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#include <asm/exec.h>
  43
  44typedef char *elf_caddr_t;
  45
  46#if 0
  47#define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
  48#else
  49#define kdebug(fmt, ...) do {} while(0)
  50#endif
  51
  52#if 0
  53#define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
  54#else
  55#define kdcore(fmt, ...) do {} while(0)
  56#endif
  57
  58MODULE_LICENSE("GPL");
  59
  60static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
  61static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
  62static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
  63			      struct mm_struct *, const char *);
  64
  65static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
  66				   struct elf_fdpic_params *,
  67				   struct elf_fdpic_params *);
  68
  69#ifndef CONFIG_MMU
  70static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
  71					    unsigned long *);
  72static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
  73						   struct file *,
  74						   struct mm_struct *);
  75#endif
  76
  77static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
  78					     struct file *, struct mm_struct *);
  79
  80#ifdef CONFIG_ELF_CORE
  81static int elf_fdpic_core_dump(struct coredump_params *cprm);
  82#endif
  83
  84static struct linux_binfmt elf_fdpic_format = {
  85	.module		= THIS_MODULE,
  86	.load_binary	= load_elf_fdpic_binary,
  87#ifdef CONFIG_ELF_CORE
  88	.core_dump	= elf_fdpic_core_dump,
  89#endif
  90	.min_coredump	= ELF_EXEC_PAGESIZE,
 
  91};
  92
  93static int __init init_elf_fdpic_binfmt(void)
  94{
  95	register_binfmt(&elf_fdpic_format);
  96	return 0;
  97}
  98
  99static void __exit exit_elf_fdpic_binfmt(void)
 100{
 101	unregister_binfmt(&elf_fdpic_format);
 102}
 103
 104core_initcall(init_elf_fdpic_binfmt);
 105module_exit(exit_elf_fdpic_binfmt);
 106
 107static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
 108{
 109	if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
 110		return 0;
 111	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
 112		return 0;
 113	if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
 114		return 0;
 115	if (!file->f_op || !file->f_op->mmap)
 116		return 0;
 117	return 1;
 118}
 119
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 120/*****************************************************************************/
 121/*
 122 * read the program headers table into memory
 123 */
 124static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
 125				 struct file *file)
 126{
 127	struct elf32_phdr *phdr;
 128	unsigned long size;
 129	int retval, loop;
 
 130
 131	if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
 132		return -ENOMEM;
 133	if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
 134		return -ENOMEM;
 135
 136	size = params->hdr.e_phnum * sizeof(struct elf_phdr);
 137	params->phdrs = kmalloc(size, GFP_KERNEL);
 138	if (!params->phdrs)
 139		return -ENOMEM;
 140
 141	retval = kernel_read(file, params->hdr.e_phoff,
 142			     (char *) params->phdrs, size);
 143	if (unlikely(retval != size))
 144		return retval < 0 ? retval : -ENOEXEC;
 145
 146	/* determine stack size for this binary */
 147	phdr = params->phdrs;
 148	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
 149		if (phdr->p_type != PT_GNU_STACK)
 150			continue;
 151
 152		if (phdr->p_flags & PF_X)
 153			params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
 154		else
 155			params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
 156
 157		params->stack_size = phdr->p_memsz;
 158		break;
 159	}
 160
 161	return 0;
 162}
 163
 164/*****************************************************************************/
 165/*
 166 * load an fdpic binary into various bits of memory
 167 */
 168static int load_elf_fdpic_binary(struct linux_binprm *bprm,
 169				 struct pt_regs *regs)
 170{
 171	struct elf_fdpic_params exec_params, interp_params;
 
 172	struct elf_phdr *phdr;
 173	unsigned long stack_size, entryaddr;
 174#ifdef ELF_FDPIC_PLAT_INIT
 175	unsigned long dynaddr;
 176#endif
 177#ifndef CONFIG_MMU
 178	unsigned long stack_prot;
 179#endif
 180	struct file *interpreter = NULL; /* to shut gcc up */
 181	char *interpreter_name = NULL;
 182	int executable_stack;
 183	int retval, i;
 
 184
 185	kdebug("____ LOAD %d ____", current->pid);
 186
 187	memset(&exec_params, 0, sizeof(exec_params));
 188	memset(&interp_params, 0, sizeof(interp_params));
 189
 190	exec_params.hdr = *(struct elfhdr *) bprm->buf;
 191	exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
 192
 193	/* check that this is a binary we know how to deal with */
 194	retval = -ENOEXEC;
 195	if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
 196		goto error;
 
 
 
 
 
 
 
 
 
 
 197
 198	/* read the program header table */
 199	retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
 200	if (retval < 0)
 201		goto error;
 202
 203	/* scan for a program header that specifies an interpreter */
 204	phdr = exec_params.phdrs;
 205
 206	for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
 207		switch (phdr->p_type) {
 208		case PT_INTERP:
 209			retval = -ENOMEM;
 210			if (phdr->p_filesz > PATH_MAX)
 211				goto error;
 212			retval = -ENOENT;
 213			if (phdr->p_filesz < 2)
 214				goto error;
 215
 216			/* read the name of the interpreter into memory */
 217			interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
 218			if (!interpreter_name)
 219				goto error;
 220
 221			retval = kernel_read(bprm->file,
 222					     phdr->p_offset,
 223					     interpreter_name,
 224					     phdr->p_filesz);
 225			if (unlikely(retval != phdr->p_filesz)) {
 226				if (retval >= 0)
 227					retval = -ENOEXEC;
 228				goto error;
 229			}
 230
 231			retval = -ENOENT;
 232			if (interpreter_name[phdr->p_filesz - 1] != '\0')
 233				goto error;
 234
 235			kdebug("Using ELF interpreter %s", interpreter_name);
 236
 237			/* replace the program with the interpreter */
 238			interpreter = open_exec(interpreter_name);
 239			retval = PTR_ERR(interpreter);
 240			if (IS_ERR(interpreter)) {
 241				interpreter = NULL;
 242				goto error;
 243			}
 244
 245			/*
 246			 * If the binary is not readable then enforce
 247			 * mm->dumpable = 0 regardless of the interpreter's
 248			 * permissions.
 249			 */
 250			would_dump(bprm, interpreter);
 251
 252			retval = kernel_read(interpreter, 0, bprm->buf,
 253					     BINPRM_BUF_SIZE);
 
 254			if (unlikely(retval != BINPRM_BUF_SIZE)) {
 255				if (retval >= 0)
 256					retval = -ENOEXEC;
 257				goto error;
 258			}
 259
 260			interp_params.hdr = *((struct elfhdr *) bprm->buf);
 261			break;
 262
 263		case PT_LOAD:
 264#ifdef CONFIG_MMU
 265			if (exec_params.load_addr == 0)
 266				exec_params.load_addr = phdr->p_vaddr;
 267#endif
 268			break;
 269		}
 270
 271	}
 272
 273	if (elf_check_const_displacement(&exec_params.hdr))
 274		exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
 275
 276	/* perform insanity checks on the interpreter */
 277	if (interpreter_name) {
 278		retval = -ELIBBAD;
 279		if (!is_elf_fdpic(&interp_params.hdr, interpreter))
 280			goto error;
 281
 282		interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
 283
 284		/* read the interpreter's program header table */
 285		retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
 286		if (retval < 0)
 287			goto error;
 288	}
 289
 290	stack_size = exec_params.stack_size;
 291	if (exec_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
 292		executable_stack = EXSTACK_ENABLE_X;
 293	else if (exec_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
 294		executable_stack = EXSTACK_DISABLE_X;
 295	else
 296		executable_stack = EXSTACK_DEFAULT;
 297
 298	if (stack_size == 0) {
 299		stack_size = interp_params.stack_size;
 300		if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
 301			executable_stack = EXSTACK_ENABLE_X;
 302		else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
 303			executable_stack = EXSTACK_DISABLE_X;
 304		else
 305			executable_stack = EXSTACK_DEFAULT;
 306	}
 307
 308	retval = -ENOEXEC;
 309	if (stack_size == 0)
 310		goto error;
 311
 312	if (elf_check_const_displacement(&interp_params.hdr))
 313		interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
 314
 315	/* flush all traces of the currently running executable */
 316	retval = flush_old_exec(bprm);
 317	if (retval)
 318		goto error;
 319
 320	/* there's now no turning back... the old userspace image is dead,
 321	 * defunct, deceased, etc. after this point we have to exit via
 322	 * error_kill */
 323	set_personality(PER_LINUX_FDPIC);
 
