xref: /openbmc/linux/fs/binfmt_elf_fdpic.c (revision 52cdded0)
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 #endif
87 	.min_coredump	= ELF_EXEC_PAGESIZE,
88 };
89 
90 static int __init init_elf_fdpic_binfmt(void)
91 {
92 	register_binfmt(&elf_fdpic_format);
93 	return 0;
94 }
95 
96 static void __exit exit_elf_fdpic_binfmt(void)
97 {
98 	unregister_binfmt(&elf_fdpic_format);
99 }
100 
101 core_initcall(init_elf_fdpic_binfmt);
102 module_exit(exit_elf_fdpic_binfmt);
103 
104 static int is_elf(struct elfhdr *hdr, struct file *file)
105 {
106 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
107 		return 0;
108 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
109 		return 0;
110 	if (!elf_check_arch(hdr))
111 		return 0;
112 	if (!file->f_op->mmap)
113 		return 0;
114 	return 1;
115 }
116 
117 #ifndef elf_check_fdpic
118 #define elf_check_fdpic(x) 0
119 #endif
120 
121 #ifndef elf_check_const_displacement
122 #define elf_check_const_displacement(x) 0
123 #endif
124 
125 static int is_constdisp(struct elfhdr *hdr)
126 {
127 	if (!elf_check_fdpic(hdr))
128 		return 1;
129 	if (elf_check_const_displacement(hdr))
130 		return 1;
131 	return 0;
132 }
133 
134 /*****************************************************************************/
135 /*
136  * read the program headers table into memory
137  */
138 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
139 				 struct file *file)
140 {
141 	struct elf32_phdr *phdr;
142 	unsigned long size;
143 	int retval, loop;
144 	loff_t pos = params->hdr.e_phoff;
145 
146 	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
147 		return -ENOMEM;
148 	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
149 		return -ENOMEM;
150 
151 	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
152 	params->phdrs = kmalloc(size, GFP_KERNEL);
153 	if (!params->phdrs)
154 		return -ENOMEM;
155 
156 	retval = kernel_read(file, params->phdrs, size, &pos);
157 	if (unlikely(retval != size))
158 		return retval < 0 ? retval : -ENOEXEC;
159 
160 	/* determine stack size for this binary */
161 	phdr = params->phdrs;
162 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
163 		if (phdr->p_type != PT_GNU_STACK)
164 			continue;
165 
166 		if (phdr->p_flags & PF_X)
167 			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
168 		else
169 			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
170 
171 		params->stack_size = phdr->p_memsz;
172 		break;
173 	}
174 
175 	return 0;
176 }
177 
178 /*****************************************************************************/
179 /*
180  * load an fdpic binary into various bits of memory
181  */
182 static int load_elf_fdpic_binary(struct linux_binprm *bprm)
183 {
184 	struct elf_fdpic_params exec_params, interp_params;
185 	struct pt_regs *regs = current_pt_regs();
186 	struct elf_phdr *phdr;
187 	unsigned long stack_size, entryaddr;
188 #ifdef ELF_FDPIC_PLAT_INIT
189 	unsigned long dynaddr;
190 #endif
191 #ifndef CONFIG_MMU
192 	unsigned long stack_prot;
193 #endif
194 	struct file *interpreter = NULL; /* to shut gcc up */
195 	char *interpreter_name = NULL;
196 	int executable_stack;
197 	int retval, i;
198 	loff_t pos;
199 
200 	kdebug("____ LOAD %d ____", current->pid);
201 
202 	memset(&exec_params, 0, sizeof(exec_params));
203 	memset(&interp_params, 0, sizeof(interp_params));
204 
205 	exec_params.hdr = *(struct elfhdr *) bprm->buf;
206 	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
207 
208 	/* check that this is a binary we know how to deal with */
209 	retval = -ENOEXEC;
210 	if (!is_elf(&exec_params.hdr, bprm->file))
211 		goto error;
212 	if (!elf_check_fdpic(&exec_params.hdr)) {
213 #ifdef CONFIG_MMU
214 		/* binfmt_elf handles non-fdpic elf except on nommu */
215 		goto error;
216 #else
217 		/* nommu can only load ET_DYN (PIE) ELF */
218 		if (exec_params.hdr.e_type != ET_DYN)
219 			goto error;
220 #endif
221 	}
222 
223 	/* read the program header table */
224 	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
225 	if (retval < 0)
226 		goto error;
227 
228 	/* scan for a program header that specifies an interpreter */
229 	phdr = exec_params.phdrs;
230 
231 	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
232 		switch (phdr->p_type) {
233 		case PT_INTERP:
234 			retval = -ENOMEM;
235 			if (phdr->p_filesz > PATH_MAX)
236 				goto error;
237 			retval = -ENOENT;
238 			if (phdr->p_filesz < 2)
239 				goto error;
240 
241 			/* read the name of the interpreter into memory */
242 			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
243 			if (!interpreter_name)
244 				goto error;
245 
246 			pos = phdr->p_offset;
247 			retval = kernel_read(bprm->file, interpreter_name,
248 					     phdr->p_filesz, &pos);
249 			if (unlikely(retval != phdr->p_filesz)) {
250 				if (retval >= 0)
251 					retval = -ENOEXEC;
252 				goto error;
253 			}
254 
255 			retval = -ENOENT;
256 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
257 				goto error;
258 
259 			kdebug("Using ELF interpreter %s", interpreter_name);
260 
261 			/* replace the program with the interpreter */
262 			interpreter = open_exec(interpreter_name);
263 			retval = PTR_ERR(interpreter);
264 			if (IS_ERR(interpreter)) {
265 				interpreter = NULL;
266 				goto error;
267 			}
268 
269 			/*
270 			 * If the binary is not readable then enforce
271 			 * mm->dumpable = 0 regardless of the interpreter's
272 			 * permissions.
