xref: /openbmc/linux/fs/binfmt_elf_fdpic.c (revision 930beb5a)
1 /* binfmt_elf_fdpic.c: FDPIC ELF binary format
2  *
3  * Copyright (C) 2003, 2004, 2006 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  * Derived from binfmt_elf.c
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License
9  * as published by the Free Software Foundation; either version
10  * 2 of the License, or (at your option) any later version.
11  */
12 
13 #include <linux/module.h>
14 
15 #include <linux/fs.h>
16 #include <linux/stat.h>
17 #include <linux/sched.h>
18 #include <linux/mm.h>
19 #include <linux/mman.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/slab.h>
27 #include <linux/pagemap.h>
28 #include <linux/security.h>
29 #include <linux/highmem.h>
30 #include <linux/highuid.h>
31 #include <linux/personality.h>
32 #include <linux/ptrace.h>
33 #include <linux/init.h>
34 #include <linux/elf.h>
35 #include <linux/elf-fdpic.h>
36 #include <linux/elfcore.h>
37 #include <linux/coredump.h>
38 
39 #include <asm/uaccess.h>
40 #include <asm/param.h>
41 #include <asm/pgalloc.h>
42 
43 typedef char *elf_caddr_t;
44 
45 #if 0
46 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
47 #else
48 #define kdebug(fmt, ...) do {} while(0)
49 #endif
50 
51 #if 0
52 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
53 #else
54 #define kdcore(fmt, ...) do {} while(0)
55 #endif
56 
57 MODULE_LICENSE("GPL");
58 
59 static int load_elf_fdpic_binary(struct linux_binprm *);
60 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
61 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
62 			      struct mm_struct *, const char *);
63 
64 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
65 				   struct elf_fdpic_params *,
66 				   struct elf_fdpic_params *);
67 
68 #ifndef CONFIG_MMU
69 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
70 					    unsigned long *);
71 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
72 						   struct file *,
73 						   struct mm_struct *);
74 #endif
75 
76 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
77 					     struct file *, struct mm_struct *);
78 
79 #ifdef CONFIG_ELF_CORE
80 static int elf_fdpic_core_dump(struct coredump_params *cprm);
81 #endif
82 
83 static struct linux_binfmt elf_fdpic_format = {
84 	.module		= THIS_MODULE,
85 	.load_binary	= load_elf_fdpic_binary,
86 #ifdef CONFIG_ELF_CORE
87 	.core_dump	= elf_fdpic_core_dump,
88 #endif
89 	.min_coredump	= ELF_EXEC_PAGESIZE,
90 };
91 
92 static int __init init_elf_fdpic_binfmt(void)
93 {
94 	register_binfmt(&elf_fdpic_format);
95 	return 0;
96 }
97 
98 static void __exit exit_elf_fdpic_binfmt(void)
99 {
100 	unregister_binfmt(&elf_fdpic_format);
101 }
102 
103 core_initcall(init_elf_fdpic_binfmt);
104 module_exit(exit_elf_fdpic_binfmt);
105 
106 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
107 {
108 	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
109 		return 0;
110 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
111 		return 0;
112 	if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
113 		return 0;
114 	if (!file->f_op->mmap)
115 		return 0;
116 	return 1;
117 }
118 
119 /*****************************************************************************/
120 /*
121  * read the program headers table into memory
122  */
123 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
124 				 struct file *file)
125 {
126 	struct elf32_phdr *phdr;
127 	unsigned long size;
128 	int retval, loop;
129 
130 	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
131 		return -ENOMEM;
132 	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
133 		return -ENOMEM;
134 
135 	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
136 	params->phdrs = kmalloc(size, GFP_KERNEL);
137 	if (!params->phdrs)
138 		return -ENOMEM;
139 
140 	retval = kernel_read(file, params->hdr.e_phoff,
141 			     (char *) params->phdrs, size);
142 	if (unlikely(retval != size))
143 		return retval < 0 ? retval : -ENOEXEC;
144 
145 	/* determine stack size for this binary */
146 	phdr = params->phdrs;
147 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
148 		if (phdr->p_type != PT_GNU_STACK)
149 			continue;
150 
151 		if (phdr->p_flags & PF_X)
152 			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
153 		else
154 			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
155 
156 		params->stack_size = phdr->p_memsz;
157 		break;
158 	}
159 
160 	return 0;
161 }
162 
163 /*****************************************************************************/
164 /*
165  * load an fdpic binary into various bits of memory
166  */
167 static int load_elf_fdpic_binary(struct linux_binprm *bprm)
168 {
169 	struct elf_fdpic_params exec_params, interp_params;
170 	struct pt_regs *regs = current_pt_regs();
171 	struct elf_phdr *phdr;
172 	unsigned long stack_size, entryaddr;
173 #ifdef ELF_FDPIC_PLAT_INIT
174 	unsigned long dynaddr;
175 #endif
176 #ifndef CONFIG_MMU
177 	unsigned long stack_prot;
178 #endif
179 	struct file *interpreter = NULL; /* to shut gcc up */
180 	char *interpreter_name = NULL;
181 	int executable_stack;
182 	int retval, i;
183 
184 	kdebug("____ LOAD %d ____", current->pid);
185 
186 	memset(&exec_params, 0, sizeof(exec_params));
187 	memset(&interp_params, 0, sizeof(interp_params));
188 
189 	exec_params.hdr = *(struct elfhdr *) bprm->buf;
190 	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
191 
192 	/* check that this is a binary we know how to deal with */
193 	retval = -ENOEXEC;
194 	if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
195 		goto error;
196 
197 	/* read the program header table */
198 	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
199 	if (retval < 0)
200 		goto error;
201 
202 	/* scan for a program header that specifies an interpreter */
203 	phdr = exec_params.phdrs;
204 
205 	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
206 		switch (phdr->p_type) {
207 		case PT_INTERP:
208 			retval = -ENOMEM;
209 			if (phdr->p_filesz > PATH_MAX)
210 				goto error;
211 			retval = -ENOENT;
212 			if (phdr->p_filesz < 2)
213 				goto error;
214 
215 			/* read the name of the interpreter into memory */
216 			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
217 			if (!interpreter_name)
218 				goto error;
219 
220 			retval = kernel_read(bprm->file,
221 					     phdr->p_offset,
222 					     interpreter_name,
223 					     phdr->p_filesz);
224 			if (unlikely(retval != phdr->p_filesz)) {
225 				if (retval >= 0)
226 					retval = -ENOEXEC;
227 				goto error;
228 			}
229 
230 			retval = -ENOENT;
231 			if (interpreter_name[phdr->p_filesz - 1] != '\0')
232 				goto error;
233 
234 			kdebug("Using ELF interpreter %s", interpreter_name);
235 
236 			/* replace the program with the interpreter */
237 			interpreter = open_exec(interpreter_name);
238 			retval = PTR_ERR(interpreter);
239 			if (IS_ERR(interpreter)) {
240 				interpreter = NULL;
241 				goto error;
242 			}
243 
244 			/*
245 			 * If the binary is not readable then enforce
246 			 * mm->dumpable = 0 regardless of the interpreter's
247 			 * permissions.
