fs/binfmt_elf_fdpic.c

   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/highmem.h>
  29 #include <linux/highuid.h>
  30 #include <linux/personality.h>
  31 #include <linux/ptrace.h>
  32 #include <linux/init.h>
  33 #include <linux/elf.h>
  34 #include <linux/elf-fdpic.h>
  35 #include <linux/elfcore.h>
  36
  37 #include <asm/uaccess.h>
  38 #include <asm/param.h>
  39 #include <asm/pgalloc.h>
  40
  41 typedef char *elf_caddr_t;
  42
  43 #if 0
  44 #define kdebug(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
  45 #else
  46 #define kdebug(fmt, ...) do {} while(0)
  47 #endif
  48
  49 #if 0
  50 #define kdcore(fmt, ...) printk("FDPIC "fmt"\n" ,##__VA_ARGS__ )
  51 #else
  52 #define kdcore(fmt, ...) do {} while(0)
  53 #endif
  54
  55 MODULE_LICENSE("GPL");
  56
  57 static int load_elf_fdpic_binary(struct linux_binprm *, struct pt_regs *);
  58 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *, struct file *);
  59 static int elf_fdpic_map_file(struct elf_fdpic_params *, struct file *,
  60                               struct mm_struct *, const char *);
  61
  62 static int create_elf_fdpic_tables(struct linux_binprm *, struct mm_struct *,
  63                                    struct elf_fdpic_params *,
  64                                    struct elf_fdpic_params *);
  65
  66 #ifndef CONFIG_MMU
  67 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *,
  68                                             unsigned long *);
  69 static int elf_fdpic_map_file_constdisp_on_uclinux(struct elf_fdpic_params *,
  70                                                    struct file *,
  71                                                    struct mm_struct *);
  72 #endif
  73
  74 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *,
  75                                              struct file *, struct mm_struct *);
  76
  77 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
  78 static int elf_fdpic_core_dump(long, struct pt_regs *, struct file *, unsigned long limit);
  79 #endif
  80
  81 static struct linux_binfmt elf_fdpic_format = {
  82         .module         = THIS_MODULE,
  83         .load_binary    = load_elf_fdpic_binary,
  84 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
  85         .core_dump      = elf_fdpic_core_dump,
  86 #endif
  87         .min_coredump   = ELF_EXEC_PAGESIZE,
  88 };
  89
  90 static int __init init_elf_fdpic_binfmt(void)
  91 {
  92         return register_binfmt(&elf_fdpic_format);
  93 }
  94
  95 static void __exit exit_elf_fdpic_binfmt(void)
  96 {
  97         unregister_binfmt(&elf_fdpic_format);
  98 }
  99
 100 core_initcall(init_elf_fdpic_binfmt);
 101 module_exit(exit_elf_fdpic_binfmt);
 102
 103 static int is_elf_fdpic(struct elfhdr *hdr, struct file *file)
 104 {
 105         if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0)
 106                 return 0;
 107         if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN)
 108                 return 0;
 109         if (!elf_check_arch(hdr) || !elf_check_fdpic(hdr))
 110                 return 0;
 111         if (!file->f_op || !file->f_op->mmap)
 112                 return 0;
 113         return 1;
 114 }
 115
 116 /*****************************************************************************/
 117 /*
 118  * read the program headers table into memory
 119  */
 120 static int elf_fdpic_fetch_phdrs(struct elf_fdpic_params *params,
 121                                  struct file *file)
 122 {
 123         struct elf32_phdr *phdr;
 124         unsigned long size;
 125         int retval, loop;
 126
 127         if (params->hdr.e_phentsize != sizeof(struct elf_phdr))
 128                 return -ENOMEM;
 129         if (params->hdr.e_phnum > 65536U / sizeof(struct elf_phdr))
 130                 return -ENOMEM;
 131
 132         size = params->hdr.e_phnum * sizeof(struct elf_phdr);
 133         params->phdrs = kmalloc(size, GFP_KERNEL);
 134         if (!params->phdrs)
 135                 return -ENOMEM;
 136
 137         retval = kernel_read(file, params->hdr.e_phoff,
 138                              (char *) params->phdrs, size);
 139         if (unlikely(retval != size))
 140                 return retval < 0 ? retval : -ENOEXEC;
 141
 142         /* determine stack size for this binary */
 143         phdr = params->phdrs;
 144         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
 145                 if (phdr->p_type != PT_GNU_STACK)
 146                         continue;
 147
 148                 if (phdr->p_flags & PF_X)
 149                         params->flags |= ELF_FDPIC_FLAG_EXEC_STACK;
 150                 else
 151                         params->flags |= ELF_FDPIC_FLAG_NOEXEC_STACK;
 152
 153                 params->stack_size = phdr->p_memsz;
 154                 break;
 155         }
 156
 157         return 0;
 158 }
 159
 160 /*****************************************************************************/
 161 /*
 162  * load an fdpic binary into various bits of memory
 163  */
 164 static int load_elf_fdpic_binary(struct linux_binprm *bprm,
 165                                  struct pt_regs *regs)
 166 {
 167         struct elf_fdpic_params exec_params, interp_params;
 168         struct elf_phdr *phdr;
 169         unsigned long stack_size, entryaddr;
 170 #ifndef CONFIG_MMU
 171         unsigned long fullsize;
 172 #endif
 173 #ifdef ELF_FDPIC_PLAT_INIT
 174         unsigned long dynaddr;
 175 #endif
 176         struct file *interpreter = NULL; /* to shut gcc up */
 177         char *interpreter_name = NULL;
 178         int executable_stack;
 179         int retval, i;
 180
 181         kdebug("____ LOAD %d ____", current->pid);
 182
 183         memset(&exec_params, 0, sizeof(exec_params));
 184         memset(&interp_params, 0, sizeof(interp_params));
 185
 186         exec_params.hdr = *(struct elfhdr *) bprm->buf;
 187         exec_params.flags = ELF_FDPIC_FLAG_PRESENT | ELF_FDPIC_FLAG_EXECUTABLE;
 188
 189         /* check that this is a binary we know how to deal with */
 190         retval = -ENOEXEC;
 191         if (!is_elf_fdpic(&exec_params.hdr, bprm->file))
 192                 goto error;
 193
 194         /* read the program header table */
 195         retval = elf_fdpic_fetch_phdrs(&exec_params, bprm->file);
 196         if (retval < 0)
 197                 goto error;
 198
 199         /* scan for a program header that specifies an interpreter */
 200         phdr = exec_params.phdrs;
 201
 202         for (i = 0; i < exec_params.hdr.e_phnum; i++, phdr++) {
 203                 switch (phdr->p_type) {
 204                 case PT_INTERP:
 205                         retval = -ENOMEM;
 206                         if (phdr->p_filesz > PATH_MAX)
 207                                 goto error;
 208                         retval = -ENOENT;
 209                         if (phdr->p_filesz < 2)
 210                                 goto error;
 211
 212                         /* read the name of the interpreter into memory */
 213                         interpreter_name = kmalloc(phdr->p_filesz, GFP_KERNEL);
 214                         if (!interpreter_name)
 215                                 goto error;
 216
 217                         retval = kernel_read(bprm->file,
 218                                              phdr->p_offset,
 219                                              interpreter_name,
 220                                              phdr->p_filesz);
 221                         if (unlikely(retval != phdr->p_filesz)) {
 222                                 if (retval >= 0)
 223                                         retval = -ENOEXEC;
 224                                 goto error;
 225                         }
 226
 227                         retval = -ENOENT;
 228                         if (interpreter_name[phdr->p_filesz - 1] != '\0')
 229                                 goto error;
 230
 231                         kdebug("Using ELF interpreter %s", interpreter_name);
 232
 233                         /* replace the program with the interpreter */
 234                         interpreter = open_exec(interpreter_name);
 235                         retval = PTR_ERR(interpreter);
 236                         if (IS_ERR(interpreter)) {
 237                                 interpreter = NULL;
 238                                 goto error;
 239                         }
 240
 241                         /*
 242                          * If the binary is not readable then enforce
 243                          * mm->dumpable = 0 regardless of the interpreter's
 244                          * permissions.
