.
272 			 */
273 			would_dump(bprm, interpreter);
274 
275 			pos = 0;
276 			retval = kernel_read(interpreter, bprm->buf,
277 					BINPRM_BUF_SIZE, &pos);
278 			if (unlikely(retval != BINPRM_BUF_SIZE)) {
279 				if (retval >= 0)
280 					retval = -ENOEXEC;
281 				goto error;
282 			}
283 
284 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
285 			break;
286 
287 		case PT_LOAD:
288 #ifdef CONFIG_MMU
289 			if (exec_params.load_addr == 0)
290 				exec_params.load_addr = phdr->p_vaddr;
291 #endif
292 			break;
293 		}
294 
295 	}
296 
297 	if (is_constdisp(&exec_params.hdr))
298 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
299 
300 	/* perform insanity checks on the interpreter */
301 	if (interpreter_name) {
302 		retval = -ELIBBAD;
303 		if (!is_elf(&interp_params.hdr, interpreter))
304 			goto error;
305 
306 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
307 
308 		/* read the interpreter's program header table */
309 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
310 		if (retval < 0)
311 			goto error;
312 	}
313 
314 	stack_size = exec_params.stack_size;
315 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
316 		executable_stack = EXSTACK_ENABLE_X;
317 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
318 		executable_stack = EXSTACK_DISABLE_X;
319 	else
320 		executable_stack = EXSTACK_DEFAULT;
321 
322 	if (stack_size == 0) {
323 		stack_size = interp_params.stack_size;
324 		if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
325 			executable_stack = EXSTACK_ENABLE_X;
326 		else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
327 			executable_stack = EXSTACK_DISABLE_X;
328 		else
329 			executable_stack = EXSTACK_DEFAULT;
330 	}
331 
332 	retval = -ENOEXEC;
333 	if (stack_size == 0)
334 		stack_size = 131072UL; /* same as exec.c's default commit */
335 
336 	if (is_constdisp(&interp_params.hdr))
337 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
338 
339 	/* flush all traces of the currently running executable */
340 	retval = begin_new_exec(bprm);
341 	if (retval)
342 		goto error;
343 
344 	/* there's now no turning back... the old userspace image is dead,
345 	 * defunct, deceased, etc.
346 	 */
347 	if (elf_check_fdpic(&exec_params.hdr))
348 		set_personality(PER_LINUX_FDPIC);
349 	else
350 		set_personality(PER_LINUX);
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 		allow_write_access(interpreter);
398 		fput(interpreter);
399 		interpreter = NULL;
400 	}
401 
402 #ifdef CONFIG_MMU
403 	if (!current->mm->start_brk)
404 		current->mm->start_brk = current->mm->end_data;
405 
406 	current->mm->brk = current->mm->start_brk =
407 		PAGE_ALIGN(current->mm->start_brk);
408 
409 #else
410 	/* create a stack area and zero-size brk area */
411 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
412 	if (stack_size < PAGE_SIZE * 2)
413 		stack_size = PAGE_SIZE * 2;
414 
415 	stack_prot = PROT_READ | PROT_WRITE;
416 	if (executable_stack == EXSTACK_ENABLE_X ||
417 	    (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
418 		stack_prot |= PROT_EXEC;
419 
420 	current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
421 					 MAP_PRIVATE | MAP_ANONYMOUS |
422 					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
423 					 0);
424 
425 	if (IS_ERR_VALUE(current->mm->start_brk)) {
426 		retval = current->mm->start_brk;
427 		current->mm->start_brk = 0;
428 		goto error;
429 	}
430 
431 	current->mm->brk = current->mm->start_brk;
432 	current->mm->context.end_brk = current->mm->start_brk;
433 	current->mm->start_stack = current->mm->start_brk + stack_size;
434 #endif
435 
436 	if (create_elf_fdpic_tables(bprm, current->mm,
437 				    &exec_params, &interp_params) < 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 		allow_write_access(interpreter);
470 		fput(interpreter);
471 	}
472 	kfree(interpreter_name);
473 	kfree(exec_params.phdrs);
474 	kfree(exec_params.loadmap);
475 	kfree(interp_params.phdrs);
476 	kfree(interp_params.loadmap);
477 	return retval;
478 }
479 
480 /*****************************************************************************/
481 
482 #ifndef ELF_BASE_PLATFORM
483 /*
484  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
485  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
486  * will be copied to the user stack in the same manner as AT_PLATFORM.
