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/mm.h>
  16 #include <linux/mman.h>
  17 #include <linux/errno.h>
  18 #include <linux/signal.h>
  19 #include <linux/binfmts.h>
  20 #include <linux/string.h>
  21 #include <linux/file.h>
  22 #include <linux/slab.h>
  23 #include <linux/personality.h>
  24 #include <linux/elfcore.h>
  25 #include <linux/init.h>
  26 #include <linux/highuid.h>
  27 #include <linux/compiler.h>
  28 #include <linux/highmem.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/security.h>
  31 #include <linux/random.h>
  32 #include <linux/elf.h>
  33 #include <linux/utsname.h>
  34 #include <linux/coredump.h>
  35 #include <asm/uaccess.h>
  36 #include <asm/param.h>
  37 #include <asm/page.h>
  38
  39 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
  40 static int load_elf_library(struct file *);
  41 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
  42                                 int, int, unsigned long);
  43
  44 /*
  45  * If we don't support core dumping, then supply a NULL so we
  46  * don't even try.
  47  */
  48 #ifdef CONFIG_ELF_CORE
  49 static int elf_core_dump(struct coredump_params *cprm);
  50 #else
  51 #define elf_core_dump   NULL
  52 #endif
  53
  54 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
  55 #define ELF_MIN_ALIGN   ELF_EXEC_PAGESIZE
  56 #else
  57 #define ELF_MIN_ALIGN   PAGE_SIZE
  58 #endif
  59
  60 #ifndef ELF_CORE_EFLAGS
  61 #define ELF_CORE_EFLAGS 0
  62 #endif
  63
  64 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
  65 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
  66 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
  67
  68 static struct linux_binfmt elf_format = {
  69         .module         = THIS_MODULE,
  70         .load_binary    = load_elf_binary,
  71         .load_shlib     = load_elf_library,
  72         .core_dump      = elf_core_dump,
  73         .min_coredump   = ELF_EXEC_PAGESIZE,
  74 };
  75
  76 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
  77
  78 static int set_brk(unsigned long start, unsigned long end)
  79 {
  80         start = ELF_PAGEALIGN(start);
  81         end = ELF_PAGEALIGN(end);
  82         if (end > start) {
  83                 unsigned long addr;
  84                 down_write(&current->mm->mmap_sem);
  85                 addr = do_brk(start, end - start);
  86                 up_write(&current->mm->mmap_sem);
  87                 if (BAD_ADDR(addr))
  88                         return addr;
  89         }
  90         current->mm->start_brk = current->mm->brk = end;
  91         return 0;
  92 }
  93
  94 /* We need to explicitly zero any fractional pages
  95    after the data section (i.e. bss).  This would
  96    contain the junk from the file that should not
  97    be in memory
  98  */
  99 static int padzero(unsigned long elf_bss)
 100 {
 101         unsigned long nbyte;
 102
 103         nbyte = ELF_PAGEOFFSET(elf_bss);
 104         if (nbyte) {
 105                 nbyte = ELF_MIN_ALIGN - nbyte;
 106                 if (clear_user((void __user *) elf_bss, nbyte))
 107                         return -EFAULT;
 108         }
 109         return 0;
 110 }
 111
 112 /* Let's use some macros to make this stack manipulation a little clearer */
 113 #ifdef CONFIG_STACK_GROWSUP
 114 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
 115 #define STACK_ROUND(sp, items) \
 116         ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
 117 #define STACK_ALLOC(sp, len) ({ \
 118         elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
 119         old_sp; })
 120 #else
 121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
 122 #define STACK_ROUND(sp, items) \
 123         (((unsigned long) (sp - items)) &~ 15UL)
 124 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
 125 #endif
 126
 127 #ifndef ELF_BASE_PLATFORM
 128 /*
 129  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
 130  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
 131  * will be copied to the user stack in the same manner as AT_PLATFORM.
 132  */
 133 #define ELF_BASE_PLATFORM NULL
 134 #endif
 135
 136 static int
 137 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
 138                 unsigned long load_addr, unsigned long interp_load_addr)
 139 {
 140         unsigned long p = bprm->p;
 141         int argc = bprm->argc;
 142         int envc = bprm->envc;
 143         elf_addr_t __user *argv;
 144         elf_addr_t __user *envp;
 145         elf_addr_t __user *sp;
 146         elf_addr_t __user *u_platform;
 147         elf_addr_t __user *u_base_platform;
 148         elf_addr_t __user *u_rand_bytes;
 149         const char *k_platform = ELF_PLATFORM;
 150         const char *k_base_platform = ELF_BASE_PLATFORM;
 151         unsigned char k_rand_bytes[16];
 152         int items;
 153         elf_addr_t *elf_info;
 154         int ei_index = 0;
 155         const struct cred *cred = current_cred();
 156         struct vm_area_struct *vma;
 157
 158         /*
 159          * In some cases (e.g. Hyper-Threading), we want to avoid L1
 160          * evictions by the processes running on the same package. One
 161          * thing we can do is to shuffle the initial stack for them.
 162          */
 163
 164         p = arch_align_stack(p);
 165
 166         /*
 167          * If this architecture has a platform capability string, copy it
 168          * to userspace.  In some cases (Sparc), this info is impossible
 169          * for userspace to get any other way, in others (i386) it is
 170          * merely difficult.
 171          */
 172         u_platform = NULL;
 173         if (k_platform) {
 174                 size_t len = strlen(k_platform) + 1;
 175
 176                 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 177                 if (__copy_to_user(u_platform, k_platform, len))
 178                         return -EFAULT;
 179         }
 180
 181         /*
 182          * If this architecture has a "base" platform capability
 183          * string, copy it to userspace.
 184          */
 185         u_base_platform = NULL;
 186         if (k_base_platform) {
 187                 size_t len = strlen(k_base_platform) + 1;
 188
 189                 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 190                 if (__copy_to_user(u_base_platform, k_base_platform, len))
 191                         return -EFAULT;
 192         }
 193
 194         /*
 195          * Generate 16 random bytes for userspace PRNG seeding.
 196          */
 197         get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
 198         u_rand_bytes = (elf_addr_t __user *)
 199                        STACK_ALLOC(p, sizeof(k_rand_bytes));
 200         if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
 201                 return -EFAULT;
 202
 203         /* Create the ELF interpreter info */
 204         elf_info = (elf_addr_t *)current->mm->saved_auxv;
 205         /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
 206 #define NEW_AUX_ENT(id, val) \
 207         do { \
 208                 elf_info[ei_index++] = id; \
 209                 elf_info[ei_index++] = val; \
 210         } while (0)
 211
 212 #ifdef ARCH_DLINFO
 213         /*
 214          * ARCH_DLINFO must come first so PPC can do its special alignment of
 215          * AUXV.
 216          * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
 217          * ARCH_DLINFO changes
 218          */
 219         ARCH_DLINFO;
 220 #endif
 221         NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
 222         NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
 223         NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
 224         NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
 225         NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
 226         NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
 227         NEW_AUX_ENT(AT_BASE, interp_load_addr);
 228         NEW_AUX_ENT(AT_FLAGS, 0);
 229         NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
 230         NEW_AUX_ENT(AT_UID, cred->uid);
 231         NEW_AUX_ENT(AT_EUID, cred->euid);
 232         NEW_AUX_ENT(AT_GID, cred->gid);
 233         NEW_AUX_ENT(AT_EGID, cred->egid);
 234         NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
 235         NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
 236         NEW_AUX_ENT(AT_EXECFN, bprm->exec);
 237         if (k_platform) {
 238                 NEW_AUX_ENT(AT_PLATFORM,
 239                             (elf_addr_t)(unsigned long)u_platform);
 240         }
 241         if (k_base_platform) {
 242                 NEW_AUX_ENT(AT_BASE_PLATFORM,
 243                             (elf_addr_t)(unsigned long)u_base_platform);
 244         }
 245         if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
 246                 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
 247         }
 248 #undef NEW_AUX_ENT
 249         /* AT_NULL is zero; clear the rest too */
 250         memset(&elf_info[ei_index], 0,
 251                sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
 252
 253         /* And advance past the AT_NULL entry.  */
 254         ei_index += 2;
 255
 256         sp = STACK_ADD(p, ei_index);
 257
 258         items = (argc + 1) + (envc + 1) + 1;
 259         bprm->p = STACK_ROUND(sp, items);
 260
 261         /* Point sp at the lowest address on the stack */
 262 #ifdef CONFIG_STACK_GROWSUP
 263         sp = (elf_addr_t __user *)bprm->p - items - ei_index;
 264         bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
 265 #else
 266         sp = (elf_addr_t __user *)bprm->p;
 267 #endif
 268
 269
 270         /*
 271          * Grow the stack manually; some architectures have a limit on how
 272          * far ahead a user-space access may be in order to grow the stack.