 
 324	if (elf_read_implies_exec(&exec_params.hdr, executable_stack))
 325		current->personality |= READ_IMPLIES_EXEC;
 326
 327	setup_new_exec(bprm);
 328
 329	set_binfmt(&elf_fdpic_format);
 330
 331	current->mm->start_code = 0;
 332	current->mm->end_code = 0;
 333	current->mm->start_stack = 0;
 334	current->mm->start_data = 0;
 335	current->mm->end_data = 0;
 336	current->mm->context.exec_fdpic_loadmap = 0;
 337	current->mm->context.interp_fdpic_loadmap = 0;
 338
 339#ifdef CONFIG_MMU
 340	elf_fdpic_arch_lay_out_mm(&exec_params,
 341				  &interp_params,
 342				  &current->mm->start_stack,
 343				  &current->mm->start_brk);
 344
 345	retval = setup_arg_pages(bprm, current->mm->start_stack,
 346				 executable_stack);
 347	if (retval < 0) {
 348		send_sig(SIGKILL, current, 0);
 349		goto error_kill;
 350	}
 
 
 
 351#endif
 352
 353	/* load the executable and interpreter into memory */
 354	retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
 355				    "executable");
 356	if (retval < 0)
 357		goto error_kill;
 358
 359	if (interpreter_name) {
 360		retval = elf_fdpic_map_file(&interp_params, interpreter,
 361					    current->mm, "interpreter");
 362		if (retval < 0) {
 363			printk(KERN_ERR "Unable to load interpreter\n");
 364			goto error_kill;
 365		}
 366
 367		allow_write_access(interpreter);
 368		fput(interpreter);
 369		interpreter = NULL;
 370	}
 371
 372#ifdef CONFIG_MMU
 373	if (!current->mm->start_brk)
 374		current->mm->start_brk = current->mm->end_data;
 375
 376	current->mm->brk = current->mm->start_brk =
 377		PAGE_ALIGN(current->mm->start_brk);
 378
 379#else
 380	/* create a stack and brk area big enough for everyone
 381	 * - the brk heap starts at the bottom and works up
 382	 * - the stack starts at the top and works down
 383	 */
 384	stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
 385	if (stack_size < PAGE_SIZE * 2)
 386		stack_size = PAGE_SIZE * 2;
 387
 388	stack_prot = PROT_READ | PROT_WRITE;
 389	if (executable_stack == EXSTACK_ENABLE_X ||
 390	    (executable_stack == EXSTACK_DEFAULT && VM_STACK_FLAGS & VM_EXEC))
 391		stack_prot |= PROT_EXEC;
 392
 393	current->mm->start_brk = vm_mmap(NULL, 0, stack_size, stack_prot,
 394					 MAP_PRIVATE | MAP_ANONYMOUS |
 395					 MAP_UNINITIALIZED | MAP_GROWSDOWN,
 396					 0);
 397
 398	if (IS_ERR_VALUE(current->mm->start_brk)) {
 399		retval = current->mm->start_brk;
 400		current->mm->start_brk = 0;
 401		goto error_kill;
 402	}
 403
 404	current->mm->brk = current->mm->start_brk;
 405	current->mm->context.end_brk = current->mm->start_brk;
 406	current->mm->context.end_brk +=
 407		(stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
 408	current->mm->start_stack = current->mm->start_brk + stack_size;
 409#endif
 410
 411	install_exec_creds(bprm);
 412	if (create_elf_fdpic_tables(bprm, current->mm,
 413				    &exec_params, &interp_params) < 0)
 414		goto error_kill;
 415
 416	kdebug("- start_code  %lx", current->mm->start_code);
 417	kdebug("- end_code    %lx", current->mm->end_code);
 418	kdebug("- start_data  %lx", current->mm->start_data);
 419	kdebug("- end_data    %lx", current->mm->end_data);
 420	kdebug("- start_brk   %lx", current->mm->start_brk);
 421	kdebug("- brk         %lx", current->mm->brk);
 422	kdebug("- start_stack %lx", current->mm->start_stack);
 423
 424#ifdef ELF_FDPIC_PLAT_INIT
 425	/*
 426	 * The ABI may specify that certain registers be set up in special
 427	 * ways (on i386 %edx is the address of a DT_FINI function, for
 428	 * example.  This macro performs whatever initialization to
 429	 * the regs structure is required.
 430	 */
 431	dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
 432	ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
 433			    dynaddr);
 434#endif
 435
 
 436	/* everything is now ready... get the userspace context ready to roll */
 437	entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
 438	start_thread(regs, entryaddr, current->mm->start_stack);
 439
 440	retval = 0;
 441
 442error:
 443	if (interpreter) {
 444		allow_write_access(interpreter);
 445		fput(interpreter);
 446	}
 447	kfree(interpreter_name);
 448	kfree(exec_params.phdrs);
 449	kfree(exec_params.loadmap);
 450	kfree(interp_params.phdrs);
 451	kfree(interp_params.loadmap);
 452	return retval;
 453
 454	/* unrecoverable error - kill the process */
 455error_kill:
 456	send_sig(SIGSEGV, current, 0);
 457	goto error;
 458
 459}
 460
 461/*****************************************************************************/
 462
 463#ifndef ELF_BASE_PLATFORM
 464/*
 465 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
 466 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
 467 * will be copied to the user stack in the same manner as AT_PLATFORM.
 468 */
 469#define ELF_BASE_PLATFORM NULL
 470#endif
 471
 472/*
 473 * present useful information to the program by shovelling it onto the new
 474 * process's stack
 475 */
 476static int create_elf_fdpic_tables(struct linux_binprm *bprm,
 477				   struct mm_struct *mm,
 478				   struct elf_fdpic_params *exec_params,
 479				   struct elf_fdpic_params *interp_params)
 480{
 481	const struct cred *cred = current_cred();
 482	unsigned long sp, csp, nitems;
 483	elf_caddr_t __user *argv, *envp;
 484	size_t platform_len = 0, len;
 485	char *k_platform, *k_base_platform;
 486	char __user *u_platform, *u_base_platform, *p;
 487	long hwcap;
 488	int loop;
 489	int nr;	/* reset for each csp adjustment */
 
 
 490
 491#ifdef CONFIG_MMU
 492	/* In some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
 493	 * by the processes running on the same package. One thing we can do is
 494	 * to shuffle the initial stack for them, so we give the architecture
 495	 * an opportunity to do so here.
 496	 */
 497	sp = arch_align_stack(bprm->p);
 498#else
 499	sp = mm->start_stack;
 500
 501	/* stack the program arguments and environment */
 502	if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
 503		return -EFAULT;
 