487  */
488 #define ELF_BASE_PLATFORM NULL
489 #endif
490 
491 /*
492  * present useful information to the program by shovelling it onto the new
493  * process's stack
494  */
495 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
496 				   struct mm_struct *mm,
497 				   struct elf_fdpic_params *exec_params,
498 				   struct elf_fdpic_params *interp_params)
499 {
500 	const struct cred *cred = current_cred();
501 	unsigned long sp, csp, nitems;
502 	elf_caddr_t __user *argv, *envp;
503 	size_t platform_len = 0, len;
504 	char *k_platform, *k_base_platform;
505 	char __user *u_platform, *u_base_platform, *p;
506 	int loop;
507 	int nr;	/* reset for each csp adjustment */
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 elf32_fdpic_loadmap);
561 	len += sizeof(struct elf32_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 elf32_fdpic_loadmap);
572 		len += sizeof(struct elf32_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 
593 	csp = sp;
594 	sp -= nitems * 2 * sizeof(unsigned long);
595 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
596 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
597 	sp -= 1 * sizeof(unsigned long);		/* argc */
598 
599 	csp -= sp & 15UL;
600 	sp -= sp & 15UL;
601 
602 	/* put the ELF interpreter info on the stack */
603 #define NEW_AUX_ENT(id, val)						\
604 	do {								\
605 		struct { unsigned long _id, _val; } __user *ent, v;	\
606 									\
607 		ent = (void __user *) csp;				\
608 		v._id = (id);						\
609 		v._val = (val);						\
610 		if (copy_to_user(ent + nr, &v, sizeof(v)))		\
611 			return -EFAULT;					\
612 		nr++;							\
613 	} while (0)
614 
615 	nr = 0;
616 	csp -= 2 * sizeof(unsigned long);
617 	NEW_AUX_ENT(AT_NULL, 0);
618 	if (k_platform) {
619 		nr = 0;
620 		csp -= 2 * sizeof(unsigned long);
621 		NEW_AUX_ENT(AT_PLATFORM,
622 			    (elf_addr_t) (unsigned long) u_platform);
623 	}
624 
625 	if (k_base_platform) {
626 		nr = 0;
627 		csp -= 2 * sizeof(unsigned long);
628 		NEW_AUX_ENT(AT_BASE_PLATFORM,
629 			    (elf_addr_t) (unsigned long) u_base_platform);
630 	}
631 
632 	if (bprm->have_execfd) {
633 		nr = 0;
634 		csp -= 2 * sizeof(unsigned long);
635 		NEW_AUX_ENT(AT_EXECFD, bprm->execfd);
636 	}
637 
638 	nr = 0;
639 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
640 	NEW_AUX_ENT(AT_HWCAP,	ELF_HWCAP);
641 #ifdef ELF_HWCAP2
642 	NEW_AUX_ENT(AT_HWCAP2,	ELF_HWCAP2);
643 #endif
644 	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
645 	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
646 	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
647 	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
648 	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
649 	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
650 	NEW_AUX_ENT(AT_FLAGS,	0);
651 	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
652 	NEW_AUX_ENT(AT_UID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
653 	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
654 	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
655 	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
656 	NEW_AUX_ENT(AT_SECURE,	bprm->secureexec);
657 	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
658 
659 #ifdef ARCH_DLINFO
660 	nr = 0;
661 	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
662 
663 	/* ARCH_DLINFO must come last so platform specific code can enforce
664 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
665 	 */
666 	ARCH_DLINFO;
667 #endif
668 #undef NEW_AUX_ENT
669 
670 	/* allocate room for argv[] and envv[] */
671 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
672 	envp = (elf_caddr_t __user *) csp;
673 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
674 	argv = (elf_caddr_t __user *) csp;
675 
676 	/* stack argc */
677 	csp -= sizeof(unsigned long);
678 	if (put_user(bprm->argc, (unsigned long __user *) csp))
679 		return -EFAULT;
680 
681 	BUG_ON(csp != sp);
682 
683 	/* fill in the argv[] array */
684 #ifdef CONFIG_MMU
685 	current->mm->arg_start = bprm->p;
686 #else
687 	current->mm->arg_start = current->mm->start_stack -
688 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
689 #endif
690 
691 	p = (char __user *) current->mm->arg_start;
692 	for (loop = bprm->argc; loop > 0; loop--) {
693 		if (put_user((elf_caddr_t) p, argv++))
694 			return -EFAULT;
695 		len = strnlen_user(p, MAX_ARG_STRLEN);
696 		if (!len || len > MAX_ARG_STRLEN)
697 			return -EINVAL;
698 		p += len;
699 	}
700 	if (put_user(NULL, argv))
701 		return -EFAULT;
702 	current->mm->arg_end = (unsigned long) p;
703 
704 	/* fill in the envv[] array */
705 	current->mm->env_start = (unsigned long) p;
706 	for (loop = bprm->envc; loop > 0; loop--) {
707 		if (put_user((elf_caddr_t)(unsigned long) p, envp++))
708 			return -EFAULT;
709 		len = strnlen_user(p, MAX_ARG_STRLEN);
710 		if (!len || len > MAX_ARG_STRLEN)
711 			return -EINVAL;
712 		p += len;
713 	}
714 	if (put_user(NULL, envp))
715 		return -EFAULT;
716 	current->mm->env_end = (unsigned long) p;
717 
718 	mm->start_stack = (unsigned long) sp;
719 	return 0;
720 }
721 
722 /*****************************************************************************/
723 /*
724  * load the appropriate binary image (executable or interpreter) into memory
725  * - we assume no MMU is available
726  * - if no other PIC bits are set in params->hdr->e_flags
727  *   - we assume that the LOADable segments in the binary are independently relocatable
728  *   - we assume R/O executable segments are shareable
729  * - else
730  *   - we assume the loadable parts of the image to require fixed displacement
731  *   - the image is not shareable
732  */
733 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
734 			      struct file *file,
735 			      struct mm_struct *mm,
736 			      const char *what)
737 {
738 	struct elf32_fdpic_loadmap *loadmap;
739 #ifdef CONFIG_MMU
740 	struct elf32_fdpic_loadseg *mseg;
741 #endif
742 	struct elf32_fdpic_loadseg *seg;
743 	struct elf32_phdr *phdr;
744 	unsigned long load_addr, stop;
745 	unsigned nloads, tmp;
