git.oblomov.eu Git - linux-2.6/blob - fs/binfmt_elf.c

   1 /*
   2  * linux/fs/binfmt_elf.c
   3  *
   4  * These are the functions used to load ELF format executables as used
   5  * on SVr4 machines.  Information on the format may be found in the book
   6  * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
   7  * Tools".
   8  *
   9  * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/kernel.h>
  14 #include <linux/fs.h>
  15 #include <linux/stat.h>
  16 #include <linux/time.h>
  17 #include <linux/mm.h>
  18 #include <linux/mman.h>
  19 #include <linux/a.out.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/ptrace.h>
  27 #include <linux/slab.h>
  28 #include <linux/shm.h>
  29 #include <linux/personality.h>
  30 #include <linux/elfcore.h>
  31 #include <linux/init.h>
  32 #include <linux/highuid.h>
  33 #include <linux/smp.h>
  34 #include <linux/compiler.h>
  35 #include <linux/highmem.h>
  36 #include <linux/pagemap.h>
  37 #include <linux/security.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/random.h>
  40 #include <linux/elf.h>
  41 #include <linux/utsname.h>
  42 #include <asm/uaccess.h>
  43 #include <asm/param.h>
  44 #include <asm/page.h>
  45
  46 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
  47 static int load_elf_library(struct file *);
  48 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
  49                                 int, int, unsigned long);
  50
  51 /*
  52  * If we don't support core dumping, then supply a NULL so we
  53  * don't even try.
  54  */
  55 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
  56 static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, unsigned long limit);
  57 #else
  58 #define elf_core_dump   NULL
  59 #endif
  60
  61 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
  62 #define ELF_MIN_ALIGN   ELF_EXEC_PAGESIZE
  63 #else
  64 #define ELF_MIN_ALIGN   PAGE_SIZE
  65 #endif
  66
  67 #ifndef ELF_CORE_EFLAGS
  68 #define ELF_CORE_EFLAGS 0
  69 #endif
  70
  71 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
  72 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
  73 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
  74
  75 static struct linux_binfmt elf_format = {
  76                 .module         = THIS_MODULE,
  77                 .load_binary    = load_elf_binary,
  78                 .load_shlib     = load_elf_library,
  79                 .core_dump      = elf_core_dump,
  80                 .min_coredump   = ELF_EXEC_PAGESIZE,
  81                 .hasvdso        = 1
  82 };
  83
  84 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
  85
  86 static int set_brk(unsigned long start, unsigned long end)
  87 {
  88         start = ELF_PAGEALIGN(start);
  89         end = ELF_PAGEALIGN(end);
  90         if (end > start) {
  91                 unsigned long addr;
  92                 down_write(&current->mm->mmap_sem);
  93                 addr = do_brk(start, end - start);
  94                 up_write(&current->mm->mmap_sem);
  95                 if (BAD_ADDR(addr))
  96                         return addr;
  97         }
  98         current->mm->start_brk = current->mm->brk = end;
  99         return 0;
 100 }
 101
 102 /* We need to explicitly zero any fractional pages
 103    after the data section (i.e. bss).  This would
 104    contain the junk from the file that should not
 105    be in memory
 106  */
 107 static int padzero(unsigned long elf_bss)
 108 {
 109         unsigned long nbyte;
 110
 111         nbyte = ELF_PAGEOFFSET(elf_bss);
 112         if (nbyte) {
 113                 nbyte = ELF_MIN_ALIGN - nbyte;
 114                 if (clear_user((void __user *) elf_bss, nbyte))
 115                         return -EFAULT;
 116         }
 117         return 0;
 118 }
 119
 120 /* Let's use some macros to make this stack manipulation a little clearer */
 121 #ifdef CONFIG_STACK_GROWSUP
 122 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
 123 #define STACK_ROUND(sp, items) \
 124         ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
 125 #define STACK_ALLOC(sp, len) ({ \
 126         elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
 127         old_sp; })
 128 #else
 129 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
 130 #define STACK_ROUND(sp, items) \
 131         (((unsigned long) (sp - items)) &~ 15UL)
 132 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
 133 #endif
 134
 135 static int
 136 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
 137                 unsigned long load_addr, unsigned long interp_load_addr)
 138 {
 139         unsigned long p = bprm->p;
 140         int argc = bprm->argc;
 141         int envc = bprm->envc;
 142         elf_addr_t __user *argv;
 143         elf_addr_t __user *envp;
 144         elf_addr_t __user *sp;
 145         elf_addr_t __user *u_platform;
 146         const char *k_platform = ELF_PLATFORM;
 147         int items;
 148         elf_addr_t *elf_info;
 149         int ei_index = 0;
 150         struct task_struct *tsk = current;
 151         struct vm_area_struct *vma;
 152
 153         /*
 154          * In some cases (e.g. Hyper-Threading), we want to avoid L1
 155          * evictions by the processes running on the same package. One
 156          * thing we can do is to shuffle the initial stack for them.
 157          */
 158
 159         p = arch_align_stack(p);
 160
 161         /*
 162          * If this architecture has a platform capability string, copy it
 163          * to userspace.  In some cases (Sparc), this info is impossible
 164          * for userspace to get any other way, in others (i386) it is
 165          * merely difficult.
 166          */
 167         u_platform = NULL;
 168         if (k_platform) {
 169                 size_t len = strlen(k_platform) + 1;
 170
 171                 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 172                 if (__copy_to_user(u_platform, k_platform, len))
 173                         return -EFAULT;
 174         }
 175
 176         /* Create the ELF interpreter info */
 177         elf_info = (elf_addr_t *)current->mm->saved_auxv;
 178         /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
 179 #define NEW_AUX_ENT(id, val) \
 180         do { \
 181                 elf_info[ei_index++] = id; \
 182                 elf_info[ei_index++] = val; \
 183         } while (0)
 184
 185 #ifdef ARCH_DLINFO
 186         /*
 187          * ARCH_DLINFO must come first so PPC can do its special alignment of
 188          * AUXV.
 189          * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
 190          * ARCH_DLINFO changes
 191          */
 192         ARCH_DLINFO;
 193 #endif
 194         NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
 195         NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
 196         NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
 197         NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
 198         NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
 199         NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
 200         NEW_AUX_ENT(AT_BASE, interp_load_addr);
 201         NEW_AUX_ENT(AT_FLAGS, 0);
 202         NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
 203         NEW_AUX_ENT(AT_UID, tsk->uid);
 204         NEW_AUX_ENT(AT_EUID, tsk->euid);
 205         NEW_AUX_ENT(AT_GID, tsk->gid);
 206         NEW_AUX_ENT(AT_EGID, tsk->egid);
 207         NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
 208         if (k_platform) {
 209                 NEW_AUX_ENT(AT_PLATFORM,
 210                             (elf_addr_t)(unsigned long)u_platform);
 211         }
 212         if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
 213                 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
 214         }
 215 #undef NEW_AUX_ENT
 216         /* AT_NULL is zero; clear the rest too */
 217         memset(&elf_info[ei_index], 0,
 218                sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
 219
 220         /* And advance past the AT_NULL entry.  */
 221         ei_index += 2;
 222
 223         sp = STACK_ADD(p, ei_index);
 224
 225         items = (argc + 1) + (envc + 1) + 1;
 226         bprm->p = STACK_ROUND(sp, items);
 227
 228         /* Point sp at the lowest address on the stack */
 229 #ifdef CONFIG_STACK_GROWSUP
 230         sp = (elf_addr_t __user *)bprm->p - items - ei_index;
 231         bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
 232 #else
 233         sp = (elf_addr_t __user *)bprm->p;
 234 #endif
 235
 236
 237         /*
 238          * Grow the stack manually; some architectures have a limit on how
 239          * far ahead a user-space access may be in order to grow the stack.