 273          */
 274         vma = find_extend_vma(current->mm, bprm->p);
 275         if (!vma)
 276                 return -EFAULT;
 277
 278         /* Now, let's put argc (and argv, envp if appropriate) on the stack */
 279         if (__put_user(argc, sp++))
 280                 return -EFAULT;
 281         argv = sp;
 282         envp = argv + argc + 1;
 283
 284         /* Populate argv and envp */
 285         p = current->mm->arg_end = current->mm->arg_start;
 286         while (argc-- > 0) {
 287                 size_t len;
 288                 if (__put_user((elf_addr_t)p, argv++))
 289                         return -EFAULT;
 290                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 291                 if (!len || len > MAX_ARG_STRLEN)
 292                         return -EINVAL;
 293                 p += len;
 294         }
 295         if (__put_user(0, argv))
 296                 return -EFAULT;
 297         current->mm->arg_end = current->mm->env_start = p;
 298         while (envc-- > 0) {
 299                 size_t len;
 300                 if (__put_user((elf_addr_t)p, envp++))
 301                         return -EFAULT;
 302                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 303                 if (!len || len > MAX_ARG_STRLEN)
 304                         return -EINVAL;
 305                 p += len;
 306         }
 307         if (__put_user(0, envp))
 308                 return -EFAULT;
 309         current->mm->env_end = p;
 310
 311         /* Put the elf_info on the stack in the right place.  */
 312         sp = (elf_addr_t __user *)envp + 1;
 313         if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
 314                 return -EFAULT;
 315         return 0;
 316 }
 317
 318 static unsigned long elf_map(struct file *filep, unsigned long addr,
 319                 struct elf_phdr *eppnt, int prot, int type,
 320                 unsigned long total_size)
 321 {
 322         unsigned long map_addr;
 323         unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
 324         unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
 325         addr = ELF_PAGESTART(addr);
 326         size = ELF_PAGEALIGN(size);
 327
 328         /* mmap() will return -EINVAL if given a zero size, but a
 329          * segment with zero filesize is perfectly valid */
 330         if (!size)
 331                 return addr;
 332
 333         down_write(&current->mm->mmap_sem);
 334         /*
 335         * total_size is the size of the ELF (interpreter) image.
 336         * The _first_ mmap needs to know the full size, otherwise
 337         * randomization might put this image into an overlapping
 338         * position with the ELF binary image. (since size < total_size)
 339         * So we first map the 'big' image - and unmap the remainder at
 340         * the end. (which unmap is needed for ELF images with holes.)
 341         */
 342         if (total_size) {
 343                 total_size = ELF_PAGEALIGN(total_size);
 344                 map_addr = do_mmap(filep, addr, total_size, prot, type, off);
 345                 if (!BAD_ADDR(map_addr))
 346                         do_munmap(current->mm, map_addr+size, total_size-size);
 347         } else
 348                 map_addr = do_mmap(filep, addr, size, prot, type, off);
 349
 350         up_write(&current->mm->mmap_sem);
 351         return(map_addr);
 352 }
 353
 354 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
 355 {
 356         int i, first_idx = -1, last_idx = -1;
 357
 358         for (i = 0; i < nr; i++) {
 359                 if (cmds[i].p_type == PT_LOAD) {
 360                         last_idx = i;
 361                         if (first_idx == -1)
 362                                 first_idx = i;
 363                 }
 364         }
 365         if (first_idx == -1)
 366                 return 0;
 367
 368         return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
 369                                 ELF_PAGESTART(cmds[first_idx].p_vaddr);
 370 }
 371
 372
 373 /* This is much more generalized than the library routine read function,
 374    so we keep this separate.  Technically the library read function
 375    is only provided so that we can read a.out libraries that have
 376    an ELF header */
 377
 378 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
 379                 struct file *interpreter, unsigned long *interp_map_addr,
 380                 unsigned long no_base)
 381 {
 382         struct elf_phdr *elf_phdata;
 383         struct elf_phdr *eppnt;
 384         unsigned long load_addr = 0;
 385         int load_addr_set = 0;
 386         unsigned long last_bss = 0, elf_bss = 0;
 387         unsigned long error = ~0UL;
 388         unsigned long total_size;
 389         int retval, i, size;
 390
 391         /* First of all, some simple consistency checks */
 392         if (interp_elf_ex->e_type != ET_EXEC &&
 393             interp_elf_ex->e_type != ET_DYN)
 394                 goto out;
 395         if (!elf_check_arch(interp_elf_ex))
 396                 goto out;
 397         if (!interpreter->f_op || !interpreter->f_op->mmap)
 398                 goto out;
 399
 400         /*
 401          * If the size of this structure has changed, then punt, since
 402          * we will be doing the wrong thing.
 403          */
 404         if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
 405                 goto out;
 406         if (interp_elf_ex->e_phnum < 1 ||
 407                 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
 408                 goto out;
 409
 410         /* Now read in all of the header information */
 411         size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
 412         if (size > ELF_MIN_ALIGN)
 413                 goto out;
 414         elf_phdata = kmalloc(size, GFP_KERNEL);
 415         if (!elf_phdata)
 416                 goto out;
 417
 418         retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
 419                              (char *)elf_phdata, size);
 420         error = -EIO;
 421         if (retval != size) {
 422                 if (retval < 0)
 423                         error = retval;
 424                 goto out_close;
 425         }
 426
 427         total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
 428         if (!total_size) {
 429                 error = -EINVAL;
 430                 goto out_close;
 431         }
 432
 433         eppnt = elf_phdata;
 434         for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
 435                 if (eppnt->p_type == PT_LOAD) {
 436                         int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
 437                         int elf_prot = 0;
 438                         unsigned long vaddr = 0;
 439                         unsigned long k, map_addr;
 440
 441                         if (eppnt->p_flags & PF_R)
 442                                 elf_prot = PROT_READ;
 443                         if (eppnt->p_flags & PF_W)
 444                                 elf_prot |= PROT_WRITE;
 445                         if (eppnt->p_flags & PF_X)
 446                                 elf_prot |= PROT_EXEC;
 447                         vaddr = eppnt->p_vaddr;
 448                         if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
 449                                 elf_type |= MAP_FIXED;
 450                         else if (no_base && interp_elf_ex->e_type == ET_DYN)
 451                                 load_addr = -vaddr;
 452
 453                         map_addr = elf_map(interpreter, load_addr + vaddr,
 454                                         eppnt, elf_prot, elf_type, total_size);
 455                         total_size = 0;
 456                         if (!*interp_map_addr)
 457                                 *interp_map_addr = map_addr;
 458                         error = map_addr;
 459                         if (BAD_ADDR(map_addr))
 460                                 goto out_close;
 461
 462                         if (!load_addr_set &&
 463                             interp_elf_ex->e_type == ET_DYN) {
 464                                 load_addr = map_addr - ELF_PAGESTART(vaddr);
 465                                 load_addr_set = 1;
 466                         }
 467
 468                         /*
 469                          * Check to see if the section's size will overflow the
 470                          * allowed task size. Note that p_filesz must always be
 471                          * <= p_memsize so it's only necessary to check p_memsz.
 472                          */
 473                         k = load_addr + eppnt->p_vaddr;
 474                         if (BAD_ADDR(k) ||
 475                             eppnt->p_filesz > eppnt->p_memsz ||
 476                             eppnt->p_memsz > TASK_SIZE ||
 477                             TASK_SIZE - eppnt->p_memsz < k) {
 478                                 error = -ENOMEM;
 479                                 goto out_close;
 480                         }
 481
 482                         /*
 483                          * Find the end of the file mapping for this phdr, and
 484                          * keep track of the largest address we see for this.
 485                          */
 486                         k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
 487                         if (k > elf_bss)
 488                                 elf_bss = k;
 489
 490                         /*
 491                          * Do the same thing for the memory mapping - between
 492                          * elf_bss and last_bss is the bss section.
 493                          */
 494                         k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
 495                         if (k > last_bss)
 496                                 last_bss = k;
 497                 }
 498         }
 499
 500         if (last_bss > elf_bss) {
 501                 /*
 502                  * Now fill out the bss section.  First pad the last page up
 503                  * to the page boundary, and then perform a mmap to make sure
 504                  * that there are zero-mapped pages up to and including the
 505                  * last bss page.