 504#endif
 505
 506	hwcap = ELF_HWCAP;
 507
 508	/*
 509	 * If this architecture has a platform capability string, copy it
 510	 * to userspace.  In some cases (Sparc), this info is impossible
 511	 * for userspace to get any other way, in others (i386) it is
 512	 * merely difficult.
 513	 */
 514	k_platform = ELF_PLATFORM;
 515	u_platform = NULL;
 516
 517	if (k_platform) {
 518		platform_len = strlen(k_platform) + 1;
 519		sp -= platform_len;
 520		u_platform = (char __user *) sp;
 521		if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
 522			return -EFAULT;
 523	}
 524
 525	/*
 526	 * If this architecture has a "base" platform capability
 527	 * string, copy it to userspace.
 528	 */
 529	k_base_platform = ELF_BASE_PLATFORM;
 530	u_base_platform = NULL;
 531
 532	if (k_base_platform) {
 533		platform_len = strlen(k_base_platform) + 1;
 534		sp -= platform_len;
 535		u_base_platform = (char __user *) sp;
 536		if (__copy_to_user(u_base_platform, k_base_platform, platform_len) != 0)
 537			return -EFAULT;
 538	}
 539
 540	sp &= ~7UL;
 541
 542	/* stack the load map(s) */
 543	len = sizeof(struct elf32_fdpic_loadmap);
 544	len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
 545	sp = (sp - len) & ~7UL;
 546	exec_params->map_addr = sp;
 547
 548	if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
 549		return -EFAULT;
 550
 551	current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
 552
 553	if (interp_params->loadmap) {
 554		len = sizeof(struct elf32_fdpic_loadmap);
 555		len += sizeof(struct elf32_fdpic_loadseg) *
 556			interp_params->loadmap->nsegs;
 557		sp = (sp - len) & ~7UL;
 558		interp_params->map_addr = sp;
 559
 560		if (copy_to_user((void __user *) sp, interp_params->loadmap,
 561				 len) != 0)
 562			return -EFAULT;
 563
 564		current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
 565	}
 566
 567	/* force 16 byte _final_ alignment here for generality */
 568#define DLINFO_ITEMS 15
 569
 570	nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0) +
 571		(k_base_platform ? 1 : 0) + AT_VECTOR_SIZE_ARCH;
 572
 573	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD)
 574		nitems++;
 
 
 
 575
 576	csp = sp;
 577	sp -= nitems * 2 * sizeof(unsigned long);
 578	sp -= (bprm->envc + 1) * sizeof(char *);	/* envv[] */
 579	sp -= (bprm->argc + 1) * sizeof(char *);	/* argv[] */
 580	sp -= 1 * sizeof(unsigned long);		/* argc */
 581
 582	csp -= sp & 15UL;
 583	sp -= sp & 15UL;
 584
 585	/* put the ELF interpreter info on the stack */
 586#define NEW_AUX_ENT(id, val)						\
 587	do {								\
 588		struct { unsigned long _id, _val; } __user *ent;	\
 589									\
 590		ent = (void __user *) csp;				\
 591		__put_user((id), &ent[nr]._id);				\
 592		__put_user((val), &ent[nr]._val);			\
 593		nr++;							\
 594	} while (0)
 595
 596	nr = 0;
 597	csp -= 2 * sizeof(unsigned long);
 598	NEW_AUX_ENT(AT_NULL, 0);
 599	if (k_platform) {
 600		nr = 0;
 601		csp -= 2 * sizeof(unsigned long);
 602		NEW_AUX_ENT(AT_PLATFORM,
 603			    (elf_addr_t) (unsigned long) u_platform);
 604	}
 605
 606	if (k_base_platform) {
 607		nr = 0;
 608		csp -= 2 * sizeof(unsigned long);
 609		NEW_AUX_ENT(AT_BASE_PLATFORM,
 610			    (elf_addr_t) (unsigned long) u_base_platform);
 611	}
 612
 613	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
 614		nr = 0;
 615		csp -= 2 * sizeof(unsigned long);
 616		NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
 617	}
 618
 619	nr = 0;
 620	csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
 621	NEW_AUX_ENT(AT_HWCAP,	hwcap);
 622	NEW_AUX_ENT(AT_PAGESZ,	PAGE_SIZE);
 623	NEW_AUX_ENT(AT_CLKTCK,	CLOCKS_PER_SEC);
 624	NEW_AUX_ENT(AT_PHDR,	exec_params->ph_addr);
 625	NEW_AUX_ENT(AT_PHENT,	sizeof(struct elf_phdr));
 626	NEW_AUX_ENT(AT_PHNUM,	exec_params->hdr.e_phnum);
 627	NEW_AUX_ENT(AT_BASE,	interp_params->elfhdr_addr);
 628	NEW_AUX_ENT(AT_FLAGS,	0);
 
 
 629	NEW_AUX_ENT(AT_ENTRY,	exec_params->entry_addr);
 630	NEW_AUX_ENT(AT_UID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->uid));
 631	NEW_AUX_ENT(AT_EUID,	(elf_addr_t) from_kuid_munged(cred->user_ns, cred->euid));
 632	NEW_AUX_ENT(AT_GID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->gid));
 633	NEW_AUX_ENT(AT_EGID,	(elf_addr_t) from_kgid_munged(cred->user_ns, cred->egid));
 634	NEW_AUX_ENT(AT_SECURE,	security_bprm_secureexec(bprm));
 635	NEW_AUX_ENT(AT_EXECFN,	bprm->exec);
 636
 637#ifdef ARCH_DLINFO
 638	nr = 0;
 639	csp -= AT_VECTOR_SIZE_ARCH * 2 * sizeof(unsigned long);
 640
 641	/* ARCH_DLINFO must come last so platform specific code can enforce
 642	 * special alignment requirements on the AUXV if necessary (eg. PPC).
 643	 */
 644	ARCH_DLINFO;
 645#endif
 646#undef NEW_AUX_ENT
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 647
 648	/* allocate room for argv[] and envv[] */
 649	csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
 650	envp = (elf_caddr_t __user *) csp;
 651	csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
 652	argv = (elf_caddr_t __user *) csp;
 653
 654	/* stack argc */
 655	csp -= sizeof(unsigned long);
 656	__put_user(bprm->argc, (unsigned long __user *) csp);
 
 657
 658	BUG_ON(csp != sp);
 659
 660	/* fill in the argv[] array */
 661#ifdef CONFIG_MMU
 662	current->mm->arg_start = bprm->p;
 663#else
 664	current->mm->arg_start = current->mm->start_stack -
 665		(MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
 666#endif
 667
 668	p = (char __user *) current->mm->arg_start;
 669	for (loop = bprm->argc; loop > 0; loop--) {
 670		__put_user((elf_caddr_t) p, argv++);
 
 671		len = strnlen_user(p, MAX_ARG_STRLEN);
 672		if (!len || len > MAX_ARG_STRLEN)
 673			return -EINVAL;
 674		p += len;
 675	}
 676	__put_user(NULL, argv);
 
 677	current->mm->arg_end = (unsigned long) p;
 678
 679	/* fill in the envv[] array */
 680	current->mm->env_start = (unsigned long) p;
 681	for (loop = bprm->envc; loop > 0; loop--) {
 682		__put_user((elf_caddr_t)(unsigned long) p, envp++);
 
 683		len = strnlen_user(p, MAX_ARG_STRLEN);
 684		if (!len || len > MAX_ARG_STRLEN)
 685			return -EINVAL;
 686		p += len;
 687	}
 688	__put_user(NULL, envp);
 