 240          */
 241         vma = find_extend_vma(current->mm, bprm->p);
 242         if (!vma)
 243                 return -EFAULT;
 244
 245         /* Now, let's put argc (and argv, envp if appropriate) on the stack */
 246         if (__put_user(argc, sp++))
 247                 return -EFAULT;
 248         argv = sp;
 249         envp = argv + argc + 1;
 250
 251         /* Populate argv and envp */
 252         p = current->mm->arg_end = current->mm->arg_start;
 253         while (argc-- > 0) {
 254                 size_t len;
 255                 if (__put_user((elf_addr_t)p, argv++))
 256                         return -EFAULT;
 257                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 258                 if (!len || len > MAX_ARG_STRLEN)
 259                         return 0;
 260                 p += len;
 261         }
 262         if (__put_user(0, argv))
 263                 return -EFAULT;
 264         current->mm->arg_end = current->mm->env_start = p;
 265         while (envc-- > 0) {
 266                 size_t len;
 267                 if (__put_user((elf_addr_t)p, envp++))
 268                         return -EFAULT;
 269                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 270                 if (!len || len > MAX_ARG_STRLEN)
 271                         return 0;
 272                 p += len;
 273         }
 274         if (__put_user(0, envp))
 275                 return -EFAULT;
 276         current->mm->env_end = p;
 277
 278         /* Put the elf_info on the stack in the right place.  */
 279         sp = (elf_addr_t __user *)envp + 1;
 280         if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
 281                 return -EFAULT;
 282         return 0;
 283 }
 284
 285 #ifndef elf_map
 286
 287 static unsigned long elf_map(struct file *filep, unsigned long addr,
 288                 struct elf_phdr *eppnt, int prot, int type,
 289                 unsigned long total_size)
 290 {
 291         unsigned long map_addr;
 292         unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
 293         unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
 294         addr = ELF_PAGESTART(addr);
 295         size = ELF_PAGEALIGN(size);
 296
 297         /* mmap() will return -EINVAL if given a zero size, but a
 298          * segment with zero filesize is perfectly valid */
 299         if (!size)
 300                 return addr;
 301
 302         down_write(&current->mm->mmap_sem);
 303         /*
 304         * total_size is the size of the ELF (interpreter) image.
 305         * The _first_ mmap needs to know the full size, otherwise
 306         * randomization might put this image into an overlapping
 307         * position with the ELF binary image. (since size < total_size)
 308         * So we first map the 'big' image - and unmap the remainder at
 309         * the end. (which unmap is needed for ELF images with holes.)
 310         */
 311         if (total_size) {
 312                 total_size = ELF_PAGEALIGN(total_size);
 313                 map_addr = do_mmap(filep, addr, total_size, prot, type, off);
 314                 if (!BAD_ADDR(map_addr))
 315                         do_munmap(current->mm, map_addr+size, total_size-size);
 316         } else
 317                 map_addr = do_mmap(filep, addr, size, prot, type, off);
 318
 319         up_write(&current->mm->mmap_sem);
 320         return(map_addr);
 321 }
 322
 323 #endif /* !elf_map */
 324
 325 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
 326 {
 327         int i, first_idx = -1, last_idx = -1;
 328
 329         for (i = 0; i < nr; i++) {
 330                 if (cmds[i].p_type == PT_LOAD) {
 331                         last_idx = i;
 332                         if (first_idx == -1)
 333                                 first_idx = i;
 334                 }
 335         }
 336         if (first_idx == -1)
 337                 return 0;
 338
 339         return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
 340                                 ELF_PAGESTART(cmds[first_idx].p_vaddr);
 341 }
 342
 343
 344 /* This is much more generalized than the library routine read function,
 345    so we keep this separate.  Technically the library read function
 346    is only provided so that we can read a.out libraries that have
 347    an ELF header */
 348
 349 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
 350                 struct file *interpreter, unsigned long *interp_map_addr,
 351                 unsigned long no_base)
 352 {
 353         struct elf_phdr *elf_phdata;
 354         struct elf_phdr *eppnt;
 355         unsigned long load_addr = 0;
 356         int load_addr_set = 0;
 357         unsigned long last_bss = 0, elf_bss = 0;
 358         unsigned long error = ~0UL;
 359         unsigned long total_size;
 360         int retval, i, size;
 361
 362         /* First of all, some simple consistency checks */
 363         if (interp_elf_ex->e_type != ET_EXEC &&
 364             interp_elf_ex->e_type != ET_DYN)
 365                 goto out;
 366         if (!elf_check_arch(interp_elf_ex))
 367                 goto out;
 368         if (!interpreter->f_op || !interpreter->f_op->mmap)
 369                 goto out;
 370
 371         /*
 372          * If the size of this structure has changed, then punt, since
 373          * we will be doing the wrong thing.
 374          */
 375         if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
 376                 goto out;
 377         if (interp_elf_ex->e_phnum < 1 ||
 378                 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
 379                 goto out;
 380
 381         /* Now read in all of the header information */
 382         size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
 383         if (size > ELF_MIN_ALIGN)
 384                 goto out;
 385         elf_phdata = kmalloc(size, GFP_KERNEL);
 386         if (!elf_phdata)
 387                 goto out;
 388
 389         retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
 390                              (char *)elf_phdata,size);
 391         error = -EIO;
 392         if (retval != size) {
 393                 if (retval < 0)
 394                         error = retval;
 395                 goto out_close;
 396         }
 397
 398         total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
 399         if (!total_size) {
 400                 error = -EINVAL;
 401                 goto out_close;
 402         }
 403
 404         eppnt = elf_phdata;
 405         for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
 406                 if (eppnt->p_type == PT_LOAD) {
 407                         int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
 408                         int elf_prot = 0;
 409                         unsigned long vaddr = 0;
 410                         unsigned long k, map_addr;
 411
 412                         if (eppnt->p_flags & PF_R)
 413                                 elf_prot = PROT_READ;
 414                         if (eppnt->p_flags & PF_W)
 415                                 elf_prot |= PROT_WRITE;
 416                         if (eppnt->p_flags & PF_X)
 417                                 elf_prot |= PROT_EXEC;
 418                         vaddr = eppnt->p_vaddr;
 419                         if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
 420                                 elf_type |= MAP_FIXED;
 421                         else if (no_base && interp_elf_ex->e_type == ET_DYN)
 422                                 load_addr = -vaddr;
 423
 424                         map_addr = elf_map(interpreter, load_addr + vaddr,
 425                                         eppnt, elf_prot, elf_type, total_size);
 426                         total_size = 0;
 427                         if (!*interp_map_addr)
 428                                 *interp_map_addr = map_addr;
 429                         error = map_addr;
 430                         if (BAD_ADDR(map_addr))
 431                                 goto out_close;
 432
 433                         if (!load_addr_set &&
 434                             interp_elf_ex->e_type == ET_DYN) {
 435                                 load_addr = map_addr - ELF_PAGESTART(vaddr);
 436                                 load_addr_set = 1;
 437                         }
 438
 439                         /*
 440                          * Check to see if the section's size will overflow the
 441                          * allowed task size. Note that p_filesz must always be
 442                          * <= p_memsize so it's only necessary to check p_memsz.
 443                          */
 444                         k = load_addr + eppnt->p_vaddr;
 445                         if (BAD_ADDR(k) ||
 446                             eppnt->p_filesz > eppnt->p_memsz ||
 447                             eppnt->p_memsz > TASK_SIZE ||
 448                             TASK_SIZE - eppnt->p_memsz < k) {
 449                                 error = -ENOMEM;
 450                                 goto out_close;
 451                         }
 452
 453                         /*
 454                          * Find the end of the file mapping for this phdr, and
 455                          * keep track of the largest address we see for this.
 456                          */
 457                         k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
 458                         if (k > elf_bss)
 459                                 elf_bss = k;
 460
 461                         /*
 462                          * Do the same thing for the memory mapping - between
 463                          * elf_bss and last_bss is the bss section.
 464                          */
 465                         k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
 466                         if (k > last_bss)
 467                                 last_bss = k;
 468                 }
 469         }
 470
 471         /*
 472          * Now fill out the bss section.  First pad the last page up
 473          * to the page boundary, and then perform a mmap to make sure
 474          * that there are zero-mapped pages up to and including the
 475          * last bss page.
 476          */
 477         if (padzero(elf_bss)) {
 478                 error = -EFAULT;
 479                 goto out_close;
 480         }
 481
 482         /* What we have mapped so far */
 483         elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
 484
 485         /* Map the last of the bss segment */
 486         if (last_bss > elf_bss) {
 487                 down_write(&current->mm->mmap_sem);
 488                 error = do_brk(elf_bss, last_bss - elf_bss);
 489                 up_write(&current->mm->mmap_sem);
 490                 if (BAD_ADDR(error))
 491                         goto out_close;
 492         }
 493
 494         error = load_addr;
 495
 496 out_close:
 497         kfree(elf_phdata);
 498 out:
 499         return error;
 500 }
 501
 502 /*
 503  * These are the functions used to load ELF style executables and shared
 504  * libraries.  There is no binary dependent code anywhere else.
