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->flags &= ~PF_FORKNOEXEC;
 716         current->mm->def_flags = def_flags;
 717
 718         /* Do this immediately, since STACK_TOP as used in setup_arg_pages
 719            may depend on the personality.  */
 720         SET_PERSONALITY(loc->elf_ex);
 721         if (elf_read_implies_exec(loc->elf_ex, executable_stack))
 722                 current->personality |= READ_IMPLIES_EXEC;
 723
 724         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 725                 current->flags |= PF_RANDOMIZE;
 726
 727         setup_new_exec(bprm);
 728
 729         /* Do this so that we can load the interpreter, if need be.  We will
 730            change some of these later */
 731         current->mm->free_area_cache = current->mm->mmap_base;
 732         current->mm->cached_hole_size = 0;
 733         retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
 734                                  executable_stack);
 735         if (retval < 0) {
 736                 send_sig(SIGKILL, current, 0);
 737                 goto out_free_dentry;
 738         }
 739
 740         current->mm->start_stack = bprm->p;
 741
 742         /* Now we do a little grungy work by mmapping the ELF image into
 743            the correct location in memory. */
 744         for(i = 0, elf_ppnt = elf_phdata;
 745             i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
 746                 int elf_prot = 0, elf_flags;
 747                 unsigned long k, vaddr;
 748
 749                 if (elf_ppnt->p_type != PT_LOAD)
 750                         continue;
 751
 752                 if (unlikely (elf_brk > elf_bss)) {
 753                         unsigned long nbyte;
 754
 755                         /* There was a PT_LOAD segment with p_memsz > p_filesz
 756                            before this one. Map anonymous pages, if needed,
 757                            and clear the area.  */
 758                         retval = set_brk(elf_bss + load_bias,
 759                                          elf_brk + load_bias);
 760                         if (retval) {
 761                                 send_sig(SIGKILL, current, 0);
 762                                 goto out_free_dentry;
 763                         }
 764                         nbyte = ELF_PAGEOFFSET(elf_bss);
 765                         if (nbyte) {
 766                                 nbyte = ELF_MIN_ALIGN - nbyte;
 767                                 if (nbyte > elf_brk - elf_bss)
 768                                         nbyte = elf_brk - elf_bss;
 769                                 if (clear_user((void __user *)elf_bss +
 770                                                         load_bias, nbyte)) {
 771                                         /*
 772                                          * This bss-zeroing can fail if the ELF
 773                                          * file specifies odd protections. So
 774                                          * we don't check the return value
 775                                          */
 776                                 }
 777                         }
 778                 }
 779
 780                 if (elf_ppnt->p_flags & PF_R)
 781                         elf_prot |= PROT_READ;
 782                 if (elf_ppnt->p_flags & PF_W)
 783                         elf_prot |= PROT_WRITE;
 784                 if (elf_ppnt->p_flags & PF_X)
 785                         elf_prot |= PROT_EXEC;
 786
 787                 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
 788
 789                 vaddr = elf_ppnt->p_vaddr;
 790                 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
 791                         elf_flags |= MAP_FIXED;
 792                 } else if (loc->elf_ex.e_type == ET_DYN) {
 793                         /* Try and get dynamic programs out of the way of the
 794                          * default mmap base, as well as whatever program they
 795                          * might try to exec.  This is because the brk will
 796                          * follow the loader, and is not movable.  */
 797 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
 798                         /* Memory randomization might have been switched off
 799                          * in runtime via sysctl.
 800                          * If that is the case, retain the original non-zero
 801                          * load_bias value in order to establish proper
 802                          * non-randomized mappings.
 803                          */
 804                         if (current->flags & PF_RANDOMIZE)
 805                                 load_bias = 0;
 806                         else
 807                                 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 808 #else
 809                         load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 810 #endif
 811                 }
 812
 813                 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
 814                                 elf_prot, elf_flags, 0);
 815                 if (BAD_ADDR(error)) {
 816                         send_sig(SIGKILL, current, 0);
 817                         retval = IS_ERR((void *)error) ?