 689	current->mm->env_end = (unsigned long) p;
 690
 691	mm->start_stack = (unsigned long) sp;
 692	return 0;
 693}
 694
 695/*****************************************************************************/
 696/*
 697 * transfer the program arguments and environment from the holding pages onto
 698 * the stack
 699 */
 700#ifndef CONFIG_MMU
 701static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
 702					    unsigned long *_sp)
 703{
 704	unsigned long index, stop, sp;
 705	char *src;
 706	int ret = 0;
 707
 708	stop = bprm->p >> PAGE_SHIFT;
 709	sp = *_sp;
 710
 711	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 712		src = kmap(bprm->page[index]);
 713		sp -= PAGE_SIZE;
 714		if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
 715			ret = -EFAULT;
 716		kunmap(bprm->page[index]);
 717		if (ret < 0)
 718			goto out;
 719	}
 720
 721	*_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
 722
 723out:
 724	return ret;
 725}
 726#endif
 727
 728/*****************************************************************************/
 729/*
 730 * load the appropriate binary image (executable or interpreter) into memory
 731 * - we assume no MMU is available
 732 * - if no other PIC bits are set in params->hdr->e_flags
 733 *   - we assume that the LOADable segments in the binary are independently relocatable
 734 *   - we assume R/O executable segments are shareable
 735 * - else
 736 *   - we assume the loadable parts of the image to require fixed displacement
 737 *   - the image is not shareable
 738 */
 739static int elf_fdpic_map_file(struct elf_fdpic_params *params,
 740			      struct file *file,
 741			      struct mm_struct *mm,
 742			      const char *what)
 743{
 744	struct elf32_fdpic_loadmap *loadmap;
 745#ifdef CONFIG_MMU
 746	struct elf32_fdpic_loadseg *mseg;
 
 747#endif
 748	struct elf32_fdpic_loadseg *seg;
 749	struct elf32_phdr *phdr;
 750	unsigned long load_addr, stop;
 751	unsigned nloads, tmp;
 752	size_t size;
 753	int loop, ret;
 754
 755	/* allocate a load map table */
 756	nloads = 0;
 757	for (loop = 0; loop < params->hdr.e_phnum; loop++)
 758		if (params->phdrs[loop].p_type == PT_LOAD)
 759			nloads++;
 760
 761	if (nloads == 0)
 762		return -ELIBBAD;
 763
 764	size = sizeof(*loadmap) + nloads * sizeof(*seg);
 765	loadmap = kzalloc(size, GFP_KERNEL);
 766	if (!loadmap)
 767		return -ENOMEM;
 768
 769	params->loadmap = loadmap;
 770
 771	loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
 772	loadmap->nsegs = nloads;
 773
 774	load_addr = params->load_addr;
 775	seg = loadmap->segs;
 776
 777	/* map the requested LOADs into the memory space */
 778	switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
 779	case ELF_FDPIC_FLAG_CONSTDISP:
 780	case ELF_FDPIC_FLAG_CONTIGUOUS:
 781#ifndef CONFIG_MMU
 782		ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
 783		if (ret < 0)
 784			return ret;
 785		break;
 786#endif
 787	default:
 788		ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
 789		if (ret < 0)
 790			return ret;
 791		break;
 792	}
 793
 794	/* map the entry point */
 795	if (params->hdr.e_entry) {
 796		seg = loadmap->segs;
 797		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
 798			if (params->hdr.e_entry >= seg->p_vaddr &&
 799			    params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
 800				params->entry_addr =
 801					(params->hdr.e_entry - seg->p_vaddr) +
 802					seg->addr;
 803				break;
 804			}
 805		}
 806	}
 807
 808	/* determine where the program header table has wound up if mapped */
 809	stop = params->hdr.e_phoff;
 810	stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
 811	phdr = params->phdrs;
 812
 813	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
 814		if (phdr->p_type != PT_LOAD)
 815			continue;
 816
 817		if (phdr->p_offset > params->hdr.e_phoff ||
 818		    phdr->p_offset + phdr->p_filesz < stop)
 819			continue;
 820
 821		seg = loadmap->segs;
 822		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
 823			if (phdr->p_vaddr >= seg->p_vaddr &&
 824			    phdr->p_vaddr + phdr->p_filesz <=
 825			    seg->p_vaddr + seg->p_memsz) {
 826				params->ph_addr =
 827					(phdr->p_vaddr - seg->p_vaddr) +
 828					seg->addr +
 829					params->hdr.e_phoff - phdr->p_offset;
 830				break;
 831			}
 832		}
 833		break;
 834	}
 835
 836	/* determine where the dynamic section has wound up if there is one */
 837	phdr = params->phdrs;
 838	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
 839		if (phdr->p_type != PT_DYNAMIC)
 840			continue;
 841
 842		seg = loadmap->segs;
 843		for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
 844			if (phdr->p_vaddr >= seg->p_vaddr &&
 845			    phdr->p_vaddr + phdr->p_memsz <=
 846			    seg->p_vaddr + seg->p_memsz) {
 
 
 
 847				params->dynamic_addr =
 848					(phdr->p_vaddr - seg->p_vaddr) +
 849					seg->addr;
 850
 851				/* check the dynamic section contains at least
 852				 * one item, and that the last item is a NULL
 853				 * entry */
 854				if (phdr->p_memsz == 0 ||
 855				    phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
 856					goto dynamic_error;
 857
 858				tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
 859				if (((Elf32_Dyn *)
 860				     params->dynamic_addr)[tmp - 1].d_tag != 0)
 
 861					goto dynamic_error;
 862				break;
 863			}
 864		}
 865		break;
 866	}
 867
 868	/* now elide adjacent segments in the load map on MMU linux
 869	 * - on uClinux the holes between may actually be filled with system
 870	 *   stuff or stuff from other processes
 871	 */
 872#ifdef CONFIG_MMU
 873	nloads = loadmap->nsegs;
 874	mseg = loadmap->segs;
 875	seg = mseg + 1;
 876	for (loop = 1; loop < nloads; loop++) {
 877		/* see if we have a candidate for merging */
 878		if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
 879			load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
 880			if (load_addr == (seg->addr & PAGE_MASK)) {
 881				mseg->p_memsz +=
 882					load_addr -
 883					(mseg->addr + mseg->p_memsz);
 884				mseg->p_memsz += seg->addr & ~PAGE_MASK;
 885				mseg->p_memsz += seg->p_memsz;
 886				loadmap->nsegs--;
 887				continue;
 888			}
 889		}
 890
 891		mseg++;
 892		if (mseg != seg)
 893			*mseg = *seg;
 894	}
 895#endif
 896
 897	kdebug("Mapped Object [%s]:", what);
 898	kdebug("- elfhdr   : %lx", params->elfhdr_addr);
 899	kdebug("- entry    : %lx", params->entry_addr);
 900	kdebug("- PHDR[]   : %lx", params->ph_addr);
 901	kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
 902	seg = loadmap->segs;
 903	for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
 904		kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
 905		       loop,
 906		       seg->addr, seg->addr + seg->p_memsz - 1,
 907		       seg->p_vaddr, seg->p_memsz);
 