 505  */
 506
 507 #define INTERPRETER_NONE 0
 508 #define INTERPRETER_ELF 2
 509
 510 #ifndef STACK_RND_MASK
 511 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))     /* 8MB of VA */
 512 #endif
 513
 514 static unsigned long randomize_stack_top(unsigned long stack_top)
 515 {
 516         unsigned int random_variable = 0;
 517
 518         if ((current->flags & PF_RANDOMIZE) &&
 519                 !(current->personality & ADDR_NO_RANDOMIZE)) {
 520                 random_variable = get_random_int() & STACK_RND_MASK;
 521                 random_variable <<= PAGE_SHIFT;
 522         }
 523 #ifdef CONFIG_STACK_GROWSUP
 524         return PAGE_ALIGN(stack_top) + random_variable;
 525 #else
 526         return PAGE_ALIGN(stack_top) - random_variable;
 527 #endif
 528 }
 529
 530 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
 531 {
 532         struct file *interpreter = NULL; /* to shut gcc up */
 533         unsigned long load_addr = 0, load_bias = 0;
 534         int load_addr_set = 0;
 535         char * elf_interpreter = NULL;
 536         unsigned long error;
 537         struct elf_phdr *elf_ppnt, *elf_phdata;
 538         unsigned long elf_bss, elf_brk;
 539         int elf_exec_fileno;
 540         int retval, i;
 541         unsigned int size;
 542         unsigned long elf_entry;
 543         unsigned long interp_load_addr = 0;
 544         unsigned long start_code, end_code, start_data, end_data;
 545         unsigned long reloc_func_desc = 0;
 546         struct files_struct *files;
 547         int executable_stack = EXSTACK_DEFAULT;
 548         unsigned long def_flags = 0;
 549         struct {
 550                 struct elfhdr elf_ex;
 551                 struct elfhdr interp_elf_ex;
 552                 struct exec interp_ex;
 553         } *loc;
 554
 555         loc = kmalloc(sizeof(*loc), GFP_KERNEL);
 556         if (!loc) {
 557                 retval = -ENOMEM;
 558                 goto out_ret;
 559         }
 560
 561         /* Get the exec-header */
 562         loc->elf_ex = *((struct elfhdr *)bprm->buf);
 563
 564         retval = -ENOEXEC;
 565         /* First of all, some simple consistency checks */
 566         if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 567                 goto out;
 568
 569         if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
 570                 goto out;
 571         if (!elf_check_arch(&loc->elf_ex))
 572                 goto out;
 573         if (!bprm->file->f_op||!bprm->file->f_op->mmap)
 574                 goto out;
 575
 576         /* Now read in all of the header information */
 577         if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
 578                 goto out;
 579         if (loc->elf_ex.e_phnum < 1 ||
 580                 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
 581                 goto out;
 582         size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
 583         retval = -ENOMEM;
 584         elf_phdata = kmalloc(size, GFP_KERNEL);
 585         if (!elf_phdata)
 586                 goto out;
 587
 588         retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
 589                              (char *)elf_phdata, size);
 590         if (retval != size) {
 591                 if (retval >= 0)
 592                         retval = -EIO;
 593                 goto out_free_ph;
 594         }
 595
 596         files = current->files; /* Refcounted so ok */
 597         retval = unshare_files();
 598         if (retval < 0)
 599                 goto out_free_ph;
 600         if (files == current->files) {
 601                 put_files_struct(files);
 602                 files = NULL;
 603         }
 604
 605         /* exec will make our files private anyway, but for the a.out
 606            loader stuff we need to do it earlier */
 607         retval = get_unused_fd();
 608         if (retval < 0)
 609                 goto out_free_fh;
 610         get_file(bprm->file);
 611         fd_install(elf_exec_fileno = retval, bprm->file);
 612
 613         elf_ppnt = elf_phdata;
 614         elf_bss = 0;
 615         elf_brk = 0;
 616
 617         start_code = ~0UL;
 618         end_code = 0;
 619         start_data = 0;
 620         end_data = 0;
 621
 622         for (i = 0; i < loc->elf_ex.e_phnum; i++) {
 623                 if (elf_ppnt->p_type == PT_INTERP) {
 624                         /* This is the program interpreter used for
 625                          * shared libraries - for now assume that this
 626                          * is an a.out format binary
 627                          */
 628                         retval = -ENOEXEC;
 629                         if (elf_ppnt->p_filesz > PATH_MAX ||
 630                             elf_ppnt->p_filesz < 2)
 631                                 goto out_free_file;
 632
 633                         retval = -ENOMEM;
 634                         elf_interpreter = kmalloc(elf_ppnt->p_filesz,
 635                                                   GFP_KERNEL);
 636                         if (!elf_interpreter)
 637                                 goto out_free_file;
 638
 639                         retval = kernel_read(bprm->file, elf_ppnt->p_offset,
 640                                              elf_interpreter,
 641                                              elf_ppnt->p_filesz);
 642                         if (retval != elf_ppnt->p_filesz) {
 643                                 if (retval >= 0)
 644                                         retval = -EIO;
 645                                 goto out_free_interp;
 646                         }
 647                         /* make sure path is NULL terminated */
 648                         retval = -ENOEXEC;
 649                         if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
 650                                 goto out_free_interp;
 651
 652                         /*
 653                          * The early SET_PERSONALITY here is so that the lookup
 654                          * for the interpreter happens in the namespace of the
 655                          * to-be-execed image.  SET_PERSONALITY can select an
 656                          * alternate root.
 657                          *
 658                          * However, SET_PERSONALITY is NOT allowed to switch
 659                          * this task into the new images's memory mapping
 660                          * policy - that is, TASK_SIZE must still evaluate to
 661                          * that which is appropriate to the execing application.
 662                          * This is because exit_mmap() needs to have TASK_SIZE
 663                          * evaluate to the size of the old image.
 664                          *
 665                          * So if (say) a 64-bit application is execing a 32-bit
 666                          * application it is the architecture's responsibility
 667                          * to defer changing the value of TASK_SIZE until the
 668                          * switch really is going to happen - do this in
 669                          * flush_thread().      - akpm
 670                          */
 671                         SET_PERSONALITY(loc->elf_ex, 0);
 672
 673                         interpreter = open_exec(elf_interpreter);
 674                         retval = PTR_ERR(interpreter);
 675                         if (IS_ERR(interpreter))
 676                                 goto out_free_interp;
 677
 678                         /*
 679                          * If the binary is not readable then enforce
 680                          * mm->dumpable = 0 regardless of the interpreter's
 681                          * permissions.
 682                          */
 683                         if (file_permission(interpreter, MAY_READ) < 0)
 684                                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
 685
 686                         retval = kernel_read(interpreter, 0, bprm->buf,
 687                                              BINPRM_BUF_SIZE);
 688                         if (retval != BINPRM_BUF_SIZE) {
 689                                 if (retval >= 0)
 690                                         retval = -EIO;
 691                                 goto out_free_dentry;
 692                         }
 693
 694                         /* Get the exec headers */
 695                         loc->interp_ex = *((struct exec *)bprm->buf);
 696                         loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
 697                         break;
 698                 }
 699                 elf_ppnt++;
 700         }
 701
 702         elf_ppnt = elf_phdata;
 703         for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
 704                 if (elf_ppnt->p_type == PT_GNU_STACK) {
 705                         if (elf_ppnt->p_flags & PF_X)
 706                                 executable_stack = EXSTACK_ENABLE_X;
 707                         else
 708                                 executable_stack = EXSTACK_DISABLE_X;
 709                         break;
 710                 }
 711
 712         /* Some simple consistency checks for the interpreter */
 713         if (elf_interpreter) {
 714                 retval = -ELIBBAD;
 715                 /* Not an ELF interpreter */
 716                 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 717                         goto out_free_dentry;
 718                 /* Verify the interpreter has a valid arch */
 719                 if (!elf_check_arch(&loc->interp_elf_ex))
 720                         goto out_free_dentry;
 721         } else {
 722                 /* Executables without an interpreter also need a personality  */
 723                 SET_PERSONALITY(loc->elf_ex, 0);
 724         }
 725
 726         /* Flush all traces of the currently running executable */
 727         retval = flush_old_exec(bprm);
 728         if (retval)
 729                 goto out_free_dentry;
 730
 731         /* Discard our unneeded old files struct */
 732         if (files) {
 733                 put_files_struct(files);
 734                 files = NULL;
 735         }
 736
 737         /* OK, This is the point of no return */
 738         current->flags &= ~PF_FORKNOEXEC;
 739         current->mm->def_flags = def_flags;
 740
 741         /* Do this immediately, since STACK_TOP as used in setup_arg_pages
 742            may depend on the personality.  */
 743         SET_PERSONALITY(loc->elf_ex, 0);
 744         if (elf_read_implies_exec(loc->elf_ex, executable_stack))
 745                 current->personality |= READ_IMPLIES_EXEC;
 746
 747         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 748                 current->flags |= PF_RANDOMIZE;
 749         arch_pick_mmap_layout(current->mm);
 750
 751         /* Do this so that we can load the interpreter, if need be.  We will
 752            change some of these later */
 753         current->mm->free_area_cache = current->mm->mmap_base;
 754         current->mm->cached_hole_size = 0;
 755         retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