 506                  */
 507                 if (padzero(elf_bss)) {
 508                         error = -EFAULT;
 509                         goto out_close;
 510                 }
 511
 512                 /* What we have mapped so far */
 513                 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
 514
 515                 /* Map the last of the bss segment */
 516                 down_write(&current->mm->mmap_sem);
 517                 error = do_brk(elf_bss, last_bss - elf_bss);
 518                 up_write(&current->mm->mmap_sem);
 519                 if (BAD_ADDR(error))
 520                         goto out_close;
 521         }
 522
 523         error = load_addr;
 524
 525 out_close:
 526         kfree(elf_phdata);
 527 out:
 528         return error;
 529 }
 530
 531 /*
 532  * These are the functions used to load ELF style executables and shared
 533  * libraries.  There is no binary dependent code anywhere else.
 534  */
 535
 536 #define INTERPRETER_NONE 0
 537 #define INTERPRETER_ELF 2
 538
 539 #ifndef STACK_RND_MASK
 540 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))     /* 8MB of VA */
 541 #endif
 542
 543 static unsigned long randomize_stack_top(unsigned long stack_top)
 544 {
 545         unsigned int random_variable = 0;
 546
 547         if ((current->flags & PF_RANDOMIZE) &&
 548                 !(current->personality & ADDR_NO_RANDOMIZE)) {
 549                 random_variable = get_random_int() & STACK_RND_MASK;
 550                 random_variable <<= PAGE_SHIFT;
 551         }
 552 #ifdef CONFIG_STACK_GROWSUP
 553         return PAGE_ALIGN(stack_top) + random_variable;
 554 #else
 555         return PAGE_ALIGN(stack_top) - random_variable;
 556 #endif
 557 }
 558
 559 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
 560 {
 561         struct file *interpreter = NULL; /* to shut gcc up */
 562         unsigned long load_addr = 0, load_bias = 0;
 563         int load_addr_set = 0;
 564         char * elf_interpreter = NULL;
 565         unsigned long error;
 566         struct elf_phdr *elf_ppnt, *elf_phdata;
 567         unsigned long elf_bss, elf_brk;
 568         int retval, i;
 569         unsigned int size;
 570         unsigned long elf_entry;
 571         unsigned long interp_load_addr = 0;
 572         unsigned long start_code, end_code, start_data, end_data;
 573         unsigned long reloc_func_desc __maybe_unused = 0;
 574         int executable_stack = EXSTACK_DEFAULT;
 575         unsigned long def_flags = 0;
 576         struct {
 577                 struct elfhdr elf_ex;
 578                 struct elfhdr interp_elf_ex;
 579         } *loc;
 580
 581         loc = kmalloc(sizeof(*loc), GFP_KERNEL);
 582         if (!loc) {
 583                 retval = -ENOMEM;
 584                 goto out_ret;
 585         }
 586
 587         /* Get the exec-header */
 588         loc->elf_ex = *((struct elfhdr *)bprm->buf);
 589
 590         retval = -ENOEXEC;
 591         /* First of all, some simple consistency checks */
 592         if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 593                 goto out;
 594
 595         if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
 596                 goto out;
 597         if (!elf_check_arch(&loc->elf_ex))
 598                 goto out;
 599         if (!bprm->file->f_op || !bprm->file->f_op->mmap)
 600                 goto out;
 601
 602         /* Now read in all of the header information */
 603         if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
 604                 goto out;
 605         if (loc->elf_ex.e_phnum < 1 ||
 606                 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
 607                 goto out;
 608         size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
 609         retval = -ENOMEM;
 610         elf_phdata = kmalloc(size, GFP_KERNEL);
 611         if (!elf_phdata)
 612                 goto out;
 613
 614         retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
 615                              (char *)elf_phdata, size);
 616         if (retval != size) {
 617                 if (retval >= 0)
 618                         retval = -EIO;
 619                 goto out_free_ph;
 620         }
 621
 622         elf_ppnt = elf_phdata;
 623         elf_bss = 0;
 624         elf_brk = 0;
 625
 626         start_code = ~0UL;
 627         end_code = 0;
 628         start_data = 0;
 629         end_data = 0;
 630
 631         for (i = 0; i < loc->elf_ex.e_phnum; i++) {
 632                 if (elf_ppnt->p_type == PT_INTERP) {
 633                         /* This is the program interpreter used for
 634                          * shared libraries - for now assume that this
 635                          * is an a.out format binary
 636                          */
 637                         retval = -ENOEXEC;
 638                         if (elf_ppnt->p_filesz > PATH_MAX ||
 639                             elf_ppnt->p_filesz < 2)
 640                                 goto out_free_ph;
 641
 642                         retval = -ENOMEM;
 643                         elf_interpreter = kmalloc(elf_ppnt->p_filesz,
 644                                                   GFP_KERNEL);
 645                         if (!elf_interpreter)
 646                                 goto out_free_ph;
 647
 648                         retval = kernel_read(bprm->file, elf_ppnt->p_offset,
 649                                              elf_interpreter,
 650                                              elf_ppnt->p_filesz);
 651                         if (retval != elf_ppnt->p_filesz) {
 652                                 if (retval >= 0)
 653                                         retval = -EIO;
 654                                 goto out_free_interp;
 655                         }
 656                         /* make sure path is NULL terminated */
 657                         retval = -ENOEXEC;
 658                         if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
 659                                 goto out_free_interp;
 660
 661                         interpreter = open_exec(elf_interpreter);
 662                         retval = PTR_ERR(interpreter);
 663                         if (IS_ERR(interpreter))
 664                                 goto out_free_interp;
 665
 666                         /*
 667                          * If the binary is not readable then enforce
 668                          * mm->dumpable = 0 regardless of the interpreter's
 669                          * permissions.
 670                          */
 671                         would_dump(bprm, interpreter);
 672
 673                         retval = kernel_read(interpreter, 0, bprm->buf,
 674                                              BINPRM_BUF_SIZE);
 675                         if (retval != BINPRM_BUF_SIZE) {
 676                                 if (retval >= 0)
 677                                         retval = -EIO;
 678                                 goto out_free_dentry;
 679                         }
 680
 681                         /* Get the exec headers */
 682                         loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
 683                         break;
 684                 }
 685                 elf_ppnt++;
 686         }
 687
 688         elf_ppnt = elf_phdata;
 689         for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
 690                 if (elf_ppnt->p_type == PT_GNU_STACK) {
 691                         if (elf_ppnt->p_flags & PF_X)
 692                                 executable_stack = EXSTACK_ENABLE_X;
 693                         else
 694                                 executable_stack = EXSTACK_DISABLE_X;
 695                         break;
 696                 }
 697
 698         /* Some simple consistency checks for the interpreter */
 699         if (elf_interpreter) {
 700                 retval = -ELIBBAD;
 701                 /* Not an ELF interpreter */
 702                 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 703                         goto out_free_dentry;
 704                 /* Verify the interpreter has a valid arch */
 705                 if (!elf_check_arch(&loc->interp_elf_ex))
 706                         goto out_free_dentry;
 707         }
 708
 709         /* Flush all traces of the currently running executable */
 710         retval = flush_old_exec(bprm);
 711         if (retval)
 712                 goto out_free_dentry;
 713
 714         /* OK, This is the point of no return */
 715         current->mm->def_flags = def_flags;
 716
 717         /* Do this immediately, since STACK_TOP as used in setup_arg_pages
 718            may depend on the personality.  */
 719         SET_PERSONALITY(loc->elf_ex);
 720         if (elf_read_implies_exec(loc->elf_ex, executable_stack))
 721                 current->personality |= READ_IMPLIES_EXEC;
 722
 723         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 724                 current->flags |= PF_RANDOMIZE;
 725
 726         setup_new_exec(bprm);
 727
 728         /* Do this so that we can load the interpreter, if need be.  We will
 729            change some of these later */
 730         current->mm->free_area_cache = current->mm->mmap_base;
 731         current->mm->cached_hole_size = 0;
 732         retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
 733                                  executable_stack);
 734         if (retval < 0) {
 735                 send_sig(SIGKILL, current, 0);
 736                 goto out_free_dentry;
 737         }
 738
 739         current->mm->start_stack = bprm->p;
 740
 741         /* Now we do a little grungy work by mmapping the ELF image into
 742            the correct location in memory. */
 743         for(i = 0, elf_ppnt = elf_phdata;
 744             i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
 745                 int elf_prot = 0, elf_flags;
 746                 unsigned long k, vaddr;
 747
 748                 if (elf_ppnt->p_type != PT_LOAD)
 749                         continue;
 750
 751                 if (unlikely (elf_brk > elf_bss)) {
 752                         unsigned long nbyte;
 753
 754                         /* There was a PT_LOAD segment with p_memsz > p_filesz
 755                            before this one. Map anonymous pages, if needed,
 756                            and clear the area.  */
 757                         retval = set_brk(elf_bss + load_bias,
 758                                          elf_brk + load_bias);
 759                         if (retval) {
 760                                 send_sig(SIGKILL, current, 0);
 761                                 goto out_free_dentry;
 762                         }
 763                         nbyte = ELF_PAGEOFFSET(elf_bss);
 764                         if (nbyte) {
 765                                 nbyte = ELF_MIN_ALIGN - nbyte;
 766                                 if (nbyte > elf_brk - elf_bss)
 767                                         nbyte = elf_brk - elf_bss;
 768                                 if (clear_user((void __user *)elf_bss +
 769                                                         load_bias, nbyte)) {
 770                                         /*
 771                                          * This bss-zeroing can fail if the ELF
 772                                          * file specifies odd protections. So
 773                                          * we don't check the return value
 774                                          */
 775                                 }
 776                         }
 777                 }
 778
 779                 if (elf_ppnt->p_flags & PF_R)
 780                         elf_prot |= PROT_READ;
 781                 if (elf_ppnt->p_flags & PF_W)
 782                         elf_prot |= PROT_WRITE;
 783                 if (elf_ppnt->p_flags & PF_X)
 784                         elf_prot |= PROT_EXEC;
 785
 786                 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
 787
 788                 vaddr = elf_ppnt->p_vaddr;
 789                 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
 790                         elf_flags |= MAP_FIXED;
 791                 } else if (loc->elf_ex.e_type == ET_DYN) {
 792                         /* Try and get dynamic programs out of the way of the
 793                          * default mmap base, as well as whatever program they
 794                          * might try to exec.  This is because the brk will
 795                          * follow the loader, and is not movable.  */
 796 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
 797                         /* Memory randomization might have been switched off
 798                          * in runtime via sysctl.