273 			 */
274 			would_dump(bprm, interpreter);
275 
276 			pos = 0;
277 			retval = kernel_read(interpreter, bprm->buf,
278 					BINPRM_BUF_SIZE, &pos);
279 			if (unlikely(retval != BINPRM_BUF_SIZE)) {
280 				if (retval >= 0)
281 					retval = -ENOEXEC;
282 				goto error;
283 			}
284 
285 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
286 			break;
287 
288 		case PT_LOAD:
289 #ifdef CONFIG_MMU
290 			if (exec_params.load_addr == 0)
291 				exec_params.load_addr = phdr->p_vaddr;
292 #endif
293 			break;
294 		}
295 
296 	}
297 
298 	if (is_constdisp(&exec_params.hdr))
299 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
300 
301 	/* perform insanity checks on the interpreter */
302 	if (interpreter_name) {
303 		retval = -ELIBBAD;
304 		if (!is_elf(&interp_params.hdr, interpreter))
305 			goto error;
306 
307 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
308 
309 		/* read the interpreter's program header table */
310 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
311 		if (retval < 0)
312 			goto error;
313 	}
314 
315 	stack_size = exec_params.stack_size;
316 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
317 		executable_stack = EXSTACK_ENABLE_X;
318 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
319 		executable_stack = EXSTACK_DISABLE_X;
320 	else
321 		executable_stack = EXSTACK_DEFAULT;
322 
323 	if (stack_size == 0) {
324 		stack_size = interp_params.stack_size;
325 		if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
326 			executable_stack = EXSTACK_ENABLE_X;
327 		else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
328 			executable_stack = EXSTACK_DISABLE_X;
329 		else
330 			executable_stack = EXSTACK_DEFAULT;
331 	}
332 
333 	retval = -ENOEXEC;
334 	if (stack_size == 0)
335 		stack_size = 131072UL; /* same as exec.c's default commit */
336 
337 	if (is_constdisp(&interp_params.hdr))
338 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
339 
340 	/* flush all traces of the currently running executable */
341 	retval = 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 	if (elf_check_fdpic(&exec_params.hdr))
349 		set_personality(PER_LINUX_FDPIC);
350 	else
351 		set_personality(PER_LINUX);
352 	if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
353 		current->personality |= READ_IMPLIES_EXEC;
354 
355 	setup_new_exec(bprm);
356 
357 	set_binfmt(&elf_fdpic_format);
358 
359 	current->mm->start_code = 0;
360 	current->mm->end_code = 0;
361 	current->mm->start_stack = 0;
362 	current->mm->start_data = 0;
363 	current->mm->end_data = 0;
364 	current->mm->context.exec_fdpic_loadmap = 0;
365 	current->mm->context.interp_fdpic_loadmap = 0;
366 
367 #ifdef CONFIG_MMU
368 	elf_fdpic_arch_lay_out_mm(&exec_params,
369 				  &interp_params,
370 				  &current->mm->start_stack,
371 				  &current->mm->start_brk);
372 
373 	retval = setup_arg_pages(bprm, current->mm->start_stack,
374 				 executable_stack);
375 	if (retval < 0)
376 		goto error;
377 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
378 	retval = arch_setup_additional_pages(bprm, !!interpreter_name);
379 	if (retval < 0)
380 		goto error;
381 #endif
382 #endif
383 
384 	/* load the executable and interpreter into memory */
385 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
386 				    "executable");
387 	if (retval < 0)
388 		goto error;
389 
390 	if (interpreter_name) {
391 		retval = elf_fdpic_map_file(&interp_params, interpreter,
392 					    current->mm, "interpreter");
393 		if (retval < 0) {
394 			printk(KERN_ERR "Unable to load interpreter\n");
395 			goto error;
396 		}
397 
398 		allow_write_access(interpreter);
399 		fput(interpreter);
400 		interpreter = NULL;
401 	}
402 
403 #ifdef CONFIG_MMU
404 	if (!current->mm->start_brk)
405 		current->mm->start_brk = current->mm->end_data;
406 
407 	current->mm->brk = current->mm->start_brk =
408 		PAGE_ALIGN(current->mm->start_brk);
409 
410 #else
411 	/* create a stack area and zero-size brk area */
412 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
413 	if (stack_size < PAGE_SIZE * 2)
414 		stack_size = PAGE_SIZE * 2;
415 
416 	stack_prot = PROT_READ | PROT_WRITE;
417 	if (executable_stack == EXSTACK_ENABLE_X ||
418 	    (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
419 		stack_prot |= PROT_EXEC;
420 
421 	current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
422 					 MAP_PRIVATE | MAP_ANONYMOUS |
423 					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
424 					 0);
425 
426 	if (IS_ERR_VALUE(current->mm->start_brk)) {
427 		retval = current->mm->start_brk;
428 		current->mm->start_brk = 0;
429 		goto error;
430 	}
431 
432 	current->mm->brk = current->mm->start_brk;
433 	current->mm->context.end_brk = current->mm->start_brk;
434 	current->mm->start_stack = current->mm->start_brk + stack_size;
435 #endif
436 
437 	if (create_elf_fdpic_tables(bprm, current->mm,
438 				    &exec_params, &interp_params) < 0)
439 		goto error;
440 
441 	kdebug("- start_code  %lx", current->mm->start_code);
442 	kdebug("- end_code    %lx", current->mm->end_code);
443 	kdebug("- start_data  %lx", current->mm->start_data);
444 	kdebug("- end_data    %lx", current->mm->end_data);
445 	kdebug("- start_brk   %lx", current->mm->start_brk);
446 	kdebug("- brk         %lx", current->mm->brk);
447 	kdebug("- start_stack %lx", current->mm->start_stack);
448 
449 #ifdef ELF_FDPIC_PLAT_INIT
450 	/*
451 	 * The ABI may specify that certain registers be set up in special
452 	 * ways (on i386 %edx is the address of a DT_FINI function, for
453 	 * example.  This macro performs whatever initialization to
454 	 * the regs structure is required.
455 	 */
456 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
457 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
458 			    dynaddr);
459 #endif
460 
461 	finalize_exec(bprm);
462 	/* everything is now ready... get the userspace context ready to roll */
463 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
464 	start_thread(regs, entryaddr, current->mm->start_stack);
465 
466 	retval = 0;
467 
468 error:
469 	if (interpreter) {
470 		allow_write_access(interpreter);
471 		fput(interpreter);
472 	}
473 	kfree(interpreter_name);
474 	kfree(exec_params.phdrs);
475 	kfree(exec_params.loadmap);
476 	kfree(interp_params.phdrs);
477 	kfree(interp_params.loadmap);
478 	return retval;
479 }
480 
481 /*****************************************************************************/
482 
483 #ifndef ELF_BASE_PLATFORM
484 /*
485  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
486  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
487  * will be copied to the user stack in the same manner as AT_PLATFORM.
488  */
489 #define ELF_BASE_PLATFORM NULL
490 #endif
491 
492 /*
493  * present useful information to the program by shovelling it onto the new
494  * process's stack
495  */
496 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
497 				   struct mm_struct *mm,
498 				   struct elf_fdpic_params *exec_params,
499 				   struct elf_fdpic_params *interp_params)
500 {
501 	const struct cred *cred = current_cred();
502 	unsigned long sp, csp, nitems;
503 	elf_caddr_t __user *argv, *envp;
504 	size_t platform_len = 0, len;
505 	char *k_platform, *k_base_platform;
506 	char __user *u_platform, *u_base_platform, *p;
507 	int loop;
508 	int nr;	/* reset for each csp adjustment */
509 
510 #ifdef CONFIG_MMU
511 	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
512 	 * by the processes running on the same package. One thing we can do is
513 	 * to shuffle the initial stack for them, so we give the architecture
514 	 * an opportunity to do so here.
515 	 */
516 	sp = arch_align_stack(bprm->p);
517 #else
518 	sp = mm->start_stack;
519 
520 	/* stack the program arguments and environment */
521 	if (transfer_args_to_stack(bprm, &sp) < 0)
522 		return -EFAULT;
523 	sp &= ~15;
524 #endif
525 
526 	/*
527 	 * If this architecture has a platform capability string, copy it
528 	 * to userspace.  In some cases (Sparc), this info is impossible
529 	 * for userspace to get any other way, in others (i386) it is
530 	 * merely difficult.