 245                          */
 246                         if (file_permission(interpreter, MAY_READ) < 0)
 247                                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
 248
 249                         retval = kernel_read(interpreter, 0, bprm->buf,
 250                                              BINPRM_BUF_SIZE);
 251                         if (unlikely(retval != BINPRM_BUF_SIZE)) {
 252                                 if (retval >= 0)
 253                                         retval = -ENOEXEC;
 254                                 goto error;
 255                         }
 256
 257                         interp_params.hdr = *((struct elfhdr *) bprm->buf);
 258                         break;
 259
 260                 case PT_LOAD:
 261 #ifdef CONFIG_MMU
 262                         if (exec_params.load_addr == 0)
 263                                 exec_params.load_addr = phdr->p_vaddr;
 264 #endif
 265                         break;
 266                 }
 267
 268         }
 269
 270         if (elf_check_const_displacement(&exec_params.hdr))
 271                 exec_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
 272
 273         /* perform insanity checks on the interpreter */
 274         if (interpreter_name) {
 275                 retval = -ELIBBAD;
 276                 if (!is_elf_fdpic(&interp_params.hdr, interpreter))
 277                         goto error;
 278
 279                 interp_params.flags = ELF_FDPIC_FLAG_PRESENT;
 280
 281                 /* read the interpreter's program header table */
 282                 retval = elf_fdpic_fetch_phdrs(&interp_params, interpreter);
 283                 if (retval < 0)
 284                         goto error;
 285         }
 286
 287         stack_size = exec_params.stack_size;
 288         if (stack_size < interp_params.stack_size)
 289                 stack_size = interp_params.stack_size;
 290
 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 if (interp_params.flags & ELF_FDPIC_FLAG_EXEC_STACK)
 296                 executable_stack = EXSTACK_ENABLE_X;
 297         else if (interp_params.flags & ELF_FDPIC_FLAG_NOEXEC_STACK)
 298                 executable_stack = EXSTACK_DISABLE_X;
 299         else
 300                 executable_stack = EXSTACK_DEFAULT;
 301
 302         retval = -ENOEXEC;
 303         if (stack_size == 0)
 304                 goto error;
 305
 306         if (elf_check_const_displacement(&interp_params.hdr))
 307                 interp_params.flags |= ELF_FDPIC_FLAG_CONSTDISP;
 308
 309         /* flush all traces of the currently running executable */
 310         retval = flush_old_exec(bprm);
 311         if (retval)
 312                 goto error;
 313
 314         /* there's now no turning back... the old userspace image is dead,
 315          * defunct, deceased, etc. after this point we have to exit via
 316          * error_kill */
 317         set_personality(PER_LINUX_FDPIC);
 318         set_binfmt(&elf_fdpic_format);
 319
 320         current->mm->start_code = 0;
 321         current->mm->end_code = 0;
 322         current->mm->start_stack = 0;
 323         current->mm->start_data = 0;
 324         current->mm->end_data = 0;
 325         current->mm->context.exec_fdpic_loadmap = 0;
 326         current->mm->context.interp_fdpic_loadmap = 0;
 327
 328         current->flags &= ~PF_FORKNOEXEC;
 329
 330 #ifdef CONFIG_MMU
 331         elf_fdpic_arch_lay_out_mm(&exec_params,
 332                                   &interp_params,
 333                                   &current->mm->start_stack,
 334                                   &current->mm->start_brk);
 335
 336         retval = setup_arg_pages(bprm, current->mm->start_stack,
 337                                  executable_stack);
 338         if (retval < 0) {
 339                 send_sig(SIGKILL, current, 0);
 340                 goto error_kill;
 341         }
 342 #endif
 343
 344         /* load the executable and interpreter into memory */
 345         retval = elf_fdpic_map_file(&exec_params, bprm->file, current->mm,
 346                                     "executable");
 347         if (retval < 0)
 348                 goto error_kill;
 349
 350         if (interpreter_name) {
 351                 retval = elf_fdpic_map_file(&interp_params, interpreter,
 352                                             current->mm, "interpreter");
 353                 if (retval < 0) {
 354                         printk(KERN_ERR "Unable to load interpreter\n");
 355                         goto error_kill;
 356                 }
 357
 358                 allow_write_access(interpreter);
 359                 fput(interpreter);
 360                 interpreter = NULL;
 361         }
 362
 363 #ifdef CONFIG_MMU
 364         if (!current->mm->start_brk)
 365                 current->mm->start_brk = current->mm->end_data;
 366
 367         current->mm->brk = current->mm->start_brk =
 368                 PAGE_ALIGN(current->mm->start_brk);
 369
 370 #else
 371         /* create a stack and brk area big enough for everyone
 372          * - the brk heap starts at the bottom and works up
 373          * - the stack starts at the top and works down
 374          */
 375         stack_size = (stack_size + PAGE_SIZE - 1) & PAGE_MASK;
 376         if (stack_size < PAGE_SIZE * 2)
 377                 stack_size = PAGE_SIZE * 2;
 378
 379         down_write(&current->mm->mmap_sem);
 380         current->mm->start_brk = do_mmap(NULL, 0, stack_size,
 381                                          PROT_READ | PROT_WRITE | PROT_EXEC,
 382                                          MAP_PRIVATE | MAP_ANONYMOUS | MAP_GROWSDOWN,
 383                                          0);
 384
 385         if (IS_ERR_VALUE(current->mm->start_brk)) {
 386                 up_write(&current->mm->mmap_sem);
 387                 retval = current->mm->start_brk;
 388                 current->mm->start_brk = 0;
 389                 goto error_kill;
 390         }
 391
 392         /* expand the stack mapping to use up the entire allocation granule */
 393         fullsize = kobjsize((char *) current->mm->start_brk);
 394         if (!IS_ERR_VALUE(do_mremap(current->mm->start_brk, stack_size,
 395                                     fullsize, 0, 0)))
 396                 stack_size = fullsize;
 397         up_write(&current->mm->mmap_sem);
 398
 399         current->mm->brk = current->mm->start_brk;
 400         current->mm->context.end_brk = current->mm->start_brk;
 401         current->mm->context.end_brk +=
 402                 (stack_size > PAGE_SIZE) ? (stack_size - PAGE_SIZE) : 0;
 403         current->mm->start_stack = current->mm->start_brk + stack_size;
 404 #endif
 405
 406         compute_creds(bprm);
 407         current->flags &= ~PF_FORKNOEXEC;
 408         if (create_elf_fdpic_tables(bprm, current->mm,
 409                                     &exec_params, &interp_params) < 0)
 410                 goto error_kill;
 411
 412         kdebug("- start_code  %lx", current->mm->start_code);
 413         kdebug("- end_code    %lx", current->mm->end_code);
 414         kdebug("- start_data  %lx", current->mm->start_data);
 415         kdebug("- end_data    %lx", current->mm->end_data);
 416         kdebug("- start_brk   %lx", current->mm->start_brk);
 417         kdebug("- brk         %lx", current->mm->brk);
 418         kdebug("- start_stack %lx", current->mm->start_stack);
 419
 420 #ifdef ELF_FDPIC_PLAT_INIT
 421         /*
 422          * The ABI may specify that certain registers be set up in special
 423          * ways (on i386 %edx is the address of a DT_FINI function, for
 424          * example.  This macro performs whatever initialization to
 425          * the regs structure is required.
 426          */
 427         dynaddr = interp_params.dynamic_addr ?: exec_params.dynamic_addr;
 428         ELF_FDPIC_PLAT_INIT(regs, exec_params.map_addr, interp_params.map_addr,
 429                             dynaddr);
 430 #endif
 431
 432         /* everything is now ready... get the userspace context ready to roll */
 433         entryaddr = interp_params.entry_addr ?: exec_params.entry_addr;
 434         start_thread(regs, entryaddr, current->mm->start_stack);
 435
 436         retval = 0;
 437
 438 error:
 439         if (interpreter) {
 440                 allow_write_access(interpreter);
 441                 fput(interpreter);
 442         }
 443         kfree(interpreter_name);
 444         kfree(exec_params.phdrs);
 445         kfree(exec_params.loadmap);
 446         kfree(interp_params.phdrs);
 447         kfree(interp_params.loadmap);
 448         return retval;
 449
 450         /* unrecoverable error - kill the process */
 451 error_kill:
 452         send_sig(SIGSEGV, current, 0);
 453         goto error;
 454
 455 }
 456
 457 /*****************************************************************************/
 458 /*
 459  * present useful information to the program
 460  */
 461 static int create_elf_fdpic_tables(struct linux_binprm *bprm,
 462                                    struct mm_struct *mm,
 463                                    struct elf_fdpic_params *exec_params,
 464                                    struct elf_fdpic_params *interp_params)
 465 {
 466         unsigned long sp, csp, nitems;
 467         elf_caddr_t __user *argv, *envp;
 468         size_t platform_len = 0, len;
 469         char *k_platform;
 470         char __user *u_platform, *p;
 471         long hwcap;
 472         int loop;
 473
 474         /* we're going to shovel a whole load of stuff onto the stack */
 475 #ifdef CONFIG_MMU
 476         sp = bprm->p;
 477 #else
 478         sp = mm->start_stack;
 479
 480         /* stack the program arguments and environment */
 481         if (elf_fdpic_transfer_args_to_stack(bprm, &sp) < 0)
 482                 return -EFAULT;
 483 #endif
 484
 485         /* get hold of platform and hardware capabilities masks for the machine
 486          * we are running on.  In some cases (Sparc), this info is impossible
 487          * to get, in others (i386) it is merely difficult.