 799                          * If that is the case, retain the original non-zero
 800                          * load_bias value in order to establish proper
 801                          * non-randomized mappings.
 802                          */
 803                         if (current->flags & PF_RANDOMIZE)
 804                                 load_bias = 0;
 805                         else
 806                                 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 807 #else
 808                         load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 809 #endif
 810                 }
 811
 812                 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
 813                                 elf_prot, elf_flags, 0);
 814                 if (BAD_ADDR(error)) {
 815                         send_sig(SIGKILL, current, 0);
 816                         retval = IS_ERR((void *)error) ?
 817                                 PTR_ERR((void*)error) : -EINVAL;
 818                         goto out_free_dentry;
 819                 }
 820
 821                 if (!load_addr_set) {
 822                         load_addr_set = 1;
 823                         load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
 824                         if (loc->elf_ex.e_type == ET_DYN) {
 825                                 load_bias += error -
 826                                              ELF_PAGESTART(load_bias + vaddr);
 827                                 load_addr += load_bias;
 828                                 reloc_func_desc = load_bias;
 829                         }
 830                 }
 831                 k = elf_ppnt->p_vaddr;
 832                 if (k < start_code)
 833                         start_code = k;
 834                 if (start_data < k)
 835                         start_data = k;
 836
 837                 /*
 838                  * Check to see if the section's size will overflow the
 839                  * allowed task size. Note that p_filesz must always be
 840                  * <= p_memsz so it is only necessary to check p_memsz.
 841                  */
 842                 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
 843                     elf_ppnt->p_memsz > TASK_SIZE ||
 844                     TASK_SIZE - elf_ppnt->p_memsz < k) {
 845                         /* set_brk can never work. Avoid overflows. */
 846                         send_sig(SIGKILL, current, 0);
 847                         retval = -EINVAL;
 848                         goto out_free_dentry;
 849                 }
 850
 851                 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
 852
 853                 if (k > elf_bss)
 854                         elf_bss = k;
 855                 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
 856                         end_code = k;
 857                 if (end_data < k)
 858                         end_data = k;
 859                 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
 860                 if (k > elf_brk)
 861                         elf_brk = k;
 862         }
 863
 864         loc->elf_ex.e_entry += load_bias;
 865         elf_bss += load_bias;
 866         elf_brk += load_bias;
 867         start_code += load_bias;
 868         end_code += load_bias;
 869         start_data += load_bias;
 870         end_data += load_bias;
 871
 872         /* Calling set_brk effectively mmaps the pages that we need
 873          * for the bss and break sections.  We must do this before
 874          * mapping in the interpreter, to make sure it doesn't wind
 875          * up getting placed where the bss needs to go.
 876          */
 877         retval = set_brk(elf_bss, elf_brk);
 878         if (retval) {
 879                 send_sig(SIGKILL, current, 0);
 880                 goto out_free_dentry;
 881         }
 882         if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
 883                 send_sig(SIGSEGV, current, 0);
 884                 retval = -EFAULT; /* Nobody gets to see this, but.. */
 885                 goto out_free_dentry;
 886         }
 887
 888         if (elf_interpreter) {
 889                 unsigned long uninitialized_var(interp_map_addr);
 890
 891                 elf_entry = load_elf_interp(&loc->interp_elf_ex,
 892                                             interpreter,
 893                                             &interp_map_addr,
 894                                             load_bias);
 895                 if (!IS_ERR((void *)elf_entry)) {
 896                         /*
 897                          * load_elf_interp() returns relocation
 898                          * adjustment
 899                          */
 900                         interp_load_addr = elf_entry;
 901                         elf_entry += loc->interp_elf_ex.e_entry;
 902                 }
 903                 if (BAD_ADDR(elf_entry)) {
 904                         force_sig(SIGSEGV, current);
 905                         retval = IS_ERR((void *)elf_entry) ?
 906                                         (int)elf_entry : -EINVAL;
 907                         goto out_free_dentry;
 908                 }
 909                 reloc_func_desc = interp_load_addr;
 910
 911                 allow_write_access(interpreter);
 912                 fput(interpreter);
 913                 kfree(elf_interpreter);
 914         } else {
 915                 elf_entry = loc->elf_ex.e_entry;
 916                 if (BAD_ADDR(elf_entry)) {
 917                         force_sig(SIGSEGV, current);
 918                         retval = -EINVAL;
 919                         goto out_free_dentry;
 920                 }
 921         }
 922
 923         kfree(elf_phdata);
 924
 925         set_binfmt(&elf_format);
 926
 927 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
 928         retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
 929         if (retval < 0) {
 930                 send_sig(SIGKILL, current, 0);
 931                 goto out;
 932         }
 933 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
 934
 935         install_exec_creds(bprm);
 936         retval = create_elf_tables(bprm, &loc->elf_ex,
 937                           load_addr, interp_load_addr);
 938         if (retval < 0) {
 939                 send_sig(SIGKILL, current, 0);
 940                 goto out;
 941         }
 942         /* N.B. passed_fileno might not be initialized? */
 943         current->mm->end_code = end_code;
 944         current->mm->start_code = start_code;
 945         current->mm->start_data = start_data;
 946         current->mm->end_data = end_data;
 947         current->mm->start_stack = bprm->p;
 948
 949 #ifdef arch_randomize_brk
 950         if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
 951                 current->mm->brk = current->mm->start_brk =
 952                         arch_randomize_brk(current->mm);
 953 #ifdef CONFIG_COMPAT_BRK
 954                 current->brk_randomized = 1;
 955 #endif
 956         }
 957 #endif
 958
 959         if (current->personality & MMAP_PAGE_ZERO) {
 960                 /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
 961                    and some applications "depend" upon this behavior.
 962                    Since we do not have the power to recompile these, we
 963                    emulate the SVr4 behavior. Sigh. */
 964                 down_write(&current->mm->mmap_sem);
 965                 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
 966                                 MAP_FIXED | MAP_PRIVATE, 0);
 967                 up_write(&current->mm->mmap_sem);
 968         }
 969
 970 #ifdef ELF_PLAT_INIT
 971         /*
 972          * The ABI may specify that certain registers be set up in special
 973          * ways (on i386 %edx is the address of a DT_FINI function, for
 974          * example.  In addition, it may also specify (eg, PowerPC64 ELF)
 975          * that the e_entry field is the address of the function descriptor
 976          * for the startup routine, rather than the address of the startup
 977          * routine itself.  This macro performs whatever initialization to
 978          * the regs structure is required as well as any relocations to the
 979          * function descriptor entries when executing dynamically links apps.