248 			 */
249 			would_dump(bprm, interpreter);
250 
251 			retval = kernel_read(interpreter, 0, bprm->buf,
252 					     BINPRM_BUF_SIZE);
253 			if (unlikely(retval != BINPRM_BUF_SIZE)) {
254 				if (retval >= 0)
255 					retval = -ENOEXEC;
256 				goto error;
257 			}
258 
259 			interp_params.hdr = *((struct elfhdr *) bprm->buf);
260 			break;
261 
262 		case PT_LOAD:
263 #ifdef CONFIG_MMU
264 			if (exec_params.load_addr == 0)
265 				exec_params.load_addr = phdr->p_vaddr;
266 #endif
267 			break;
268 		}
269 
270 	}
271 
272 	if (elf_check_const_displacement(&exec_params.hdr))
273 		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
274 
275 	/* perform insanity checks on the interpreter */
276 	if (interpreter_name) {
277 		retval = -ELIBBAD;
278 		if (!is_elf_fdpic(&interp_params.hdr, interpreter))
279 			goto error;
280 
281 		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
282 
283 		/* read the interpreter's program header table */
284 		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
285 		if (retval < 0)
286 			goto error;
287 	}
288 
289 	stack_size = exec_params.stack_size;
290 	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
291 		executable_stack = EXSTACK_ENABLE_X;
292 	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
293 		executable_stack = EXSTACK_DISABLE_X;
294 	else
295 		executable_stack = EXSTACK_DEFAULT;
296 
297 	if (stack_size == 0) {
298 		stack_size = interp_params.stack_size;
299 		if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
300 			executable_stack = EXSTACK_ENABLE_X;
301 		else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
302 			executable_stack = EXSTACK_DISABLE_X;
303 		else
304 			executable_stack = EXSTACK_DEFAULT;
305 	}
306 
307 	retval = -ENOEXEC;
308 	if (stack_size == 0)
309 		goto error;
310 
311 	if (elf_check_const_displacement(&interp_params.hdr))
312 		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
313 
314 	/* flush all traces of the currently running executable */
315 	retval = flush_old_exec(bprm);
316 	if (retval)
317 		goto error;
318 
319 	/* there's now no turning back... the old userspace image is dead,
320 	 * defunct, deceased, etc. after this point we have to exit via
321 	 * error_kill */
322 	set_personality(PER_LINUX_FDPIC);
323 	if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
324 		current->personality |= READ_IMPLIES_EXEC;
325 
326 	setup_new_exec(bprm);
327 
328 	set_binfmt(&elf_fdpic_format);
329 
330 	current->mm->start_code = 0;
331 	current->mm->end_code = 0;
332 	current->mm->start_stack = 0;
333 	current->mm->start_data = 0;
334 	current->mm->end_data = 0;
335 	current->mm->context.exec_fdpic_loadmap = 0;
336 	current->mm->context.interp_fdpic_loadmap = 0;
337 
338 #ifdef CONFIG_MMU
339 	elf_fdpic_arch_lay_out_mm(&exec_params,
340 				  &interp_params,
341 				  &current->mm->start_stack,
342 				  &current->mm->start_brk);
343 
344 	retval = setup_arg_pages(bprm, current->mm->start_stack,
345 				 executable_stack);
346 	if (retval < 0) {
347 		send_sig(SIGKILL, current, 0);
348 		goto error_kill;
349 	}
350 #endif
351 
352 	/* load the executable and interpreter into memory */
353 	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
354 				    "executable");
355 	if (retval < 0)
356 		goto error_kill;
357 
358 	if (interpreter_name) {
359 		retval = elf_fdpic_map_file(&interp_params, interpreter,
360 					    current->mm, "interpreter");
361 		if (retval < 0) {
362 			printk(KERN_ERR "Unable to load interpreter\n");
363 			goto error_kill;
364 		}
365 
366 		allow_write_access(interpreter);
367 		fput(interpreter);
368 		interpreter = NULL;
369 	}
370 
371 #ifdef CONFIG_MMU
372 	if (!current->mm->start_brk)
373 		current->mm->start_brk = current->mm->end_data;
374 
375 	current->mm->brk = current->mm->start_brk =
376 		PAGE_ALIGN(current->mm->start_brk);
377 
378 #else
379 	/* create a stack and brk area big enough for everyone
380 	 * - the brk heap starts at the bottom and works up
381 	 * - the stack starts at the top and works down
382 	 */
383 	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
384 	if (stack_size < PAGE_SIZE * 2)
385 		stack_size = PAGE_SIZE * 2;
386 
387 	stack_prot = PROT_READ | PROT_WRITE;
388 	if (executable_stack == EXSTACK_ENABLE_X ||
389 	    (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
390 		stack_prot |= PROT_EXEC;
391 
392 	current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
393 					 MAP_PRIVATE | MAP_ANONYMOUS |
394 					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
395 					 0);
396 
397 	if (IS_ERR_VALUE(current->mm->start_brk)) {
398 		retval = current->mm->start_brk;
399 		current->mm->start_brk = 0;
400 		goto error_kill;
401 	}
402 
403 	current->mm->brk = current->mm->start_brk;
404 	current->mm->context.end_brk = current->mm->start_brk;
405 	current->mm->context.end_brk +=
406 		(stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
407 	current->mm->start_stack = current->mm->start_brk + stack_size;
408 #endif
409 
410 	install_exec_creds(bprm);
411 	if (create_elf_fdpic_tables(bprm, current->mm,
412 				    &exec_params, &interp_params) < 0)
413 		goto error_kill;
414 
415 	kdebug("- start_code  %lx", current->mm->start_code);
416 	kdebug("- end_code    %lx", current->mm->end_code);
417 	kdebug("- start_data  %lx", current->mm->start_data);
418 	kdebug("- end_data    %lx", current->mm->end_data);
419 	kdebug("- start_brk   %lx", current->mm->start_brk);
420 	kdebug("- brk         %lx", current->mm->brk);
421 	kdebug("- start_stack %lx", current->mm->start_stack);
422 
423 #ifdef ELF_FDPIC_PLAT_INIT
424 	/*
425 	 * The ABI may specify that certain registers be set up in special
426 	 * ways (on i386 %edx is the address of a DT_FINI function, for
427 	 * example.  This macro performs whatever initialization to
428 	 * the regs structure is required.