 488          */
 489         hwcap = ELF_HWCAP;
 490         k_platform = ELF_PLATFORM;
 491         u_platform = NULL;
 492
 493         if (k_platform) {
 494                 platform_len = strlen(k_platform) + 1;
 495                 sp -= platform_len;
 496                 u_platform = (char __user *) sp;
 497                 if (__copy_to_user(u_platform, k_platform, platform_len) != 0)
 498                         return -EFAULT;
 499         }
 500
 501 #if defined(__i386__) && defined(CONFIG_SMP)
 502         /* in some cases (e.g. Hyper-Threading), we want to avoid L1 evictions
 503          * by the processes running on the same package. One thing we can do is
 504          * to shuffle the initial stack for them.
 505          *
 506          * the conditionals here are unneeded, but kept in to make the code
 507          * behaviour the same as pre change unless we have hyperthreaded
 508          * processors. This keeps Mr Marcelo Person happier but should be
 509          * removed for 2.5
 510          */
 511         if (smp_num_siblings > 1)
 512                 sp = sp - ((current->pid % 64) << 7);
 513 #endif
 514
 515         sp &= ~7UL;
 516
 517         /* stack the load map(s) */
 518         len = sizeof(struct elf32_fdpic_loadmap);
 519         len += sizeof(struct elf32_fdpic_loadseg) * exec_params->loadmap->nsegs;
 520         sp = (sp - len) & ~7UL;
 521         exec_params->map_addr = sp;
 522
 523         if (copy_to_user((void __user *) sp, exec_params->loadmap, len) != 0)
 524                 return -EFAULT;
 525
 526         current->mm->context.exec_fdpic_loadmap = (unsigned long) sp;
 527
 528         if (interp_params->loadmap) {
 529                 len = sizeof(struct elf32_fdpic_loadmap);
 530                 len += sizeof(struct elf32_fdpic_loadseg) *
 531                         interp_params->loadmap->nsegs;
 532                 sp = (sp - len) & ~7UL;
 533                 interp_params->map_addr = sp;
 534
 535                 if (copy_to_user((void __user *) sp, interp_params->loadmap,
 536                                  len) != 0)
 537                         return -EFAULT;
 538
 539                 current->mm->context.interp_fdpic_loadmap = (unsigned long) sp;
 540         }
 541
 542         /* force 16 byte _final_ alignment here for generality */
 543 #define DLINFO_ITEMS 13
 544
 545         nitems = 1 + DLINFO_ITEMS + (k_platform ? 1 : 0);
 546 #ifdef DLINFO_ARCH_ITEMS
 547         nitems += DLINFO_ARCH_ITEMS;
 548 #endif
 549
 550         csp = sp;
 551         sp -= nitems * 2 * sizeof(unsigned long);
 552         sp -= (bprm->envc + 1) * sizeof(char *);        /* envv[] */
 553         sp -= (bprm->argc + 1) * sizeof(char *);        /* argv[] */
 554         sp -= 1 * sizeof(unsigned long);                /* argc */
 555
 556         csp -= sp & 15UL;
 557         sp -= sp & 15UL;
 558
 559         /* put the ELF interpreter info on the stack */
 560 #define NEW_AUX_ENT(nr, id, val)                                        \
 561         do {                                                            \
 562                 struct { unsigned long _id, _val; } __user *ent;        \
 563                                                                         \
 564                 ent = (void __user *) csp;                              \
 565                 __put_user((id), &ent[nr]._id);                         \
 566                 __put_user((val), &ent[nr]._val);                       \
 567         } while (0)
 568
 569         csp -= 2 * sizeof(unsigned long);
 570         NEW_AUX_ENT(0, AT_NULL, 0);
 571         if (k_platform) {
 572                 csp -= 2 * sizeof(unsigned long);
 573                 NEW_AUX_ENT(0, AT_PLATFORM,
 574                             (elf_addr_t) (unsigned long) u_platform);
 575         }
 576
 577         csp -= DLINFO_ITEMS * 2 * sizeof(unsigned long);
 578         NEW_AUX_ENT( 0, AT_HWCAP,       hwcap);
 579         NEW_AUX_ENT( 1, AT_PAGESZ,      PAGE_SIZE);
 580         NEW_AUX_ENT( 2, AT_CLKTCK,      CLOCKS_PER_SEC);
 581         NEW_AUX_ENT( 3, AT_PHDR,        exec_params->ph_addr);
 582         NEW_AUX_ENT( 4, AT_PHENT,       sizeof(struct elf_phdr));
 583         NEW_AUX_ENT( 5, AT_PHNUM,       exec_params->hdr.e_phnum);
 584         NEW_AUX_ENT( 6, AT_BASE,        interp_params->elfhdr_addr);
 585         NEW_AUX_ENT( 7, AT_FLAGS,       0);
 586         NEW_AUX_ENT( 8, AT_ENTRY,       exec_params->entry_addr);
 587         NEW_AUX_ENT( 9, AT_UID,         (elf_addr_t) current->uid);
 588         NEW_AUX_ENT(10, AT_EUID,        (elf_addr_t) current->euid);
 589         NEW_AUX_ENT(11, AT_GID,         (elf_addr_t) current->gid);
 590         NEW_AUX_ENT(12, AT_EGID,        (elf_addr_t) current->egid);
 591
 592 #ifdef ARCH_DLINFO
 593         /* ARCH_DLINFO must come last so platform specific code can enforce
 594          * special alignment requirements on the AUXV if necessary (eg. PPC).