 980          */
 981         ELF_PLAT_INIT(regs, reloc_func_desc);
 982 #endif
 983
 984         start_thread(regs, elf_entry, bprm->p);
 985         retval = 0;
 986 out:
 987         kfree(loc);
 988 out_ret:
 989         return retval;
 990
 991         /* error cleanup */
 992 out_free_dentry:
 993         allow_write_access(interpreter);
 994         if (interpreter)
 995                 fput(interpreter);
 996 out_free_interp:
 997         kfree(elf_interpreter);
 998 out_free_ph:
 999         kfree(elf_phdata);
1000         goto out;
1001 }
1002
1003 /* This is really simpleminded and specialized - we are loading an
1004    a.out library that is given an ELF header. */
1005 static int load_elf_library(struct file *file)
1006 {
1007         struct elf_phdr *elf_phdata;
1008         struct elf_phdr *eppnt;
1009         unsigned long elf_bss, bss, len;
1010         int retval, error, i, j;
1011         struct elfhdr elf_ex;
1012
1013         error = -ENOEXEC;
1014         retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1015         if (retval != sizeof(elf_ex))
1016                 goto out;
1017
1018         if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1019                 goto out;
1020
1021         /* First of all, some simple consistency checks */
1022         if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1023             !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1024                 goto out;
1025
1026         /* Now read in all of the header information */
1027
1028         j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1029         /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1030
1031         error = -ENOMEM;
1032         elf_phdata = kmalloc(j, GFP_KERNEL);
1033         if (!elf_phdata)
1034                 goto out;
1035
1036         eppnt = elf_phdata;
1037         error = -ENOEXEC;
1038         retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1039         if (retval != j)
1040                 goto out_free_ph;
1041
1042         for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1043                 if ((eppnt + i)->p_type == PT_LOAD)
1044                         j++;
1045         if (j != 1)
1046                 goto out_free_ph;
1047
1048         while (eppnt->p_type != PT_LOAD)
1049                 eppnt++;
1050
1051         /* Now use mmap to map the library into memory. */
1052         down_write(&current->mm->mmap_sem);
1053         error = do_mmap(file,
1054                         ELF_PAGESTART(eppnt->p_vaddr),
1055                         (eppnt->p_filesz +
1056                          ELF_PAGEOFFSET(eppnt->p_vaddr)),
1057                         PROT_READ | PROT_WRITE | PROT_EXEC,
1058                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1059                         (eppnt->p_offset -
1060                          ELF_PAGEOFFSET(eppnt->p_vaddr)));
1061         up_write(&current->mm->mmap_sem);
1062         if (error != ELF_PAGESTART(eppnt->p_vaddr))
1063                 goto out_free_ph;
1064
1065         elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1066         if (padzero(elf_bss)) {
1067                 error = -EFAULT;
1068                 goto out_free_ph;
1069         }
1070
1071         len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1072                             ELF_MIN_ALIGN - 1);
1073         bss = eppnt->p_memsz + eppnt->p_vaddr;
1074         if (bss > len) {
1075                 down_write(&current->mm->mmap_sem);
1076                 do_brk(len, bss - len);
1077                 up_write(&current->mm->mmap_sem);
1078         }
1079         error = 0;
1080
1081 out_free_ph:
1082         kfree(elf_phdata);
1083 out:
1084         return error;
1085 }
1086
1087 #ifdef CONFIG_ELF_CORE
1088 /*
1089  * ELF core dumper
1090  *
1091  * Modelled on fs/exec.c:aout_core_dump()
1092  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1093  */
1094
1095 /*
1096  * Decide what to dump of a segment, part, all or none.
1097  */
1098 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1099                                    unsigned long mm_flags)
1100 {
1101 #define FILTER(type)    (mm_flags & (1UL << MMF_DUMP_##type))
1102
1103         /* The vma can be set up to tell us the answer directly.  */
1104         if (vma->vm_flags & VM_ALWAYSDUMP)
1105                 goto whole;
1106
1107         /* Hugetlb memory check */
1108         if (vma->vm_flags & VM_HUGETLB) {
1109                 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1110                         goto whole;
1111                 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1112                         goto whole;
1113         }
1114
1115         /* Do not dump I/O mapped devices or special mappings */
1116         if (vma->vm_flags & (VM_IO | VM_RESERVED))
1117                 return 0;
1118
1119         /* By default, dump shared memory if mapped from an anonymous file. */
1120         if (vma->vm_flags & VM_SHARED) {
1121                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1122                     FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1123                         goto whole;
1124                 return 0;
1125         }
1126
1127         /* Dump segments that have been written to.  */
1128         if (vma->anon_vma && FILTER(ANON_PRIVATE))
1129                 goto whole;
1130         if (vma->vm_file == NULL)
1131                 return 0;
1132
1133         if (FILTER(MAPPED_PRIVATE))
1134                 goto whole;
1135
1136         /*
1137          * If this looks like the beginning of a DSO or executable mapping,
1138          * check for an ELF header.  If we find one, dump the first page to
1139          * aid in determining what was mapped here.
1140          */
1141         if (FILTER(ELF_HEADERS) &&
1142             vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1143                 u32 __user *header = (u32 __user *) vma->vm_start;
1144                 u32 word;
1145                 mm_segment_t fs = get_fs();
1146                 /*
1147                  * Doing it this way gets the constant folded by GCC.
1148                  */
1149                 union {
1150                         u32 cmp;
1151                         char elfmag[SELFMAG];
1152                 } magic;
1153                 BUILD_BUG_ON(SELFMAG != sizeof word);
1154                 magic.elfmag[EI_MAG0] = ELFMAG0;
1155                 magic.elfmag[EI_MAG1] = ELFMAG1;
1156                 magic.elfmag[EI_MAG2] = ELFMAG2;
1157                 magic.elfmag[EI_MAG3] = ELFMAG3;
1158                 /*
1159                  * Switch to the user "segment" for get_user(),
1160                  * then put back what elf_core_dump() had in place.
1161                  */
1162                 set_fs(USER_DS);
1163                 if (unlikely(get_user(word, header)))
1164                         word = 0;
1165                 set_fs(fs);
1166                 if (word == magic.cmp)
1167                         return PAGE_SIZE;
1168         }
1169
1170 #undef  FILTER
1171
1172         return 0;
1173
1174 whole:
1175         return vma->vm_end - vma->vm_start;
1176 }
1177
1178 /* An ELF note in memory */
1179 struct memelfnote
1180 {
1181         const char *name;
1182         int type;
1183         unsigned int datasz;
1184         void *data;
1185 };
1186
1187 static int notesize(struct memelfnote *en)
1188 {
1189         int sz;
1190
1191         sz = sizeof(struct elf_note);
1192         sz += roundup(strlen(en->name) + 1, 4);
1193         sz += roundup(en->datasz, 4);
1194
1195         return sz;
1196 }
1197
1198 #define DUMP_WRITE(addr, nr, foffset)   \
1199         do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1200
1201 static int alignfile(struct file *file, loff_t *foffset)
1202 {
1203         static const char buf[4] = { 0, };
1204         DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1205         return 1;
1206 }
1207
1208 static int writenote(struct memelfnote *men, struct file *file,
1209                         loff_t *foffset)
1210 {
1211         struct elf_note en;
1212         en.n_namesz = strlen(men->name) + 1;
1213         en.n_descsz = men->datasz;
1214         en.n_type = men->type;
1215
1216         DUMP_WRITE(&en, sizeof(en), foffset);
1217         DUMP_WRITE(men->name, en.n_namesz, foffset);
1218         if (!alignfile(file, foffset))
1219                 return 0;
1220         DUMP_WRITE(men->data, men->datasz, foffset);
1221         if (!alignfile(file, foffset))
1222                 return 0;
1223
1224         return 1;
1225 }
1226 #undef DUMP_WRITE
1227
1228 static void fill_elf_header(struct elfhdr *elf, int segs,
1229                             u16 machine, u32 flags, u8 osabi)
1230 {
1231         memset(elf, 0, sizeof(*elf));
1232
1233         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1234         elf->e_ident[EI_CLASS] = ELF_CLASS;
1235         elf->e_ident[EI_DATA] = ELF_DATA;
1236         elf->e_ident[EI_VERSION] = EV_CURRENT;
1237         elf->e_ident[EI_OSABI] = ELF_OSABI;
1238
1239         elf->e_type = ET_CORE;
1240         elf->e_machine = machine;
1241         elf->e_version = EV_CURRENT;
1242         elf->e_phoff = sizeof(struct elfhdr);
1243         elf->e_flags = flags;
1244         elf->e_ehsize = sizeof(struct elfhdr);
1245         elf->e_phentsize = sizeof(struct elf_phdr);
1246         elf->e_phnum = segs;
1247
1248         return;
1249 }
1250
1251 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1252 {
1253         phdr->p_type = PT_NOTE;
1254         phdr->p_offset = offset;
1255         phdr->p_vaddr = 0;
1256         phdr->p_paddr = 0;
1257         phdr->p_filesz = sz;
1258         phdr->p_memsz = 0;
1259         phdr->p_flags = 0;
1260         phdr->p_align = 0;
1261         return;
1262 }
1263
1264 static void fill_note(struct memelfnote *note, const char *name, int type,
1265                 unsigned int sz, void *data)
1266 {
1267         note->name = name;
1268         note->type = type;
1269         note->datasz = sz;
1270         note->data = data;
1271         return;
1272 }
1273
1274 /*
1275  * fill up all the fields in prstatus from the given task struct, except
1276  * registers which need to be filled up separately.