746 	size_t size;
747 	int loop, ret;
748 
749 	/* allocate a load map table */
750 	nloads = 0;
751 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
752 		if (params->phdrs[loop].p_type == PT_LOAD)
753 			nloads++;
754 
755 	if (nloads == 0)
756 		return -ELIBBAD;
757 
758 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
759 	loadmap = kzalloc(size, GFP_KERNEL);
760 	if (!loadmap)
761 		return -ENOMEM;
762 
763 	params->loadmap = loadmap;
764 
765 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
766 	loadmap->nsegs = nloads;
767 
768 	load_addr = params->load_addr;
769 	seg = loadmap->segs;
770 
771 	/* map the requested LOADs into the memory space */
772 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
773 	case ELF_FDPIC_FLAG_CONSTDISP:
774 	case ELF_FDPIC_FLAG_CONTIGUOUS:
775 #ifndef CONFIG_MMU
776 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
777 		if (ret < 0)
778 			return ret;
779 		break;
780 #endif
781 	default:
782 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
783 		if (ret < 0)
784 			return ret;
785 		break;
786 	}
787 
788 	/* map the entry point */
789 	if (params->hdr.e_entry) {
790 		seg = loadmap->segs;
791 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
792 			if (params->hdr.e_entry >= seg->p_vaddr &&
793 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
794 				params->entry_addr =
795 					(params->hdr.e_entry - seg->p_vaddr) +
796 					seg->addr;
797 				break;
798 			}
799 		}
800 	}
801 
802 	/* determine where the program header table has wound up if mapped */
803 	stop = params->hdr.e_phoff;
804 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
805 	phdr = params->phdrs;
806 
807 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
808 		if (phdr->p_type != PT_LOAD)
809 			continue;
810 
811 		if (phdr->p_offset > params->hdr.e_phoff ||
812 		    phdr->p_offset + phdr->p_filesz < stop)
813 			continue;
814 
815 		seg = loadmap->segs;
816 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
817 			if (phdr->p_vaddr >= seg->p_vaddr &&
818 			    phdr->p_vaddr + phdr->p_filesz <=
819 			    seg->p_vaddr + seg->p_memsz) {
820 				params->ph_addr =
821 					(phdr->p_vaddr - seg->p_vaddr) +
822 					seg->addr +
823 					params->hdr.e_phoff - phdr->p_offset;
824 				break;
825 			}
826 		}
827 		break;
828 	}
829 
830 	/* determine where the dynamic section has wound up if there is one */
831 	phdr = params->phdrs;
832 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
833 		if (phdr->p_type != PT_DYNAMIC)
834 			continue;
835 
836 		seg = loadmap->segs;
837 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
838 			if (phdr->p_vaddr >= seg->p_vaddr &&
839 			    phdr->p_vaddr + phdr->p_memsz <=
840 			    seg->p_vaddr + seg->p_memsz) {
841 				Elf32_Dyn __user *dyn;
842 				Elf32_Sword d_tag;
843 
844 				params->dynamic_addr =
845 					(phdr->p_vaddr - seg->p_vaddr) +
846 					seg->addr;
847 
848 				/* check the dynamic section contains at least
849 				 * one item, and that the last item is a NULL
850 				 * entry */
851 				if (phdr->p_memsz == 0 ||
852 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
853 					goto dynamic_error;
854 
855 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
856 				dyn = (Elf32_Dyn __user *)params->dynamic_addr;
857 				if (get_user(d_tag, &dyn[tmp - 1].d_tag) ||
858 				    d_tag != 0)
859 					goto dynamic_error;
860 				break;
861 			}
862 		}
863 		break;
864 	}
865 
866 	/* now elide adjacent segments in the load map on MMU linux
867 	 * - on uClinux the holes between may actually be filled with system
868 	 *   stuff or stuff from other processes
869 	 */
870 #ifdef CONFIG_MMU
871 	nloads = loadmap->nsegs;
872 	mseg = loadmap->segs;
873 	seg = mseg + 1;
874 	for (loop = 1; loop < nloads; loop++) {
875 		/* see if we have a candidate for merging */
876 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
877 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
878 			if (load_addr == (seg->addr & PAGE_MASK)) {
879 				mseg->p_memsz +=
880 					load_addr -
881 					(mseg->addr + mseg->p_memsz);
882 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
883 				mseg->p_memsz += seg->p_memsz;
884 				loadmap->nsegs--;
885 				continue;
886 			}
887 		}
888 
889 		mseg++;
890 		if (mseg != seg)
891 			*mseg = *seg;
892 	}
893 #endif
894 
895 	kdebug("Mapped Object [%s]:", what);
896 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
897 	kdebug("- entry    : %lx", params->entry_addr);
898 	kdebug("- PHDR[]   : %lx", params->ph_addr);
899 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
900 	seg = loadmap->segs;
901 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
902 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
903 		       loop,
904 		       seg->addr, seg->addr + seg->p_memsz - 1,
905 		       seg->p_vaddr, seg->p_memsz);
906 
907 	return 0;
908 
909 dynamic_error:
910 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
911 	       what, file_inode(file)->i_ino);
912 	return -ELIBBAD;
913 }
914 
915 /*****************************************************************************/
916 /*
917  * map a file with constant displacement under uClinux
918  */
919 #ifndef CONFIG_MMU
920 static int elf_fdpic_map_file_constdisp_on_uclinux(
921 	struct elf_fdpic_params *params,
922 	struct file *file,
923 	struct mm_struct *mm)
924 {
925 	struct elf32_fdpic_loadseg *seg;
926 	struct elf32_phdr *phdr;
927 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
928 	int loop, ret;
929 
930 	load_addr = params->load_addr;
931 	seg = params->loadmap->segs;
932 
933 	/* determine the bounds of the contiguous overall allocation we must
934 	 * make */
935 	phdr = params->phdrs;
936 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
937 		if (params->phdrs[loop].p_type != PT_LOAD)
938 			continue;
939 
940 		if (base > phdr->p_vaddr)
941 			base = phdr->p_vaddr;
942 		if (top < phdr->p_vaddr + phdr->p_memsz)
943 			top = phdr->p_vaddr + phdr->p_memsz;
944 	}
945 
946 	/* allocate one big anon block for everything */