429 	 */
430 	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
431 	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
432 			    dynaddr);
433 #endif
434 
435 	/* everything is now ready... get the userspace context ready to roll */
436 	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
437 	start_thread(regs, entryaddr, current->mm->start_stack);
438 
439 	retval = 0;
440 
441 error:
442 	if (interpreter) {
443 		allow_write_access(interpreter);
444 		fput(interpreter);
445 	}
446 	kfree(interpreter_name);
447 	kfree(exec_params.phdrs);
448 	kfree(exec_params.loadmap);
449 	kfree(interp_params.phdrs);
450 	kfree(interp_params.loadmap);
451 	return retval;
452 
453 	/* unrecoverable error - kill the process */
454 error_kill:
455 	send_sig(SIGSEGV, current, 0);
456 	goto error;
457 
458 }
459 
460 /*****************************************************************************/
461 
462 #ifndef ELF_BASE_PLATFORM
463 /*
464  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
465  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
466  * will be copied to the user stack in the same manner as AT_PLATFORM.
467  */
468 #define ELF_BASE_PLATFORM NULL
469 #endif
470 
471 /*
472  * present useful information to the program by shovelling it onto the new
473  * process's stack
474  */
475 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
476 				   struct mm_struct *mm,
477 				   struct elf_fdpic_params *exec_params,
478 				   struct elf_fdpic_params *interp_params)
479 {
480 	const struct cred *cred = current_cred();
481 	unsigned long sp, csp, nitems;
482 	elf_caddr_t __user *argv, *envp;
483 	size_t platform_len = 0, len;
484 	char *k_platform, *k_base_platform;
485 	char __user *u_platform, *u_base_platform, *p;
486 	int loop;
487 	int nr;	/* reset for each csp adjustment */
488 
489 #ifdef CONFIG_MMU
490 	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
491 	 * by the processes running on the same package. One thing we can do is
492 	 * to shuffle the initial stack for them, so we give the architecture
493 	 * an opportunity to do so here.
494 	 */
495 	sp = arch_align_stack(bprm->p);
496 #else
497 	sp = mm->start_stack;
498 
499 	/* stack the program arguments and environment */
500 	if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
501 		return -EFAULT;
502 #endif
503 
504 	/*
505 	 * If this architecture has a platform capability string, copy it
506 	 * to userspace.  In some cases (Sparc), this info is impossible
507 	 * for userspace to get any other way, in others (i386) it is
508 	 * merely difficult.
509 	 */
510 	k_platform = ELF_PLATFORM;
511 	u_platform = NULL;
512 
513 	if (k_platform) {
514 		platform_len = strlen(k_platform) + 1;
515 		sp -= platform_len;
516 		u_platform = (char __user *) sp;
517 		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
518 			return -EFAULT;
519 	}
520 
521 	/*
522 	 * If this architecture has a "base" platform capability
523 	 * string, copy it to userspace.