1277  */
1278 static void fill_prstatus(struct elf_prstatus *prstatus,
1279                 struct task_struct *p, long signr)
1280 {
1281         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1282         prstatus->pr_sigpend = p->pending.signal.sig[0];
1283         prstatus->pr_sighold = p->blocked.sig[0];
1284         rcu_read_lock();
1285         prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1286         rcu_read_unlock();
1287         prstatus->pr_pid = task_pid_vnr(p);
1288         prstatus->pr_pgrp = task_pgrp_vnr(p);
1289         prstatus->pr_sid = task_session_vnr(p);
1290         if (thread_group_leader(p)) {
1291                 struct task_cputime cputime;
1292
1293                 /*
1294                  * This is the record for the group leader.  It shows the
1295                  * group-wide total, not its individual thread total.
1296                  */
1297                 thread_group_cputime(p, &cputime);
1298                 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1299                 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1300         } else {
1301                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1302                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1303         }
1304         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1305         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1306 }
1307
1308 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1309                        struct mm_struct *mm)
1310 {
1311         const struct cred *cred;
1312         unsigned int i, len;
1313
1314         /* first copy the parameters from user space */
1315         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1316
1317         len = mm->arg_end - mm->arg_start;
1318         if (len >= ELF_PRARGSZ)
1319                 len = ELF_PRARGSZ-1;
1320         if (copy_from_user(&psinfo->pr_psargs,
1321                            (const char __user *)mm->arg_start, len))
1322                 return -EFAULT;
1323         for(i = 0; i < len; i++)
1324                 if (psinfo->pr_psargs[i] == 0)
1325                         psinfo->pr_psargs[i] = ' ';
1326         psinfo->pr_psargs[len] = 0;
1327
1328         rcu_read_lock();
1329         psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1330         rcu_read_unlock();
1331         psinfo->pr_pid = task_pid_vnr(p);
1332         psinfo->pr_pgrp = task_pgrp_vnr(p);
1333         psinfo->pr_sid = task_session_vnr(p);
1334
1335         i = p->state ? ffz(~p->state) + 1 : 0;
1336         psinfo->pr_state = i;
1337         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1338         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1339         psinfo->pr_nice = task_nice(p);
1340         psinfo->pr_flag = p->flags;
1341         rcu_read_lock();
1342         cred = __task_cred(p);
1343         SET_UID(psinfo->pr_uid, cred->uid);
1344         SET_GID(psinfo->pr_gid, cred->gid);
1345         rcu_read_unlock();
1346         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1347
1348         return 0;
1349 }
1350
1351 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1352 {
1353         elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1354         int i = 0;
1355         do
1356                 i += 2;
1357         while (auxv[i - 2] != AT_NULL);
1358         fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1359 }
1360
1361 #ifdef CORE_DUMP_USE_REGSET
1362 #include <linux/regset.h>
1363
1364 struct elf_thread_core_info {
1365         struct elf_thread_core_info *next;
1366         struct task_struct *task;
1367         struct elf_prstatus prstatus;
1368         struct memelfnote notes[0];
1369 };
1370
1371 struct elf_note_info {
1372         struct elf_thread_core_info *thread;
1373         struct memelfnote psinfo;
1374         struct memelfnote auxv;
1375         size_t size;
1376         int thread_notes;
1377 };
1378
1379 /*
1380  * When a regset has a writeback hook, we call it on each thread before
1381  * dumping user memory.  On register window machines, this makes sure the
1382  * user memory backing the register data is up to date before we read it.
1383  */
1384 static void do_thread_regset_writeback(struct task_struct *task,
1385                                        const struct user_regset *regset)
1386 {
1387         if (regset->writeback)
1388                 regset->writeback(task, regset, 1);
1389 }
1390
1391 static int fill_thread_core_info(struct elf_thread_core_info *t,
1392                                  const struct user_regset_view *view,
1393                                  long signr, size_t *total)
1394 {
1395         unsigned int i;
1396
1397         /*
1398          * NT_PRSTATUS is the one special case, because the regset data
1399          * goes into the pr_reg field inside the note contents, rather
1400          * than being the whole note contents.  We fill the reset in here.
1401          * We assume that regset 0 is NT_PRSTATUS.
1402          */
1403         fill_prstatus(&t->prstatus, t->task, signr);
1404         (void) view->regsets[0].get(t->task, &view->regsets[0],
1405                                     0, sizeof(t->prstatus.pr_reg),
1406                                     &t->prstatus.pr_reg, NULL);
1407
1408         fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1409                   sizeof(t->prstatus), &t->prstatus);
1410         *total += notesize(&t->notes[0]);
1411
1412         do_thread_regset_writeback(t->task, &view->regsets[0]);
1413
1414         /*
1415          * Each other regset might generate a note too.  For each regset
1416          * that has no core_note_type or is inactive, we leave t->notes[i]
1417          * all zero and we'll know to skip writing it later.
1418          */
1419         for (i = 1; i < view->n; ++i) {
1420                 const struct user_regset *regset = &view->regsets[i];
1421                 do_thread_regset_writeback(t->task, regset);
1422                 if (regset->core_note_type && regset->get &&
1423                     (!regset->active || regset->active(t->task, regset))) {
1424                         int ret;
1425                         size_t size = regset->n * regset->size;
1426                         void *data = kmalloc(size, GFP_KERNEL);
1427                         if (unlikely(!data))
1428                                 return 0;
1429                         ret = regset->get(t->task, regset,
1430                                           0, size, data, NULL);
1431                         if (unlikely(ret))
1432                                 kfree(data);
1433                         else {
1434                                 if (regset->core_note_type != NT_PRFPREG)
1435                                         fill_note(&t->notes[i], "LINUX",
1436                                                   regset->core_note_type,
1437                                                   size, data);
1438                                 else {
1439                                         t->prstatus.pr_fpvalid = 1;
1440                                         fill_note(&t->notes[i], "CORE",
1441                                                   NT_PRFPREG, size, data);
1442                                 }
1443                                 *total += notesize(&t->notes[i]);
1444                         }
1445                 }
1446         }
1447
1448         return 1;
1449 }
1450
1451 static int fill_note_info(struct elfhdr *elf, int phdrs,
1452                           struct elf_note_info *info,
1453                           long signr, struct pt_regs *regs)
1454 {
1455         struct task_struct *dump_task = current;
1456         const struct user_regset_view *view = task_user_regset_view(dump_task);
1457         struct elf_thread_core_info *t;
1458         struct elf_prpsinfo *psinfo;
1459         struct core_thread *ct;
1460         unsigned int i;
1461
1462         info->size = 0;
1463         info->thread = NULL;
1464
1465         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1466         if (psinfo == NULL)
1467                 return 0;
1468
1469         fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1470
1471         /*
1472          * Figure out how many notes we're going to need for each thread.
1473          */
1474         info->thread_notes = 0;
1475         for (i = 0; i < view->n; ++i)
1476                 if (view->regsets[i].core_note_type != 0)
1477                         ++info->thread_notes;
1478
1479         /*
1480          * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1481          * since it is our one special case.
1482          */
1483         if (unlikely(info->thread_notes == 0) ||
1484             unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1485                 WARN_ON(1);
1486                 return 0;
1487         }
1488
1489         /*
1490          * Initialize the ELF file header.
1491          */
1492         fill_elf_header(elf, phdrs,
1493                         view->e_machine, view->e_flags, view->ei_osabi);
1494
1495         /*
1496          * Allocate a structure for each thread.
1497          */
1498         for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1499                 t = kzalloc(offsetof(struct elf_thread_core_info,
1500                                      notes[info->thread_notes]),
1501                             GFP_KERNEL);
1502                 if (unlikely(!t))
1503                         return 0;
1504
1505                 t->task = ct->task;
1506                 if (ct->task == dump_task || !info->thread) {
1507                         t->next = info->thread;
1508                         info->thread = t;
1509                 } else {
1510                         /*
1511                          * Make sure to keep the original task at
1512                          * the head of the list.
1513                          */
1514                         t->next = info->thread->next;
1515                         info->thread->next = t;
1516                 }
1517         }
1518
1519         /*
1520          * Now fill in each thread's information.
1521          */
1522         for (t = info->thread; t != NULL; t = t->next)
1523                 if (!fill_thread_core_info(t, view, signr, &info->size))
1524                         return 0;
1525
1526         /*
1527          * Fill in the two process-wide notes.