947 	mflags = MAP_PRIVATE;
948 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
949 		mflags |= MAP_EXECUTABLE;
950 
951 	maddr = vm_mmap(NULL, load_addr, top - base,
952 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
953 	if (IS_ERR_VALUE(maddr))
954 		return (int) maddr;
955 
956 	if (load_addr != 0)
957 		load_addr += PAGE_ALIGN(top - base);
958 
959 	/* and then load the file segments into it */
960 	phdr = params->phdrs;
961 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
962 		if (params->phdrs[loop].p_type != PT_LOAD)
963 			continue;
964 
965 		seg->addr = maddr + (phdr->p_vaddr - base);
966 		seg->p_vaddr = phdr->p_vaddr;
967 		seg->p_memsz = phdr->p_memsz;
968 
969 		ret = read_code(file, seg->addr, phdr->p_offset,
970 				       phdr->p_filesz);
971 		if (ret < 0)
972 			return ret;
973 
974 		/* map the ELF header address if in this segment */
975 		if (phdr->p_offset == 0)
976 			params->elfhdr_addr = seg->addr;
977 
978 		/* clear any space allocated but not loaded */
979 		if (phdr->p_filesz < phdr->p_memsz) {
980 			if (clear_user((void *) (seg->addr + phdr->p_filesz),
981 				       phdr->p_memsz - phdr->p_filesz))
982 				return -EFAULT;
983 		}
984 
985 		if (mm) {
986 			if (phdr->p_flags & PF_X) {
987 				if (!mm->start_code) {
988 					mm->start_code = seg->addr;
989 					mm->end_code = seg->addr +
990 						phdr->p_memsz;
991 				}
992 			} else if (!mm->start_data) {
993 				mm->start_data = seg->addr;
994 				mm->end_data = seg->addr + phdr->p_memsz;
995 			}
996 		}
997 
998 		seg++;
999 	}
1000 
1001 	return 0;
1002 }
1003 #endif
1004 
1005 /*****************************************************************************/
1006 /*
1007  * map a binary by direct mmap() of the individual PT_LOAD segments
1008  */
1009 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1010 					     struct file *file,
1011 					     struct mm_struct *mm)
1012 {
1013 	struct elf32_fdpic_loadseg *seg;
1014 	struct elf32_phdr *phdr;
1015 	unsigned long load_addr, delta_vaddr;
1016 	int loop, dvset;
1017 
1018 	load_addr = params->load_addr;
1019 	delta_vaddr = 0;
1020 	dvset = 0;
1021 
1022 	seg = params->loadmap->segs;
1023 
1024 	/* deal with each load segment separately */
1025 	phdr = params->phdrs;
1026 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1027 		unsigned long maddr, disp, excess, excess1;
1028 		int prot = 0, flags;
1029 
1030 		if (phdr->p_type != PT_LOAD)
1031 			continue;
1032 
1033 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1034 		       (unsigned long) phdr->p_vaddr,
1035 		       (unsigned long) phdr->p_offset,
1036 		       (unsigned long) phdr->p_filesz,
1037 		       (unsigned long) phdr->p_memsz);
1038 
1039 		/* determine the mapping parameters */
1040 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1041 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1042 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1043 
1044 		flags = MAP_PRIVATE | MAP_DENYWRITE;
1045 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1046 			flags |= MAP_EXECUTABLE;
1047 
1048 		maddr = 0;
1049 
1050 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1051 		case ELF_FDPIC_FLAG_INDEPENDENT:
1052 			/* PT_LOADs are independently locatable */
1053 			break;
1054 
1055 		case ELF_FDPIC_FLAG_HONOURVADDR:
1056 			/* the specified virtual address must be honoured */
1057 			maddr = phdr->p_vaddr;
1058 			flags |= MAP_FIXED;
1059 			break;
1060 
1061 		case ELF_FDPIC_FLAG_CONSTDISP:
1062 			/* constant displacement
1063 			 * - can be mapped anywhere, but must be mapped as a
1064 			 *   unit
1065 			 */
1066 			if (!dvset) {
1067 				maddr = load_addr;
1068 				delta_vaddr = phdr->p_vaddr;
1069 				dvset = 1;
1070 			} else {
1071 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1072 				flags |= MAP_FIXED;
1073 			}
1074 			break;
1075 
1076 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1077 			/* contiguity handled later */
1078 			break;
1079 
1080 		default:
1081 			BUG();
1082 		}
1083 
1084 		maddr &= PAGE_MASK;
1085 
1086 		/* create the mapping */
1087 		disp = phdr->p_vaddr & ~PAGE_MASK;
1088 		maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1089 				phdr->p_offset - disp);
1090 
1091 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1092 		       loop, phdr->p_memsz + disp, prot, flags,
1093 		       phdr->p_offset - disp, maddr);
1094 
1095 		if (IS_ERR_VALUE(maddr))
1096 			return (int) maddr;
1097 
1098 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1099 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1100 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1101 
1102 		seg->addr = maddr + disp;
1103 		seg->p_vaddr = phdr->p_vaddr;
1104 		seg->p_memsz = phdr->p_memsz;
1105 
1106 		/* map the ELF header address if in this segment */
1107 		if (phdr->p_offset == 0)
1108 			params->elfhdr_addr = seg->addr;
1109 
1110 		/* clear the bit between beginning of mapping and beginning of
1111 		 * PT_LOAD */
1112 		if (prot & PROT_WRITE && disp > 0) {
1113 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1114 			if (clear_user((void __user *) maddr, disp))
1115 				return -EFAULT;
1116 			maddr += disp;
1117 		}
1118 
1119 		/* clear any space allocated but not loaded
1120 		 * - on uClinux we can just clear the lot
1121 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1122 		 *   extant in the file
1123 		 */
1124 		excess = phdr->p_memsz - phdr->p_filesz;
1125 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1126 
1127 #ifdef CONFIG_MMU
1128 		if (excess > excess1) {
1129 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1130 			unsigned long xmaddr;
1131 
1132 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1133 			xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1134 					 prot, flags, 0);
1135 
1136 			kdebug("mmap[%d] <anon>"
1137 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1138 			       loop, xaddr, excess - excess1, prot, flags,
1139 			       xmaddr);
1140 
1141 			if (xmaddr != xaddr)
1142 				return -ENOMEM;
1143 		}
1144 