524 	 */
525 	k_base_platform = ELF_BASE_PLATFORM;
526 	u_base_platform = NULL;
527 
528 	if (k_base_platform) {
529 		platform_len = strlen(k_base_platform) + 1;
530 		sp -= platform_len;
531 		u_base_platform = (char __user *) sp;
532 		if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
533 			return -EFAULT;
534 	}
535 
536 	sp &= ~7UL;
537 
538 	/* stack the load map(s) */
539 	len = sizeof(struct elf32_fdpic_loadmap);
540 	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
541 	sp = (sp - len) & ~7UL;
542 	exec_params->map_addr = sp;
543 
544 	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
545 		return -EFAULT;
546 
547 	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
548 
549 	if (interp_params->loadmap) {
550 		len = sizeof(struct elf32_fdpic_loadmap);
551 		len += sizeof(struct elf32_fdpic_loadseg) *
552 			interp_params->loadmap->nsegs;
553 		sp = (sp - len) & ~7UL;
554 		interp_params->map_addr = sp;
555 
556 		if (copy_to_user((void __user *) sp, interp_params->loadmap,
557 				 len) != 0)
558 			return -EFAULT;
559 
560 		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
561 	}
562 
563 	/* force 16 byte _final_ alignment here for generality */
564 #define DLINFO_ITEMS 15
565 
566 	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
567 		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
568 
569 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
570 		nitems++;
571 
572 	csp = sp;
573 	sp -= nitems * 2 * sizeof(unsigned long);
574 	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
575 	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
576 	sp -= 1 * sizeof(unsigned long);		/* argc */
577 
578 	csp -= sp & 15UL;
579 	sp -= sp & 15UL;
580 
581 	/* put the ELF interpreter info on the stack */
582 #define NEW_AUX_ENT(id, val)						\
583 	do {								\
584 		struct { unsigned long _id, _val; } __user *ent;	\
585 									\
586 		ent = (void __user *) csp;				\
587 		__put_user((id), &ent[nr]._id);				\
588 		__put_user((val), &ent[nr]._val);			\
589 		nr++;							\
590 	} while (0)
591 
592 	nr = 0;
593 	csp -= 2 * sizeof(unsigned long);
594 	NEW_AUX_ENT(AT_NULL, 0);
595 	if (k_platform) {
596 		nr = 0;
597 		csp -= 2 * sizeof(unsigned long);
598 		NEW_AUX_ENT(AT_PLATFORM,
599 			    (elf_addr_t) (unsigned long) u_platform);
600 	}
601 
602 	if (k_base_platform) {
603 		nr = 0;
604 		csp -= 2 * sizeof(unsigned long);
605 		NEW_AUX_ENT(AT_BASE_PLATFORM,
606 			    (elf_addr_t) (unsigned long) u_base_platform);
607 	}
608 
609 	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
610 		nr = 0;
611 		csp -= 2 * sizeof(unsigned long);
612 		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
613 	}
614 
615 	nr = 0;
616 	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
617 	NEW_AUX_ENT(AT_HWCAP,	ELF_HWCAP);
618 #ifdef ELF_HWCAP2
619 	NEW_AUX_ENT(AT_HWCAP2,	ELF_HWCAP2);
620 #endif
621 	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
622 	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
623 	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
624 	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
625 	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
626 	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
627 	NEW_AUX_ENT(AT_FLAGS,	0);
628 	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
629 	NEW_AUX_ENT(AT_UID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
630 	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
631 	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
632 	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
633 	NEW_AUX_ENT(AT_SECURE,	security_bprm_secureexec(bprm));
634 	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
635 
636 #ifdef ARCH_DLINFO
637 	nr = 0;
638 	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
639 
640 	/* ARCH_DLINFO must come last so platform specific code can enforce
641 	 * special alignment requirements on the AUXV if necessary (eg. PPC).
642 	 */
643 	ARCH_DLINFO;
644 #endif
645 #undef NEW_AUX_ENT
646 
647 	/* allocate room for argv[] and envv[] */
648 	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
649 	envp = (elf_caddr_t __user *) csp;
650 	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
651 	argv = (elf_caddr_t __user *) csp;
652 
653 	/* stack argc */
654 	csp -= sizeof(unsigned long);
655 	__put_user(bprm->argc, (unsigned long __user *) csp);
656 
657 	BUG_ON(csp != sp);
658 
659 	/* fill in the argv[] array */
660 #ifdef CONFIG_MMU
661 	current->mm->arg_start = bprm->p;
662 #else
663 	current->mm->arg_start = current->mm->start_stack -
664 		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
665 #endif
666 
667 	p = (char __user *) current->mm->arg_start;
668 	for (loop = bprm->argc; loop > 0; loop--) {
669 		__put_user((elf_caddr_t) p, argv++);
670 		len = strnlen_user(p, MAX_ARG_STRLEN);
671 		if (!len || len > MAX_ARG_STRLEN)
672 			return -EINVAL;
673 		p += len;
674 	}
675 	__put_user(NULL, argv);
676 	current->mm->arg_end = (unsigned long) p;
677 
678 	/* fill in the envv[] array */
679 	current->mm->env_start = (unsigned long) p;
680 	for (loop = bprm->envc; loop > 0; loop--) {
681 		__put_user((elf_caddr_t)(unsigned long) p, envp++);
682 		len = strnlen_user(p, MAX_ARG_STRLEN);
683 		if (!len || len > MAX_ARG_STRLEN)
684 			return -EINVAL;
685 		p += len;
686 	}
687 	__put_user(NULL, envp);
688 	current->mm->env_end = (unsigned long) p;
689 
690 	mm->start_stack = (unsigned long) sp;
691 	return 0;
692 }
693 
694 /*****************************************************************************/
695 /*
696  * transfer the program arguments and environment from the holding pages onto
697  * the stack
698  */
699 #ifndef CONFIG_MMU
700 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
701 					    unsigned long *_sp)
702 {
703 	unsigned long index, stop, sp;
704 	char *src;
705 	int ret = 0;
706 
707 	stop = bprm->p >> PAGE_SHIFT;
708 	sp = *_sp;
709 
710 	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
711 		src = kmap(bprm->page[index]);
712 		sp -= PAGE_SIZE;
713 		if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
714 			ret = -EFAULT;
715 		kunmap(bprm->page[index]);