1528          */
1529         fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1530         info->size += notesize(&info->psinfo);
1531
1532         fill_auxv_note(&info->auxv, current->mm);
1533         info->size += notesize(&info->auxv);
1534
1535         return 1;
1536 }
1537
1538 static size_t get_note_info_size(struct elf_note_info *info)
1539 {
1540         return info->size;
1541 }
1542
1543 /*
1544  * Write all the notes for each thread.  When writing the first thread, the
1545  * process-wide notes are interleaved after the first thread-specific note.
1546  */
1547 static int write_note_info(struct elf_note_info *info,
1548                            struct file *file, loff_t *foffset)
1549 {
1550         bool first = 1;
1551         struct elf_thread_core_info *t = info->thread;
1552
1553         do {
1554                 int i;
1555
1556                 if (!writenote(&t->notes[0], file, foffset))
1557                         return 0;
1558
1559                 if (first && !writenote(&info->psinfo, file, foffset))
1560                         return 0;
1561                 if (first && !writenote(&info->auxv, file, foffset))
1562                         return 0;
1563
1564                 for (i = 1; i < info->thread_notes; ++i)
1565                         if (t->notes[i].data &&
1566                             !writenote(&t->notes[i], file, foffset))
1567                                 return 0;
1568
1569                 first = 0;
1570                 t = t->next;
1571         } while (t);
1572
1573         return 1;
1574 }
1575
1576 static void free_note_info(struct elf_note_info *info)
1577 {
1578         struct elf_thread_core_info *threads = info->thread;
1579         while (threads) {
1580                 unsigned int i;
1581                 struct elf_thread_core_info *t = threads;
1582                 threads = t->next;
1583                 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1584                 for (i = 1; i < info->thread_notes; ++i)
1585                         kfree(t->notes[i].data);
1586                 kfree(t);
1587         }
1588         kfree(info->psinfo.data);
1589 }
1590
1591 #else
1592
1593 /* Here is the structure in which status of each thread is captured. */
1594 struct elf_thread_status
1595 {
1596         struct list_head list;
1597         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1598         elf_fpregset_t fpu;             /* NT_PRFPREG */
1599         struct task_struct *thread;
1600 #ifdef ELF_CORE_COPY_XFPREGS
1601         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1602 #endif
1603         struct memelfnote notes[3];
1604         int num_notes;
1605 };
1606
1607 /*
1608  * In order to add the specific thread information for the elf file format,
1609  * we need to keep a linked list of every threads pr_status and then create
1610  * a single section for them in the final core file.
1611  */
1612 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1613 {
1614         int sz = 0;
1615         struct task_struct *p = t->thread;
1616         t->num_notes = 0;
1617
1618         fill_prstatus(&t->prstatus, p, signr);
1619         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1620
1621         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1622                   &(t->prstatus));
1623         t->num_notes++;
1624         sz += notesize(&t->notes[0]);
1625
1626         if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1627                                                                 &t->fpu))) {
1628                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1629                           &(t->fpu));
1630                 t->num_notes++;
1631                 sz += notesize(&t->notes[1]);
1632         }
1633
1634 #ifdef ELF_CORE_COPY_XFPREGS
1635         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1636                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1637                           sizeof(t->xfpu), &t->xfpu);
1638                 t->num_notes++;
1639                 sz += notesize(&t->notes[2]);
1640         }
1641 #endif
1642         return sz;
1643 }
1644
1645 struct elf_note_info {
1646         struct memelfnote *notes;
1647         struct elf_prstatus *prstatus;  /* NT_PRSTATUS */
1648         struct elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
1649         struct list_head thread_list;
1650         elf_fpregset_t *fpu;
1651 #ifdef ELF_CORE_COPY_XFPREGS
1652         elf_fpxregset_t *xfpu;
1653 #endif
1654         int thread_status_size;
1655         int numnote;
1656 };
1657
1658 static int elf_note_info_init(struct elf_note_info *info)
1659 {
1660         memset(info, 0, sizeof(*info));
1661         INIT_LIST_HEAD(&info->thread_list);
1662
1663         /* Allocate space for six ELF notes */
1664         info->notes = kmalloc(6 * sizeof(struct memelfnote), GFP_KERNEL);
1665         if (!info->notes)
1666                 return 0;
1667         info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1668         if (!info->psinfo)
1669                 goto notes_free;
1670         info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1671         if (!info->prstatus)
1672                 goto psinfo_free;
1673         info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1674         if (!info->fpu)
1675                 goto prstatus_free;
1676 #ifdef ELF_CORE_COPY_XFPREGS
1677         info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1678         if (!info->xfpu)
1679                 goto fpu_free;
1680 #endif
1681         return 1;
1682 #ifdef ELF_CORE_COPY_XFPREGS
1683  fpu_free:
1684         kfree(info->fpu);
1685 #endif
1686  prstatus_free:
1687         kfree(info->prstatus);
1688  psinfo_free:
1689         kfree(info->psinfo);
1690  notes_free:
1691         kfree(info->notes);
1692         return 0;
1693 }
1694
1695 static int fill_note_info(struct elfhdr *elf, int phdrs,
1696                           struct elf_note_info *info,
1697                           long signr, struct pt_regs *regs)
1698 {
1699         struct list_head *t;
1700
1701         if (!elf_note_info_init(info))
1702                 return 0;
1703
1704         if (signr) {
1705                 struct core_thread *ct;
1706                 struct elf_thread_status *ets;
1707
1708                 for (ct = current->mm->core_state->dumper.next;
1709                                                 ct; ct = ct->next) {
1710                         ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1711                         if (!ets)
1712                                 return 0;
1713
1714                         ets->thread = ct->task;
1715                         list_add(&ets->list, &info->thread_list);
1716                 }
1717
1718                 list_for_each(t, &info->thread_list) {
1719                         int sz;
1720
1721                         ets = list_entry(t, struct elf_thread_status, list);
1722                         sz = elf_dump_thread_status(signr, ets);
1723                         info->thread_status_size += sz;
1724                 }
1725         }
1726         /* now collect the dump for the current */
1727         memset(info->prstatus, 0, sizeof(*info->prstatus));
1728         fill_prstatus(info->prstatus, current, signr);
1729         elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1730
1731         /* Set up header */
1732         fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1733
1734         /*
1735          * Set up the notes in similar form to SVR4 core dumps made
1736          * with info from their /proc.
1737          */
1738
1739         fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1740                   sizeof(*info->prstatus), info->prstatus);
1741         fill_psinfo(info->psinfo, current->group_leader, current->mm);
1742         fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1743                   sizeof(*info->psinfo), info->psinfo);
1744
1745         info->numnote = 2;
1746
1747         fill_auxv_note(&info->notes[info->numnote++], current->mm);
1748
1749         /* Try to dump the FPU. */
1750         info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1751                                                                info->fpu);
1752         if (info->prstatus->pr_fpvalid)
1753                 fill_note(info->notes + info->numnote++,
1754                           "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1755 #ifdef ELF_CORE_COPY_XFPREGS
1756         if (elf_core_copy_task_xfpregs(current, info->xfpu))
1757                 fill_note(info->notes + info->numnote++,
1758                           "LINUX", ELF_CORE_XFPREG_TYPE,
1759                           sizeof(*info->xfpu), info->xfpu);
1760 #endif
1761
1762         return 1;
1763 }
1764
1765 static size_t get_note_info_size(struct elf_note_info *info)
1766 {
1767         int sz = 0;
1768         int i;
1769
1770         for (i = 0; i < info->numnote; i++)
1771                 sz += notesize(info->notes + i);
1772
1773         sz += info->thread_status_size;
1774
1775         return sz;
1776 }
1777
1778 static int write_note_info(struct elf_note_info *info,
1779                            struct file *file, loff_t *foffset)
1780 {
1781         int i;
1782         struct list_head *t;
1783
1784         for (i = 0; i < info->numnote; i++)
1785                 if (!writenote(info->notes + i, file, foffset))
1786                         return 0;
1787
1788         /* write out the thread status notes section */
1789         list_for_each(t, &info->thread_list) {
1790                 struct elf_thread_status *tmp =
1791                                 list_entry(t, struct elf_thread_status, list);
1792
1793                 for (i = 0; i < tmp->num_notes; i++)
1794                         if (!writenote(&tmp->notes[i], file, foffset))
1795                                 return 0;
1796         }
1797
1798         return 1;
1799 }
1800
1801 static void free_note_info(struct elf_note_info *info)
1802 {
1803         while (!list_empty(&info->thread_list)) {
1804                 struct list_head *tmp = info->thread_list.next;
1805                 list_del(tmp);
1806                 kfree(list_entry(tmp, struct elf_thread_status, list));
1807         }
1808
1809         kfree(info->prstatus);
1810         kfree(info->psinfo);
1811         kfree(info->notes);
1812         kfree(info->fpu);
1813 #ifdef ELF_CORE_COPY_XFPREGS
1814         kfree(info->xfpu);
1815 #endif
1816 }
1817
1818 #endif
1819
1820 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1821                                         struct vm_area_struct *gate_vma)
1822 {
1823         struct vm_area_struct *ret = tsk->mm->mmap;
1824
1825         if (ret)
1826                 return ret;
1827         return gate_vma;
1828 }
1829 /*
1830  * Helper function for iterating across a vma list.  It ensures that the caller
1831  * will visit `gate_vma' prior to terminating the search.