531 	 */
532 	k_platform = ELF_PLATFORM;
533 	u_platform = NULL;
534 
535 	if (k_platform) {
536 		platform_len = strlen(k_platform) + 1;
537 		sp -= platform_len;
538 		u_platform = (char __user *) sp;
539 		if (copy_to_user(u_platform, k_platform, platform_len) != 0)
540 			return -EFAULT;
541 	}
542 
543 	/*
544 	 * If this architecture has a "base" platform capability
545 	 * string, copy it to userspace.
546 	 */
547 	k_base_platform = ELF_BASE_PLATFORM;
548 	u_base_platform = NULL;
549 
550 	if (k_base_platform) {
551 		platform_len = strlen(k_base_platform) + 1;
552 		sp -= platform_len;
553 		u_base_platform = (char __user *) sp;
554 		if (copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
555 			return -EFAULT;
556 	}
557 
558 	sp &= ~7UL;
559 
560 	/* stack the load map(s) */
561 	len = sizeof(struct elf32_fdpic_loadmap);
562 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
563 	sp = (sp - len) & ~7UL;
564 	exec_params->map_addr = sp;
565 
566 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
567 		return -EFAULT;
568 
569 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
570 
571 	if (interp_params->loadmap) {
572 		len = sizeof(struct elf32_fdpic_loadmap);
573 		len += sizeof(struct elf32_fdpic_loadseg) *
574 			interp_params->loadmap->nsegs;
575 		sp = (sp - len) & ~7UL;
576 		interp_params->map_addr = sp;
577 
578 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
579 				 len) != 0)
580 			return -EFAULT;
581 
582 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
583 	}
584 
585 	/* force 16 byte _final_ alignment here for generality */
586 #define DLINFO_ITEMS 15
587 
588 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
589 		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
590 
591 	if (bprm->have_execfd)
592 		nitems++;
593 
594 	csp = sp;
595 	sp -= nitems * 2 * sizeof(unsigned long);
596 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
597 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
598 	sp -= 1 * sizeof(unsigned long);		/* argc */
599 
600 	csp -= sp & 15UL;
601 	sp -= sp & 15UL;
602 
603 	/* put the ELF interpreter info on the stack */
604 #define NEW_AUX_ENT(id, val)						\
605 	do {								\
606 		struct { unsigned long _id, _val; } __user *ent, v;	\
607 									\
608 		ent = (void __user *) csp;				\
609 		v._id = (id);						\
610 		v._val = (val);						\
611 		if (copy_to_user(ent + nr, &v, sizeof(v)))		\
612 			return -EFAULT;					\
613 		nr++;							\
614 	} while (0)
615 
616 	nr = 0;
617 	csp -= 2 * sizeof(unsigned long);
618 	NEW_AUX_ENT(AT_NULL, 0);
619 	if (k_platform) {
620 		nr = 0;
621 		csp -= 2 * sizeof(unsigned long);
622 		NEW_AUX_ENT(AT_PLATFORM,
623 			    (elf_addr_t) (unsigned long) u_platform);
624 	}
625 
626 	if (k_base_platform) {
627 		nr = 0;
628 		csp -= 2 * sizeof(unsigned long);
629 		NEW_AUX_ENT(AT_BASE_PLATFORM,
630 			    (elf_addr_t) (unsigned long) u_base_platform);
631 	}
632 
633 	if (bprm->have_execfd) {
634 		nr = 0;
635 		csp -= 2 * sizeof(unsigned long);
636 		NEW_AUX_ENT(AT_EXECFD, bprm->execfd);
637 	}
638 
639 	nr = 0;
640 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
641 	NEW_AUX_ENT(AT_HWCAP,	ELF_HWCAP);
642 #ifdef ELF_HWCAP2
643 	NEW_AUX_ENT(AT_HWCAP2,	ELF_HWCAP2);
644 #endif
645 	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
646 	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
647 	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
648 	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
649 	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
650 	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
651 	NEW_AUX_ENT(AT_FLAGS,	0);
652 	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
653 	NEW_AUX_ENT(AT_UID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
654 	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
655 	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
656 	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
657 	NEW_AUX_ENT(AT_SECURE,	bprm->secureexec);
658 	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
659 
660 #ifdef ARCH_DLINFO
661 	nr = 0;
662 	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
663 
664 	/* ARCH_DLINFO must come last so platform specific code can enforce
665 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
666 	 */
667 	ARCH_DLINFO;
668 #endif
669 #undef NEW_AUX_ENT
670 
671 	/* allocate room for argv[] and envv[] */
672 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
673 	envp = (elf_caddr_t __user *) csp;
674 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
675 	argv = (elf_caddr_t __user *) csp;
676 
677 	/* stack argc */
678 	csp -= sizeof(unsigned long);
679 	if (put_user(bprm->argc, (unsigned long __user *) csp))
680 		return -EFAULT;
681 
682 	BUG_ON(csp != sp);
683 
684 	/* fill in the argv[] array */
685 #ifdef CONFIG_MMU
686 	current->mm->arg_start = bprm->p;
687 #else
688 	current->mm->arg_start = current->mm->start_stack -
689 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
690 #endif
691 
692 	p = (char __user *) current->mm->arg_start;
693 	for (loop = bprm->argc; loop > 0; loop--) {
694 		if (put_user((elf_caddr_t) p, argv++))
695 			return -EFAULT;
696 		len = strnlen_user(p, MAX_ARG_STRLEN);
697 		if (!len || len > MAX_ARG_STRLEN)
698 			return -EINVAL;
699 		p += len;
700 	}
701 	if (put_user(NULL, argv))
702 		return -EFAULT;
703 	current->mm->arg_end = (unsigned long) p;
704 
705 	/* fill in the envv[] array */
706 	current->mm->env_start = (unsigned long) p;
707 	for (loop = bprm->envc; loop > 0; loop--) {
708 		if (put_user((elf_caddr_t)(unsigned long) p, envp++))
709 			return -EFAULT;
710 		len = strnlen_user(p, MAX_ARG_STRLEN);
711 		if (!len || len > MAX_ARG_STRLEN)
712 			return -EINVAL;
713 		p += len;
714 	}
715 	if (put_user(NULL, envp))
716 		return -EFAULT;
717 	current->mm->env_end = (unsigned long) p;
718 
719 	mm->start_stack = (unsigned long) sp;
720 	return 0;
721 }
722 
723 /*****************************************************************************/
724 /*
725  * load the appropriate binary image (executable or interpreter) into memory
726  * - we assume no MMU is available
727  * - if no other PIC bits are set in params->hdr->e_flags
728  *   - we assume that the LOADable segments in the binary are independently relocatable
729  *   - we assume R/O executable segments are shareable
730  * - else
731  *   - we assume the loadable parts of the image to require fixed displacement
732  *   - the image is not shareable
733  */
734 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
735 			      struct file *file,
736 			      struct mm_struct *mm,
737 			      const char *what)
738 {
739 	struct elf32_fdpic_loadmap *loadmap;
740 #ifdef CONFIG_MMU
741 	struct elf32_fdpic_loadseg *mseg;
742 #endif
743 	struct elf32_fdpic_loadseg *seg;
744 	struct elf32_phdr *phdr;
745 	unsigned long load_addr, stop;
746 	unsigned nloads, tmp;
747 	size_t size;
748 	int loop, ret;
749 
750 	/* allocate a load map table */
751 	nloads = 0;
752 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