 595          */
 596         ARCH_DLINFO;
 597 #endif
 598 #undef NEW_AUX_ENT
 599
 600         /* allocate room for argv[] and envv[] */
 601         csp -= (bprm->envc + 1) * sizeof(elf_caddr_t);
 602         envp = (elf_caddr_t __user *) csp;
 603         csp -= (bprm->argc + 1) * sizeof(elf_caddr_t);
 604         argv = (elf_caddr_t __user *) csp;
 605
 606         /* stack argc */
 607         csp -= sizeof(unsigned long);
 608         __put_user(bprm->argc, (unsigned long __user *) csp);
 609
 610         BUG_ON(csp != sp);
 611
 612         /* fill in the argv[] array */
 613 #ifdef CONFIG_MMU
 614         current->mm->arg_start = bprm->p;
 615 #else
 616         current->mm->arg_start = current->mm->start_stack -
 617                 (MAX_ARG_PAGES * PAGE_SIZE - bprm->p);
 618 #endif
 619
 620         p = (char __user *) current->mm->arg_start;
 621         for (loop = bprm->argc; loop > 0; loop--) {
 622                 __put_user((elf_caddr_t) p, argv++);
 623                 len = strnlen_user(p, MAX_ARG_STRLEN);
 624                 if (!len || len > MAX_ARG_STRLEN)
 625                         return -EINVAL;
 626                 p += len;
 627         }
 628         __put_user(NULL, argv);
 629         current->mm->arg_end = (unsigned long) p;
 630
 631         /* fill in the envv[] array */
 632         current->mm->env_start = (unsigned long) p;
 633         for (loop = bprm->envc; loop > 0; loop--) {
 634                 __put_user((elf_caddr_t)(unsigned long) p, envp++);
 635                 len = strnlen_user(p, MAX_ARG_STRLEN);
 636                 if (!len || len > MAX_ARG_STRLEN)
 637                         return -EINVAL;
 638                 p += len;
 639         }
 640         __put_user(NULL, envp);
 641         current->mm->env_end = (unsigned long) p;
 642
 643         mm->start_stack = (unsigned long) sp;
 644         return 0;
 645 }
 646
 647 /*****************************************************************************/
 648 /*
 649  * transfer the program arguments and environment from the holding pages onto
 650  * the stack
 651  */
 652 #ifndef CONFIG_MMU
 653 static int elf_fdpic_transfer_args_to_stack(struct linux_binprm *bprm,
 654                                             unsigned long *_sp)
 655 {
 656         unsigned long index, stop, sp;
 657         char *src;
 658         int ret = 0;
 659
 660         stop = bprm->p >> PAGE_SHIFT;
 661         sp = *_sp;
 662
 663         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 664                 src = kmap(bprm->page[index]);
 665                 sp -= PAGE_SIZE;
 666                 if (copy_to_user((void *) sp, src, PAGE_SIZE) != 0)
 667                         ret = -EFAULT;
 668                 kunmap(bprm->page[index]);
 669                 if (ret < 0)
 670                         goto out;
 671         }
 672
 673         *_sp = (*_sp - (MAX_ARG_PAGES * PAGE_SIZE - bprm->p)) & ~15;
 674
 675 out:
 676         return ret;
 677 }
 678 #endif
 679
 680 /*****************************************************************************/
 681 /*
 682  * load the appropriate binary image (executable or interpreter) into memory
 683  * - we assume no MMU is available
 684  * - if no other PIC bits are set in params->hdr->e_flags
 685  *   - we assume that the LOADable segments in the binary are independently relocatable
 686  *   - we assume R/O executable segments are shareable
 687  * - else
 688  *   - we assume the loadable parts of the image to require fixed displacement
 689  *   - the image is not shareable
 690  */
 691 static int elf_fdpic_map_file(struct elf_fdpic_params *params,
 692                               struct file *file,
 693                               struct mm_struct *mm,
 694                               const char *what)
 695 {
 696         struct elf32_fdpic_loadmap *loadmap;
 697 #ifdef CONFIG_MMU
 698         struct elf32_fdpic_loadseg *mseg;
 699 #endif
 700         struct elf32_fdpic_loadseg *seg;
 701         struct elf32_phdr *phdr;
 702         unsigned long load_addr, stop;
 703         unsigned nloads, tmp;
 704         size_t size;
 705         int loop, ret;
 706
 707         /* allocate a load map table */
 708         nloads = 0;
 709         for (loop = 0; loop < params->hdr.e_phnum; loop++)
 710                 if (params->phdrs[loop].p_type == PT_LOAD)
 711                         nloads++;
 712
 713         if (nloads == 0)
 714                 return -ELIBBAD;
 715
 716         size = sizeof(*loadmap) + nloads * sizeof(*seg);
 717         loadmap = kzalloc(size, GFP_KERNEL);
 718         if (!loadmap)
 719                 return -ENOMEM;
 720
 721         params->loadmap = loadmap;
 722
 723         loadmap->version = ELF32_FDPIC_LOADMAP_VERSION;
 724         loadmap->nsegs = nloads;
 725
 726         load_addr = params->load_addr;
 727         seg = loadmap->segs;
 728
 729         /* map the requested LOADs into the memory space */
 730         switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
 731         case ELF_FDPIC_FLAG_CONSTDISP:
 732         case ELF_FDPIC_FLAG_CONTIGUOUS:
 733 #ifndef CONFIG_MMU
 734                 ret = elf_fdpic_map_file_constdisp_on_uclinux(params, file, mm);
 735                 if (ret < 0)
 736                         return ret;
 737                 break;
 738 #endif
 739         default:
 740                 ret = elf_fdpic_map_file_by_direct_mmap(params, file, mm);
 741                 if (ret < 0)
 742                         return ret;
 743                 break;
 744         }
 745
 746         /* map the entry point */
 747         if (params->hdr.e_entry) {
 748                 seg = loadmap->segs;
 749                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
 750                         if (params->hdr.e_entry >= seg->p_vaddr &&
 751                             params->hdr.e_entry < seg->p_vaddr + seg->p_memsz) {
 752                                 params->entry_addr =
 753                                         (params->hdr.e_entry - seg->p_vaddr) +
 754                                         seg->addr;
 755                                 break;
 756                         }
 757                 }
 758         }
 759
 760         /* determine where the program header table has wound up if mapped */
 761         stop = params->hdr.e_phoff;
 762         stop += params->hdr.e_phnum * sizeof (struct elf_phdr);
 763         phdr = params->phdrs;
 764
 765         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
 766                 if (phdr->p_type != PT_LOAD)
 767                         continue;
 768
 769                 if (phdr->p_offset > params->hdr.e_phoff ||
 770                     phdr->p_offset + phdr->p_filesz < stop)
 771                         continue;
 772
 773                 seg = loadmap->segs;
 774                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
 775                         if (phdr->p_vaddr >= seg->p_vaddr &&
 776                             phdr->p_vaddr + phdr->p_filesz <=
 777                             seg->p_vaddr + seg->p_memsz) {
 778                                 params->ph_addr =
 779                                         (phdr->p_vaddr - seg->p_vaddr) +
 780                                         seg->addr +
 781                                         params->hdr.e_phoff - phdr->p_offset;
 782                                 break;
 783                         }
 784                 }
 785                 break;
 786         }
 787
 788         /* determine where the dynamic section has wound up if there is one */
 789         phdr = params->phdrs;
 790         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
 791                 if (phdr->p_type != PT_DYNAMIC)
 792                         continue;
 793
 794                 seg = loadmap->segs;
 795                 for (loop = loadmap->nsegs; loop > 0; loop--, seg++) {
 796                         if (phdr->p_vaddr >= seg->p_vaddr &&
 797                             phdr->p_vaddr + phdr->p_memsz <=
 798                             seg->p_vaddr + seg->p_memsz) {
 799                                 params->dynamic_addr =
 800                                         (phdr->p_vaddr - seg->p_vaddr) +
 801                                         seg->addr;
 802
 803                                 /* check the dynamic section contains at least
 804                                  * one item, and that the last item is a NULL
 805                                  * entry */
 806                                 if (phdr->p_memsz == 0 ||
 807                                     phdr->p_memsz % sizeof(Elf32_Dyn) != 0)
 808                                         goto dynamic_error;
 809
 810                                 tmp = phdr->p_memsz / sizeof(Elf32_Dyn);
 811                                 if (((Elf32_Dyn *)
 812                                      params->dynamic_addr)[tmp - 1].d_tag != 0)
 813                                         goto dynamic_error;
 814                                 break;
 815                         }
 816                 }
 817                 break;
 818         }
 819
 820         /* now elide adjacent segments in the load map on MMU linux
 821          * - on uClinux the holes between may actually be filled with system
 822          *   stuff or stuff from other processes
 823          */
 824 #ifdef CONFIG_MMU
 825         nloads = loadmap->nsegs;
 826         mseg = loadmap->segs;
 827         seg = mseg + 1;
 828         for (loop = 1; loop < nloads; loop++) {
 829                 /* see if we have a candidate for merging */
 830                 if (seg->p_vaddr - mseg->p_vaddr == seg->addr - mseg->addr) {
 831                         load_addr = PAGE_ALIGN(mseg->addr + mseg->p_memsz);
 832                         if (load_addr == (seg->addr & PAGE_MASK)) {
 833                                 mseg->p_memsz +=
 834                                         load_addr -