 
 908
 909	return 0;
 910
 911dynamic_error:
 912	printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
 913	       what, file->f_path.dentry->d_inode->i_ino);
 914	return -ELIBBAD;
 915}
 916
 917/*****************************************************************************/
 918/*
 919 * map a file with constant displacement under uClinux
 920 */
 921#ifndef CONFIG_MMU
 922static int elf_fdpic_map_file_constdisp_on_uclinux(
 923	struct elf_fdpic_params *params,
 924	struct file *file,
 925	struct mm_struct *mm)
 926{
 927	struct elf32_fdpic_loadseg *seg;
 928	struct elf32_phdr *phdr;
 929	unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
 930	loff_t fpos;
 931	int loop, ret;
 932
 933	load_addr = params->load_addr;
 934	seg = params->loadmap->segs;
 935
 936	/* determine the bounds of the contiguous overall allocation we must
 937	 * make */
 938	phdr = params->phdrs;
 939	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
 940		if (params->phdrs[loop].p_type != PT_LOAD)
 941			continue;
 942
 943		if (base > phdr->p_vaddr)
 944			base = phdr->p_vaddr;
 945		if (top < phdr->p_vaddr + phdr->p_memsz)
 946			top = phdr->p_vaddr + phdr->p_memsz;
 947	}
 948
 949	/* allocate one big anon block for everything */
 950	mflags = MAP_PRIVATE;
 951	if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
 952		mflags |= MAP_EXECUTABLE;
 953
 954	maddr = vm_mmap(NULL, load_addr, top - base,
 955			PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
 956	if (IS_ERR_VALUE(maddr))
 957		return (int) maddr;
 958
 959	if (load_addr != 0)
 960		load_addr += PAGE_ALIGN(top - base);
 961
 962	/* and then load the file segments into it */
 963	phdr = params->phdrs;
 964	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
 965		if (params->phdrs[loop].p_type != PT_LOAD)
 966			continue;
 967
 968		fpos = phdr->p_offset;
 969
 970		seg->addr = maddr + (phdr->p_vaddr - base);
 971		seg->p_vaddr = phdr->p_vaddr;
 972		seg->p_memsz = phdr->p_memsz;
 973
 974		ret = file->f_op->read(file, (void *) seg->addr,
 975				       phdr->p_filesz, &fpos);
 976		if (ret < 0)
 977			return ret;
 978
 979		/* map the ELF header address if in this segment */
 980		if (phdr->p_offset == 0)
 981			params->elfhdr_addr = seg->addr;
 982
 983		/* clear any space allocated but not loaded */
 984		if (phdr->p_filesz < phdr->p_memsz) {
 985			if (clear_user((void *) (seg->addr + phdr->p_filesz),
 986				       phdr->p_memsz - phdr->p_filesz))
 987				return -EFAULT;
 988		}
 989
 990		if (mm) {
 991			if (phdr->p_flags & PF_X) {
 992				if (!mm->start_code) {
 993					mm->start_code = seg->addr;
 994					mm->end_code = seg->addr +
 995						phdr->p_memsz;
 996				}
 997			} else if (!mm->start_data) {
 998				mm->start_data = seg->addr;
 999				mm->end_data = seg->addr + phdr->p_memsz;
1000			}
1001		}
1002
1003		seg++;
1004	}
1005
1006	return 0;
1007}
1008#endif
1009
1010/*****************************************************************************/
1011/*
1012 * map a binary by direct mmap() of the individual PT_LOAD segments
1013 */
1014static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
1015					     struct file *file,
1016					     struct mm_struct *mm)
1017{
1018	struct elf32_fdpic_loadseg *seg;
1019	struct elf32_phdr *phdr;
1020	unsigned long load_addr, delta_vaddr;
1021	int loop, dvset;
1022
1023	load_addr = params->load_addr;
1024	delta_vaddr = 0;
1025	dvset = 0;
1026
1027	seg = params->loadmap->segs;
1028
1029	/* deal with each load segment separately */
1030	phdr = params->phdrs;
1031	for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
1032		unsigned long maddr, disp, excess, excess1;
1033		int prot = 0, flags;
1034
1035		if (phdr->p_type != PT_LOAD)
1036			continue;
1037
1038		kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
1039		       (unsigned long) phdr->p_vaddr,
1040		       (unsigned long) phdr->p_offset,
1041		       (unsigned long) phdr->p_filesz,
1042		       (unsigned long) phdr->p_memsz);
1043
1044		/* determine the mapping parameters */
1045		if (phdr->p_flags & PF_R) prot |= PROT_READ;
1046		if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1047		if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1048
1049		flags = MAP_PRIVATE | MAP_DENYWRITE;
1050		if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1051			flags |= MAP_EXECUTABLE;
1052
1053		maddr = 0;
1054
1055		switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1056		case ELF_FDPIC_FLAG_INDEPENDENT:
1057			/* PT_LOADs are independently locatable */
1058			break;
1059
1060		case ELF_FDPIC_FLAG_HONOURVADDR:
1061			/* the specified virtual address must be honoured */
1062			maddr = phdr->p_vaddr;
1063			flags |= MAP_FIXED;
1064			break;
1065
1066		case ELF_FDPIC_FLAG_CONSTDISP:
1067			/* constant displacement
1068			 * - can be mapped anywhere, but must be mapped as a
1069			 *   unit
1070			 */
1071			if (!dvset) {
1072				maddr = load_addr;
1073				delta_vaddr = phdr->p_vaddr;
1074				dvset = 1;
1075			} else {
1076				maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1077				flags |= MAP_FIXED;
1078			}
1079			break;
1080
1081		case ELF_FDPIC_FLAG_CONTIGUOUS:
1082			/* contiguity handled later */
1083			break;
1084
1085		default:
1086			BUG();
1087		}
1088
1089		maddr &= PAGE_MASK;
1090
1091		/* create the mapping */
1092		disp = phdr->p_vaddr & ~PAGE_MASK;
1093		maddr = vm_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1094				phdr->p_offset - disp);
1095
1096		kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1097		       loop, phdr->p_memsz + disp, prot, flags,
1098		       phdr->p_offset - disp, maddr);
 
1099
1100		if (IS_ERR_VALUE(maddr))
1101			return (int) maddr;
1102
1103		if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1104		    ELF_FDPIC_FLAG_CONTIGUOUS)
1105			load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1106
1107		seg->addr = maddr + disp;
1108		seg->p_vaddr = phdr->p_vaddr;
1109		seg->p_memsz = phdr->p_memsz;
1110
1111		/* map the ELF header address if in this segment */
1112		if (phdr->p_offset == 0)
1113			params->elfhdr_addr = seg->addr;
1114
1115		/* clear the bit between beginning of mapping and beginning of
1116		 * PT_LOAD */
1117		if (prot & PROT_WRITE && disp > 0) {
1118			kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1119			if (clear_user((void __user *) maddr, disp))
1120				return -EFAULT;
1121			maddr += disp;
1122		}
1123
1124		/* clear any space allocated but not loaded
1125		 * - on uClinux we can just clear the lot
1126		 * - on MMU linux we'll get a SIGBUS beyond the last page
1127		 *   extant in the file
1128		 */
1129		excess = phdr->p_memsz - phdr->p_filesz;
1130		excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1131
1132#ifdef CONFIG_MMU
1133		if (excess > excess1) {
1134			unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1135			unsigned long xmaddr;
1136
1137			flags |= MAP_FIXED | MAP_ANONYMOUS;
1138			xmaddr = vm_mmap(NULL, xaddr, excess - excess1,
1139					 prot, flags, 0);
1140
1141			kdebug("mmap[%d] <anon>"
1142			       " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1143			       loop, xaddr, excess - excess1, prot, flags,
1144			       xmaddr);
1145
1146			if (xmaddr != xaddr)
1147				return -ENOMEM;
1148		}
1149
1150		if (prot & PROT_WRITE && excess1 > 0) {
1151			kdebug("clear[%d] ad=%lx sz=%lx",
1152			       loop, maddr + phdr->p_filesz, excess1);
1153			if (clear_user((void __user *) maddr + phdr->p_filesz,
1154				       excess1))
1155				return -EFAULT;
1156		}
1157
1158#else
1159		if (excess > 0) {
1160			kdebug("clear[%d] ad=%lx sz=%lx",
1161			       loop, maddr + phdr->p_filesz, excess);
 