 756                                  executable_stack);
 757         if (retval < 0) {
 758                 send_sig(SIGKILL, current, 0);
 759                 goto out_free_dentry;
 760         }
 761
 762         current->mm->start_stack = bprm->p;
 763
 764         /* Now we do a little grungy work by mmaping the ELF image into
 765            the correct location in memory. */
 766         for(i = 0, elf_ppnt = elf_phdata;
 767             i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
 768                 int elf_prot = 0, elf_flags;
 769                 unsigned long k, vaddr;
 770
 771                 if (elf_ppnt->p_type != PT_LOAD)
 772                         continue;
 773
 774                 if (unlikely (elf_brk > elf_bss)) {
 775                         unsigned long nbyte;
 776
 777                         /* There was a PT_LOAD segment with p_memsz > p_filesz
 778                            before this one. Map anonymous pages, if needed,
 779                            and clear the area.  */
 780                         retval = set_brk (elf_bss + load_bias,
 781                                           elf_brk + load_bias);
 782                         if (retval) {
 783                                 send_sig(SIGKILL, current, 0);
 784                                 goto out_free_dentry;
 785                         }
 786                         nbyte = ELF_PAGEOFFSET(elf_bss);
 787                         if (nbyte) {
 788                                 nbyte = ELF_MIN_ALIGN - nbyte;
 789                                 if (nbyte > elf_brk - elf_bss)
 790                                         nbyte = elf_brk - elf_bss;
 791                                 if (clear_user((void __user *)elf_bss +
 792                                                         load_bias, nbyte)) {
 793                                         /*
 794                                          * This bss-zeroing can fail if the ELF
 795                                          * file specifies odd protections. So
 796                                          * we don't check the return value
 797                                          */
 798                                 }
 799                         }
 800                 }
 801
 802                 if (elf_ppnt->p_flags & PF_R)
 803                         elf_prot |= PROT_READ;
 804                 if (elf_ppnt->p_flags & PF_W)
 805                         elf_prot |= PROT_WRITE;
 806                 if (elf_ppnt->p_flags & PF_X)
 807                         elf_prot |= PROT_EXEC;
 808
 809                 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
 810
 811                 vaddr = elf_ppnt->p_vaddr;
 812                 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
 813                         elf_flags |= MAP_FIXED;
 814                 } else if (loc->elf_ex.e_type == ET_DYN) {
 815                         /* Try and get dynamic programs out of the way of the
 816                          * default mmap base, as well as whatever program they
 817                          * might try to exec.  This is because the brk will
 818                          * follow the loader, and is not movable.  */
 819 #ifdef CONFIG_X86
 820                         load_bias = 0;
 821 #else
 822                         load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 823 #endif
 824                 }
 825
 826                 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
 827                                 elf_prot, elf_flags, 0);
 828                 if (BAD_ADDR(error)) {
 829                         send_sig(SIGKILL, current, 0);
 830                         retval = IS_ERR((void *)error) ?
 831                                 PTR_ERR((void*)error) : -EINVAL;
 832                         goto out_free_dentry;
 833                 }
 834
 835                 if (!load_addr_set) {
 836                         load_addr_set = 1;
 837                         load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
 838                         if (loc->elf_ex.e_type == ET_DYN) {
 839                                 load_bias += error -
 840                                              ELF_PAGESTART(load_bias + vaddr);
 841                                 load_addr += load_bias;
 842                                 reloc_func_desc = load_bias;
 843                         }
 844                 }
 845                 k = elf_ppnt->p_vaddr;
 846                 if (k < start_code)
 847                         start_code = k;
 848                 if (start_data < k)
 849                         start_data = k;
 850
 851                 /*
 852                  * Check to see if the section's size will overflow the
 853                  * allowed task size. Note that p_filesz must always be
 854                  * <= p_memsz so it is only necessary to check p_memsz.
 855                  */
 856                 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
 857                     elf_ppnt->p_memsz > TASK_SIZE ||
 858                     TASK_SIZE - elf_ppnt->p_memsz < k) {
 859                         /* set_brk can never work. Avoid overflows. */
 860                         send_sig(SIGKILL, current, 0);
 861                         retval = -EINVAL;
 862                         goto out_free_dentry;
 863                 }
 864
 865                 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
 866
 867                 if (k > elf_bss)
 868                         elf_bss = k;
 869                 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
 870                         end_code = k;
 871                 if (end_data < k)
 872                         end_data = k;
 873                 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
 874                 if (k > elf_brk)
 875                         elf_brk = k;
 876         }
 877
 878         loc->elf_ex.e_entry += load_bias;
 879         elf_bss += load_bias;
 880         elf_brk += load_bias;
 881         start_code += load_bias;
 882         end_code += load_bias;
 883         start_data += load_bias;
 884         end_data += load_bias;
 885
 886         /* Calling set_brk effectively mmaps the pages that we need
 887          * for the bss and break sections.  We must do this before
 888          * mapping in the interpreter, to make sure it doesn't wind
 889          * up getting placed where the bss needs to go.
 890          */
 891         retval = set_brk(elf_bss, elf_brk);
 892         if (retval) {
 893                 send_sig(SIGKILL, current, 0);
 894                 goto out_free_dentry;
 895         }
 896         if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
 897                 send_sig(SIGSEGV, current, 0);
 898                 retval = -EFAULT; /* Nobody gets to see this, but.. */
 899                 goto out_free_dentry;
 900         }
 901
 902         if (elf_interpreter) {
 903                 unsigned long uninitialized_var(interp_map_addr);
 904
 905                 elf_entry = load_elf_interp(&loc->interp_elf_ex,
 906                                             interpreter,
 907                                             &interp_map_addr,
 908                                             load_bias);
 909                 if (!IS_ERR((void *)elf_entry)) {
 910                         /*
 911                          * load_elf_interp() returns relocation
 912                          * adjustment
 913                          */
 914                         interp_load_addr = elf_entry;
 915                         elf_entry += loc->interp_elf_ex.e_entry;
 916                 }
 917                 if (BAD_ADDR(elf_entry)) {
 918                         force_sig(SIGSEGV, current);
 919                         retval = IS_ERR((void *)elf_entry) ?
 920                                         (int)elf_entry : -EINVAL;
 921                         goto out_free_dentry;
 922                 }
 923                 reloc_func_desc = interp_load_addr;
 924
 925                 allow_write_access(interpreter);
 926                 fput(interpreter);
 927                 kfree(elf_interpreter);
 928         } else {
 929                 elf_entry = loc->elf_ex.e_entry;
 930                 if (BAD_ADDR(elf_entry)) {
 931                         force_sig(SIGSEGV, current);
 932                         retval = -EINVAL;
 933                         goto out_free_dentry;
 934                 }
 935         }
 936
 937         kfree(elf_phdata);
 938
 939         sys_close(elf_exec_fileno);
 940
 941         set_binfmt(&elf_format);
 942
 943 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
 944         retval = arch_setup_additional_pages(bprm, executable_stack);
 945         if (retval < 0) {
 946                 send_sig(SIGKILL, current, 0);
 947                 goto out;
 948         }
 949 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
 950
 951         compute_creds(bprm);
 952         current->flags &= ~PF_FORKNOEXEC;
 953         retval = create_elf_tables(bprm, &loc->elf_ex,
 954                           load_addr, interp_load_addr);
 955         if (retval < 0) {
 956                 send_sig(SIGKILL, current, 0);
 957                 goto out;
 958         }
 959         /* N.B. passed_fileno might not be initialized? */
 960         current->mm->end_code = end_code;
 961         current->mm->start_code = start_code;
 962         current->mm->start_data = start_data;
 963         current->mm->end_data = end_data;
 964         current->mm->start_stack = bprm->p;
 965
 966 #ifdef arch_randomize_brk
 967         if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1))
 968                 current->mm->brk = current->mm->start_brk =
 969                         arch_randomize_brk(current->mm);
 970 #endif
 971
 972         if (current->personality & MMAP_PAGE_ZERO) {
 973                 /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
 974                    and some applications "depend" upon this behavior.
 975                    Since we do not have the power to recompile these, we
 976                    emulate the SVr4 behavior. Sigh. */
 977                 down_write(&current->mm->mmap_sem);
 978                 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
 979                                 MAP_FIXED | MAP_PRIVATE, 0);
 980                 up_write(&current->mm->mmap_sem);
 981         }
 982
 983 #ifdef ELF_PLAT_INIT
 984         /*
 985          * The ABI may specify that certain registers be set up in special
 986          * ways (on i386 %edx is the address of a DT_FINI function, for
 987          * example.  In addition, it may also specify (eg, PowerPC64 ELF)
 988          * that the e_entry field is the address of the function descriptor
 989          * for the startup routine, rather than the address of the startup
 990          * routine itself.  This macro performs whatever initialization to
 991          * the regs structure is required as well as any relocations to the
 992          * function descriptor entries when executing dynamically links apps.