 835                                         (mseg->addr + mseg->p_memsz);
 836                                 mseg->p_memsz += seg->addr & ~PAGE_MASK;
 837                                 mseg->p_memsz += seg->p_memsz;
 838                                 loadmap->nsegs--;
 839                                 continue;
 840                         }
 841                 }
 842
 843                 mseg++;
 844                 if (mseg != seg)
 845                         *mseg = *seg;
 846         }
 847 #endif
 848
 849         kdebug("Mapped Object [%s]:", what);
 850         kdebug("- elfhdr   : %lx", params->elfhdr_addr);
 851         kdebug("- entry    : %lx", params->entry_addr);
 852         kdebug("- PHDR[]   : %lx", params->ph_addr);
 853         kdebug("- DYNAMIC[]: %lx", params->dynamic_addr);
 854         seg = loadmap->segs;
 855         for (loop = 0; loop < loadmap->nsegs; loop++, seg++)
 856                 kdebug("- LOAD[%d] : %08x-%08x [va=%x ms=%x]",
 857                        loop,
 858                        seg->addr, seg->addr + seg->p_memsz - 1,
 859                        seg->p_vaddr, seg->p_memsz);
 860
 861         return 0;
 862
 863 dynamic_error:
 864         printk("ELF FDPIC %s with invalid DYNAMIC section (inode=%lu)\n",
 865                what, file->f_path.dentry->d_inode->i_ino);
 866         return -ELIBBAD;
 867 }
 868
 869 /*****************************************************************************/
 870 /*
 871  * map a file with constant displacement under uClinux
 872  */
 873 #ifndef CONFIG_MMU
 874 static int elf_fdpic_map_file_constdisp_on_uclinux(
 875         struct elf_fdpic_params *params,
 876         struct file *file,
 877         struct mm_struct *mm)
 878 {
 879         struct elf32_fdpic_loadseg *seg;
 880         struct elf32_phdr *phdr;
 881         unsigned long load_addr, base = ULONG_MAX, top = 0, maddr = 0, mflags;
 882         loff_t fpos;
 883         int loop, ret;
 884
 885         load_addr = params->load_addr;
 886         seg = params->loadmap->segs;
 887
 888         /* determine the bounds of the contiguous overall allocation we must
 889          * make */
 890         phdr = params->phdrs;
 891         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
 892                 if (params->phdrs[loop].p_type != PT_LOAD)
 893                         continue;
 894
 895                 if (base > phdr->p_vaddr)
 896                         base = phdr->p_vaddr;
 897                 if (top < phdr->p_vaddr + phdr->p_memsz)
 898                         top = phdr->p_vaddr + phdr->p_memsz;
 899         }
 900
 901         /* allocate one big anon block for everything */
 902         mflags = MAP_PRIVATE;
 903         if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
 904                 mflags |= MAP_EXECUTABLE;
 905
 906         down_write(&mm->mmap_sem);
 907         maddr = do_mmap(NULL, load_addr, top - base,
 908                         PROT_READ | PROT_WRITE | PROT_EXEC, mflags, 0);
 909         up_write(&mm->mmap_sem);
 910         if (IS_ERR_VALUE(maddr))
 911                 return (int) maddr;
 912
 913         if (load_addr != 0)
 914                 load_addr += PAGE_ALIGN(top - base);
 915
 916         /* and then load the file segments into it */
 917         phdr = params->phdrs;
 918         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
 919                 if (params->phdrs[loop].p_type != PT_LOAD)
 920                         continue;
 921
 922                 fpos = phdr->p_offset;
 923
 924                 seg->addr = maddr + (phdr->p_vaddr - base);
 925                 seg->p_vaddr = phdr->p_vaddr;
 926                 seg->p_memsz = phdr->p_memsz;
 927
 928                 ret = file->f_op->read(file, (void *) seg->addr,
 929                                        phdr->p_filesz, &fpos);
 930                 if (ret < 0)
 931                         return ret;
 932
 933                 /* map the ELF header address if in this segment */
 934                 if (phdr->p_offset == 0)
 935                         params->elfhdr_addr = seg->addr;
 936
 937                 /* clear any space allocated but not loaded */
 938                 if (phdr->p_filesz < phdr->p_memsz)
 939                         clear_user((void *) (seg->addr + phdr->p_filesz),
 940                                    phdr->p_memsz - phdr->p_filesz);
 941
 942                 if (mm) {
 943                         if (phdr->p_flags & PF_X) {
 944                                 if (!mm->start_code) {
 945                                         mm->start_code = seg->addr;
 946                                         mm->end_code = seg->addr +
 947                                                 phdr->p_memsz;
 948                                 }
 949                         } else if (!mm->start_data) {
 950                                 mm->start_data = seg->addr;
 951 #ifndef CONFIG_MMU
 952                                 mm->end_data = seg->addr + phdr->p_memsz;
 953 #endif
 954                         }
 955
 956 #ifdef CONFIG_MMU
 957                         if (seg->addr + phdr->p_memsz > mm->end_data)
 958                                 mm->end_data = seg->addr + phdr->p_memsz;
 959 #endif
 960                 }
 961
 962                 seg++;
 963         }
 964
 965         return 0;
 966 }
 967 #endif
 968
 969 /*****************************************************************************/
 970 /*
 971  * map a binary by direct mmap() of the individual PT_LOAD segments
 972  */
 973 static int elf_fdpic_map_file_by_direct_mmap(struct elf_fdpic_params *params,
 974                                              struct file *file,
 975                                              struct mm_struct *mm)
 976 {
 977         struct elf32_fdpic_loadseg *seg;
 978         struct elf32_phdr *phdr;
 979         unsigned long load_addr, delta_vaddr;
 980         int loop, dvset;
 981
 982         load_addr = params->load_addr;
 983         delta_vaddr = 0;
 984         dvset = 0;
 985
 986         seg = params->loadmap->segs;
 987
 988         /* deal with each load segment separately */
 989         phdr = params->phdrs;
 990         for (loop = 0; loop < params->hdr.e_phnum; loop++, phdr++) {
 991                 unsigned long maddr, disp, excess, excess1;
 992                 int prot = 0, flags;
 993
 994                 if (phdr->p_type != PT_LOAD)
 995                         continue;
 996
 997                 kdebug("[LOAD] va=%lx of=%lx fs=%lx ms=%lx",
 998                        (unsigned long) phdr->p_vaddr,
 999                        (unsigned long) phdr->p_offset,
1000                        (unsigned long) phdr->p_filesz,
1001                        (unsigned long) phdr->p_memsz);
1002
1003                 /* determine the mapping parameters */
1004                 if (phdr->p_flags & PF_R) prot |= PROT_READ;
1005                 if (phdr->p_flags & PF_W) prot |= PROT_WRITE;
1006                 if (phdr->p_flags & PF_X) prot |= PROT_EXEC;
1007
1008                 flags = MAP_PRIVATE | MAP_DENYWRITE;
1009                 if (params->flags & ELF_FDPIC_FLAG_EXECUTABLE)
1010                         flags |= MAP_EXECUTABLE;
1011
1012                 maddr = 0;
1013
1014                 switch (params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) {
1015                 case ELF_FDPIC_FLAG_INDEPENDENT:
1016                         /* PT_LOADs are independently locatable */
1017                         break;
1018
1019                 case ELF_FDPIC_FLAG_HONOURVADDR:
1020                         /* the specified virtual address must be honoured */
1021                         maddr = phdr->p_vaddr;
1022                         flags |= MAP_FIXED;
1023                         break;
1024
1025                 case ELF_FDPIC_FLAG_CONSTDISP:
1026                         /* constant displacement
1027                          * - can be mapped anywhere, but must be mapped as a
1028                          *   unit
1029                          */
1030                         if (!dvset) {
1031                                 maddr = load_addr;
1032                                 delta_vaddr = phdr->p_vaddr;
1033                                 dvset = 1;
1034                         } else {
1035                                 maddr = load_addr + phdr->p_vaddr - delta_vaddr;
1036                                 flags |= MAP_FIXED;
1037                         }
1038                         break;
1039
1040                 case ELF_FDPIC_FLAG_CONTIGUOUS:
1041                         /* contiguity handled later */
1042                         break;
1043
1044                 default:
1045                         BUG();
1046                 }
1047
1048                 maddr &= PAGE_MASK;
1049
1050                 /* create the mapping */
1051                 disp = phdr->p_vaddr & ~PAGE_MASK;
1052                 down_write(&mm->mmap_sem);
1053                 maddr = do_mmap(file, maddr, phdr->p_memsz + disp, prot, flags,
1054                                 phdr->p_offset - disp);
1055                 up_write(&mm->mmap_sem);
1056
1057                 kdebug("mmap[%d] <file> sz=%lx pr=%x fl=%x of=%lx --> %08lx",
1058                        loop, phdr->p_memsz + disp, prot, flags,
1059                        phdr->p_offset - disp, maddr);
1060
1061                 if (IS_ERR_VALUE(maddr))
1062                         return (int) maddr;
1063
1064                 if ((params->flags & ELF_FDPIC_FLAG_ARRANGEMENT) ==
1065                     ELF_FDPIC_FLAG_CONTIGUOUS)
1066                         load_addr += PAGE_ALIGN(phdr->p_memsz + disp);
1067
1068                 seg->addr = maddr + disp;
1069                 seg->p_vaddr = phdr->p_vaddr;
1070                 seg->p_memsz = phdr->p_memsz;
1071
1072                 /* map the ELF header address if in this segment */
1073                 if (phdr->p_offset == 0)
1074                         params->elfhdr_addr = seg->addr;
1075