753 		if (params->phdrs[loop].p_type == PT_LOAD)
754 			nloads++;
755 
756 	if (nloads == 0)
757 		return -ELIBBAD;
758 
759 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
760 	loadmap = kzalloc(size, GFP_KERNEL);
761 	if (!loadmap)
762 		return -ENOMEM;
763 
764 	params->loadmap = loadmap;
765 
766 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
767 	loadmap->nsegs = nloads;
768 
769 	load_addr = params->load_addr;
770 	seg = loadmap->segs;
771 
772 	/* map the requested LOADs into the memory space */
773 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
774 	case ELF_FDPIC_FLAG_CONSTDISP:
775 	case ELF_FDPIC_FLAG_CONTIGUOUS:
776 #ifndef CONFIG_MMU
777 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
778 		if (ret < 0)
779 			return ret;
780 		break;
781 #endif
782 	default:
783 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
784 		if (ret < 0)
785 			return ret;
786 		break;
787 	}
788 
789 	/* map the entry point */
790 	if (params->hdr.e_entry) {
791 		seg = loadmap->segs;
792 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
793 			if (params->hdr.e_entry >= seg->p_vaddr &&
794 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
795 				params->entry_addr =
796 					(params->hdr.e_entry - seg->p_vaddr) +
797 					seg->addr;
798 				break;
799 			}
800 		}
801 	}
802 
803 	/* determine where the program header table has wound up if mapped */
804 	stop = params->hdr.e_phoff;
805 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
806 	phdr = params->phdrs;
807 
808 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
809 		if (phdr->p_type != PT_LOAD)
810 			continue;
811 
812 		if (phdr->p_offset > params->hdr.e_phoff ||
813 		    phdr->p_offset + phdr->p_filesz < stop)
814 			continue;
815 
816 		seg = loadmap->segs;
817 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
818 			if (phdr->p_vaddr >= seg->p_vaddr &&
819 			    phdr->p_vaddr + phdr->p_filesz <=
820 			    seg->p_vaddr + seg->p_memsz) {
821 				params->ph_addr =
822 					(phdr->p_vaddr - seg->p_vaddr) +
823 					seg->addr +
824 					params->hdr.e_phoff - phdr->p_offset;
825 				break;
826 			}
827 		}
828 		break;
829 	}
830 
831 	/* determine where the dynamic section has wound up if there is one */
832 	phdr = params->phdrs;
833 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
834 		if (phdr->p_type != PT_DYNAMIC)
835 			continue;
836 
837 		seg = loadmap->segs;
838 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
839 			if (phdr->p_vaddr >= seg->p_vaddr &&
840 			    phdr->p_vaddr + phdr->p_memsz <=
841 			    seg->p_vaddr + seg->p_memsz) {
842 				Elf32_Dyn __user *dyn;
843 				Elf32_Sword d_tag;
844 
845 				params->dynamic_addr =
846 					(phdr->p_vaddr - seg->p_vaddr) +
847 					seg->addr;
848 
849 				/* check the dynamic section contains at least
850 				 * one item, and that the last item is a NULL
851 				 * entry */
852 				if (phdr->p_memsz == 0 ||
853 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
854 					goto dynamic_error;
855 
856 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
857 				dyn = (Elf32_Dyn __user *)params->dynamic_addr;
858 				if (get_user(d_tag, &dyn[tmp - 1].d_tag) ||
859 				    d_tag != 0)
860 					goto dynamic_error;
861 				break;
862 			}
863 		}
864 		break;
865 	}
866 
867 	/* now elide adjacent segments in the load map on MMU linux
868 	 * - on uClinux the holes between may actually be filled with system
869 	 *   stuff or stuff from other processes
870 	 */
871 #ifdef CONFIG_MMU
872 	nloads = loadmap->nsegs;
873 	mseg = loadmap->segs;
874 	seg = mseg + 1;
875 	for (loop = 1; loop < nloads; loop++) {
876 		/* see if we have a candidate for merging */
877 		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
878 			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
879 			if (load_addr == (seg->addr & PAGE_MASK)) {
880 				mseg->p_memsz +=
881 					load_addr -
882 					(mseg->addr + mseg->p_memsz);
883 				mseg->p_memsz += seg->addr & ~PAGE_MASK;
884 				mseg->p_memsz += seg->p_memsz;
885 				loadmap->nsegs--;
886 				continue;
887 			}
888 		}
889 
890 		mseg++;
891 		if (mseg != seg)
892 			*mseg = *seg;
893 	}
894 #endif
895 
896 	kdebug("Mapped Object [%s]:", what);
897 	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
898 	kdebug("- entry    : %lx", params->entry_addr);
899 	kdebug("- PHDR[]   : %lx", params->ph_addr);
900 	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
901 	seg = loadmap->segs;
902 	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
903 		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
904 		       loop,
905 		       seg->addr, seg->addr + seg->p_memsz - 1,
906 		       seg->p_vaddr, seg->p_memsz);
907 
908 	return 0;
909 
910 dynamic_error:
911 	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
912 	       what, file_inode(file)->i_ino);
913 	return -ELIBBAD;
914 }
915 
916 /*****************************************************************************/
917 /*
918  * map a file with constant displacement under uClinux
919  */
920 #ifndef CONFIG_MMU
921 static int elf_fdpic_map_file_constdisp_on_uclinux(
922 	struct elf_fdpic_params *params,
923 	struct file *file,
924 	struct mm_struct *mm)
925 {
926 	struct elf32_fdpic_loadseg *seg;
927 	struct elf32_phdr *phdr;
928 	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
929 	int loop, ret;
930 
931 	load_addr = params->load_addr;
932 	seg = params->loadmap->segs;
933 
934 	/* determine the bounds of the contiguous overall allocation we must
935 	 * make */
936 	phdr = params->phdrs;
937 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
938 		if (params->phdrs[loop].p_type != PT_LOAD)
939 			continue;
940 
941 		if (base > phdr->p_vaddr)
942 			base = phdr->p_vaddr;
943 		if (top < phdr->p_vaddr + phdr->p_memsz)
944 			top = phdr->p_vaddr + phdr->p_memsz;
945 	}
946 
947 	/* allocate one big anon block for everything */
948 	mflags = MAP_PRIVATE;
949 	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
950 		mflags |= MAP_EXECUTABLE;
951 
952 	maddr = vm_mmap(NULL, load_addr, top - base,
953 			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
954 	if (IS_ERR_VALUE(maddr))
955 		return (int) maddr;
956 
957 	if (load_addr != 0)
958 		load_addr += PAGE_ALIGN(top - base);
959 
960 	/* and then load the file segments into it */
961 	phdr = params->phdrs;
962 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
963 		if (params->phdrs[loop].p_type != PT_LOAD)
964 			continue;
965 
966 		seg->addr = maddr + (phdr->p_vaddr - base);
967 		seg->p_vaddr = phdr->p_vaddr;
968 		seg->p_memsz = phdr->p_memsz;
969 
970 		ret = read_code(file, seg->addr, phdr->p_offset,
971 				       phdr->p_filesz);
972 		if (ret < 0)
973 			return ret;
974 
975 		/* map the ELF header address if in this segment */
976 		if (phdr->p_offset == 0)
977 			params->elfhdr_addr = seg->addr;
978 
979 		/* clear any space allocated but not loaded */
980 		if (phdr->p_filesz < phdr->p_memsz) {
981 			if (clear_user((void *) (seg->addr + phdr->p_filesz),
982 				       phdr->p_memsz - phdr->p_filesz))
983 				return -EFAULT;
984 		}
985 