716 		if (ret < 0)
717 			goto out;
718 	}
719 
720 	*_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
721 
722 out:
723 	return ret;
724 }
725 #endif
726 
727 /*****************************************************************************/
728 /*
729  * load the appropriate binary image (executable or interpreter) into memory
730  * - we assume no MMU is available
731  * - if no other PIC bits are set in params->hdr->e_flags
732  *   - we assume that the LOADable segments in the binary are independently relocatable
733  *   - we assume R/O executable segments are shareable
734  * - else
735  *   - we assume the loadable parts of the image to require fixed displacement
736  *   - the image is not shareable
737  */
738 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
739 			      struct file *file,
740 			      struct mm_struct *mm,
741 			      const char *what)
742 {
743 	struct elf32_fdpic_loadmap *loadmap;
744 #ifdef CONFIG_MMU
745 	struct elf32_fdpic_loadseg *mseg;
746 #endif
747 	struct elf32_fdpic_loadseg *seg;
748 	struct elf32_phdr *phdr;
749 	unsigned long load_addr, stop;
750 	unsigned nloads, tmp;
751 	size_t size;
752 	int loop, ret;
753 
754 	/* allocate a load map table */
755 	nloads = 0;
756 	for (loop = 0; loop < params->hdr.e_phnum; loop++)
757 		if (params->phdrs[loop].p_type == PT_LOAD)
758 			nloads++;
759 
760 	if (nloads == 0)
761 		return -ELIBBAD;
762 
763 	size = sizeof(*loadmap) + nloads * sizeof(*seg);
764 	loadmap = kzalloc(size, GFP_KERNEL);
765 	if (!loadmap)
766 		return -ENOMEM;
767 
768 	params->loadmap = loadmap;
769 
770 	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
771 	loadmap->nsegs = nloads;
772 
773 	load_addr = params->load_addr;
774 	seg = loadmap->segs;
775 
776 	/* map the requested LOADs into the memory space */
777 	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
778 	case ELF_FDPIC_FLAG_CONSTDISP:
779 	case ELF_FDPIC_FLAG_CONTIGUOUS:
780 #ifndef CONFIG_MMU
781 		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
782 		if (ret < 0)
783 			return ret;
784 		break;
785 #endif
786 	default:
787 		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
788 		if (ret < 0)
789 			return ret;
790 		break;
791 	}
792 
793 	/* map the entry point */
794 	if (params->hdr.e_entry) {
795 		seg = loadmap->segs;
796 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
797 			if (params->hdr.e_entry >= seg->p_vaddr &&
798 			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
799 				params->entry_addr =
800 					(params->hdr.e_entry - seg->p_vaddr) +
801 					seg->addr;
802 				break;
803 			}
804 		}
805 	}
806 
807 	/* determine where the program header table has wound up if mapped */
808 	stop = params->hdr.e_phoff;
809 	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
810 	phdr = params->phdrs;
811 
812 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
813 		if (phdr->p_type != PT_LOAD)
814 			continue;
815 
816 		if (phdr->p_offset > params->hdr.e_phoff ||
817 		    phdr->p_offset + phdr->p_filesz < stop)
818 			continue;
819 
820 		seg = loadmap->segs;
821 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
822 			if (phdr->p_vaddr >= seg->p_vaddr &&
823 			    phdr->p_vaddr + phdr->p_filesz <=
824 			    seg->p_vaddr + seg->p_memsz) {
825 				params->ph_addr =
826 					(phdr->p_vaddr - seg->p_vaddr) +
827 					seg->addr +
828 					params->hdr.e_phoff - phdr->p_offset;
829 				break;
830 			}
831 		}
832 		break;
833 	}
834 
835 	/* determine where the dynamic section has wound up if there is one */
836 	phdr = params->phdrs;
837 	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
838 		if (phdr->p_type != PT_DYNAMIC)
839 			continue;
840 
841 		seg = loadmap->segs;
842 		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
843 			if (phdr->p_vaddr >= seg->p_vaddr &&
844 			    phdr->p_vaddr + phdr->p_memsz <=
845 			    seg->p_vaddr + seg->p_memsz) {
846 				params->dynamic_addr =
847 					(phdr->p_vaddr - seg->p_vaddr) +
848 					seg->addr;
849 
850 				/* check the dynamic section contains at least
851 				 * one item, and that the last item is a NULL
852 				 * entry */
853 				if (phdr->p_memsz == 0 ||
854 				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
855 					goto dynamic_error;
856 
857 				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
858 				if (((Elf32_Dyn *)
859 				     params->dynamic_addr)[tmp - 1].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 /*
1193  * Decide whether a segment is worth dumping; default is yes to be
1194  * sure (missing info is worse than too much; etc).
1195  * Personally I'd include everything, and use the coredump limit...
1196  *
1197  * I think we should skip something. But I am not sure how. H.J.
1198  */
1199 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1200 {
1201 	int dump_ok;
1202 
1203 	/* Do not dump I/O mapped devices or special mappings */
1204 	if (vma->vm_flags & VM_IO) {
1205 		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1206 		return 0;
1207 	}
1208 
1209 	/* If we may not read the contents, don't allow us to dump
1210 	 * them either. "dump_write()" can't handle it anyway.
1211 	 */
1212 	if (!(vma->vm_flags & VM_READ)) {
1213 		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1214 		return 0;
1215 	}
1216 
1217 	/* By default, dump shared memory if mapped from an anonymous file. */
1218 	if (vma->vm_flags & VM_SHARED) {
1219 		if (file_inode(vma->vm_file)->i_nlink == 0) {
1220 			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1221 			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1222 			       vma->vm_flags, dump_ok ? "yes" : "no");
1223 			return dump_ok;
1224 		}
1225 
1226 		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1227 		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1228 		       vma->vm_flags, dump_ok ? "yes" : "no");
1229 		return dump_ok;
1230 	}
1231 
1232 #ifdef CONFIG_MMU
1233 	/* By default, if it hasn't been written to, don't write it out */
1234 	if (!vma->anon_vma) {
1235 		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1236 		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1237 		       vma->vm_flags, dump_ok ? "yes" : "no");