1076                 /* clear the bit between beginning of mapping and beginning of
1077                  * PT_LOAD */
1078                 if (prot & PROT_WRITE && disp > 0) {
1079                         kdebug("clear[%d] ad=%lx sz=%lx", loop, maddr, disp);
1080                         clear_user((void __user *) maddr, disp);
1081                         maddr += disp;
1082                 }
1083
1084                 /* clear any space allocated but not loaded
1085                  * - on uClinux we can just clear the lot
1086                  * - on MMU linux we'll get a SIGBUS beyond the last page
1087                  *   extant in the file
1088                  */
1089                 excess = phdr->p_memsz - phdr->p_filesz;
1090                 excess1 = PAGE_SIZE - ((maddr + phdr->p_filesz) & ~PAGE_MASK);
1091
1092 #ifdef CONFIG_MMU
1093                 if (excess > excess1) {
1094                         unsigned long xaddr = maddr + phdr->p_filesz + excess1;
1095                         unsigned long xmaddr;
1096
1097                         flags |= MAP_FIXED | MAP_ANONYMOUS;
1098                         down_write(&mm->mmap_sem);
1099                         xmaddr = do_mmap(NULL, xaddr, excess - excess1,
1100                                          prot, flags, 0);
1101                         up_write(&mm->mmap_sem);
1102
1103                         kdebug("mmap[%d] <anon>"
1104                                " ad=%lx sz=%lx pr=%x fl=%x of=0 --> %08lx",
1105                                loop, xaddr, excess - excess1, prot, flags,
1106                                xmaddr);
1107
1108                         if (xmaddr != xaddr)
1109                                 return -ENOMEM;
1110                 }
1111
1112                 if (prot & PROT_WRITE && excess1 > 0) {
1113                         kdebug("clear[%d] ad=%lx sz=%lx",
1114                                loop, maddr + phdr->p_filesz, excess1);
1115                         clear_user((void __user *) maddr + phdr->p_filesz,
1116                                    excess1);
1117                 }
1118
1119 #else
1120                 if (excess > 0) {
1121                         kdebug("clear[%d] ad=%lx sz=%lx",
1122                                loop, maddr + phdr->p_filesz, excess);
1123                         clear_user((void *) maddr + phdr->p_filesz, excess);
1124                 }
1125 #endif
1126
1127                 if (mm) {
1128                         if (phdr->p_flags & PF_X) {
1129                                 if (!mm->start_code) {
1130                                         mm->start_code = maddr;
1131                                         mm->end_code = maddr + phdr->p_memsz;
1132                                 }
1133                         } else if (!mm->start_data) {
1134                                 mm->start_data = maddr;
1135                                 mm->end_data = maddr + phdr->p_memsz;
1136                         }
1137                 }
1138
1139                 seg++;
1140         }
1141
1142         return 0;
1143 }
1144
1145 /*****************************************************************************/
1146 /*
1147  * ELF-FDPIC core dumper
1148  *
1149  * Modelled on fs/exec.c:aout_core_dump()
1150  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1151  *
1152  * Modelled on fs/binfmt_elf.c core dumper
1153  */
1154 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1155
1156 /*
1157  * These are the only things you should do on a core-file: use only these
1158  * functions to write out all the necessary info.
1159  */
1160 static int dump_write(struct file *file, const void *addr, int nr)
1161 {
1162         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1163 }
1164
1165 static int dump_seek(struct file *file, loff_t off)
1166 {
1167         if (file->f_op->llseek) {
1168                 if (file->f_op->llseek(file, off, SEEK_SET) != off)
1169                         return 0;
1170         } else {
1171                 file->f_pos = off;
1172         }
1173         return 1;
1174 }
1175
1176 /*
1177  * Decide whether a segment is worth dumping; default is yes to be
1178  * sure (missing info is worse than too much; etc).
1179  * Personally I'd include everything, and use the coredump limit...
1180  *
1181  * I think we should skip something. But I am not sure how. H.J.
1182  */
1183 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1184 {
1185         int dump_ok;
1186
1187         /* Do not dump I/O mapped devices or special mappings */
1188         if (vma->vm_flags & (VM_IO | VM_RESERVED)) {
1189                 kdcore("%08lx: %08lx: no (IO)", vma->vm_start, vma->vm_flags);
1190                 return 0;
1191         }
1192
1193         /* If we may not read the contents, don't allow us to dump
1194          * them either. "dump_write()" can't handle it anyway.
1195          */
1196         if (!(vma->vm_flags & VM_READ)) {
1197                 kdcore("%08lx: %08lx: no (!read)", vma->vm_start, vma->vm_flags);
1198                 return 0;
1199         }
1200
1201         /* By default, dump shared memory if mapped from an anonymous file. */
1202         if (vma->vm_flags & VM_SHARED) {
1203                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0) {
1204                         dump_ok = test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1205                         kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1206                                vma->vm_flags, dump_ok ? "yes" : "no");
1207                         return dump_ok;
1208                 }
1209
1210                 dump_ok = test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1211                 kdcore("%08lx: %08lx: %s (share)", vma->vm_start,
1212                        vma->vm_flags, dump_ok ? "yes" : "no");
1213                 return dump_ok;
1214         }
1215
1216 #ifdef CONFIG_MMU
1217         /* By default, if it hasn't been written to, don't write it out */
1218         if (!vma->anon_vma) {
1219                 dump_ok = test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1220                 kdcore("%08lx: %08lx: %s (!anon)", vma->vm_start,
1221                        vma->vm_flags, dump_ok ? "yes" : "no");
1222                 return dump_ok;
1223         }
1224 #endif
1225
1226         dump_ok = test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1227         kdcore("%08lx: %08lx: %s", vma->vm_start, vma->vm_flags,
1228                dump_ok ? "yes" : "no");
1229         return dump_ok;
1230 }
1231
1232 /* An ELF note in memory */
1233 struct memelfnote
1234 {
1235         const char *name;
1236         int type;
1237         unsigned int datasz;
1238         void *data;
1239 };
1240
1241 static int notesize(struct memelfnote *en)
1242 {
1243         int sz;
1244
1245         sz = sizeof(struct elf_note);
1246         sz += roundup(strlen(en->name) + 1, 4);
1247         sz += roundup(en->datasz, 4);
1248
1249         return sz;
1250 }
1251
1252 /* #define DEBUG */
1253
1254 #define DUMP_WRITE(addr, nr)    \
1255         do { if (!dump_write(file, (addr), (nr))) return 0; } while(0)
1256 #define DUMP_SEEK(off)  \
1257         do { if (!dump_seek(file, (off))) return 0; } while(0)
1258
1259 static int writenote(struct memelfnote *men, struct file *file)
1260 {
1261         struct elf_note en;
1262
1263         en.n_namesz = strlen(men->name) + 1;
1264         en.n_descsz = men->datasz;
1265         en.n_type = men->type;
1266
1267         DUMP_WRITE(&en, sizeof(en));
1268         DUMP_WRITE(men->name, en.n_namesz);
1269         /* XXX - cast from long long to long to avoid need for libgcc.a */
1270         DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));      /* XXX */
1271         DUMP_WRITE(men->data, men->datasz);
1272         DUMP_SEEK(roundup((unsigned long)file->f_pos, 4));      /* XXX */
1273
1274         return 1;
1275 }
1276 #undef DUMP_WRITE
1277 #undef DUMP_SEEK
1278
1279 #define DUMP_WRITE(addr, nr)    \
1280         if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1281                 goto end_coredump;
1282 #define DUMP_SEEK(off)  \
1283         if (!dump_seek(file, (off))) \
1284                 goto end_coredump;
1285
1286 static inline void fill_elf_fdpic_header(struct elfhdr *elf, int segs)
1287 {
1288         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1289         elf->e_ident[EI_CLASS] = ELF_CLASS;
1290         elf->e_ident[EI_DATA] = ELF_DATA;
1291         elf->e_ident[EI_VERSION] = EV_CURRENT;
1292         elf->e_ident[EI_OSABI] = ELF_OSABI;
1293         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1294
1295         elf->e_type = ET_CORE;
1296         elf->e_machine = ELF_ARCH;
1297         elf->e_version = EV_CURRENT;
1298         elf->e_entry = 0;
1299         elf->e_phoff = sizeof(struct elfhdr);
1300         elf->e_shoff = 0;
1301         elf->e_flags = ELF_FDPIC_CORE_EFLAGS;
1302         elf->e_ehsize = sizeof(struct elfhdr);
1303         elf->e_phentsize = sizeof(struct elf_phdr);
1304         elf->e_phnum = segs;
1305         elf->e_shentsize = 0;
1306         elf->e_shnum = 0;
1307         elf->e_shstrndx = 0;
1308         return;
1309 }
1310
1311 static inline void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1312 {
1313         phdr->p_type = PT_NOTE;
1314         phdr->p_offset = offset;
1315         phdr->p_vaddr = 0;
1316         phdr->p_paddr = 0;
1317         phdr->p_filesz = sz;
1318         phdr->p_memsz = 0;
1319         phdr->p_flags = 0;
1320         phdr->p_align = 0;
1321         return;
1322 }
1323
1324 static inline void fill_note(struct memelfnote *note, const char *name, int type,
1325                 unsigned int sz, void *data)
1326 {
1327         note->name = name;
1328         note->type = type;
1329         note->datasz = sz;
1330         note->data = data;
1331         return;
1332 }
1333
1334 /*
1335  * fill up all the fields in prstatus from the given task struct, except
1336  * registers which need to be filled up seperately.