986 		if (mm) {
987 			if (phdr->p_flags & PF_X) {
988 				if (!mm->start_code) {
989 					mm->start_code = seg->addr;
990 					mm->end_code = seg->addr +
991 						phdr->p_memsz;
992 				}
993 			} else if (!mm->start_data) {
994 				mm->start_data = seg->addr;
995 				mm->end_data = seg->addr + phdr->p_memsz;
996 			}
997 		}
998 
999 		seg++;
1000 	}
1001 
1002 	return 0;
1003 }
1004 #endif
1005 
1006 /*****************************************************************************/
1007 /*
1008  * map a binary by direct mmap() of the individual PT_LOAD segments
1009  */
1010 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1011 					     struct file *file,
1012 					     struct mm_struct *mm)
1013 {
1014 	struct elf32_fdpic_loadseg *seg;
1015 	struct elf32_phdr *phdr;
1016 	unsigned long load_addr, delta_vaddr;
1017 	int loop, dvset;
1018 
1019 	load_addr = params->load_addr;
1020 	delta_vaddr = 0;
1021 	dvset = 0;
1022 
1023 	seg = params->loadmap->segs;
1024 
1025 	/* deal with each load segment separately */
1026 	phdr = params->phdrs;
1027 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1028 		unsigned long maddr, disp, excess, excess1;
1029 		int prot = 0, flags;
1030 
1031 		if (phdr->p_type != PT_LOAD)
1032 			continue;
1033 
1034 		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1035 		       (unsigned long) phdr->p_vaddr,
1036 		       (unsigned long) phdr->p_offset,
1037 		       (unsigned long) phdr->p_filesz,
1038 		       (unsigned long) phdr->p_memsz);
1039 
1040 		/* determine the mapping parameters */
1041 		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1042 		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1043 		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1044 
1045 		flags = MAP_PRIVATE | MAP_DENYWRITE;
1046 		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1047 			flags |= MAP_EXECUTABLE;
1048 
1049 		maddr = 0;
1050 
1051 		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1052 		case ELF_FDPIC_FLAG_INDEPENDENT:
1053 			/* PT_LOADs are independently locatable */
1054 			break;
1055 
1056 		case ELF_FDPIC_FLAG_HONOURVADDR:
1057 			/* the specified virtual address must be honoured */
1058 			maddr = phdr->p_vaddr;
1059 			flags |= MAP_FIXED;
1060 			break;
1061 
1062 		case ELF_FDPIC_FLAG_CONSTDISP:
1063 			/* constant displacement
1064 			 * - can be mapped anywhere, but must be mapped as a
1065 			 *   unit
1066 			 */
1067 			if (!dvset) {
1068 				maddr = load_addr;
1069 				delta_vaddr = phdr->p_vaddr;
1070 				dvset = 1;
1071 			} else {
1072 				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1073 				flags |= MAP_FIXED;
1074 			}
1075 			break;
1076 
1077 		case ELF_FDPIC_FLAG_CONTIGUOUS:
1078 			/* contiguity handled later */
1079 			break;
1080 
1081 		default:
1082 			BUG();
1083 		}
1084 
1085 		maddr &= PAGE_MASK;
1086 
1087 		/* create the mapping */
1088 		disp = phdr->p_vaddr & ~PAGE_MASK;
1089 		maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1090 				phdr->p_offset - disp);
1091 
1092 		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1093 		       loop, phdr->p_memsz + disp, prot, flags,
1094 		       phdr->p_offset - disp, maddr);
1095 
1096 		if (IS_ERR_VALUE(maddr))
1097 			return (int) maddr;
1098 
1099 		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1100 		    ELF_FDPIC_FLAG_CONTIGUOUS)
1101 			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1102 
1103 		seg->addr = maddr + disp;
1104 		seg->p_vaddr = phdr->p_vaddr;
1105 		seg->p_memsz = phdr->p_memsz;
1106 
1107 		/* map the ELF header address if in this segment */
1108 		if (phdr->p_offset == 0)
1109 			params->elfhdr_addr = seg->addr;
1110 
1111 		/* clear the bit between beginning of mapping and beginning of
1112 		 * PT_LOAD */
1113 		if (prot & PROT_WRITE && disp > 0) {
1114 			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1115 			if (clear_user((void __user *) maddr, disp))
1116 				return -EFAULT;
1117 			maddr += disp;
1118 		}
1119 
1120 		/* clear any space allocated but not loaded
1121 		 * - on uClinux we can just clear the lot
1122 		 * - on MMU linux we'll get a SIGBUS beyond the last page
1123 		 *   extant in the file
1124 		 */
1125 		excess = phdr->p_memsz - phdr->p_filesz;
1126 		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1127 
1128 #ifdef CONFIG_MMU
1129 		if (excess > excess1) {
1130 			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1131 			unsigned long xmaddr;
1132 
1133 			flags |= MAP_FIXED | MAP_ANONYMOUS;
1134 			xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1135 					 prot, flags, 0);
1136 
1137 			kdebug("mmap[%d] <anon>"
1138 			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1139 			       loop, xaddr, excess - excess1, prot, flags,
1140 			       xmaddr);
1141 
1142 			if (xmaddr != xaddr)
1143 				return -ENOMEM;
1144 		}
1145 
1146 		if (prot & PROT_WRITE && excess1 > 0) {
1147 			kdebug("clear[%d] ad=%lx sz=%lx",
1148 			       loop, maddr + phdr->p_filesz, excess1);
1149 			if (clear_user((void __user *) maddr + phdr->p_filesz,
1150 				       excess1))
1151 				return -EFAULT;
1152 		}
1153 
1154 #else
1155 		if (excess > 0) {
1156 			kdebug("clear[%d] ad=%lx sz=%lx",
1157 			       loop, maddr + phdr->p_filesz, excess);
1158 			if (clear_user((void *) maddr + phdr->p_filesz, excess))
1159 				return -EFAULT;
1160 		}
1161 #endif
1162 
1163 		if (mm) {
1164 			if (phdr->p_flags & PF_X) {
1165 				if (!mm->start_code) {
1166 					mm->start_code = maddr;
1167 					mm->end_code = maddr + phdr->p_memsz;
1168 				}
1169 			} else if (!mm->start_data) {
1170 				mm->start_data = maddr;
1171 				mm->end_data = maddr + phdr->p_memsz;
1172 			}
1173 		}
1174 
1175 		seg++;
1176 	}
1177 
1178 	return 0;
1179 }
1180 
1181 /*****************************************************************************/
1182 /*
1183  * ELF-FDPIC core dumper
1184  *
1185  * Modelled on fs/exec.c:aout_core_dump()
1186  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1187  *
1188  * Modelled on fs/binfmt_elf.c core dumper
1189  */
1190 #ifdef CONFIG_ELF_CORE
1191 
1192 struct elf_prstatus_fdpic
1193 {
1194 	struct elf_siginfo pr_info;	/* Info associated with signal */
1195 	short	pr_cursig;		/* Current signal */
1196 	unsigned long pr_sigpend;	/* Set of pending signals */
1197 	unsigned long pr_sighold;	/* Set of held signals */
1198 	pid_t	pr_pid;
1199 	pid_t	pr_ppid;
1200 	pid_t	pr_pgrp;
1201 	pid_t	pr_sid;
1202 	struct __kernel_old_timeval pr_utime;	/* User time */
1203 	struct __kernel_old_timeval pr_stime;	/* System time */
1204 	struct __kernel_old_timeval pr_cutime;	/* Cumulative user time */
1205 	struct __kernel_old_timeval pr_cstime;	/* Cumulative system time */
1206 	elf_gregset_t pr_reg;	/* GP registers */
1207 	/* When using FDPIC, the loadmap addresses need to be communicated
1208 	 * to GDB in order for GDB to do the necessary relocations.  The
1209 	 * fields (below) used to communicate this information are placed
1210 	 * immediately after ``pr_reg'', so that the loadmap addresses may
1211 	 * be viewed as part of the register set if so desired.