 993          */
 994         ELF_PLAT_INIT(regs, reloc_func_desc);
 995 #endif
 996
 997         start_thread(regs, elf_entry, bprm->p);
 998         if (unlikely(current->ptrace & PT_PTRACED)) {
 999                 if (current->ptrace & PT_TRACE_EXEC)
1000                         ptrace_notify ((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
1001                 else
1002                         send_sig(SIGTRAP, current, 0);
1003         }
1004         retval = 0;
1005 out:
1006         kfree(loc);
1007 out_ret:
1008         return retval;
1009
1010         /* error cleanup */
1011 out_free_dentry:
1012         allow_write_access(interpreter);
1013         if (interpreter)
1014                 fput(interpreter);
1015 out_free_interp:
1016         kfree(elf_interpreter);
1017 out_free_file:
1018         sys_close(elf_exec_fileno);
1019 out_free_fh:
1020         if (files)
1021                 reset_files_struct(current, files);
1022 out_free_ph:
1023         kfree(elf_phdata);
1024         goto out;
1025 }
1026
1027 /* This is really simpleminded and specialized - we are loading an
1028    a.out library that is given an ELF header. */
1029 static int load_elf_library(struct file *file)
1030 {
1031         struct elf_phdr *elf_phdata;
1032         struct elf_phdr *eppnt;
1033         unsigned long elf_bss, bss, len;
1034         int retval, error, i, j;
1035         struct elfhdr elf_ex;
1036
1037         error = -ENOEXEC;
1038         retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1039         if (retval != sizeof(elf_ex))
1040                 goto out;
1041
1042         if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1043                 goto out;
1044
1045         /* First of all, some simple consistency checks */
1046         if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1047             !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1048                 goto out;
1049
1050         /* Now read in all of the header information */
1051
1052         j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1053         /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1054
1055         error = -ENOMEM;
1056         elf_phdata = kmalloc(j, GFP_KERNEL);
1057         if (!elf_phdata)
1058                 goto out;
1059
1060         eppnt = elf_phdata;
1061         error = -ENOEXEC;
1062         retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1063         if (retval != j)
1064                 goto out_free_ph;
1065
1066         for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1067                 if ((eppnt + i)->p_type == PT_LOAD)
1068                         j++;
1069         if (j != 1)
1070                 goto out_free_ph;
1071
1072         while (eppnt->p_type != PT_LOAD)
1073                 eppnt++;
1074
1075         /* Now use mmap to map the library into memory. */
1076         down_write(&current->mm->mmap_sem);
1077         error = do_mmap(file,
1078                         ELF_PAGESTART(eppnt->p_vaddr),
1079                         (eppnt->p_filesz +
1080                          ELF_PAGEOFFSET(eppnt->p_vaddr)),
1081                         PROT_READ | PROT_WRITE | PROT_EXEC,
1082                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1083                         (eppnt->p_offset -
1084                          ELF_PAGEOFFSET(eppnt->p_vaddr)));
1085         up_write(&current->mm->mmap_sem);
1086         if (error != ELF_PAGESTART(eppnt->p_vaddr))
1087                 goto out_free_ph;
1088
1089         elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1090         if (padzero(elf_bss)) {
1091                 error = -EFAULT;
1092                 goto out_free_ph;
1093         }
1094
1095         len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1096                             ELF_MIN_ALIGN - 1);
1097         bss = eppnt->p_memsz + eppnt->p_vaddr;
1098         if (bss > len) {
1099                 down_write(&current->mm->mmap_sem);
1100                 do_brk(len, bss - len);
1101                 up_write(&current->mm->mmap_sem);
1102         }
1103         error = 0;
1104
1105 out_free_ph:
1106         kfree(elf_phdata);
1107 out:
1108         return error;
1109 }
1110
1111 /*
1112  * Note that some platforms still use traditional core dumps and not
1113  * the ELF core dump.  Each platform can select it as appropriate.
1114  */
1115 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1116
1117 /*
1118  * ELF core dumper
1119  *
1120  * Modelled on fs/exec.c:aout_core_dump()
1121  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1122  */
1123 /*
1124  * These are the only things you should do on a core-file: use only these
1125  * functions to write out all the necessary info.
1126  */
1127 static int dump_write(struct file *file, const void *addr, int nr)
1128 {
1129         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1130 }
1131
1132 static int dump_seek(struct file *file, loff_t off)
1133 {
1134         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
1135                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
1136                         return 0;
1137         } else {
1138                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
1139                 if (!buf)
1140                         return 0;
1141                 while (off > 0) {
1142                         unsigned long n = off;
1143                         if (n > PAGE_SIZE)
1144                                 n = PAGE_SIZE;
1145                         if (!dump_write(file, buf, n))
1146                                 return 0;
1147                         off -= n;
1148                 }
1149                 free_page((unsigned long)buf);
1150         }
1151         return 1;
1152 }
1153
1154 /*
1155  * Decide what to dump of a segment, part, all or none.
1156  */
1157 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1158                                    unsigned long mm_flags)
1159 {
1160         /* The vma can be set up to tell us the answer directly.  */
1161         if (vma->vm_flags & VM_ALWAYSDUMP)
1162                 goto whole;
1163
1164         /* Do not dump I/O mapped devices or special mappings */
1165         if (vma->vm_flags & (VM_IO | VM_RESERVED))
1166                 return 0;
1167
1168 #define FILTER(type)    (mm_flags & (1UL << MMF_DUMP_##type))
1169
1170         /* By default, dump shared memory if mapped from an anonymous file. */
1171         if (vma->vm_flags & VM_SHARED) {
1172                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1173                     FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1174                         goto whole;
1175                 return 0;
1176         }
1177
1178         /* Dump segments that have been written to.  */
1179         if (vma->anon_vma && FILTER(ANON_PRIVATE))
1180                 goto whole;
1181         if (vma->vm_file == NULL)
1182                 return 0;
1183
1184         if (FILTER(MAPPED_PRIVATE))
1185                 goto whole;
1186
1187         /*
1188          * If this looks like the beginning of a DSO or executable mapping,
1189          * check for an ELF header.  If we find one, dump the first page to
1190          * aid in determining what was mapped here.
1191          */
1192         if (FILTER(ELF_HEADERS) && vma->vm_file != NULL && vma->vm_pgoff == 0) {
1193                 u32 __user *header = (u32 __user *) vma->vm_start;
1194                 u32 word;
1195                 /*
1196                  * Doing it this way gets the constant folded by GCC.
1197                  */
1198                 union {
1199                         u32 cmp;
1200                         char elfmag[SELFMAG];
1201                 } magic;
1202                 BUILD_BUG_ON(SELFMAG != sizeof word);
1203                 magic.elfmag[EI_MAG0] = ELFMAG0;
1204                 magic.elfmag[EI_MAG1] = ELFMAG1;
1205                 magic.elfmag[EI_MAG2] = ELFMAG2;
1206                 magic.elfmag[EI_MAG3] = ELFMAG3;
1207                 if (get_user(word, header) == 0 && word == magic.cmp)
1208                         return PAGE_SIZE;
1209         }
1210
1211 #undef  FILTER
1212
1213         return 0;
1214
1215 whole:
1216         return vma->vm_end - vma->vm_start;
1217 }
1218
1219 /* An ELF note in memory */
1220 struct memelfnote
1221 {
1222         const char *name;
1223         int type;
1224         unsigned int datasz;
1225         void *data;
1226 };
1227
1228 static int notesize(struct memelfnote *en)
1229 {
1230         int sz;
1231
1232         sz = sizeof(struct elf_note);
1233         sz += roundup(strlen(en->name) + 1, 4);
1234         sz += roundup(en->datasz, 4);
1235
1236         return sz;
1237 }
1238
1239 #define DUMP_WRITE(addr, nr, foffset)   \
1240         do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1241
1242 static int alignfile(struct file *file, loff_t *foffset)
1243 {
1244         static const char buf[4] = { 0, };
1245         DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1246         return 1;
1247 }
1248
1249 static int writenote(struct memelfnote *men, struct file *file,
1250                         loff_t *foffset)
1251 {
1252         struct elf_note en;
1253         en.n_namesz = strlen(men->name) + 1;
1254         en.n_descsz = men->datasz;
1255         en.n_type = men->type;
1256
1257         DUMP_WRITE(&en, sizeof(en), foffset);
1258         DUMP_WRITE(men->name, en.n_namesz, foffset);
1259         if (!alignfile(file, foffset))
1260                 return 0;
1261         DUMP_WRITE(men->data, men->datasz, foffset);
1262         if (!alignfile(file, foffset))
1263                 return 0;
1264
1265         return 1;
1266 }
1267 #undef DUMP_WRITE
1268
1269 #define DUMP_WRITE(addr, nr)    \
1270         if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1271                 goto end_coredump;
1272 #define DUMP_SEEK(off)  \
1273         if (!dump_seek(file, (off))) \
1274                 goto end_coredump;
1275
1276 static void fill_elf_header(struct elfhdr *elf, int segs,
1277                             u16 machine, u32 flags, u8 osabi)
1278 {
1279         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1280         elf->e_ident[EI_CLASS] = ELF_CLASS;
1281         elf->e_ident[EI_DATA] = ELF_DATA;
1282         elf->e_ident[EI_VERSION] = EV_CURRENT;
1283         elf->e_ident[EI_OSABI] = ELF_OSABI;
1284         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1285
1286         elf->e_type = ET_CORE;
1287         elf->e_machine = machine;
1288         elf->e_version = EV_CURRENT;
1289         elf->e_entry = 0;
1290         elf->e_phoff = sizeof(struct elfhdr);
1291         elf->e_shoff = 0;
1292         elf->e_flags = flags;
1293         elf->e_ehsize = sizeof(struct elfhdr);
1294         elf->e_phentsize = sizeof(struct elf_phdr);
1295         elf->e_phnum = segs;
1296         elf->e_shentsize = 0;
1297         elf->e_shnum = 0;
1298         elf->e_shstrndx = 0;
1299         return;
1300 }
1301
1302 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1303 {
1304         phdr->p_type = PT_NOTE;
1305         phdr->p_offset = offset;
1306         phdr->p_vaddr = 0;
1307         phdr->p_paddr = 0;
1308         phdr->p_filesz = sz;
1309         phdr->p_memsz = 0;
1310         phdr->p_flags = 0;
1311         phdr->p_align = 0;
1312         return;
1313 }
1314
1315 static void fill_note(struct memelfnote *note, const char *name, int type,
1316                 unsigned int sz, void *data)
1317 {
1318         note->name = name;
1319         note->type = type;
1320         note->datasz = sz;
1321         note->data = data;
1322         return;
1323 }
1324
1325 /*
1326  * fill up all the fields in prstatus from the given task struct, except
1327  * registers which need to be filled up separately.