 818                                 PTR_ERR((void*)error) : -EINVAL;
 819                         goto out_free_dentry;
 820                 }
 821
 822                 if (!load_addr_set) {
 823                         load_addr_set = 1;
 824                         load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
 825                         if (loc->elf_ex.e_type == ET_DYN) {
 826                                 load_bias += error -
 827                                              ELF_PAGESTART(load_bias + vaddr);
 828                                 load_addr += load_bias;
 829                                 reloc_func_desc = load_bias;
 830                         }
 831                 }
 832                 k = elf_ppnt->p_vaddr;
 833                 if (k < start_code)
 834                         start_code = k;
 835                 if (start_data < k)
 836                         start_data = k;
 837
 838                 /*
 839                  * Check to see if the section's size will overflow the
 840                  * allowed task size. Note that p_filesz must always be
 841                  * <= p_memsz so it is only necessary to check p_memsz.
 842                  */
 843                 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
 844                     elf_ppnt->p_memsz > TASK_SIZE ||
 845                     TASK_SIZE - elf_ppnt->p_memsz < k) {
 846                         /* set_brk can never work. Avoid overflows. */
 847                         send_sig(SIGKILL, current, 0);
 848                         retval = -EINVAL;
 849                         goto out_free_dentry;
 850                 }
 851
 852                 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
 853
 854                 if (k > elf_bss)
 855                         elf_bss = k;
 856                 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
 857                         end_code = k;
 858                 if (end_data < k)
 859                         end_data = k;
 860                 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
 861                 if (k > elf_brk)
 862                         elf_brk = k;
 863         }
 864
 865         loc->elf_ex.e_entry += load_bias;
 866         elf_bss += load_bias;
 867         elf_brk += load_bias;
 868         start_code += load_bias;
 869         end_code += load_bias;
 870         start_data += load_bias;
 871         end_data += load_bias;
 872
 873         /* Calling set_brk effectively mmaps the pages that we need
 874          * for the bss and break sections.  We must do this before
 875          * mapping in the interpreter, to make sure it doesn't wind
 876          * up getting placed where the bss needs to go.
 877          */
 878         retval = set_brk(elf_bss, elf_brk);
 879         if (retval) {
 880                 send_sig(SIGKILL, current, 0);
 881                 goto out_free_dentry;
 882         }
 883         if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
 884                 send_sig(SIGSEGV, current, 0);
 885                 retval = -EFAULT; /* Nobody gets to see this, but.. */
 886                 goto out_free_dentry;
 887         }
 888
 889         if (elf_interpreter) {
 890                 unsigned long uninitialized_var(interp_map_addr);
 891
 892                 elf_entry = load_elf_interp(&loc->interp_elf_ex,
 893                                             interpreter,
 894                                             &interp_map_addr,
 895                                             load_bias);
 896                 if (!IS_ERR((void *)elf_entry)) {
 897                         /*
 898                          * load_elf_interp() returns relocation
 899                          * adjustment
 900                          */
 901                         interp_load_addr = elf_entry;
 902                         elf_entry += loc->interp_elf_ex.e_entry;
 903                 }
 904                 if (BAD_ADDR(elf_entry)) {
 905                         force_sig(SIGSEGV, current);
 906                         retval = IS_ERR((void *)elf_entry) ?