1212 	 */
1213 	unsigned long pr_exec_fdpic_loadmap;
1214 	unsigned long pr_interp_fdpic_loadmap;
1215 	int pr_fpvalid;		/* True if math co-processor being used.  */
1216 };
1217 
1218 /*
1219  * Decide whether a segment is worth dumping; default is yes to be
1220  * sure (missing info is worse than too much; etc).
1221  * Personally I'd include everything, and use the coredump limit...
1222  *
1223  * I think we should skip something. But I am not sure how. H.J.
1224  */
1225 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1226 {
1227 	int dump_ok;
1228 
1229 	/* Do not dump I/O mapped devices or special mappings */
1230 	if (vma->vm_flags & VM_IO) {
1231 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1232 		return 0;
1233 	}
1234 
1235 	/* If we may not read the contents, don't allow us to dump
1236 	 * them either. "dump_write()" can't handle it anyway.
1237 	 */
1238 	if (!(vma->vm_flags & VM_READ)) {
1239 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1240 		return 0;
1241 	}
1242 
1243 	/* support for DAX */
1244 	if (vma_is_dax(vma)) {
1245 		if (vma->vm_flags & VM_SHARED) {
1246 			dump_ok = test_bit(MMF_DUMP_DAX_SHARED, &mm_flags);
1247 			kdcore("%08lx: %08lx: %s (DAX shared)", vma->vm_start,
1248 			       vma->vm_flags, dump_ok ? "yes" : "no");
1249 		} else {
1250 			dump_ok = test_bit(MMF_DUMP_DAX_PRIVATE, &mm_flags);
1251 			kdcore("%08lx: %08lx: %s (DAX private)", vma->vm_start,
1252 			       vma->vm_flags, dump_ok ? "yes" : "no");
1253 		}
1254 		return dump_ok;
1255 	}
1256 
1257 	/* By default, dump shared memory if mapped from an anonymous file. */
1258 	if (vma->vm_flags & VM_SHARED) {
1259 		if (file_inode(vma->vm_file)->i_nlink == 0) {
1260 			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1261 			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1262 			       vma->vm_flags, dump_ok ? "yes" : "no");
1263 			return dump_ok;
1264 		}
1265 
1266 		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1267 		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1268 		       vma->vm_flags, dump_ok ? "yes" : "no");
1269 		return dump_ok;
1270 	}
1271 
1272 #ifdef CONFIG_MMU
1273 	/* By default, if it hasn't been written to, don't write it out */
1274 	if (!vma->anon_vma) {
1275 		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1276 		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1277 		       vma->vm_flags, dump_ok ? "yes" : "no");
1278 		return dump_ok;
1279 	}
1280 #endif
1281 
1282 	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1283 	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1284 	       dump_ok ? "yes" : "no");
1285 	return dump_ok;
1286 }
1287 
1288 /* An ELF note in memory */
1289 struct memelfnote
1290 {
1291 	const char *name;
1292 	int type;
1293 	unsigned int datasz;
1294 	void *data;
1295 };
1296 
1297 static int notesize(struct memelfnote *en)
1298 {
1299 	int sz;
1300 
1301 	sz = sizeof(struct elf_note);
1302 	sz += roundup(strlen(en->name) + 1, 4);
1303 	sz += roundup(en->datasz, 4);
1304 
1305 	return sz;
1306 }
1307 
1308 /* #define DEBUG */
1309 
1310 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1311 {
1312 	struct elf_note en;
1313 	en.n_namesz = strlen(men->name) + 1;
1314 	en.n_descsz = men->datasz;
1315 	en.n_type = men->type;
1316 
1317 	return dump_emit(cprm, &en, sizeof(en)) &&
1318 		dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1319 		dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1320 }
1321 
1322 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1323 {
1324 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1325 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1326 	elf->e_ident[EI_DATA] = ELF_DATA;
1327 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1328 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1329 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1330 
1331 	elf->e_type = ET_CORE;
1332 	elf->e_machine = ELF_ARCH;
1333 	elf->e_version = EV_CURRENT;
1334 	elf->e_entry = 0;
1335 	elf->e_phoff = sizeof(struct elfhdr);
1336 	elf->e_shoff = 0;
1337 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1338 	elf->e_ehsize = sizeof(struct elfhdr);
1339 	elf->e_phentsize = sizeof(struct elf_phdr);
1340 	elf->e_phnum = segs;
1341 	elf->e_shentsize = 0;
1342 	elf->e_shnum = 0;
1343 	elf->e_shstrndx = 0;
1344 	return;
1345 }
1346 
1347 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1348 {
1349 	phdr->p_type = PT_NOTE;
1350 	phdr->p_offset = offset;
1351 	phdr->p_vaddr = 0;
1352 	phdr->p_paddr = 0;
1353 	phdr->p_filesz = sz;
1354 	phdr->p_memsz = 0;
1355 	phdr->p_flags = 0;
1356 	phdr->p_align = 0;
1357 	return;
1358 }
1359 
1360 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1361 		unsigned int sz, void *data)
1362 {
1363 	note->name = name;
1364 	note->type = type;
1365 	note->datasz = sz;
1366 	note->data = data;
1367 	return;
1368 }
1369 
1370 /*
1371  * fill up all the fields in prstatus from the given task struct, except
1372  * registers which need to be filled up separately.