1145 		if (prot & PROT_WRITE && excess1 > 0) {
1146 			kdebug("clear[%d] ad=%lx sz=%lx",
1147 			       loop, maddr + phdr->p_filesz, excess1);
1148 			if (clear_user((void __user *) maddr + phdr->p_filesz,
1149 				       excess1))
1150 				return -EFAULT;
1151 		}
1152 
1153 #else
1154 		if (excess > 0) {
1155 			kdebug("clear[%d] ad=%lx sz=%lx",
1156 			       loop, maddr + phdr->p_filesz, excess);
1157 			if (clear_user((void *) maddr + phdr->p_filesz, excess))
1158 				return -EFAULT;
1159 		}
1160 #endif
1161 
1162 		if (mm) {
1163 			if (phdr->p_flags & PF_X) {
1164 				if (!mm->start_code) {
1165 					mm->start_code = maddr;
1166 					mm->end_code = maddr + phdr->p_memsz;
1167 				}
1168 			} else if (!mm->start_data) {
1169 				mm->start_data = maddr;
1170 				mm->end_data = maddr + phdr->p_memsz;
1171 			}
1172 		}
1173 
1174 		seg++;
1175 	}
1176 
1177 	return 0;
1178 }
1179 
1180 /*****************************************************************************/
1181 /*
1182  * ELF-FDPIC core dumper
1183  *
1184  * Modelled on fs/exec.c:aout_core_dump()
1185  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1186  *
1187  * Modelled on fs/binfmt_elf.c core dumper
1188  */
1189 #ifdef CONFIG_ELF_CORE
1190 
1191 /*
1192  * Decide whether a segment is worth dumping; default is yes to be
1193  * sure (missing info is worse than too much; etc).
1194  * Personally I'd include everything, and use the coredump limit...
1195  *
1196  * I think we should skip something. But I am not sure how. H.J.
1197  */
1198 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1199 {
1200 	int dump_ok;
1201 
1202 	/* Do not dump I/O mapped devices or special mappings */
1203 	if (vma->vm_flags & VM_IO) {
1204 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1205 		return 0;
1206 	}
1207 
1208 	/* If we may not read the contents, don't allow us to dump
1209 	 * them either. "dump_write()" can't handle it anyway.
1210 	 */
1211 	if (!(vma->vm_flags & VM_READ)) {
1212 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1213 		return 0;
1214 	}
1215 
1216 	/* support for DAX */
1217 	if (vma_is_dax(vma)) {
1218 		if (vma->vm_flags & VM_SHARED) {
1219 			dump_ok = test_bit(MMF_DUMP_DAX_SHARED, &mm_flags);
1220 			kdcore("%08lx: %08lx: %s (DAX shared)", vma->vm_start,
1221 			       vma->vm_flags, dump_ok ? "yes" : "no");
1222 		} else {
1223 			dump_ok = test_bit(MMF_DUMP_DAX_PRIVATE, &mm_flags);
1224 			kdcore("%08lx: %08lx: %s (DAX private)", vma->vm_start,
1225 			       vma->vm_flags, dump_ok ? "yes" : "no");
1226 		}
1227 		return dump_ok;
1228 	}
1229 
1230 	/* By default, dump shared memory if mapped from an anonymous file. */
1231 	if (vma->vm_flags & VM_SHARED) {
1232 		if (file_inode(vma->vm_file)->i_nlink == 0) {
1233 			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1234 			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1235 			       vma->vm_flags, dump_ok ? "yes" : "no");
1236 			return dump_ok;
1237 		}
1238 
1239 		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1240 		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1241 		       vma->vm_flags, dump_ok ? "yes" : "no");
1242 		return dump_ok;
1243 	}
1244 
1245 #ifdef CONFIG_MMU
1246 	/* By default, if it hasn't been written to, don't write it out */
1247 	if (!vma->anon_vma) {
1248 		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1249 		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1250 		       vma->vm_flags, dump_ok ? "yes" : "no");
1251 		return dump_ok;
1252 	}
1253 #endif
1254 
1255 	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1256 	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1257 	       dump_ok ? "yes" : "no");
1258 	return dump_ok;
1259 }
1260 
1261 /* An ELF note in memory */
1262 struct memelfnote
1263 {
1264 	const char *name;
1265 	int type;
1266 	unsigned int datasz;
1267 	void *data;
1268 };
1269 
1270 static int notesize(struct memelfnote *en)
1271 {
1272 	int sz;
1273 
1274 	sz = sizeof(struct elf_note);
1275 	sz += roundup(strlen(en->name) + 1, 4);
1276 	sz += roundup(en->datasz, 4);
1277 
1278 	return sz;
1279 }
1280 
1281 /* #define DEBUG */
1282 
1283 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1284 {
1285 	struct elf_note en;
1286 	en.n_namesz = strlen(men->name) + 1;
1287 	en.n_descsz = men->datasz;
1288 	en.n_type = men->type;
1289 
1290 	return dump_emit(cprm, &en, sizeof(en)) &&
1291 		dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1292 		dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1293 }
1294 
1295 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1296 {
1297 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1298 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1299 	elf->e_ident[EI_DATA] = ELF_DATA;
1300 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1301 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1302 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1303 
1304 	elf->e_type = ET_CORE;
1305 	elf->e_machine = ELF_ARCH;
1306 	elf->e_version = EV_CURRENT;
1307 	elf->e_entry = 0;
1308 	elf->e_phoff = sizeof(struct elfhdr);
1309 	elf->e_shoff = 0;
1310 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1311 	elf->e_ehsize = sizeof(struct elfhdr);
1312 	elf->e_phentsize = sizeof(struct elf_phdr);
1313 	elf->e_phnum = segs;
1314 	elf->e_shentsize = 0;
1315 	elf->e_shnum = 0;
1316 	elf->e_shstrndx = 0;
1317 	return;
1318 }
1319 
1320 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1321 {
1322 	phdr->p_type = PT_NOTE;
1323 	phdr->p_offset = offset;
1324 	phdr->p_vaddr = 0;
1325 	phdr->p_paddr = 0;
1326 	phdr->p_filesz = sz;
1327 	phdr->p_memsz = 0;
1328 	phdr->p_flags = 0;
1329 	phdr->p_align = 0;
1330 	return;
1331 }
1332 
1333 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1334 		unsigned int sz, void *data)
1335 {
1336 	note->name = name;
1337 	note->type = type;
1338 	note->datasz = sz;
1339 	note->data = data;
1340 	return;
1341 }
1342 
1343 /*
1344  * fill up all the fields in prstatus from the given task struct, except
1345  * registers which need to be filled up separately.