 907                                         (int)elf_entry : -EINVAL;
 908                         goto out_free_dentry;
 909                 }
 910                 reloc_func_desc = interp_load_addr;
 911
 912                 allow_write_access(interpreter);
 913                 fput(interpreter);
 914                 kfree(elf_interpreter);
 915         } else {
 916                 elf_entry = loc->elf_ex.e_entry;
 917                 if (BAD_ADDR(elf_entry)) {
 918                         force_sig(SIGSEGV, current);
 919                         retval = -EINVAL;
 920                         goto out_free_dentry;
 921                 }
 922         }
 923
 924         kfree(elf_phdata);
 925
 926         set_binfmt(&elf_format);
 927
 928 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
 929         retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
 930         if (retval < 0) {
 931                 send_sig(SIGKILL, current, 0);
 932                 goto out;
 933         }
 934 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
 935
 936         install_exec_creds(bprm);
 937         current->flags &= ~PF_FORKNOEXEC;
 938         retval = create_elf_tables(bprm, &loc->elf_ex,
 939                           load_addr, interp_load_addr);
 940         if (retval < 0) {
 941                 send_sig(SIGKILL, current, 0);
 942                 goto out;
 943         }
 944         /* N.B. passed_fileno might not be initialized? */
 945         current->mm->end_code = end_code;
 946         current->mm->start_code = start_code;
 947         current->mm->start_data = start_data;
 948         current->mm->end_data = end_data;
 949         current->mm->start_stack = bprm->p;
 950
 951 #ifdef arch_randomize_brk
 952         if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
 953                 current->mm->brk = current->mm->start_brk =
 954                         arch_randomize_brk(current->mm);
 955 #ifdef CONFIG_COMPAT_BRK
 956                 current->brk_randomized = 1;
 957 #endif
 958         }
 959 #endif
 960
 961         if (current->personality & MMAP_PAGE_ZERO) {
 962                 /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
 963                    and some applications "depend" upon this behavior.
 964                    Since we do not have the power to recompile these, we
 965                    emulate the SVr4 behavior. Sigh. */
 966                 down_write(&current->mm->mmap_sem);
 967                 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
 968                                 MAP_FIXED | MAP_PRIVATE, 0);
 969                 up_write(&current->mm->mmap_sem);
 970         }
 971
 972 #ifdef ELF_PLAT_INIT
 973         /*
 974          * The ABI may specify that certain registers be set up in special
 975          * ways (on i386 %edx is the address of a DT_FINI function, for
 976          * example.  In addition, it may also specify (eg, PowerPC64 ELF)
 977          * that the e_entry field is the address of the function descriptor
 978          * for the startup routine, rather than the address of the startup
 979          * routine itself.  This macro performs whatever initialization to
 980          * the regs structure is required as well as any relocations to the
 981          * function descriptor entries when executing dynamically links apps.
 982          */
 983         ELF_PLAT_INIT(regs, reloc_func_desc);
 984 #endif
 985
 986         start_thread(regs, elf_entry, bprm->p);
 987         retval = 0;
 988 out:
 989         kfree(loc);
 990 out_ret:
 991         return retval;
 992
 993         /* error cleanup */
 994 out_free_dentry:
 995         allow_write_access(interpreter);
 996         if (interpreter)
 997                 fput(interpreter);
 998 out_free_interp:
 999         kfree(elf_interpreter);
1000 out_free_ph:
1001         kfree(elf_phdata);
1002         goto out;
1003 }
1004
1005 /* This is really simpleminded and specialized - we are loading an
1006    a.out library that is given an ELF header. */
1007 static int load_elf_library(struct file *file)
1008 {
1009         struct elf_phdr *elf_phdata;
1010         struct elf_phdr *eppnt;
1011         unsigned long elf_bss, bss, len;
1012         int retval, error, i, j;
1013         struct elfhdr elf_ex;
1014
1015         error = -ENOEXEC;
1016         retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1017         if (retval != sizeof(elf_ex))
1018                 goto out;
1019
1020         if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1021                 goto out;
1022
1023         /* First of all, some simple consistency checks */
1024         if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1025             !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1026                 goto out;
1027
1028         /* Now read in all of the header information */
1029
1030         j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1031         /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1032
1033         error = -ENOMEM;
1034         elf_phdata = kmalloc(j, GFP_KERNEL);
1035         if (!elf_phdata)
1036                 goto out;