1162			if (clear_user((void *) maddr + phdr->p_filesz, excess))
1163				return -EFAULT;
1164		}
1165#endif
1166
1167		if (mm) {
1168			if (phdr->p_flags & PF_X) {
1169				if (!mm->start_code) {
1170					mm->start_code = maddr;
1171					mm->end_code = maddr + phdr->p_memsz;
1172				}
1173			} else if (!mm->start_data) {
1174				mm->start_data = maddr;
1175				mm->end_data = maddr + phdr->p_memsz;
1176			}
1177		}
1178
1179		seg++;
1180	}
1181
1182	return 0;
1183}
1184
1185/*****************************************************************************/
1186/*
1187 * ELF-FDPIC core dumper
1188 *
1189 * Modelled on fs/exec.c:aout_core_dump()
1190 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1191 *
1192 * Modelled on fs/binfmt_elf.c core dumper
1193 */
1194#ifdef CONFIG_ELF_CORE
1195
1196/*
1197 * Decide whether a segment is worth dumping; default is yes to be
1198 * sure (missing info is worse than too much; etc).
1199 * Personally I'd include everything, and use the coredump limit...
1200 *
1201 * I think we should skip something. But I am not sure how. H.J.
1202 */
1203static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1204{
1205	int dump_ok;
1206
1207	/* Do not dump I/O mapped devices or special mappings */
1208	if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1209		kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1210		return 0;
1211	}
1212
1213	/* If we may not read the contents, don't allow us to dump
1214	 * them either. "dump_write()" can't handle it anyway.
1215	 */
1216	if (!(vma->vm_flags & VM_READ)) {
1217		kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1218		return 0;
1219	}
1220
1221	/* By default, dump shared memory if mapped from an anonymous file. */
1222	if (vma->vm_flags & VM_SHARED) {
1223		if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1224			dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1225			kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1226			       vma->vm_flags, dump_ok ? "yes" : "no");
1227			return dump_ok;
1228		}
1229
1230		dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1231		kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1232		       vma->vm_flags, dump_ok ? "yes" : "no");
1233		return dump_ok;
1234	}
1235
1236#ifdef CONFIG_MMU
1237	/* By default, if it hasn't been written to, don't write it out */
1238	if (!vma->anon_vma) {
1239		dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1240		kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1241		       vma->vm_flags, dump_ok ? "yes" : "no");
1242		return dump_ok;
1243	}
1244#endif
1245
1246	dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1247	kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1248	       dump_ok ? "yes" : "no");
1249	return dump_ok;
1250}
1251
1252/* An ELF note in memory */
1253struct memelfnote
1254{
1255	const char *name;
1256	int type;
1257	unsigned int datasz;
1258	void *data;
1259};
1260
1261static int notesize(struct memelfnote *en)
1262{
1263	int sz;
1264
1265	sz = sizeof(struct elf_note);
1266	sz += roundup(strlen(en->name) + 1, 4);
1267	sz += roundup(en->datasz, 4);
1268
1269	return sz;
1270}
1271
1272/* #define DEBUG */
1273
1274#define DUMP_WRITE(addr, nr, foffset)	\
1275	do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1276
1277static int alignfile(struct file *file, loff_t *foffset)
1278{
1279	static const char buf[4] = { 0, };
1280	DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1281	return 1;
1282}
1283
1284static int writenote(struct memelfnote *men, struct file *file,
1285			loff_t *foffset)
1286{
1287	struct elf_note en;
1288	en.n_namesz = strlen(men->name) + 1;
1289	en.n_descsz = men->datasz;
1290	en.n_type = men->type;
1291
1292	DUMP_WRITE(&en, sizeof(en), foffset);
1293	DUMP_WRITE(men->name, en.n_namesz, foffset);
1294	if (!alignfile(file, foffset))
1295		return 0;
1296	DUMP_WRITE(men->data, men->datasz, foffset);
1297	if (!alignfile(file, foffset))
1298		return 0;
1299
1300	return 1;
1301}
1302#undef DUMP_WRITE
1303
1304static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1305{
1306	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1307	elf->e_ident[EI_CLASS] = ELF_CLASS;
1308	elf->e_ident[EI_DATA] = ELF_DATA;
1309	elf->e_ident[EI_VERSION] = EV_CURRENT;
1310	elf->e_ident[EI_OSABI] = ELF_OSABI;
1311	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1312
1313	elf->e_type = ET_CORE;
1314	elf->e_machine = ELF_ARCH;
1315	elf->e_version = EV_CURRENT;
1316	elf->e_entry = 0;
1317	elf->e_phoff = sizeof(struct elfhdr);
1318	elf->e_shoff = 0;
1319	elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1320	elf->e_ehsize = sizeof(struct elfhdr);
1321	elf->e_phentsize = sizeof(struct elf_phdr);
1322	elf->e_phnum = segs;
1323	elf->e_shentsize = 0;
1324	elf->e_shnum = 0;
1325	elf->e_shstrndx = 0;
1326	return;
1327}
1328
1329static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1330{
1331	phdr->p_type = PT_NOTE;
1332	phdr->p_offset = offset;
1333	phdr->p_vaddr = 0;
1334	phdr->p_paddr = 0;
1335	phdr->p_filesz = sz;
1336	phdr->p_memsz = 0;
1337	phdr->p_flags = 0;
1338	phdr->p_align = 0;
1339	return;
1340}
1341
1342static inline void fill_note(struct memelfnote *note, const char *name, int type,
1343		unsigned int sz, void *data)
1344{
1345	note->name = name;
1346	note->type = type;
1347	note->datasz = sz;
1348	note->data = data;
1349	return;
1350}
1351
1352/*
1353 * fill up all the fields in prstatus from the given task struct, except
1354 * registers which need to be filled up separately.
1355 */
1356static void fill_prstatus(struct elf_prstatus *prstatus,
1357			  struct task_struct *p, long signr)
1358{
1359	prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1360	prstatus->pr_sigpend = p->pending.signal.sig[0];
1361	prstatus->pr_sighold = p->blocked.sig[0];
1362	rcu_read_lock();
1363	prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1364	rcu_read_unlock();
1365	prstatus->pr_pid = task_pid_vnr(p);
1366	prstatus->pr_pgrp = task_pgrp_vnr(p);
1367	prstatus->pr_sid = task_session_vnr(p);
1368	if (thread_group_leader(p)) {
1369		struct task_cputime cputime;
1370
1371		/*
1372		 * This is the record for the group leader.  It shows the
1373		 * group-wide total, not its individual thread total.
1374		 */
1375		thread_group_cputime(p, &cputime);
1376		cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1377		cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1378	} else {
1379		cputime_to_timeval(p->utime, &prstatus->pr_utime);
1380		cputime_to_timeval(p->stime, &prstatus->pr_stime);
1381	}
1382	cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1383	cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1384
1385	prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1386	prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
 
 
 
 
1387}
1388
1389static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1390		       struct mm_struct *mm)
1391{
1392	const struct cred *cred;
1393	unsigned int i, len;
 