1328  */
1329 static void fill_prstatus(struct elf_prstatus *prstatus,
1330                 struct task_struct *p, long signr)
1331 {
1332         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1333         prstatus->pr_sigpend = p->pending.signal.sig[0];
1334         prstatus->pr_sighold = p->blocked.sig[0];
1335         prstatus->pr_pid = task_pid_vnr(p);
1336         prstatus->pr_ppid = task_pid_vnr(p->real_parent);
1337         prstatus->pr_pgrp = task_pgrp_vnr(p);
1338         prstatus->pr_sid = task_session_vnr(p);
1339         if (thread_group_leader(p)) {
1340                 /*
1341                  * This is the record for the group leader.  Add in the
1342                  * cumulative times of previous dead threads.  This total
1343                  * won't include the time of each live thread whose state
1344                  * is included in the core dump.  The final total reported
1345                  * to our parent process when it calls wait4 will include
1346                  * those sums as well as the little bit more time it takes
1347                  * this and each other thread to finish dying after the
1348                  * core dump synchronization phase.
1349                  */
1350                 cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1351                                    &prstatus->pr_utime);
1352                 cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1353                                    &prstatus->pr_stime);
1354         } else {
1355                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1356                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1357         }
1358         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1359         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1360 }
1361
1362 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1363                        struct mm_struct *mm)
1364 {
1365         unsigned int i, len;
1366
1367         /* first copy the parameters from user space */
1368         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1369
1370         len = mm->arg_end - mm->arg_start;
1371         if (len >= ELF_PRARGSZ)
1372                 len = ELF_PRARGSZ-1;
1373         if (copy_from_user(&psinfo->pr_psargs,
1374                            (const char __user *)mm->arg_start, len))
1375                 return -EFAULT;
1376         for(i = 0; i < len; i++)
1377                 if (psinfo->pr_psargs[i] == 0)
1378                         psinfo->pr_psargs[i] = ' ';
1379         psinfo->pr_psargs[len] = 0;
1380
1381         psinfo->pr_pid = task_pid_vnr(p);
1382         psinfo->pr_ppid = task_pid_vnr(p->real_parent);
1383         psinfo->pr_pgrp = task_pgrp_vnr(p);
1384         psinfo->pr_sid = task_session_vnr(p);
1385
1386         i = p->state ? ffz(~p->state) + 1 : 0;
1387         psinfo->pr_state = i;
1388         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1389         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1390         psinfo->pr_nice = task_nice(p);
1391         psinfo->pr_flag = p->flags;
1392         SET_UID(psinfo->pr_uid, p->uid);
1393         SET_GID(psinfo->pr_gid, p->gid);
1394         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1395
1396         return 0;
1397 }
1398
1399 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1400 {
1401         elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1402         int i = 0;
1403         do
1404                 i += 2;
1405         while (auxv[i - 2] != AT_NULL);
1406         fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1407 }
1408
1409 #ifdef CORE_DUMP_USE_REGSET
1410 #include <linux/regset.h>
1411
1412 struct elf_thread_core_info {
1413         struct elf_thread_core_info *next;
1414         struct task_struct *task;
1415         struct elf_prstatus prstatus;
1416         struct memelfnote notes[0];
1417 };
1418
1419 struct elf_note_info {
1420         struct elf_thread_core_info *thread;
1421         struct memelfnote psinfo;
1422         struct memelfnote auxv;
1423         size_t size;
1424         int thread_notes;
1425 };
1426
1427 static int fill_thread_core_info(struct elf_thread_core_info *t,
1428                                  const struct user_regset_view *view,
1429                                  long signr, size_t *total)
1430 {
1431         unsigned int i;
1432
1433         /*
1434          * NT_PRSTATUS is the one special case, because the regset data
1435          * goes into the pr_reg field inside the note contents, rather
1436          * than being the whole note contents.  We fill the reset in here.
1437          * We assume that regset 0 is NT_PRSTATUS.
1438          */
1439         fill_prstatus(&t->prstatus, t->task, signr);
1440         (void) view->regsets[0].get(t->task, &view->regsets[0],
1441                                     0, sizeof(t->prstatus.pr_reg),
1442                                     &t->prstatus.pr_reg, NULL);
1443
1444         fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1445                   sizeof(t->prstatus), &t->prstatus);
1446         *total += notesize(&t->notes[0]);
1447
1448         /*
1449          * Each other regset might generate a note too.  For each regset
1450          * that has no core_note_type or is inactive, we leave t->notes[i]
1451          * all zero and we'll know to skip writing it later.
1452          */
1453         for (i = 1; i < view->n; ++i) {
1454                 const struct user_regset *regset = &view->regsets[i];
1455                 if (regset->core_note_type &&
1456                     (!regset->active || regset->active(t->task, regset))) {
1457                         int ret;
1458                         size_t size = regset->n * regset->size;
1459                         void *data = kmalloc(size, GFP_KERNEL);
1460                         if (unlikely(!data))
1461                                 return 0;
1462                         ret = regset->get(t->task, regset,
1463                                           0, size, data, NULL);
1464                         if (unlikely(ret))
1465                                 kfree(data);
1466                         else {
1467                                 if (regset->core_note_type != NT_PRFPREG)
1468                                         fill_note(&t->notes[i], "LINUX",
1469                                                   regset->core_note_type,
1470                                                   size, data);
1471                                 else {
1472                                         t->prstatus.pr_fpvalid = 1;
1473                                         fill_note(&t->notes[i], "CORE",
1474                                                   NT_PRFPREG, size, data);
1475                                 }
1476                                 *total += notesize(&t->notes[i]);
1477                         }
1478                 }
1479         }
1480
1481         return 1;
1482 }
1483
1484 static int fill_note_info(struct elfhdr *elf, int phdrs,
1485                           struct elf_note_info *info,
1486                           long signr, struct pt_regs *regs)
1487 {
1488         struct task_struct *dump_task = current;
1489         const struct user_regset_view *view = task_user_regset_view(dump_task);
1490         struct elf_thread_core_info *t;
1491         struct elf_prpsinfo *psinfo;
1492         struct task_struct *g, *p;
1493         unsigned int i;
1494
1495         info->size = 0;
1496         info->thread = NULL;
1497
1498         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1499         fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1500
1501         if (psinfo == NULL)
1502                 return 0;
1503
1504         /*
1505          * Figure out how many notes we're going to need for each thread.
1506          */
1507         info->thread_notes = 0;
1508         for (i = 0; i < view->n; ++i)
1509                 if (view->regsets[i].core_note_type != 0)
1510                         ++info->thread_notes;
1511
1512         /*
1513          * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1514          * since it is our one special case.
1515          */
1516         if (unlikely(info->thread_notes == 0) ||
1517             unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1518                 WARN_ON(1);
1519                 return 0;
1520         }
1521
1522         /*
1523          * Initialize the ELF file header.