1832  */
1833 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1834                                         struct vm_area_struct *gate_vma)
1835 {
1836         struct vm_area_struct *ret;
1837
1838         ret = this_vma->vm_next;
1839         if (ret)
1840                 return ret;
1841         if (this_vma == gate_vma)
1842                 return NULL;
1843         return gate_vma;
1844 }
1845
1846 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1847                              elf_addr_t e_shoff, int segs)
1848 {
1849         elf->e_shoff = e_shoff;
1850         elf->e_shentsize = sizeof(*shdr4extnum);
1851         elf->e_shnum = 1;
1852         elf->e_shstrndx = SHN_UNDEF;
1853
1854         memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1855
1856         shdr4extnum->sh_type = SHT_NULL;
1857         shdr4extnum->sh_size = elf->e_shnum;
1858         shdr4extnum->sh_link = elf->e_shstrndx;
1859         shdr4extnum->sh_info = segs;
1860 }
1861
1862 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
1863                                      unsigned long mm_flags)
1864 {
1865         struct vm_area_struct *vma;
1866         size_t size = 0;
1867
1868         for (vma = first_vma(current, gate_vma); vma != NULL;
1869              vma = next_vma(vma, gate_vma))
1870                 size += vma_dump_size(vma, mm_flags);
1871         return size;
1872 }
1873
1874 /*
1875  * Actual dumper
1876  *
1877  * This is a two-pass process; first we find the offsets of the bits,
1878  * and then they are actually written out.  If we run out of core limit
1879  * we just truncate.
1880  */
1881 static int elf_core_dump(struct coredump_params *cprm)
1882 {
1883         int has_dumped = 0;
1884         mm_segment_t fs;
1885         int segs;
1886         size_t size = 0;
1887         struct vm_area_struct *vma, *gate_vma;
1888         struct elfhdr *elf = NULL;
1889         loff_t offset = 0, dataoff, foffset;
1890         struct elf_note_info info;
1891         struct elf_phdr *phdr4note = NULL;
1892         struct elf_shdr *shdr4extnum = NULL;
1893         Elf_Half e_phnum;
1894         elf_addr_t e_shoff;
1895
1896         /*
1897          * We no longer stop all VM operations.
1898          *
1899          * This is because those proceses that could possibly change map_count
1900          * or the mmap / vma pages are now blocked in do_exit on current
1901          * finishing this core dump.
1902          *
1903          * Only ptrace can touch these memory addresses, but it doesn't change
1904          * the map_count or the pages allocated. So no possibility of crashing
1905          * exists while dumping the mm->vm_next areas to the core file.
1906          */
1907
1908         /* alloc memory for large data structures: too large to be on stack */
1909         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1910         if (!elf)
1911                 goto out;
1912         /*
1913          * The number of segs are recored into ELF header as 16bit value.
1914          * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
1915          */
1916         segs = current->mm->map_count;
1917         segs += elf_core_extra_phdrs();
1918
1919         gate_vma = get_gate_vma(current->mm);
1920         if (gate_vma != NULL)
1921                 segs++;
1922
1923         /* for notes section */
1924         segs++;
1925
1926         /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1927          * this, kernel supports extended numbering. Have a look at
1928          * include/linux/elf.h for further information. */
1929         e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1930
1931         /*
1932          * Collect all the non-memory information about the process for the
1933          * notes.  This also sets up the file header.
1934          */
1935         if (!fill_note_info(elf, e_phnum, &info, cprm->signr, cprm->regs))
1936                 goto cleanup;
1937
1938         has_dumped = 1;
1939         current->flags |= PF_DUMPCORE;
1940
1941         fs = get_fs();
1942         set_fs(KERNEL_DS);
1943
1944         offset += sizeof(*elf);                         /* Elf header */
1945         offset += segs * sizeof(struct elf_phdr);       /* Program headers */
1946         foffset = offset;
1947
1948         /* Write notes phdr entry */
1949         {
1950                 size_t sz = get_note_info_size(&info);
1951
1952                 sz += elf_coredump_extra_notes_size();
1953
1954                 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1955                 if (!phdr4note)
1956                         goto end_coredump;
1957
1958                 fill_elf_note_phdr(phdr4note, sz, offset);
1959                 offset += sz;
1960         }
1961
1962         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1963
1964         offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
1965         offset += elf_core_extra_data_size();
1966         e_shoff = offset;
1967
1968         if (e_phnum == PN_XNUM) {
1969                 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1970                 if (!shdr4extnum)
1971                         goto end_coredump;
1972                 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1973         }
1974
1975         offset = dataoff;
1976
1977         size += sizeof(*elf);
1978         if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
1979                 goto end_coredump;
1980
1981         size += sizeof(*phdr4note);
1982         if (size > cprm->limit
1983             || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
1984                 goto end_coredump;
1985
1986         /* Write program headers for segments dump */
1987         for (vma = first_vma(current, gate_vma); vma != NULL;
1988                         vma = next_vma(vma, gate_vma)) {
1989                 struct elf_phdr phdr;
1990
1991                 phdr.p_type = PT_LOAD;
1992                 phdr.p_offset = offset;
1993                 phdr.p_vaddr = vma->vm_start;
1994                 phdr.p_paddr = 0;
1995                 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
1996                 phdr.p_memsz = vma->vm_end - vma->vm_start;
1997                 offset += phdr.p_filesz;
1998                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1999                 if (vma->vm_flags & VM_WRITE)
2000                         phdr.p_flags |= PF_W;
2001                 if (vma->vm_flags & VM_EXEC)
2002                         phdr.p_flags |= PF_X;
2003                 phdr.p_align = ELF_EXEC_PAGESIZE;
2004
2005                 size += sizeof(phdr);
2006                 if (size > cprm->limit
2007                     || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2008                         goto end_coredump;
2009         }
2010
2011         if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2012                 goto end_coredump;
2013
2014         /* write out the notes section */
2015         if (!write_note_info(&info, cprm->file, &foffset))
2016                 goto end_coredump;
2017
2018         if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2019                 goto end_coredump;
2020
2021         /* Align to page */
2022         if (!dump_seek(cprm->file, dataoff - foffset))
2023                 goto end_coredump;
2024
2025         for (vma = first_vma(current, gate_vma); vma != NULL;
2026                         vma = next_vma(vma, gate_vma)) {
2027                 unsigned long addr;
2028                 unsigned long end;
2029
2030                 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2031
2032                 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2033                         struct page *page;
2034                         int stop;
2035
2036                         page = get_dump_page(addr);
2037                         if (page) {
2038                                 void *kaddr = kmap(page);
2039                                 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2040                                         !dump_write(cprm->file, kaddr,
2041                                                     PAGE_SIZE);
2042                                 kunmap(page);
2043                                 page_cache_release(page);
2044                         } else
2045                                 stop = !dump_seek(cprm->file, PAGE_SIZE);
2046                         if (stop)
2047                                 goto end_coredump;
2048                 }
2049         }
2050
2051         if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2052                 goto end_coredump;
2053
2054         if (e_phnum == PN_XNUM) {
2055                 size += sizeof(*shdr4extnum);
2056                 if (size > cprm->limit
2057                     || !dump_write(cprm->file, shdr4extnum,
2058                                    sizeof(*shdr4extnum)))
2059                         goto end_coredump;
2060         }
2061
2062 end_coredump:
2063         set_fs(fs);
2064
2065 cleanup:
2066         free_note_info(&info);
2067         kfree(shdr4extnum);
2068         kfree(phdr4note);
2069         kfree(elf);
2070 out:
2071         return has_dumped;
2072 }
2073
2074 #endif          /* CONFIG_ELF_CORE */
2075
2076 static int __init init_elf_binfmt(void)
2077 {
2078         register_binfmt(&elf_format);
2079         return 0;
2080 }
2081
2082 static void __exit exit_elf_binfmt(void)
2083 {
2084         /* Remove the COFF and ELF loaders. */
2085         unregister_binfmt(&elf_format);
2086 }
2087
2088 core_initcall(init_elf_binfmt);
2089 module_exit(exit_elf_binfmt);
2090 MODULE_LICENSE("GPL");