1394
1395	/* first copy the parameters from user space */
1396	memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1397
1398	len = mm->arg_end - mm->arg_start;
1399	if (len >= ELF_PRARGSZ)
1400		len = ELF_PRARGSZ - 1;
1401	if (copy_from_user(&psinfo->pr_psargs,
1402		           (const char __user *) mm->arg_start, len))
1403		return -EFAULT;
1404	for (i = 0; i < len; i++)
1405		if (psinfo->pr_psargs[i] == 0)
1406			psinfo->pr_psargs[i] = ' ';
1407	psinfo->pr_psargs[len] = 0;
1408
1409	rcu_read_lock();
1410	psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1411	rcu_read_unlock();
1412	psinfo->pr_pid = task_pid_vnr(p);
1413	psinfo->pr_pgrp = task_pgrp_vnr(p);
1414	psinfo->pr_sid = task_session_vnr(p);
1415
1416	i = p->state ? ffz(~p->state) + 1 : 0;
 
1417	psinfo->pr_state = i;
1418	psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1419	psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1420	psinfo->pr_nice = task_nice(p);
1421	psinfo->pr_flag = p->flags;
1422	rcu_read_lock();
1423	cred = __task_cred(p);
1424	SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1425	SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1426	rcu_read_unlock();
1427	strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1428
1429	return 0;
1430}
1431
1432/* Here is the structure in which status of each thread is captured. */
1433struct elf_thread_status
1434{
1435	struct list_head list;
1436	struct elf_prstatus prstatus;	/* NT_PRSTATUS */
1437	elf_fpregset_t fpu;		/* NT_PRFPREG */
1438	struct task_struct *thread;
1439#ifdef ELF_CORE_COPY_XFPREGS
1440	elf_fpxregset_t xfpu;		/* ELF_CORE_XFPREG_TYPE */
1441#endif
1442	struct memelfnote notes[3];
1443	int num_notes;
1444};
1445
1446/*
1447 * In order to add the specific thread information for the elf file format,
1448 * we need to keep a linked list of every thread's pr_status and then create
1449 * a single section for them in the final core file.
1450 */
1451static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1452{
1453	struct task_struct *p = t->thread;
1454	int sz = 0;
1455
1456	t->num_notes = 0;
1457
1458	fill_prstatus(&t->prstatus, p, signr);
1459	elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
 
 
 
 
 
 
1460
1461	fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1462		  &t->prstatus);
1463	t->num_notes++;
1464	sz += notesize(&t->notes[0]);
 
 
 
 
 
 
 
 
 
 
 
 
1465
1466	t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1467	if (t->prstatus.pr_fpvalid) {
1468		fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1469			  &t->fpu);
1470		t->num_notes++;
1471		sz += notesize(&t->notes[1]);
1472	}
1473
1474#ifdef ELF_CORE_COPY_XFPREGS
1475	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1476		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1477			  sizeof(t->xfpu), &t->xfpu);
1478		t->num_notes++;
1479		sz += notesize(&t->notes[2]);
1480	}
1481#endif
1482	return sz;
1483}
1484
1485static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1486			     elf_addr_t e_shoff, int segs)
1487{
1488	elf->e_shoff = e_shoff;
1489	elf->e_shentsize = sizeof(*shdr4extnum);
1490	elf->e_shnum = 1;
1491	elf->e_shstrndx = SHN_UNDEF;
1492
1493	memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1494
1495	shdr4extnum->sh_type = SHT_NULL;
1496	shdr4extnum->sh_size = elf->e_shnum;
1497	shdr4extnum->sh_link = elf->e_shstrndx;
1498	shdr4extnum->sh_info = segs;
1499}
1500
1501/*
1502 * dump the segments for an MMU process
1503 */
1504#ifdef CONFIG_MMU
1505static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1506			   unsigned long *limit, unsigned long mm_flags)
1507{
1508	struct vm_area_struct *vma;
1509	int err = 0;
1510
1511	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1512		unsigned long addr;
1513
1514		if (!maydump(vma, mm_flags))
1515			continue;
1516
1517		for (addr = vma->vm_start; addr < vma->vm_end;
1518							addr += PAGE_SIZE) {
1519			struct page *page = get_dump_page(addr);
1520			if (page) {
1521				void *kaddr = kmap(page);
1522				*size += PAGE_SIZE;
1523				if (*size > *limit)
1524					err = -EFBIG;
1525				else if (!dump_write(file, kaddr, PAGE_SIZE))
1526					err = -EIO;
1527				kunmap(page);
1528				page_cache_release(page);
1529			} else if (!dump_seek(file, PAGE_SIZE))
1530				err = -EFBIG;
1531			if (err)
1532				goto out;
1533		}
1534	}
1535out:
1536	return err;
1537}
1538#endif
1539
1540/*
1541 * dump the segments for a NOMMU process
1542 */
1543#ifndef CONFIG_MMU
1544static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1545			   unsigned long *limit, unsigned long mm_flags)
1546{
1547	struct vm_area_struct *vma;
1548
1549	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1550		if (!maydump(vma, mm_flags))
1551			continue;
1552
1553		if ((*size += PAGE_SIZE) > *limit)
1554			return -EFBIG;
1555
1556		if (!dump_write(file, (void *) vma->vm_start,
1557				vma->vm_end - vma->vm_start))
1558			return -EIO;
1559	}
1560
1561	return 0;
1562}
1563#endif
1564
1565static size_t elf_core_vma_data_size(unsigned long mm_flags)
1566{
1567	struct vm_area_struct *vma;
1568	size_t size = 0;
1569
1570	for (vma = current->mm->mmap; vma; vma = vma->vm_next)
1571		if (maydump(vma, mm_flags))
1572			size += vma->vm_end - vma->vm_start;
1573	return size;
1574}
1575
1576/*
1577 * Actual dumper
1578 *
1579 * This is a two-pass process; first we find the offsets of the bits,
1580 * and then they are actually written out.  If we run out of core limit
1581 * we just truncate.
1582 */
1583static int elf_fdpic_core_dump(struct coredump_params *cprm)
1584{
1585#define	NUM_NOTES	6
1586	int has_dumped = 0;
1587	mm_segment_t fs;
1588	int segs;
1589	size_t size = 0;
1590	int i;
1591	struct vm_area_struct *vma;
1592	struct elfhdr *elf = NULL;
1593	loff_t offset = 0, dataoff, foffset;
1594	int numnote;
1595	struct memelfnote *notes = NULL;
1596	struct elf_prstatus *prstatus = NULL;	/* NT_PRSTATUS */
1597	struct elf_prpsinfo *psinfo = NULL;	/* NT_PRPSINFO */
1598 	LIST_HEAD(thread_list);
1599 	struct list_head *t;
1600	elf_fpregset_t *fpu = NULL;
1601#ifdef ELF_CORE_COPY_XFPREGS
1602	elf_fpxregset_t *xfpu = NULL;
1603#endif
1604	int thread_status_size = 0;
1605	elf_addr_t *auxv;
1606	struct elf_phdr *phdr4note = NULL;
1607	struct elf_shdr *shdr4extnum = NULL;
1608	Elf_Half e_phnum;
1609	elf_addr_t e_shoff;
1610
1611	/*
1612	 * We no longer stop all VM operations.
1613	 *
1614	 * This is because those proceses that could possibly change map_count
1615	 * or the mmap / vma pages are now blocked in do_exit on current
1616	 * finishing this core dump.
1617	 *
1618	 * Only ptrace can touch these memory addresses, but it doesn't change
1619	 * the map_count or the pages allocated. So no possibility of crashing
1620	 * exists while dumping the mm->vm_next areas to the core file.
1621	 */
1622
1623	/* alloc memory for large data structures: too large to be on stack */
1624	elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1625	if (!elf)
1626		goto cleanup;
1627	prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1628	if (!prstatus)
1629		goto cleanup;
1630	psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1631	if (!psinfo)
1632		goto cleanup;
1633	notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1634	if (!notes)
1635		goto cleanup;
1636	fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1637	if (!fpu)
1638		goto cleanup;
1639#ifdef ELF_CORE_COPY_XFPREGS
1640	xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1641	if (!xfpu)
1642		goto cleanup;
1643#endif
1644
1645	if (cprm->signr) {
1646		struct core_thread *ct;
1647		struct elf_thread_status *tmp;
1648
1649		for (ct = current->mm->core_state->dumper.next;
1650						ct; ct = ct->next) {
1651			tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1652			if (!tmp)
1653				goto cleanup;
1654
1655			tmp->thread = ct->task;
1656			list_add(&tmp->list, &thread_list);
1657		}
 