1346  */
1347 static void fill_prstatus(struct elf_prstatus *prstatus,
1348 			  struct task_struct *p, long signr)
1349 {
1350 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1351 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1352 	prstatus->pr_sighold = p->blocked.sig[0];
1353 	rcu_read_lock();
1354 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1355 	rcu_read_unlock();
1356 	prstatus->pr_pid = task_pid_vnr(p);
1357 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1358 	prstatus->pr_sid = task_session_vnr(p);
1359 	if (thread_group_leader(p)) {
1360 		struct task_cputime cputime;
1361 
1362 		/*
1363 		 * This is the record for the group leader.  It shows the
1364 		 * group-wide total, not its individual thread total.
1365 		 */
1366 		thread_group_cputime(p, &cputime);
1367 		prstatus->pr_utime = ns_to_kernel_old_timeval(cputime.utime);
1368 		prstatus->pr_stime = ns_to_kernel_old_timeval(cputime.stime);
1369 	} else {
1370 		u64 utime, stime;
1371 
1372 		task_cputime(p, &utime, &stime);
1373 		prstatus->pr_utime = ns_to_kernel_old_timeval(utime);
1374 		prstatus->pr_stime = ns_to_kernel_old_timeval(stime);
1375 	}
1376 	prstatus->pr_cutime = ns_to_kernel_old_timeval(p->signal->cutime);
1377 	prstatus->pr_cstime = ns_to_kernel_old_timeval(p->signal->cstime);
1378 
1379 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1380 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1381 }
1382 
1383 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1384 		       struct mm_struct *mm)
1385 {
1386 	const struct cred *cred;
1387 	unsigned int i, len;
1388 
1389 	/* first copy the parameters from user space */
1390 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1391 
1392 	len = mm->arg_end - mm->arg_start;
1393 	if (len >= ELF_PRARGSZ)
1394 		len = ELF_PRARGSZ - 1;
1395 	if (copy_from_user(&psinfo->pr_psargs,
1396 		           (const char __user *) mm->arg_start, len))
1397 		return -EFAULT;
1398 	for (i = 0; i < len; i++)
1399 		if (psinfo->pr_psargs[i] == 0)
1400 			psinfo->pr_psargs[i] = ' ';
1401 	psinfo->pr_psargs[len] = 0;
1402 
1403 	rcu_read_lock();
1404 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1405 	rcu_read_unlock();
1406 	psinfo->pr_pid = task_pid_vnr(p);
1407 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1408 	psinfo->pr_sid = task_session_vnr(p);
1409 
1410 	i = p->state ? ffz(~p->state) + 1 : 0;
1411 	psinfo->pr_state = i;
1412 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1413 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1414 	psinfo->pr_nice = task_nice(p);
1415 	psinfo->pr_flag = p->flags;
1416 	rcu_read_lock();
1417 	cred = __task_cred(p);
1418 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1419 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1420 	rcu_read_unlock();
1421 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1422 
1423 	return 0;
1424 }
1425 
1426 /* Here is the structure in which status of each thread is captured. */
1427 struct elf_thread_status
1428 {
1429 	struct list_head list;
1430 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1431 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1432 	struct task_struct *thread;
1433 #ifdef ELF_CORE_COPY_XFPREGS
1434 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1435 #endif
1436 	struct memelfnote notes[3];
1437 	int num_notes;
1438 };
1439 
1440 /*
1441  * In order to add the specific thread information for the elf file format,
1442  * we need to keep a linked list of every thread's pr_status and then create
1443  * a single section for them in the final core file.