1373  */
1374 static void fill_prstatus(struct elf_prstatus_fdpic *prstatus,
1375 			  struct task_struct *p, long signr)
1376 {
1377 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1378 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1379 	prstatus->pr_sighold = p->blocked.sig[0];
1380 	rcu_read_lock();
1381 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1382 	rcu_read_unlock();
1383 	prstatus->pr_pid = task_pid_vnr(p);
1384 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1385 	prstatus->pr_sid = task_session_vnr(p);
1386 	if (thread_group_leader(p)) {
1387 		struct task_cputime cputime;
1388 
1389 		/*
1390 		 * This is the record for the group leader.  It shows the
1391 		 * group-wide total, not its individual thread total.
1392 		 */
1393 		thread_group_cputime(p, &cputime);
1394 		prstatus->pr_utime = ns_to_kernel_old_timeval(cputime.utime);
1395 		prstatus->pr_stime = ns_to_kernel_old_timeval(cputime.stime);
1396 	} else {
1397 		u64 utime, stime;
1398 
1399 		task_cputime(p, &utime, &stime);
1400 		prstatus->pr_utime = ns_to_kernel_old_timeval(utime);
1401 		prstatus->pr_stime = ns_to_kernel_old_timeval(stime);
1402 	}
1403 	prstatus->pr_cutime = ns_to_kernel_old_timeval(p->signal->cutime);
1404 	prstatus->pr_cstime = ns_to_kernel_old_timeval(p->signal->cstime);
1405 
1406 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1407 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1408 }
1409 
1410 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1411 		       struct mm_struct *mm)
1412 {
1413 	const struct cred *cred;
1414 	unsigned int i, len;
1415 
1416 	/* first copy the parameters from user space */
1417 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1418 
1419 	len = mm->arg_end - mm->arg_start;
1420 	if (len >= ELF_PRARGSZ)
1421 		len = ELF_PRARGSZ - 1;
1422 	if (copy_from_user(&psinfo->pr_psargs,
1423 		           (const char __user *) mm->arg_start, len))
1424 		return -EFAULT;
1425 	for (i = 0; i < len; i++)
1426 		if (psinfo->pr_psargs[i] == 0)
1427 			psinfo->pr_psargs[i] = ' ';
1428 	psinfo->pr_psargs[len] = 0;
1429 
1430 	rcu_read_lock();
1431 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1432 	rcu_read_unlock();
1433 	psinfo->pr_pid = task_pid_vnr(p);
1434 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1435 	psinfo->pr_sid = task_session_vnr(p);
1436 
1437 	i = p->state ? ffz(~p->state) + 1 : 0;
1438 	psinfo->pr_state = i;
1439 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1440 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1441 	psinfo->pr_nice = task_nice(p);
1442 	psinfo->pr_flag = p->flags;
1443 	rcu_read_lock();
1444 	cred = __task_cred(p);
1445 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1446 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1447 	rcu_read_unlock();
1448 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1449 
1450 	return 0;
1451 }
1452 
1453 /* Here is the structure in which status of each thread is captured. */
1454 struct elf_thread_status
1455 {
1456 	struct elf_thread_status *next;
1457 	struct elf_prstatus_fdpic prstatus;	/* NT_PRSTATUS */
1458 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1459 	struct memelfnote notes[2];
1460 	int num_notes;
1461 };
1462 
1463 /*
1464  * In order to add the specific thread information for the elf file format,
1465  * we need to keep a linked list of every thread's pr_status and then create
1466  * a single section for them in the final core file.
1467  */
1468 static struct elf_thread_status *elf_dump_thread_status(long signr, struct task_struct *p, int *sz)
1469 {
1470 	const struct user_regset_view *view = task_user_regset_view(p);
1471 	struct elf_thread_status *t;
1472 	int i, ret;
1473 
1474 	t = kzalloc(sizeof(struct elf_thread_status), GFP_KERNEL);
1475 	if (!t)
1476 		return t;
1477 
1478 	fill_prstatus(&t->prstatus, p, signr);
1479 	regset_get(p, &view->regsets[0],
1480 		   sizeof(t->prstatus.pr_reg), &t->prstatus.pr_reg);
1481 
1482 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1483 		  &t->prstatus);
1484 	t->num_notes++;
1485 	*sz += notesize(&t->notes[0]);
1486 
1487 	for (i = 1; i < view->n; ++i) {
1488 		const struct user_regset *regset = &view->regsets[i];
1489 		if (regset->core_note_type != NT_PRFPREG)
1490 			continue;
1491 		if (regset->active && regset->active(p, regset) <= 0)
1492 			continue;
1493 		ret = regset_get(p, regset, sizeof(t->fpu), &t->fpu);
1494 		if (ret >= 0)
1495 			t->prstatus.pr_fpvalid = 1;
1496 		break;
1497 	}
1498 
1499 	if (t->prstatus.pr_fpvalid) {
1500 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1501 			  &t->fpu);
1502 		t->num_notes++;
1503 		*sz += notesize(&t->notes[1]);
1504 	}
1505 	return t;
1506 }
1507 
1508 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1509 			     elf_addr_t e_shoff, int segs)
1510 {
1511 	elf->e_shoff = e_shoff;
1512 	elf->e_shentsize = sizeof(*shdr4extnum);
1513 	elf->e_shnum = 1;
1514 	elf->e_shstrndx = SHN_UNDEF;
1515 
1516 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1517 
1518 	shdr4extnum->sh_type = SHT_NULL;
1519 	shdr4extnum->sh_size = elf->e_shnum;
1520 	shdr4extnum->sh_link = elf->e_shstrndx;
1521 	shdr4extnum->sh_info = segs;
1522 }
1523 
1524 /*
1525  * dump the segments for an MMU process
1526  */
1527 static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1528 {
1529 	struct vm_area_struct *vma;
1530 
1531 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1532 #ifdef CONFIG_MMU
1533 		unsigned long addr;
1534 #endif
1535 
1536 		if (!maydump(vma, cprm->mm_flags))
1537 			continue;
1538 
1539 #ifdef CONFIG_MMU
1540 		for (addr = vma->vm_start; addr < vma->vm_end;
1541 							addr += PAGE_SIZE) {
1542 			bool res;
1543 			struct page *page = get_dump_page(addr);
1544 			if (page) {
1545 				void *kaddr = kmap(page);
1546 				res = dump_emit(cprm, kaddr, PAGE_SIZE);
1547 				kunmap(page);
1548 				put_page(page);
1549 			} else {
1550 				res = dump_skip(cprm, PAGE_SIZE);
1551 			}
1552 			if (!res)
1553 				return false;
1554 		}
1555 #else
1556 		if (!dump_emit(cprm, (void *) vma->vm_start,
1557 				vma->vm_end - vma->vm_start))
1558 			return false;
1559 #endif
1560 	}
1561 	return true;
1562 }
1563 
1564 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1565 {
1566 	struct vm_area_struct *vma;
1567 	size_t size = 0;
1568 
1569 	for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1570 		if (maydump(vma, mm_flags))
1571 			size += vma->vm_end - vma->vm_start;
1572 	return size;
1573 }
1574 
1575 /*
1576  * Actual dumper
1577  *
1578  * This is a two-pass process; first we find the offsets of the bits,
1579  * and then they are actually written out.  If we run out of core limit
1580  * we just truncate.