1524          */
1525         fill_elf_header(elf, phdrs,
1526                         view->e_machine, view->e_flags, view->ei_osabi);
1527
1528         /*
1529          * Allocate a structure for each thread.
1530          */
1531         rcu_read_lock();
1532         do_each_thread(g, p)
1533                 if (p->mm == dump_task->mm) {
1534                         t = kzalloc(offsetof(struct elf_thread_core_info,
1535                                              notes[info->thread_notes]),
1536                                     GFP_ATOMIC);
1537                         if (unlikely(!t)) {
1538                                 rcu_read_unlock();
1539                                 return 0;
1540                         }
1541                         t->task = p;
1542                         if (p == dump_task || !info->thread) {
1543                                 t->next = info->thread;
1544                                 info->thread = t;
1545                         } else {
1546                                 /*
1547                                  * Make sure to keep the original task at
1548                                  * the head of the list.
1549                                  */
1550                                 t->next = info->thread->next;
1551                                 info->thread->next = t;
1552                         }
1553                 }
1554         while_each_thread(g, p);
1555         rcu_read_unlock();
1556
1557         /*
1558          * Now fill in each thread's information.
1559          */
1560         for (t = info->thread; t != NULL; t = t->next)
1561                 if (!fill_thread_core_info(t, view, signr, &info->size))
1562                         return 0;
1563
1564         /*
1565          * Fill in the two process-wide notes.
1566          */
1567         fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1568         info->size += notesize(&info->psinfo);
1569
1570         fill_auxv_note(&info->auxv, current->mm);
1571         info->size += notesize(&info->auxv);
1572
1573         return 1;
1574 }
1575
1576 static size_t get_note_info_size(struct elf_note_info *info)
1577 {
1578         return info->size;
1579 }
1580
1581 /*
1582  * Write all the notes for each thread.  When writing the first thread, the
1583  * process-wide notes are interleaved after the first thread-specific note.
1584  */
1585 static int write_note_info(struct elf_note_info *info,
1586                            struct file *file, loff_t *foffset)
1587 {
1588         bool first = 1;
1589         struct elf_thread_core_info *t = info->thread;
1590
1591         do {
1592                 int i;
1593
1594                 if (!writenote(&t->notes[0], file, foffset))
1595                         return 0;
1596
1597                 if (first && !writenote(&info->psinfo, file, foffset))
1598                         return 0;
1599                 if (first && !writenote(&info->auxv, file, foffset))
1600                         return 0;
1601
1602                 for (i = 1; i < info->thread_notes; ++i)
1603                         if (t->notes[i].data &&
1604                             !writenote(&t->notes[i], file, foffset))
1605                                 return 0;
1606
1607                 first = 0;
1608                 t = t->next;
1609         } while (t);
1610
1611         return 1;
1612 }
1613
1614 static void free_note_info(struct elf_note_info *info)
1615 {
1616         struct elf_thread_core_info *threads = info->thread;
1617         while (threads) {
1618                 unsigned int i;
1619                 struct elf_thread_core_info *t = threads;
1620                 threads = t->next;
1621                 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1622                 for (i = 1; i < info->thread_notes; ++i)
1623                         kfree(t->notes[i].data);
1624                 kfree(t);
1625         }
1626         kfree(info->psinfo.data);
1627 }
1628
1629 #else
1630
1631 /* Here is the structure in which status of each thread is captured. */
1632 struct elf_thread_status
1633 {
1634         struct list_head list;
1635         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1636         elf_fpregset_t fpu;             /* NT_PRFPREG */
1637         struct task_struct *thread;
1638 #ifdef ELF_CORE_COPY_XFPREGS
1639         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1640 #endif
1641         struct memelfnote notes[3];
1642         int num_notes;
1643 };
1644
1645 /*
1646  * In order to add the specific thread information for the elf file format,
1647  * we need to keep a linked list of every threads pr_status and then create
1648  * a single section for them in the final core file.
1649  */
1650 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1651 {
1652         int sz = 0;
1653         struct task_struct *p = t->thread;
1654         t->num_notes = 0;
1655
1656         fill_prstatus(&t->prstatus, p, signr);
1657         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1658
1659         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1660                   &(t->prstatus));
1661         t->num_notes++;
1662         sz += notesize(&t->notes[0]);
1663
1664         if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1665                                                                 &t->fpu))) {
1666                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1667                           &(t->fpu));
1668                 t->num_notes++;
1669                 sz += notesize(&t->notes[1]);
1670         }
1671
1672 #ifdef ELF_CORE_COPY_XFPREGS
1673         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1674                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1675                           sizeof(t->xfpu), &t->xfpu);
1676                 t->num_notes++;
1677                 sz += notesize(&t->notes[2]);
1678         }
1679 #endif
1680         return sz;
1681 }
1682
1683 struct elf_note_info {
1684         struct memelfnote *notes;
1685         struct elf_prstatus *prstatus;  /* NT_PRSTATUS */
1686         struct elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
1687         struct list_head thread_list;
1688         elf_fpregset_t *fpu;
1689 #ifdef ELF_CORE_COPY_XFPREGS
1690         elf_fpxregset_t *xfpu;
1691 #endif
1692         int thread_status_size;
1693         int numnote;
1694 };
1695
1696 static int fill_note_info(struct elfhdr *elf, int phdrs,
1697                           struct elf_note_info *info,
1698                           long signr, struct pt_regs *regs)
1699 {
1700 #define NUM_NOTES       6
1701         struct list_head *t;
1702         struct task_struct *g, *p;
1703
1704         info->notes = NULL;
1705         info->prstatus = NULL;
1706         info->psinfo = NULL;
1707         info->fpu = NULL;
1708 #ifdef ELF_CORE_COPY_XFPREGS
1709         info->xfpu = NULL;
1710 #endif
1711         INIT_LIST_HEAD(&info->thread_list);
1712
1713         info->notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote),
1714                               GFP_KERNEL);
1715         if (!info->notes)
1716                 return 0;
1717         info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1718         if (!info->psinfo)
1719                 return 0;
1720         info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1721         if (!info->prstatus)
1722                 return 0;
1723         info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1724         if (!info->fpu)
1725                 return 0;
1726 #ifdef ELF_CORE_COPY_XFPREGS
1727         info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1728         if (!info->xfpu)
1729                 return 0;
1730 #endif
1731
1732         info->thread_status_size = 0;
1733         if (signr) {
1734                 struct elf_thread_status *tmp;
1735                 rcu_read_lock();
1736                 do_each_thread(g, p)
1737                         if (current->mm == p->mm && current != p) {
1738                                 tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC);
1739                                 if (!tmp) {
1740                                         rcu_read_unlock();
1741                                         return 0;
1742                                 }
1743                                 tmp->thread = p;
1744                                 list_add(&tmp->list, &info->thread_list);
1745                         }
1746                 while_each_thread(g, p);
1747                 rcu_read_unlock();
1748                 list_for_each(t, &info->thread_list) {
1749                         struct elf_thread_status *tmp;
1750                         int sz;
1751
1752                         tmp = list_entry(t, struct elf_thread_status, list);
1753                         sz = elf_dump_thread_status(signr, tmp);
1754                         info->thread_status_size += sz;
1755                 }
1756         }
1757         /* now collect the dump for the current */
1758         memset(info->prstatus, 0, sizeof(*info->prstatus));
1759         fill_prstatus(info->prstatus, current, signr);
1760         elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1761
1762         /* Set up header */
1763         fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1764
1765         /*
1766          * Set up the notes in similar form to SVR4 core dumps made
1767          * with info from their /proc.