1337  */
1338 static void fill_prstatus(struct elf_prstatus *prstatus,
1339                           struct task_struct *p, long signr)
1340 {
1341         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1342         prstatus->pr_sigpend = p->pending.signal.sig[0];
1343         prstatus->pr_sighold = p->blocked.sig[0];
1344         prstatus->pr_pid = task_pid_vnr(p);
1345         prstatus->pr_ppid = task_pid_vnr(p->parent);
1346         prstatus->pr_pgrp = task_pgrp_vnr(p);
1347         prstatus->pr_sid = task_session_vnr(p);
1348         if (thread_group_leader(p)) {
1349                 /*
1350                  * This is the record for the group leader.  Add in the
1351                  * cumulative times of previous dead threads.  This total
1352                  * won't include the time of each live thread whose state
1353                  * is included in the core dump.  The final total reported
1354                  * to our parent process when it calls wait4 will include
1355                  * those sums as well as the little bit more time it takes
1356                  * this and each other thread to finish dying after the
1357                  * core dump synchronization phase.
1358                  */
1359                 cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1360                                    &prstatus->pr_utime);
1361                 cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1362                                    &prstatus->pr_stime);
1363         } else {
1364                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1365                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1366         }
1367         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1368         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1369
1370         prstatus->pr_exec_fdpic_loadmap = p->mm->context.exec_fdpic_loadmap;
1371         prstatus->pr_interp_fdpic_loadmap = p->mm->context.interp_fdpic_loadmap;
1372 }
1373
1374 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1375                        struct mm_struct *mm)
1376 {
1377         unsigned int i, len;
1378
1379         /* first copy the parameters from user space */
1380         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1381
1382         len = mm->arg_end - mm->arg_start;
1383         if (len >= ELF_PRARGSZ)
1384                 len = ELF_PRARGSZ - 1;
1385         if (copy_from_user(&psinfo->pr_psargs,
1386                            (const char __user *) mm->arg_start, len))
1387                 return -EFAULT;
1388         for (i = 0; i < len; i++)
1389                 if (psinfo->pr_psargs[i] == 0)
1390                         psinfo->pr_psargs[i] = ' ';
1391         psinfo->pr_psargs[len] = 0;
1392
1393         psinfo->pr_pid = task_pid_vnr(p);
1394         psinfo->pr_ppid = task_pid_vnr(p->parent);
1395         psinfo->pr_pgrp = task_pgrp_vnr(p);
1396         psinfo->pr_sid = task_session_vnr(p);
1397
1398         i = p->state ? ffz(~p->state) + 1 : 0;
1399         psinfo->pr_state = i;
1400         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1401         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1402         psinfo->pr_nice = task_nice(p);
1403         psinfo->pr_flag = p->flags;
1404         SET_UID(psinfo->pr_uid, p->uid);
1405         SET_GID(psinfo->pr_gid, p->gid);
1406         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1407
1408         return 0;
1409 }
1410
1411 /* Here is the structure in which status of each thread is captured. */
1412 struct elf_thread_status
1413 {
1414         struct list_head list;
1415         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1416         elf_fpregset_t fpu;             /* NT_PRFPREG */
1417         struct task_struct *thread;
1418 #ifdef ELF_CORE_COPY_XFPREGS
1419         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1420 #endif
1421         struct memelfnote notes[3];
1422         int num_notes;
1423 };
1424
1425 /*
1426  * In order to add the specific thread information for the elf file format,
1427  * we need to keep a linked list of every thread's pr_status and then create
1428  * a single section for them in the final core file.
1429  */
1430 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1431 {
1432         struct task_struct *p = t->thread;
1433         int sz = 0;
1434
1435         t->num_notes = 0;
1436
1437         fill_prstatus(&t->prstatus, p, signr);
1438         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1439
1440         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1441                   &t->prstatus);
1442         t->num_notes++;
1443         sz += notesize(&t->notes[0]);
1444
1445         t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL, &t->fpu);
1446         if (t->prstatus.pr_fpvalid) {
1447                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1448                           &t->fpu);
1449                 t->num_notes++;
1450                 sz += notesize(&t->notes[1]);
1451         }
1452
1453 #ifdef ELF_CORE_COPY_XFPREGS
1454         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1455                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1456                           sizeof(t->xfpu), &t->xfpu);
1457                 t->num_notes++;
1458                 sz += notesize(&t->notes[2]);
1459         }
1460 #endif
1461         return sz;
1462 }
1463
1464 /*
1465  * dump the segments for an MMU process
1466  */
1467 #ifdef CONFIG_MMU
1468 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1469                            unsigned long *limit, unsigned long mm_flags)
1470 {
1471         struct vm_area_struct *vma;
1472
1473         for (vma = current->mm->mmap; vma; vma = vma->vm_next) {
1474                 unsigned long addr;
1475
1476                 if (!maydump(vma, mm_flags))
1477                         continue;
1478
1479                 for (addr = vma->vm_start;
1480                      addr < vma->vm_end;
1481                      addr += PAGE_SIZE
1482                      ) {
1483                         struct vm_area_struct *vma;
1484                         struct page *page;
1485
1486                         if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1487                                            &page, &vma) <= 0) {
1488                                 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1489                         }
1490                         else if (page == ZERO_PAGE(0)) {
1491                                 page_cache_release(page);
1492                                 DUMP_SEEK(file->f_pos + PAGE_SIZE);
1493                         }
1494                         else {
1495                                 void *kaddr;
1496
1497                                 flush_cache_page(vma, addr, page_to_pfn(page));
1498                                 kaddr = kmap(page);
1499                                 if ((*size += PAGE_SIZE) > *limit ||
1500                                     !dump_write(file, kaddr, PAGE_SIZE)
1501                                     ) {
1502                                         kunmap(page);
1503                                         page_cache_release(page);
1504                                         return -EIO;
1505                                 }
1506                                 kunmap(page);
1507                                 page_cache_release(page);
1508                         }
1509                 }
1510         }
1511
1512         return 0;
1513
1514 end_coredump:
1515         return -EFBIG;
1516 }
1517 #endif
1518
1519 /*
1520  * dump the segments for a NOMMU process
1521  */
1522 #ifndef CONFIG_MMU
1523 static int elf_fdpic_dump_segments(struct file *file, size_t *size,
1524                            unsigned long *limit, unsigned long mm_flags)
1525 {
1526         struct vm_list_struct *vml;
1527
1528         for (vml = current->mm->context.vmlist; vml; vml = vml->next) {
1529         struct vm_area_struct *vma = vml->vma;
1530
1531                 if (!maydump(vma, mm_flags))
1532                         continue;
1533
1534                 if ((*size += PAGE_SIZE) > *limit)
1535                         return -EFBIG;
1536
1537                 if (!dump_write(file, (void *) vma->vm_start,
1538                                 vma->vm_end - vma->vm_start))
1539                         return -EIO;
1540         }
1541
1542         return 0;
1543 }
1544 #endif
1545
1546 /*
1547  * Actual dumper
1548  *
1549  * This is a two-pass process; first we find the offsets of the bits,
1550  * and then they are actually written out.  If we run out of core limit
1551  * we just truncate.
1552  */
1553 static int elf_fdpic_core_dump(long signr, struct pt_regs *regs,
1554                                struct file *file, unsigned long limit)
1555 {
1556 #define NUM_NOTES       6
1557         int has_dumped = 0;
1558         mm_segment_t fs;
1559         int segs;
1560         size_t size = 0;
1561         int i;
1562         struct vm_area_struct *vma;
1563         struct elfhdr *elf = NULL;
1564         loff_t offset = 0, dataoff;
1565         int numnote;
1566         struct memelfnote *notes = NULL;
1567         struct elf_prstatus *prstatus = NULL;   /* NT_PRSTATUS */
1568         struct elf_prpsinfo *psinfo = NULL;     /* NT_PRPSINFO */
1569         LIST_HEAD(thread_list);
1570         struct list_head *t;
1571         elf_fpregset_t *fpu = NULL;
1572 #ifdef ELF_CORE_COPY_XFPREGS
1573         elf_fpxregset_t *xfpu = NULL;
1574 #endif
1575         int thread_status_size = 0;
1576 #ifndef CONFIG_MMU
1577         struct vm_list_struct *vml;
1578 #endif
1579         elf_addr_t *auxv;
1580         unsigned long mm_flags;
1581
1582         /*
1583          * We no longer stop all VM operations.