1581  */
1582 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1583 {
1584 	int has_dumped = 0;
1585 	int segs;
1586 	int i;
1587 	struct vm_area_struct *vma;
1588 	struct elfhdr *elf = NULL;
1589 	loff_t offset = 0, dataoff;
1590 	struct memelfnote psinfo_note, auxv_note;
1591 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1592 	struct elf_thread_status *thread_list = NULL;
1593 	int thread_status_size = 0;
1594 	elf_addr_t *auxv;
1595 	struct elf_phdr *phdr4note = NULL;
1596 	struct elf_shdr *shdr4extnum = NULL;
1597 	Elf_Half e_phnum;
1598 	elf_addr_t e_shoff;
1599 	struct core_thread *ct;
1600 	struct elf_thread_status *tmp;
1601 
1602 	/*
1603 	 * We no longer stop all VM operations.
1604 	 *
1605 	 * This is because those proceses that could possibly change map_count
1606 	 * or the mmap / vma pages are now blocked in do_exit on current
1607 	 * finishing this core dump.
1608 	 *
1609 	 * Only ptrace can touch these memory addresses, but it doesn't change
1610 	 * the map_count or the pages allocated. So no possibility of crashing
1611 	 * exists while dumping the mm->vm_next areas to the core file.
1612 	 */
1613 
1614 	/* alloc memory for large data structures: too large to be on stack */
1615 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1616 	if (!elf)
1617 		goto end_coredump;
1618 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1619 	if (!psinfo)
1620 		goto end_coredump;
1621 
1622 	for (ct = current->mm->core_state->dumper.next;
1623 					ct; ct = ct->next) {
1624 		tmp = elf_dump_thread_status(cprm->siginfo->si_signo,
1625 					     ct->task, &thread_status_size);
1626 		if (!tmp)
1627 			goto end_coredump;
1628 
1629 		tmp->next = thread_list;
1630 		thread_list = tmp;
1631 	}
1632 
1633 	/* now collect the dump for the current */
1634 	tmp = elf_dump_thread_status(cprm->siginfo->si_signo,
1635 				     current, &thread_status_size);
1636 	if (!tmp)
1637 		goto end_coredump;
1638 	tmp->next = thread_list;
1639 	thread_list = tmp;
1640 
1641 	segs = current->mm->map_count;
1642 	segs += elf_core_extra_phdrs();
1643 
1644 	/* for notes section */
1645 	segs++;
1646 
1647 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1648 	 * this, kernel supports extended numbering. Have a look at
1649 	 * include/linux/elf.h for further information. */
1650 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1651 
1652 	/* Set up header */
1653 	fill_elf_fdpic_header(elf, e_phnum);
1654 
1655 	has_dumped = 1;
1656 	/*
1657 	 * Set up the notes in similar form to SVR4 core dumps made
1658 	 * with info from their /proc.
1659 	 */
1660 
1661 	fill_psinfo(psinfo, current->group_leader, current->mm);
1662 	fill_note(&psinfo_note, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1663 	thread_status_size += notesize(&psinfo_note);
1664 
1665 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1666 	i = 0;
1667 	do
1668 		i += 2;
1669 	while (auxv[i - 2] != AT_NULL);
1670 	fill_note(&auxv_note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1671 	thread_status_size += notesize(&auxv_note);
1672 
1673 	offset = sizeof(*elf);				/* Elf header */
1674 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1675 
1676 	/* Write notes phdr entry */
1677 	phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1678 	if (!phdr4note)
1679 		goto end_coredump;
1680 
1681 	fill_elf_note_phdr(phdr4note, thread_status_size, offset);
1682 	offset += thread_status_size;
1683 
1684 	/* Page-align dumped data */
1685 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1686 
1687 	offset += elf_core_vma_data_size(cprm->mm_flags);
1688 	offset += elf_core_extra_data_size();
1689 	e_shoff = offset;
1690 
1691 	if (e_phnum == PN_XNUM) {
1692 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1693 		if (!shdr4extnum)
1694 			goto end_coredump;
1695 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1696 	}
1697 
1698 	offset = dataoff;
1699 
1700 	if (!dump_emit(cprm, elf, sizeof(*elf)))
1701 		goto end_coredump;
1702 
1703 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1704 		goto end_coredump;
1705 
1706 	/* write program headers for segments dump */
1707 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1708 		struct elf_phdr phdr;
1709 		size_t sz;
1710 
1711 		sz = vma->vm_end - vma->vm_start;
1712 
1713 		phdr.p_type = PT_LOAD;
1714 		phdr.p_offset = offset;
1715 		phdr.p_vaddr = vma->vm_start;
1716 		phdr.p_paddr = 0;
1717 		phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1718 		phdr.p_memsz = sz;
1719 		offset += phdr.p_filesz;
1720 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1721 		if (vma->vm_flags & VM_WRITE)
1722 			phdr.p_flags |= PF_W;
1723 		if (vma->vm_flags & VM_EXEC)
1724 			phdr.p_flags |= PF_X;
1725 		phdr.p_align = ELF_EXEC_PAGESIZE;
1726 
1727 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1728 			goto end_coredump;
1729 	}
1730 
1731 	if (!elf_core_write_extra_phdrs(cprm, offset))
1732 		goto end_coredump;
1733 
1734  	/* write out the notes section */
1735 	if (!writenote(thread_list->notes, cprm))
1736 		goto end_coredump;
1737 	if (!writenote(&psinfo_note, cprm))
1738 		goto end_coredump;
1739 	if (!writenote(&auxv_note, cprm))
1740 		goto end_coredump;
1741 	for (i = 1; i < thread_list->num_notes; i++)
1742 		if (!writenote(thread_list->notes + i, cprm))
1743 			goto end_coredump;
1744 
1745 	/* write out the thread status notes section */
1746 	for (tmp = thread_list->next; tmp; tmp = tmp->next) {
1747 		for (i = 0; i < tmp->num_notes; i++)
1748 			if (!writenote(&tmp->notes[i], cprm))
1749 				goto end_coredump;
1750 	}
1751 
1752 	if (!dump_skip(cprm, dataoff - cprm->pos))
1753 		goto end_coredump;
1754 
1755 	if (!elf_fdpic_dump_segments(cprm))
1756 		goto end_coredump;
1757 
1758 	if (!elf_core_write_extra_data(cprm))
1759 		goto end_coredump;
1760 
1761 	if (e_phnum == PN_XNUM) {
1762 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1763 			goto end_coredump;
1764 	}
1765 
1766 	if (cprm->file->f_pos != offset) {
1767 		/* Sanity check */
1768 		printk(KERN_WARNING
1769 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1770 		       cprm->file->f_pos, offset);
1771 	}
1772 
1773 end_coredump:
1774 	while (thread_list) {
1775 		tmp = thread_list;
1776 		thread_list = thread_list->next;
1777 		kfree(tmp);
1778 	}
1779 	kfree(phdr4note);
1780 	kfree(elf);
1781 	kfree(psinfo);
1782 	kfree(shdr4extnum);
1783 	return has_dumped;
1784 }
1785 
1786 #endif		/* CONFIG_ELF_CORE */
1787