1768          */
1769
1770         fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1771                   sizeof(*info->prstatus), info->prstatus);
1772         fill_psinfo(info->psinfo, current->group_leader, current->mm);
1773         fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1774                   sizeof(*info->psinfo), info->psinfo);
1775
1776         info->numnote = 2;
1777
1778         fill_auxv_note(&info->notes[info->numnote++], current->mm);
1779
1780         /* Try to dump the FPU. */
1781         info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1782                                                                info->fpu);
1783         if (info->prstatus->pr_fpvalid)
1784                 fill_note(info->notes + info->numnote++,
1785                           "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1786 #ifdef ELF_CORE_COPY_XFPREGS
1787         if (elf_core_copy_task_xfpregs(current, info->xfpu))
1788                 fill_note(info->notes + info->numnote++,
1789                           "LINUX", ELF_CORE_XFPREG_TYPE,
1790                           sizeof(*info->xfpu), info->xfpu);
1791 #endif
1792
1793         return 1;
1794
1795 #undef NUM_NOTES
1796 }
1797
1798 static size_t get_note_info_size(struct elf_note_info *info)
1799 {
1800         int sz = 0;
1801         int i;
1802
1803         for (i = 0; i < info->numnote; i++)
1804                 sz += notesize(info->notes + i);
1805
1806         sz += info->thread_status_size;
1807
1808         return sz;
1809 }
1810
1811 static int write_note_info(struct elf_note_info *info,
1812                            struct file *file, loff_t *foffset)
1813 {
1814         int i;
1815         struct list_head *t;
1816
1817         for (i = 0; i < info->numnote; i++)
1818                 if (!writenote(info->notes + i, file, foffset))
1819                         return 0;
1820
1821         /* write out the thread status notes section */
1822         list_for_each(t, &info->thread_list) {
1823                 struct elf_thread_status *tmp =
1824                                 list_entry(t, struct elf_thread_status, list);
1825
1826                 for (i = 0; i < tmp->num_notes; i++)
1827                         if (!writenote(&tmp->notes[i], file, foffset))
1828                                 return 0;
1829         }
1830
1831         return 1;
1832 }
1833
1834 static void free_note_info(struct elf_note_info *info)
1835 {
1836         while (!list_empty(&info->thread_list)) {
1837                 struct list_head *tmp = info->thread_list.next;
1838                 list_del(tmp);
1839                 kfree(list_entry(tmp, struct elf_thread_status, list));
1840         }
1841
1842         kfree(info->prstatus);
1843         kfree(info->psinfo);
1844         kfree(info->notes);
1845         kfree(info->fpu);
1846 #ifdef ELF_CORE_COPY_XFPREGS
1847         kfree(info->xfpu);
1848 #endif
1849 }
1850
1851 #endif
1852
1853 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1854                                         struct vm_area_struct *gate_vma)
1855 {
1856         struct vm_area_struct *ret = tsk->mm->mmap;
1857
1858         if (ret)
1859                 return ret;
1860         return gate_vma;
1861 }
1862 /*
1863  * Helper function for iterating across a vma list.  It ensures that the caller
1864  * will visit `gate_vma' prior to terminating the search.
1865  */
1866 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1867                                         struct vm_area_struct *gate_vma)
1868 {
1869         struct vm_area_struct *ret;
1870
1871         ret = this_vma->vm_next;
1872         if (ret)
1873                 return ret;
1874         if (this_vma == gate_vma)
1875                 return NULL;
1876         return gate_vma;
1877 }
1878
1879 /*
1880  * Actual dumper
1881  *
1882  * This is a two-pass process; first we find the offsets of the bits,
1883  * and then they are actually written out.  If we run out of core limit
1884  * we just truncate.
1885  */
1886 static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, unsigned long limit)
1887 {
1888         int has_dumped = 0;
1889         mm_segment_t fs;
1890         int segs;
1891         size_t size = 0;
1892         struct vm_area_struct *vma, *gate_vma;
1893         struct elfhdr *elf = NULL;
1894         loff_t offset = 0, dataoff, foffset;
1895         unsigned long mm_flags;
1896         struct elf_note_info info;
1897
1898         /*
1899          * We no longer stop all VM operations.
1900          *
1901          * This is because those proceses that could possibly change map_count
1902          * or the mmap / vma pages are now blocked in do_exit on current
1903          * finishing this core dump.
1904          *
1905          * Only ptrace can touch these memory addresses, but it doesn't change
1906          * the map_count or the pages allocated. So no possibility of crashing
1907          * exists while dumping the mm->vm_next areas to the core file.
1908          */
1909
1910         /* alloc memory for large data structures: too large to be on stack */
1911         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1912         if (!elf)
1913                 goto cleanup;
1914
1915         segs = current->mm->map_count;
1916 #ifdef ELF_CORE_EXTRA_PHDRS
1917         segs += ELF_CORE_EXTRA_PHDRS;
1918 #endif
1919
1920         gate_vma = get_gate_vma(current);
1921         if (gate_vma != NULL)
1922                 segs++;
1923
1924         /*
1925          * Collect all the non-memory information about the process for the
1926          * notes.  This also sets up the file header.
1927          */
1928         if (!fill_note_info(elf, segs + 1, /* including notes section */
1929                             &info, signr, regs))
1930                 goto cleanup;
1931
1932         has_dumped = 1;
1933         current->flags |= PF_DUMPCORE;
1934
1935         fs = get_fs();
1936         set_fs(KERNEL_DS);
1937
1938         DUMP_WRITE(elf, sizeof(*elf));
1939         offset += sizeof(*elf);                         /* Elf header */
1940         offset += (segs + 1) * sizeof(struct elf_phdr); /* Program headers */
1941         foffset = offset;
1942
1943         /* Write notes phdr entry */
1944         {
1945                 struct elf_phdr phdr;
1946                 size_t sz = get_note_info_size(&info);
1947
1948                 sz += elf_coredump_extra_notes_size();
1949
1950                 fill_elf_note_phdr(&phdr, sz, offset);
1951                 offset += sz;
1952                 DUMP_WRITE(&phdr, sizeof(phdr));
1953         }
1954
1955         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1956
1957         /*
1958          * We must use the same mm->flags while dumping core to avoid
1959          * inconsistency between the program headers and bodies, otherwise an
1960          * unusable core file can be generated.
1961          */
1962         mm_flags = current->mm->flags;
1963
1964         /* Write program headers for segments dump */
1965         for (vma = first_vma(current, gate_vma); vma != NULL;
1966                         vma = next_vma(vma, gate_vma)) {
1967                 struct elf_phdr phdr;
1968
1969                 phdr.p_type = PT_LOAD;
1970                 phdr.p_offset = offset;
1971                 phdr.p_vaddr = vma->vm_start;
1972                 phdr.p_paddr = 0;
1973                 phdr.p_filesz = vma_dump_size(vma, mm_flags);
1974                 phdr.p_memsz = vma->vm_end - vma->vm_start;
1975                 offset += phdr.p_filesz;
1976                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1977                 if (vma->vm_flags & VM_WRITE)
1978                         phdr.p_flags |= PF_W;
1979                 if (vma->vm_flags & VM_EXEC)
1980                         phdr.p_flags |= PF_X;
1981                 phdr.p_align = ELF_EXEC_PAGESIZE;
1982
1983                 DUMP_WRITE(&phdr, sizeof(phdr));
1984         }
1985
1986 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1987         ELF_CORE_WRITE_EXTRA_PHDRS;
1988 #endif
1989
1990         /* write out the notes section */
1991         if (!write_note_info(&info, file, &foffset))
1992                 goto end_coredump;
1993
1994         if (elf_coredump_extra_notes_write(file, &foffset))
1995                 goto end_coredump;
1996
1997         /* Align to page */
1998         DUMP_SEEK(dataoff - foffset);
1999
2000         for (vma = first_vma(current, gate_vma); vma != NULL;
2001                         vma = next_vma(vma, gate_vma)) {
2002                 unsigned long addr;
2003                 unsigned long end;
2004
2005                 end = vma->vm_start + vma_dump_size(vma, mm_flags);
2006
2007                 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2008                         struct page *page;
2009                         struct vm_area_struct *vma;
2010
2011                         if (get_user_pages(current, current->mm, addr, 1, 0, 1,
2012                                                 &page, &vma) <= 0) {
2013                                 DUMP_SEEK(PAGE_SIZE);
2014                         } else {
2015                                 if (page == ZERO_PAGE(0)) {
2016                                         if (!dump_seek(file, PAGE_SIZE)) {
2017                                                 page_cache_release(page);
2018                                                 goto end_coredump;
2019                                         }
2020                                 } else {
2021                                         void *kaddr;
2022                                         flush_cache_page(vma, addr,
2023                                                          page_to_pfn(page));
2024                                         kaddr = kmap(page);
2025                                         if ((size += PAGE_SIZE) > limit ||
2026                                             !dump_write(file, kaddr,
2027                                             PAGE_SIZE)) {
2028                                                 kunmap(page);
2029                                                 page_cache_release(page);
2030                                                 goto end_coredump;
2031                                         }
2032                                         kunmap(page);
2033                                 }
2034                                 page_cache_release(page);
2035                         }
2036                 }
2037         }
2038
2039 #ifdef ELF_CORE_WRITE_EXTRA_DATA
2040         ELF_CORE_WRITE_EXTRA_DATA;
2041 #endif
2042
2043 end_coredump:
2044         set_fs(fs);
2045
2046 cleanup:
2047         kfree(elf);
2048         free_note_info(&info);
2049         return has_dumped;
2050 }
2051
2052 #endif          /* USE_ELF_CORE_DUMP */
2053
2054 static int __init init_elf_binfmt(void)
2055 {
2056         return register_binfmt(&elf_format);
2057 }
2058
2059 static void __exit exit_elf_binfmt(void)
2060 {
2061         /* Remove the COFF and ELF loaders. */
2062         unregister_binfmt(&elf_format);
2063 }
2064
2065 core_initcall(init_elf_binfmt);
2066 module_exit(exit_elf_binfmt);
2067 MODULE_LICENSE("GPL");