1584          *
1585          * This is because those proceses that could possibly change map_count
1586          * or the mmap / vma pages are now blocked in do_exit on current
1587          * finishing this core dump.
1588          *
1589          * Only ptrace can touch these memory addresses, but it doesn't change
1590          * the map_count or the pages allocated. So no possibility of crashing
1591          * exists while dumping the mm->vm_next areas to the core file.
1592          */
1593
1594         /* alloc memory for large data structures: too large to be on stack */
1595         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1596         if (!elf)
1597                 goto cleanup;
1598         prstatus = kzalloc(sizeof(*prstatus), GFP_KERNEL);
1599         if (!prstatus)
1600                 goto cleanup;
1601         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1602         if (!psinfo)
1603                 goto cleanup;
1604         notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1605         if (!notes)
1606                 goto cleanup;
1607         fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1608         if (!fpu)
1609                 goto cleanup;
1610 #ifdef ELF_CORE_COPY_XFPREGS
1611         xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1612         if (!xfpu)
1613                 goto cleanup;
1614 #endif
1615
1616         if (signr) {
1617                 struct core_thread *ct;
1618                 struct elf_thread_status *tmp;
1619
1620                 for (ct = current->mm->core_state->dumper.next;
1621                                                 ct; ct = ct->next) {
1622                         tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
1623                         if (!tmp)
1624                                 goto cleanup;
1625
1626                         tmp->thread = ct->task;
1627                         list_add(&tmp->list, &thread_list);
1628                 }
1629
1630                 list_for_each(t, &thread_list) {
1631                         struct elf_thread_status *tmp;
1632                         int sz;
1633
1634                         tmp = list_entry(t, struct elf_thread_status, list);
1635                         sz = elf_dump_thread_status(signr, tmp);
1636                         thread_status_size += sz;
1637                 }
1638         }
1639
1640         /* now collect the dump for the current */
1641         fill_prstatus(prstatus, current, signr);
1642         elf_core_copy_regs(&prstatus->pr_reg, regs);
1643
1644 #ifdef CONFIG_MMU
1645         segs = current->mm->map_count;
1646 #else
1647         segs = 0;
1648         for (vml = current->mm->context.vmlist; vml; vml = vml->next)
1649             segs++;
1650 #endif
1651 #ifdef ELF_CORE_EXTRA_PHDRS
1652         segs += ELF_CORE_EXTRA_PHDRS;
1653 #endif
1654
1655         /* Set up header */
1656         fill_elf_fdpic_header(elf, segs + 1);   /* including notes section */
1657
1658         has_dumped = 1;
1659         current->flags |= PF_DUMPCORE;
1660
1661         /*
1662          * Set up the notes in similar form to SVR4 core dumps made
1663          * with info from their /proc.
1664          */
1665
1666         fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1667         fill_psinfo(psinfo, current->group_leader, current->mm);
1668         fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1669
1670         numnote = 2;
1671
1672         auxv = (elf_addr_t *) current->mm->saved_auxv;
1673
1674         i = 0;
1675         do
1676                 i += 2;
1677         while (auxv[i - 2] != AT_NULL);
1678         fill_note(&notes[numnote++], "CORE", NT_AUXV,
1679                   i * sizeof(elf_addr_t), auxv);
1680
1681         /* Try to dump the FPU. */
1682         if ((prstatus->pr_fpvalid =
1683              elf_core_copy_task_fpregs(current, regs, fpu)))
1684                 fill_note(notes + numnote++,
1685                           "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1686 #ifdef ELF_CORE_COPY_XFPREGS
1687         if (elf_core_copy_task_xfpregs(current, xfpu))
1688                 fill_note(notes + numnote++,
1689                           "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1690 #endif
1691
1692         fs = get_fs();
1693         set_fs(KERNEL_DS);
1694
1695         DUMP_WRITE(elf, sizeof(*elf));
1696         offset += sizeof(*elf);                         /* Elf header */
1697         offset += (segs+1) * sizeof(struct elf_phdr);   /* Program headers */
1698
1699         /* Write notes phdr entry */
1700         {
1701                 struct elf_phdr phdr;
1702                 int sz = 0;
1703
1704                 for (i = 0; i < numnote; i++)
1705                         sz += notesize(notes + i);
1706
1707                 sz += thread_status_size;
1708
1709                 fill_elf_note_phdr(&phdr, sz, offset);
1710                 offset += sz;
1711                 DUMP_WRITE(&phdr, sizeof(phdr));
1712         }
1713
1714         /* Page-align dumped data */
1715         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1716
1717         /*
1718          * We must use the same mm->flags while dumping core to avoid
1719          * inconsistency between the program headers and bodies, otherwise an
1720          * unusable core file can be generated.
1721          */
1722         mm_flags = current->mm->flags;
1723
1724         /* write program headers for segments dump */
1725         for (
1726 #ifdef CONFIG_MMU
1727                 vma = current->mm->mmap; vma; vma = vma->vm_next
1728 #else
1729                         vml = current->mm->context.vmlist; vml; vml = vml->next
1730 #endif
1731              ) {
1732                 struct elf_phdr phdr;
1733                 size_t sz;
1734
1735 #ifndef CONFIG_MMU
1736                 vma = vml->vma;
1737 #endif
1738
1739                 sz = vma->vm_end - vma->vm_start;
1740
1741                 phdr.p_type = PT_LOAD;
1742                 phdr.p_offset = offset;
1743                 phdr.p_vaddr = vma->vm_start;
1744                 phdr.p_paddr = 0;
1745                 phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1746                 phdr.p_memsz = sz;
1747                 offset += phdr.p_filesz;
1748                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1749                 if (vma->vm_flags & VM_WRITE)
1750                         phdr.p_flags |= PF_W;
1751                 if (vma->vm_flags & VM_EXEC)
1752                         phdr.p_flags |= PF_X;
1753                 phdr.p_align = ELF_EXEC_PAGESIZE;
1754
1755                 DUMP_WRITE(&phdr, sizeof(phdr));
1756         }
1757
1758 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1759         ELF_CORE_WRITE_EXTRA_PHDRS;
1760 #endif
1761
1762         /* write out the notes section */
1763         for (i = 0; i < numnote; i++)
1764                 if (!writenote(notes + i, file))
1765                         goto end_coredump;
1766
1767         /* write out the thread status notes section */
1768         list_for_each(t, &thread_list) {
1769                 struct elf_thread_status *tmp =
1770                                 list_entry(t, struct elf_thread_status, list);
1771
1772                 for (i = 0; i < tmp->num_notes; i++)
1773                         if (!writenote(&tmp->notes[i], file))
1774                                 goto end_coredump;
1775         }
1776
1777         DUMP_SEEK(dataoff);
1778
1779         if (elf_fdpic_dump_segments(file, &size, &limit, mm_flags) < 0)
1780                 goto end_coredump;
1781
1782 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1783         ELF_CORE_WRITE_EXTRA_DATA;
1784 #endif
1785
1786         if (file->f_pos != offset) {
1787                 /* Sanity check */
1788                 printk(KERN_WARNING
1789                        "elf_core_dump: file->f_pos (%lld) != offset (%lld)\n",
1790                        file->f_pos, offset);
1791         }
1792
1793 end_coredump:
1794         set_fs(fs);
1795
1796 cleanup:
1797         while (!list_empty(&thread_list)) {
1798                 struct list_head *tmp = thread_list.next;
1799                 list_del(tmp);
1800                 kfree(list_entry(tmp, struct elf_thread_status, list));
1801         }
1802
1803         kfree(elf);
1804         kfree(prstatus);
1805         kfree(psinfo);
1806         kfree(notes);
1807         kfree(fpu);
1808 #ifdef ELF_CORE_COPY_XFPREGS
1809         kfree(xfpu);
1810 #endif
1811         return has_dumped;
1812 #undef NUM_NOTES
1813 }
1814
1815 #endif          /* USE_ELF_CORE_DUMP */