1238 		return dump_ok;
1239 	}
1240 #endif
1241 
1242 	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1243 	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1244 	       dump_ok ? "yes" : "no");
1245 	return dump_ok;
1246 }
1247 
1248 /* An ELF note in memory */
1249 struct memelfnote
1250 {
1251 	const char *name;
1252 	int type;
1253 	unsigned int datasz;
1254 	void *data;
1255 };
1256 
1257 static int notesize(struct memelfnote *en)
1258 {
1259 	int sz;
1260 
1261 	sz = sizeof(struct elf_note);
1262 	sz += roundup(strlen(en->name) + 1, 4);
1263 	sz += roundup(en->datasz, 4);
1264 
1265 	return sz;
1266 }
1267 
1268 /* #define DEBUG */
1269 
1270 static int writenote(struct memelfnote *men, struct coredump_params *cprm)
1271 {
1272 	struct elf_note en;
1273 	en.n_namesz = strlen(men->name) + 1;
1274 	en.n_descsz = men->datasz;
1275 	en.n_type = men->type;
1276 
1277 	return dump_emit(cprm, &en, sizeof(en)) &&
1278 		dump_emit(cprm, men->name, en.n_namesz) && dump_align(cprm, 4) &&
1279 		dump_emit(cprm, men->data, men->datasz) && dump_align(cprm, 4);
1280 }
1281 
1282 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1283 {
1284 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1285 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1286 	elf->e_ident[EI_DATA] = ELF_DATA;
1287 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1288 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1289 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1290 
1291 	elf->e_type = ET_CORE;
1292 	elf->e_machine = ELF_ARCH;
1293 	elf->e_version = EV_CURRENT;
1294 	elf->e_entry = 0;
1295 	elf->e_phoff = sizeof(struct elfhdr);
1296 	elf->e_shoff = 0;
1297 	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1298 	elf->e_ehsize = sizeof(struct elfhdr);
1299 	elf->e_phentsize = sizeof(struct elf_phdr);
1300 	elf->e_phnum = segs;
1301 	elf->e_shentsize = 0;
1302 	elf->e_shnum = 0;
1303 	elf->e_shstrndx = 0;
1304 	return;
1305 }
1306 
1307 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1308 {
1309 	phdr->p_type = PT_NOTE;
1310 	phdr->p_offset = offset;
1311 	phdr->p_vaddr = 0;
1312 	phdr->p_paddr = 0;
1313 	phdr->p_filesz = sz;
1314 	phdr->p_memsz = 0;
1315 	phdr->p_flags = 0;
1316 	phdr->p_align = 0;
1317 	return;
1318 }
1319 
1320 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1321 		unsigned int sz, void *data)
1322 {
1323 	note->name = name;
1324 	note->type = type;
1325 	note->datasz = sz;
1326 	note->data = data;
1327 	return;
1328 }
1329 
1330 /*
1331  * fill up all the fields in prstatus from the given task struct, except
1332  * registers which need to be filled up separately.
1333  */
1334 static void fill_prstatus(struct elf_prstatus *prstatus,
1335 			  struct task_struct *p, long signr)
1336 {
1337 	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1338 	prstatus->pr_sigpend = p->pending.signal.sig[0];
1339 	prstatus->pr_sighold = p->blocked.sig[0];
1340 	rcu_read_lock();
1341 	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1342 	rcu_read_unlock();
1343 	prstatus->pr_pid = task_pid_vnr(p);
1344 	prstatus->pr_pgrp = task_pgrp_vnr(p);
1345 	prstatus->pr_sid = task_session_vnr(p);
1346 	if (thread_group_leader(p)) {
1347 		struct task_cputime cputime;
1348 
1349 		/*
1350 		 * This is the record for the group leader.  It shows the
1351 		 * group-wide total, not its individual thread total.
1352 		 */
1353 		thread_group_cputime(p, &cputime);
1354 		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1355 		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1356 	} else {
1357 		cputime_t utime, stime;
1358 
1359 		task_cputime(p, &utime, &stime);
1360 		cputime_to_timeval(utime, &prstatus->pr_utime);
1361 		cputime_to_timeval(stime, &prstatus->pr_stime);
1362 	}
1363 	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1364 	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1365 
1366 	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1367 	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1368 }
1369 
1370 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1371 		       struct mm_struct *mm)
1372 {
1373 	const struct cred *cred;
1374 	unsigned int i, len;
1375 
1376 	/* first copy the parameters from user space */
1377 	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1378 
1379 	len = mm->arg_end - mm->arg_start;
1380 	if (len >= ELF_PRARGSZ)
1381 		len = ELF_PRARGSZ - 1;
1382 	if (copy_from_user(&psinfo->pr_psargs,
1383 		           (const char __user *) mm->arg_start, len))
1384 		return -EFAULT;
1385 	for (i = 0; i < len; i++)
1386 		if (psinfo->pr_psargs[i] == 0)
1387 			psinfo->pr_psargs[i] = ' ';
1388 	psinfo->pr_psargs[len] = 0;
1389 
1390 	rcu_read_lock();
1391 	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1392 	rcu_read_unlock();
1393 	psinfo->pr_pid = task_pid_vnr(p);
1394 	psinfo->pr_pgrp = task_pgrp_vnr(p);
1395 	psinfo->pr_sid = task_session_vnr(p);
1396 
1397 	i = p->state ? ffz(~p->state) + 1 : 0;
1398 	psinfo->pr_state = i;
1399 	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1400 	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1401 	psinfo->pr_nice = task_nice(p);
1402 	psinfo->pr_flag = p->flags;
1403 	rcu_read_lock();
1404 	cred = __task_cred(p);
1405 	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1406 	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1407 	rcu_read_unlock();
1408 	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1409 
1410 	return 0;
1411 }
1412 
1413 /* Here is the structure in which status of each thread is captured. */
1414 struct elf_thread_status
1415 {
1416 	struct list_head list;
1417 	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1418 	elf_fpregset_t fpu;		/* NT_PRFPREG */
1419 	struct task_struct *thread;
1420 #ifdef ELF_CORE_COPY_XFPREGS
1421 	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1422 #endif
1423 	struct memelfnote notes[3];
1424 	int num_notes;
1425 };
1426 
1427 /*
1428  * In order to add the specific thread information for the elf file format,
1429  * we need to keep a linked list of every thread's pr_status and then create
1430  * a single section for them in the final core file.