1444  */
1445 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1446 {
1447 	struct task_struct *p = t->thread;
1448 	int sz = 0;
1449 
1450 	t->num_notes = 0;
1451 
1452 	fill_prstatus(&t->prstatus, p, signr);
1453 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1454 
1455 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1456 		  &t->prstatus);
1457 	t->num_notes++;
1458 	sz += notesize(&t->notes[0]);
1459 
1460 	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1461 	if (t->prstatus.pr_fpvalid) {
1462 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1463 			  &t->fpu);
1464 		t->num_notes++;
1465 		sz += notesize(&t->notes[1]);
1466 	}
1467 
1468 #ifdef ELF_CORE_COPY_XFPREGS
1469 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1470 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1471 			  sizeof(t->xfpu), &t->xfpu);
1472 		t->num_notes++;
1473 		sz += notesize(&t->notes[2]);
1474 	}
1475 #endif
1476 	return sz;
1477 }
1478 
1479 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1480 			     elf_addr_t e_shoff, int segs)
1481 {
1482 	elf->e_shoff = e_shoff;
1483 	elf->e_shentsize = sizeof(*shdr4extnum);
1484 	elf->e_shnum = 1;
1485 	elf->e_shstrndx = SHN_UNDEF;
1486 
1487 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1488 
1489 	shdr4extnum->sh_type = SHT_NULL;
1490 	shdr4extnum->sh_size = elf->e_shnum;
1491 	shdr4extnum->sh_link = elf->e_shstrndx;
1492 	shdr4extnum->sh_info = segs;
1493 }
1494 
1495 /*
1496  * dump the segments for an MMU process
1497  */
1498 static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1499 {
1500 	struct vm_area_struct *vma;
1501 
1502 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1503 #ifdef CONFIG_MMU
1504 		unsigned long addr;
1505 #endif
1506 
1507 		if (!maydump(vma, cprm->mm_flags))
1508 			continue;
1509 
1510 #ifdef CONFIG_MMU
1511 		for (addr = vma->vm_start; addr < vma->vm_end;
1512 							addr += PAGE_SIZE) {
1513 			bool res;
1514 			struct page *page = get_dump_page(addr);
1515 			if (page) {
1516 				void *kaddr = kmap(page);
1517 				res = dump_emit(cprm, kaddr, PAGE_SIZE);
1518 				kunmap(page);
1519 				put_page(page);
1520 			} else {
1521 				res = dump_skip(cprm, PAGE_SIZE);
1522 			}
1523 			if (!res)
1524 				return false;
1525 		}
1526 #else
1527 		if (!dump_emit(cprm, (void *) vma->vm_start,
1528 				vma->vm_end - vma->vm_start))
1529 			return false;
1530 #endif
1531 	}
1532 	return true;
1533 }
1534 
1535 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1536 {
1537 	struct vm_area_struct *vma;
1538 	size_t size = 0;
1539 
1540 	for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1541 		if (maydump(vma, mm_flags))
1542 			size += vma->vm_end - vma->vm_start;
1543 	return size;
1544 }
1545 
1546 /*
1547  * Actual dumper
1548  *
1549  * This is a two-pass process; first we find the offsets of the bits,
1550  * and then they are actually written out.  If we run out of core limit
1551  * we just truncate.
1552  */
1553 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1554 {
1555 #define	NUM_NOTES	6
1556 	int has_dumped = 0;
1557 	int segs;
1558 	int i;
1559 	struct vm_area_struct *vma;
1560 	struct elfhdr *elf = NULL;
1561 	loff_t offset = 0, dataoff;
1562 	int numnote;
1563 	struct memelfnote *notes = NULL;
1564 	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1565 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1566  	LIST_HEAD(thread_list);
1567  	struct list_head *t;
1568 	elf_fpregset_t *fpu = NULL;
1569 #ifdef ELF_CORE_COPY_XFPREGS
1570 	elf_fpxregset_t *xfpu = NULL;
1571 #endif
1572 	int thread_status_size = 0;
1573 	elf_addr_t *auxv;
1574 	struct elf_phdr *phdr4note = NULL;
1575 	struct elf_shdr *shdr4extnum = NULL;
1576 	Elf_Half e_phnum;
1577 	elf_addr_t e_shoff;
1578 	struct core_thread *ct;
1579 	struct elf_thread_status *tmp;
1580 
1581 	/*
1582 	 * We no longer stop all VM operations.
1583 	 *
1584 	 * This is because those proceses that could possibly change map_count
1585 	 * or the mmap / vma pages are now blocked in do_exit on current
1586 	 * finishing this core dump.
1587 	 *
1588 	 * Only ptrace can touch these memory addresses, but it doesn't change
1589 	 * the map_count or the pages allocated. So no possibility of crashing
1590 	 * exists while dumping the mm->vm_next areas to the core file.
1591 	 */
1592 
1593 	/* alloc memory for large data structures: too large to be on stack */
1594 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1595 	if (!elf)
1596 		goto end_coredump;
1597 	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1598 	if (!prstatus)
1599 		goto end_coredump;
1600 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1601 	if (!psinfo)
1602 		goto end_coredump;
1603 	notes = kmalloc_array(NUM_NOTES, sizeof(struct memelfnote),
1604 			      GFP_KERNEL);
1605 	if (!notes)
1606 		goto end_coredump;
1607 	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1608 	if (!fpu)
1609 		goto end_coredump;
1610 #ifdef ELF_CORE_COPY_XFPREGS
1611 	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1612 	if (!xfpu)
1613 		goto end_coredump;
1614 #endif
1615 
1616 	for (ct = current->mm->core_state->dumper.next;
1617 					ct; ct = ct->next) {
1618 		tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1619 		if (!tmp)
1620 			goto end_coredump;
1621 
1622 		tmp->thread = ct->task;
1623 		list_add(&tmp->list, &thread_list);
1624 	}
1625 
1626 	list_for_each(t, &thread_list) {
1627 		struct elf_thread_status *tmp;
1628 		int sz;
1629 
1630 		tmp = list_entry(t, struct elf_thread_status, list);
1631 		sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1632 		thread_status_size += sz;
1633 	}
1634 
1635 	/* now collect the dump for the current */
1636 	fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1637 	elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1638 
1639 	segs = current->mm->map_count;
1640 	segs += elf_core_extra_phdrs();
1641 
1642 	/* for notes section */
1643 	segs++;
1644 
1645 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1646 	 * this, kernel supports extended numbering. Have a look at
1647 	 * include/linux/elf.h for further information. */
1648 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1649 
1650 	/* Set up header */
1651 	fill_elf_fdpic_header(elf, e_phnum);
1652 
1653 	has_dumped = 1;
1654 	/*
1655 	 * Set up the notes in similar form to SVR4 core dumps made
1656 	 * with info from their /proc.