 
 
1658
1659		list_for_each(t, &thread_list) {
1660			struct elf_thread_status *tmp;
1661			int sz;
1662
1663			tmp = list_entry(t, struct elf_thread_status, list);
1664			sz = elf_dump_thread_status(cprm->signr, tmp);
1665			thread_status_size += sz;
1666		}
1667	}
1668
1669	/* now collect the dump for the current */
1670	fill_prstatus(prstatus, current, cprm->signr);
1671	elf_core_copy_regs(&prstatus->pr_reg, cprm->regs);
 
 
 
 
1672
1673	segs = current->mm->map_count;
1674	segs += elf_core_extra_phdrs();
1675
1676	/* for notes section */
1677	segs++;
1678
1679	/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1680	 * this, kernel supports extended numbering. Have a look at
1681	 * include/linux/elf.h for further information. */
1682	e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1683
1684	/* Set up header */
1685	fill_elf_fdpic_header(elf, e_phnum);
1686
1687	has_dumped = 1;
1688	current->flags |= PF_DUMPCORE;
1689
1690	/*
1691	 * Set up the notes in similar form to SVR4 core dumps made
1692	 * with info from their /proc.
1693	 */
1694
1695	fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1696	fill_psinfo(psinfo, current->group_leader, current->mm);
1697	fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1698
1699	numnote = 2;
1700
1701	auxv = (elf_addr_t *) current->mm->saved_auxv;
1702
1703	i = 0;
1704	do
1705		i += 2;
1706	while (auxv[i - 2] != AT_NULL);
1707	fill_note(&notes[numnote++], "CORE", NT_AUXV,
1708		  i * sizeof(elf_addr_t), auxv);
1709
1710  	/* Try to dump the FPU. */
1711	if ((prstatus->pr_fpvalid =
1712	     elf_core_copy_task_fpregs(current, cprm->regs, fpu)))
1713		fill_note(notes + numnote++,
1714			  "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1715#ifdef ELF_CORE_COPY_XFPREGS
1716	if (elf_core_copy_task_xfpregs(current, xfpu))
1717		fill_note(notes + numnote++,
1718			  "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1719#endif
1720
1721	fs = get_fs();
1722	set_fs(KERNEL_DS);
1723
1724	offset += sizeof(*elf);				/* Elf header */
1725	offset += segs * sizeof(struct elf_phdr);	/* Program headers */
1726	foffset = offset;
1727
1728	/* Write notes phdr entry */
1729	{
1730		int sz = 0;
1731
1732		for (i = 0; i < numnote; i++)
1733			sz += notesize(notes + i);
1734
1735		sz += thread_status_size;
1736
1737		phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1738		if (!phdr4note)
1739			goto end_coredump;
1740
1741		fill_elf_note_phdr(phdr4note, sz, offset);
1742		offset += sz;
1743	}
1744
1745	/* Page-align dumped data */
1746	dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1747
1748	offset += elf_core_vma_data_size(cprm->mm_flags);
1749	offset += elf_core_extra_data_size();
1750	e_shoff = offset;
1751
1752	if (e_phnum == PN_XNUM) {
1753		shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1754		if (!shdr4extnum)
1755			goto end_coredump;
1756		fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1757	}
1758
1759	offset = dataoff;
1760
1761	size += sizeof(*elf);
1762	if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
1763		goto end_coredump;
1764
1765	size += sizeof(*phdr4note);
1766	if (size > cprm->limit
1767	    || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
1768		goto end_coredump;
1769
1770	/* write program headers for segments dump */
1771	for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
 
1772		struct elf_phdr phdr;
1773		size_t sz;
1774
1775		sz = vma->vm_end - vma->vm_start;
1776
1777		phdr.p_type = PT_LOAD;
1778		phdr.p_offset = offset;
1779		phdr.p_vaddr = vma->vm_start;
1780		phdr.p_paddr = 0;
1781		phdr.p_filesz = maydump(vma, cprm->mm_flags) ? sz : 0;
1782		phdr.p_memsz = sz;
1783		offset += phdr.p_filesz;
1784		phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1785		if (vma->vm_flags & VM_WRITE)
 
 
1786			phdr.p_flags |= PF_W;
1787		if (vma->vm_flags & VM_EXEC)
1788			phdr.p_flags |= PF_X;
1789		phdr.p_align = ELF_EXEC_PAGESIZE;
1790
1791		size += sizeof(phdr);
1792		if (size > cprm->limit
1793		    || !dump_write(cprm->file, &phdr, sizeof(phdr)))
1794			goto end_coredump;
1795	}
1796
1797	if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
1798		goto end_coredump;
1799
1800 	/* write out the notes section */
1801	for (i = 0; i < numnote; i++)
1802		if (!writenote(notes + i, cprm->file, &foffset))
 
 
 
 
 
 
1803			goto end_coredump;
1804
1805	/* write out the thread status notes section */
1806	list_for_each(t, &thread_list) {
1807		struct elf_thread_status *tmp =
1808				list_entry(t, struct elf_thread_status, list);
1809
1810		for (i = 0; i < tmp->num_notes; i++)
1811			if (!writenote(&tmp->notes[i], cprm->file, &foffset))
1812				goto end_coredump;
1813	}
1814
1815	if (!dump_seek(cprm->file, dataoff - foffset))
1816		goto end_coredump;
1817
1818	if (elf_fdpic_dump_segments(cprm->file, &size, &cprm->limit,
1819				    cprm->mm_flags) < 0)
1820		goto end_coredump;
1821
1822	if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
1823		goto end_coredump;
1824
1825	if (e_phnum == PN_XNUM) {
1826		size += sizeof(*shdr4extnum);
1827		if (size > cprm->limit
1828		    || !dump_write(cprm->file, shdr4extnum,
1829				   sizeof(*shdr4extnum)))
1830			goto end_coredump;
1831	}
1832
1833	if (cprm->file->f_pos != offset) {
1834		/* Sanity check */
1835		printk(KERN_WARNING
1836		       "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1837		       cprm->file->f_pos, offset);
1838	}
1839
1840end_coredump:
1841	set_fs(fs);
1842
1843cleanup:
1844	while (!list_empty(&thread_list)) {
1845		struct list_head *tmp = thread_list.next;
1846		list_del(tmp);
1847		kfree(list_entry(tmp, struct elf_thread_status, list));
1848	}
1849	kfree(phdr4note);
1850	kfree(elf);
1851	kfree(prstatus);
1852	kfree(psinfo);
1853	kfree(notes);
1854	kfree(fpu);
1855	kfree(shdr4extnum);
1856#ifdef ELF_CORE_COPY_XFPREGS
1857	kfree(xfpu);
1858#endif
1859	return has_dumped;
1860#undef NUM_NOTES
1861}
1862
1863#endif		/* CONFIG_ELF_CORE */