1431  */
1432 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1433 {
1434 	struct task_struct *p = t->thread;
1435 	int sz = 0;
1436 
1437 	t->num_notes = 0;
1438 
1439 	fill_prstatus(&t->prstatus, p, signr);
1440 	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1441 
1442 	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1443 		  &t->prstatus);
1444 	t->num_notes++;
1445 	sz += notesize(&t->notes[0]);
1446 
1447 	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1448 	if (t->prstatus.pr_fpvalid) {
1449 		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1450 			  &t->fpu);
1451 		t->num_notes++;
1452 		sz += notesize(&t->notes[1]);
1453 	}
1454 
1455 #ifdef ELF_CORE_COPY_XFPREGS
1456 	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1457 		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1458 			  sizeof(t->xfpu), &t->xfpu);
1459 		t->num_notes++;
1460 		sz += notesize(&t->notes[2]);
1461 	}
1462 #endif
1463 	return sz;
1464 }
1465 
1466 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1467 			     elf_addr_t e_shoff, int segs)
1468 {
1469 	elf->e_shoff = e_shoff;
1470 	elf->e_shentsize = sizeof(*shdr4extnum);
1471 	elf->e_shnum = 1;
1472 	elf->e_shstrndx = SHN_UNDEF;
1473 
1474 	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1475 
1476 	shdr4extnum->sh_type = SHT_NULL;
1477 	shdr4extnum->sh_size = elf->e_shnum;
1478 	shdr4extnum->sh_link = elf->e_shstrndx;
1479 	shdr4extnum->sh_info = segs;
1480 }
1481 
1482 /*
1483  * dump the segments for an MMU process
1484  */
1485 static bool elf_fdpic_dump_segments(struct coredump_params *cprm)
1486 {
1487 	struct vm_area_struct *vma;
1488 
1489 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1490 		unsigned long addr;
1491 
1492 		if (!maydump(vma, cprm->mm_flags))
1493 			continue;
1494 
1495 #ifdef CONFIG_MMU
1496 		for (addr = vma->vm_start; addr < vma->vm_end;
1497 							addr += PAGE_SIZE) {
1498 			bool res;
1499 			struct page *page = get_dump_page(addr);
1500 			if (page) {
1501 				void *kaddr = kmap(page);
1502 				res = dump_emit(cprm, kaddr, PAGE_SIZE);
1503 				kunmap(page);
1504 				page_cache_release(page);
1505 			} else {
1506 				res = dump_skip(cprm, PAGE_SIZE);
1507 			}
1508 			if (!res)
1509 				return false;
1510 		}
1511 #else
1512 		if (!dump_emit(cprm, (void *) vma->vm_start,
1513 				vma->vm_end - vma->vm_start))
1514 			return false;
1515 #endif
1516 	}
1517 	return true;
1518 }
1519 
1520 static size_t elf_core_vma_data_size(unsigned long mm_flags)
1521 {
1522 	struct vm_area_struct *vma;
1523 	size_t size = 0;
1524 
1525 	for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1526 		if (maydump(vma, mm_flags))
1527 			size += vma->vm_end - vma->vm_start;
1528 	return size;
1529 }
1530 
1531 /*
1532  * Actual dumper
1533  *
1534  * This is a two-pass process; first we find the offsets of the bits,
1535  * and then they are actually written out.  If we run out of core limit
1536  * we just truncate.
1537  */
1538 static int elf_fdpic_core_dump(struct coredump_params *cprm)
1539 {
1540 #define	NUM_NOTES	6
1541 	int has_dumped = 0;
1542 	mm_segment_t fs;
1543 	int segs;
1544 	int i;
1545 	struct vm_area_struct *vma;
1546 	struct elfhdr *elf = NULL;
1547 	loff_t offset = 0, dataoff;
1548 	int numnote;
1549 	struct memelfnote *notes = NULL;
1550 	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1551 	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1552  	LIST_HEAD(thread_list);
1553  	struct list_head *t;
1554 	elf_fpregset_t *fpu = NULL;
1555 #ifdef ELF_CORE_COPY_XFPREGS
1556 	elf_fpxregset_t *xfpu = NULL;
1557 #endif
1558 	int thread_status_size = 0;
1559 	elf_addr_t *auxv;
1560 	struct elf_phdr *phdr4note = NULL;
1561 	struct elf_shdr *shdr4extnum = NULL;
1562 	Elf_Half e_phnum;
1563 	elf_addr_t e_shoff;
1564 	struct core_thread *ct;
1565 	struct elf_thread_status *tmp;
1566 
1567 	/*
1568 	 * We no longer stop all VM operations.
1569 	 *
1570 	 * This is because those proceses that could possibly change map_count
1571 	 * or the mmap / vma pages are now blocked in do_exit on current
1572 	 * finishing this core dump.
1573 	 *
1574 	 * Only ptrace can touch these memory addresses, but it doesn't change
1575 	 * the map_count or the pages allocated. So no possibility of crashing
1576 	 * exists while dumping the mm->vm_next areas to the core file.
1577 	 */
1578 
1579 	/* alloc memory for large data structures: too large to be on stack */
1580 	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1581 	if (!elf)
1582 		goto cleanup;
1583 	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1584 	if (!prstatus)
1585 		goto cleanup;
1586 	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1587 	if (!psinfo)
1588 		goto cleanup;
1589 	notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1590 	if (!notes)
1591 		goto cleanup;
1592 	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1593 	if (!fpu)
1594 		goto cleanup;
1595 #ifdef ELF_CORE_COPY_XFPREGS
1596 	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1597 	if (!xfpu)
1598 		goto cleanup;
1599 #endif
1600 
1601 	for (ct = current->mm->core_state->dumper.next;
1602 					ct; ct = ct->next) {
1603 		tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1604 		if (!tmp)
1605 			goto cleanup;
1606 
1607 		tmp->thread = ct->task;
1608 		list_add(&tmp->list, &thread_list);
1609 	}
1610 
1611 	list_for_each(t, &thread_list) {
1612 		struct elf_thread_status *tmp;
1613 		int sz;
1614 
1615 		tmp = list_entry(t, struct elf_thread_status, list);
1616 		sz = elf_dump_thread_status(cprm->siginfo->si_signo, tmp);
1617 		thread_status_size += sz;
1618 	}
1619 
1620 	/* now collect the dump for the current */
1621 	fill_prstatus(prstatus, current, cprm->siginfo->si_signo);
1622 	elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
1623 
1624 	segs = current->mm->map_count;
1625 	segs += elf_core_extra_phdrs();
1626 
1627 	/* for notes section */
1628 	segs++;
1629 
1630 	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1631 	 * this, kernel supports extended numbering. Have a look at
1632 	 * include/linux/elf.h for further information. */
1633 	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1634 
1635 	/* Set up header */
1636 	fill_elf_fdpic_header(elf, e_phnum);
1637 
1638 	has_dumped = 1;
1639 	/*
1640 	 * Set up the notes in similar form to SVR4 core dumps made
1641 	 * with info from their /proc.