1657 	 */
1658 
1659 	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1660 	fill_psinfo(psinfo, current->group_leader, current->mm);
1661 	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1662 
1663 	numnote = 2;
1664 
1665 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1666 
1667 	i = 0;
1668 	do
1669 		i += 2;
1670 	while (auxv[i - 2] != AT_NULL);
1671 	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1672 		  i * sizeof(elf_addr_t), auxv);
1673 
1674   	/* Try to dump the FPU. */
1675 	if ((prstatus->pr_fpvalid =
1676 	     elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1677 		fill_note(notes + numnote++,
1678 			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1679 #ifdef ELF_CORE_COPY_XFPREGS
1680 	if (elf_core_copy_task_xfpregs(current, xfpu))
1681 		fill_note(notes + numnote++,
1682 			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1683 #endif
1684 
1685 	offset += sizeof(*elf);				/* Elf header */
1686 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1687 
1688 	/* Write notes phdr entry */
1689 	{
1690 		int sz = 0;
1691 
1692 		for (i = 0; i < numnote; i++)
1693 			sz += notesize(notes + i);
1694 
1695 		sz += thread_status_size;
1696 
1697 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1698 		if (!phdr4note)
1699 			goto end_coredump;
1700 
1701 		fill_elf_note_phdr(phdr4note, sz, offset);
1702 		offset += sz;
1703 	}
1704 
1705 	/* Page-align dumped data */
1706 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1707 
1708 	offset += elf_core_vma_data_size(cprm->mm_flags);
1709 	offset += elf_core_extra_data_size();
1710 	e_shoff = offset;
1711 
1712 	if (e_phnum == PN_XNUM) {
1713 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1714 		if (!shdr4extnum)
1715 			goto end_coredump;
1716 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1717 	}
1718 
1719 	offset = dataoff;
1720 
1721 	if (!dump_emit(cprm, elf, sizeof(*elf)))
1722 		goto end_coredump;
1723 
1724 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1725 		goto end_coredump;
1726 
1727 	/* write program headers for segments dump */
1728 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1729 		struct elf_phdr phdr;
1730 		size_t sz;
1731 
1732 		sz = vma->vm_end - vma->vm_start;
1733 
1734 		phdr.p_type = PT_LOAD;
1735 		phdr.p_offset = offset;
1736 		phdr.p_vaddr = vma->vm_start;
1737 		phdr.p_paddr = 0;
1738 		phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1739 		phdr.p_memsz = sz;
1740 		offset += phdr.p_filesz;
1741 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1742 		if (vma->vm_flags & VM_WRITE)
1743 			phdr.p_flags |= PF_W;
1744 		if (vma->vm_flags & VM_EXEC)
1745 			phdr.p_flags |= PF_X;
1746 		phdr.p_align = ELF_EXEC_PAGESIZE;
1747 
1748 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1749 			goto end_coredump;
1750 	}
1751 
1752 	if (!elf_core_write_extra_phdrs(cprm, offset))
1753 		goto end_coredump;
1754 
1755  	/* write out the notes section */
1756 	for (i = 0; i < numnote; i++)
1757 		if (!writenote(notes + i, cprm))
1758 			goto end_coredump;
1759 
1760 	/* write out the thread status notes section */
1761 	list_for_each(t, &thread_list) {
1762 		struct elf_thread_status *tmp =
1763 				list_entry(t, struct elf_thread_status, list);
1764 
1765 		for (i = 0; i < tmp->num_notes; i++)
1766 			if (!writenote(&tmp->notes[i], cprm))
1767 				goto end_coredump;
1768 	}
1769 
1770 	if (!dump_skip(cprm, dataoff - cprm->pos))
1771 		goto end_coredump;
1772 
1773 	if (!elf_fdpic_dump_segments(cprm))
1774 		goto end_coredump;
1775 
1776 	if (!elf_core_write_extra_data(cprm))
1777 		goto end_coredump;
1778 
1779 	if (e_phnum == PN_XNUM) {
1780 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1781 			goto end_coredump;
1782 	}
1783 
1784 	if (cprm->file->f_pos != offset) {
1785 		/* Sanity check */
1786 		printk(KERN_WARNING
1787 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1788 		       cprm->file->f_pos, offset);
1789 	}
1790 
1791 end_coredump:
1792 	while (!list_empty(&thread_list)) {
1793 		struct list_head *tmp = thread_list.next;
1794 		list_del(tmp);
1795 		kfree(list_entry(tmp, struct elf_thread_status, list));
1796 	}
1797 	kfree(phdr4note);
1798 	kfree(elf);
1799 	kfree(prstatus);
1800 	kfree(psinfo);
1801 	kfree(notes);
1802 	kfree(fpu);
1803 	kfree(shdr4extnum);
1804 #ifdef ELF_CORE_COPY_XFPREGS
1805 	kfree(xfpu);
1806 #endif
1807 	return has_dumped;
1808 #undef NUM_NOTES
1809 }
1810 
1811 #endif		/* CONFIG_ELF_CORE */
1812