1642 	 */
1643 
1644 	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1645 	fill_psinfo(psinfo, current->group_leader, current->mm);
1646 	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1647 
1648 	numnote = 2;
1649 
1650 	auxv = (elf_addr_t *) current->mm->saved_auxv;
1651 
1652 	i = 0;
1653 	do
1654 		i += 2;
1655 	while (auxv[i - 2] != AT_NULL);
1656 	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1657 		  i * sizeof(elf_addr_t), auxv);
1658 
1659   	/* Try to dump the FPU. */
1660 	if ((prstatus->pr_fpvalid =
1661 	     elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1662 		fill_note(notes + numnote++,
1663 			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1664 #ifdef ELF_CORE_COPY_XFPREGS
1665 	if (elf_core_copy_task_xfpregs(current, xfpu))
1666 		fill_note(notes + numnote++,
1667 			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1668 #endif
1669 
1670 	fs = get_fs();
1671 	set_fs(KERNEL_DS);
1672 
1673 	offset += sizeof(*elf);				/* Elf header */
1674 	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1675 
1676 	/* Write notes phdr entry */
1677 	{
1678 		int sz = 0;
1679 
1680 		for (i = 0; i < numnote; i++)
1681 			sz += notesize(notes + i);
1682 
1683 		sz += thread_status_size;
1684 
1685 		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1686 		if (!phdr4note)
1687 			goto end_coredump;
1688 
1689 		fill_elf_note_phdr(phdr4note, sz, offset);
1690 		offset += sz;
1691 	}
1692 
1693 	/* Page-align dumped data */
1694 	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1695 
1696 	offset += elf_core_vma_data_size(cprm->mm_flags);
1697 	offset += elf_core_extra_data_size();
1698 	e_shoff = offset;
1699 
1700 	if (e_phnum == PN_XNUM) {
1701 		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1702 		if (!shdr4extnum)
1703 			goto end_coredump;
1704 		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1705 	}
1706 
1707 	offset = dataoff;
1708 
1709 	if (!dump_emit(cprm, elf, sizeof(*elf)))
1710 		goto end_coredump;
1711 
1712 	if (!dump_emit(cprm, phdr4note, sizeof(*phdr4note)))
1713 		goto end_coredump;
1714 
1715 	/* write program headers for segments dump */
1716 	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1717 		struct elf_phdr phdr;
1718 		size_t sz;
1719 
1720 		sz = vma->vm_end - vma->vm_start;
1721 
1722 		phdr.p_type = PT_LOAD;
1723 		phdr.p_offset = offset;
1724 		phdr.p_vaddr = vma->vm_start;
1725 		phdr.p_paddr = 0;
1726 		phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1727 		phdr.p_memsz = sz;
1728 		offset += phdr.p_filesz;
1729 		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1730 		if (vma->vm_flags & VM_WRITE)
1731 			phdr.p_flags |= PF_W;
1732 		if (vma->vm_flags & VM_EXEC)
1733 			phdr.p_flags |= PF_X;
1734 		phdr.p_align = ELF_EXEC_PAGESIZE;
1735 
1736 		if (!dump_emit(cprm, &phdr, sizeof(phdr)))
1737 			goto end_coredump;
1738 	}
1739 
1740 	if (!elf_core_write_extra_phdrs(cprm, offset))
1741 		goto end_coredump;
1742 
1743  	/* write out the notes section */
1744 	for (i = 0; i < numnote; i++)
1745 		if (!writenote(notes + i, cprm))
1746 			goto end_coredump;
1747 
1748 	/* write out the thread status notes section */
1749 	list_for_each(t, &thread_list) {
1750 		struct elf_thread_status *tmp =
1751 				list_entry(t, struct elf_thread_status, list);
1752 
1753 		for (i = 0; i < tmp->num_notes; i++)
1754 			if (!writenote(&tmp->notes[i], cprm))
1755 				goto end_coredump;
1756 	}
1757 
1758 	if (!dump_skip(cprm, dataoff - cprm->written))
1759 		goto end_coredump;
1760 
1761 	if (!elf_fdpic_dump_segments(cprm))
1762 		goto end_coredump;
1763 
1764 	if (!elf_core_write_extra_data(cprm))
1765 		goto end_coredump;
1766 
1767 	if (e_phnum == PN_XNUM) {
1768 		if (!dump_emit(cprm, shdr4extnum, sizeof(*shdr4extnum)))
1769 			goto end_coredump;
1770 	}
1771 
1772 	if (cprm->file->f_pos != offset) {
1773 		/* Sanity check */
1774 		printk(KERN_WARNING
1775 		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1776 		       cprm->file->f_pos, offset);
1777 	}
1778 
1779 end_coredump:
1780 	set_fs(fs);
1781 
1782 cleanup:
1783 	while (!list_empty(&thread_list)) {
1784 		struct list_head *tmp = thread_list.next;
1785 		list_del(tmp);
1786 		kfree(list_entry(tmp, struct elf_thread_status, list));
1787 	}
1788 	kfree(phdr4note);
1789 	kfree(elf);
1790 	kfree(prstatus);
1791 	kfree(psinfo);
1792 	kfree(notes);
1793 	kfree(fpu);
1794 	kfree(shdr4extnum);
1795 #ifdef ELF_CORE_COPY_XFPREGS
1796 	kfree(xfpu);
1797 #endif
1798 	return has_dumped;
1799 #undef NUM_NOTES
1800 }
1801 
1802 #endif		/* CONFIG_ELF_CORE */
1803