1037
1038         eppnt = elf_phdata;
1039         error = -ENOEXEC;
1040         retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1041         if (retval != j)
1042                 goto out_free_ph;
1043
1044         for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1045                 if ((eppnt + i)->p_type == PT_LOAD)
1046                         j++;
1047         if (j != 1)
1048                 goto out_free_ph;
1049
1050         while (eppnt->p_type != PT_LOAD)
1051                 eppnt++;
1052
1053         /* Now use mmap to map the library into memory. */
1054         down_write(&current->mm->mmap_sem);
1055         error = do_mmap(file,
1056                         ELF_PAGESTART(eppnt->p_vaddr),
1057                         (eppnt->p_filesz +
1058                          ELF_PAGEOFFSET(eppnt->p_vaddr)),
1059                         PROT_READ | PROT_WRITE | PROT_EXEC,
1060                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1061                         (eppnt->p_offset -
1062                          ELF_PAGEOFFSET(eppnt->p_vaddr)));
1063         up_write(&current->mm->mmap_sem);
1064         if (error != ELF_PAGESTART(eppnt->p_vaddr))
1065                 goto out_free_ph;
1066
1067         elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1068         if (padzero(elf_bss)) {
1069                 error = -EFAULT;
1070                 goto out_free_ph;
1071         }
1072
1073         len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1074                             ELF_MIN_ALIGN - 1);
1075         bss = eppnt->p_memsz + eppnt->p_vaddr;
1076         if (bss > len) {
1077                 down_write(&current->mm->mmap_sem);
1078                 do_brk(len, bss - len);
1079                 up_write(&current->mm->mmap_sem);
1080         }
1081         error = 0;
1082
1083 out_free_ph:
1084         kfree(elf_phdata);
1085 out:
1086         return error;
1087 }
1088
1089 #ifdef CONFIG_ELF_CORE
1090 /*
1091  * ELF core dumper
1092  *
1093  * Modelled on fs/exec.c:aout_core_dump()
1094  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1095  */
1096
1097 /*
1098  * Decide what to dump of a segment, part, all or none.
1099  */
1100 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1101                                    unsigned long mm_flags)
1102 {
1103 #define FILTER(type)    (mm_flags & (1UL << MMF_DUMP_##type))
1104
1105         /* The vma can be set up to tell us the answer directly.  */
1106         if (vma->vm_flags & VM_ALWAYSDUMP)
1107                 goto whole;
1108
1109         /* Hugetlb memory check */
1110         if (vma->vm_flags & VM_HUGETLB) {
1111                 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1112                         goto whole;
1113                 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1114                         goto whole;
1115         }
1116
1117         /* Do not dump I/O mapped devices or special mappings */
1118         if (vma->vm_flags & (VM_IO | VM_RESERVED))
1119                 return 0;
1120
1121         /* By default, dump shared memory if mapped from an anonymous file. */
1122         if (vma->vm_flags & VM_SHARED) {
1123                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1124                     FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1125                         goto whole;
1126                 return 0;
1127         }
1128
1129         /* Dump segments that have been written to.  */
1130         if (vma->anon_vma && FILTER(ANON_PRIVATE))
1131                 goto whole;
1132         if (vma->vm_file == NULL)
1133                 return 0;
1134
1135         if (FILTER(MAPPED_PRIVATE))
1136                 goto whole;
1137
1138         /*
1139          * If this looks like the beginning of a DSO or executable mapping,
1140          * check for an ELF header.  If we find one, dump the first page to
1141          * aid in determining what was mapped here.
1142          */
1143         if (FILTER(ELF_HEADERS) &&
1144             vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1145                 u32 __user *header = (u32 __user *) vma->vm_start;
1146                 u32 word;
1147                 mm_segment_t fs = get_fs();
1148                 /*
1149                  * Doing it this way gets the constant folded by GCC.
1150                  */
1151                 union {
1152                         u32 cmp;
1153                         char elfmag[SELFMAG];
1154                 } magic;
1155                 BUILD_BUG_ON(SELFMAG != sizeof word);
1156                 magic.elfmag[EI_MAG0] = ELFMAG0;
1157                 magic.elfmag[EI_MAG1] = ELFMAG1;
1158                 magic.elfmag[EI_MAG2] = ELFMAG2;
1159                 magic.elfmag[EI_MAG3] = ELFMAG3;
1160                 /*
1161                  * Switch to the user "segment" for get_user(),
1162                  * then put back what elf_core_dump() had in place.
1163                  */
1164                 set_fs(USER_DS);
1165                 if (unlikely(get_user(word, header)))
1166                         word = 0;
1167                 set_fs(fs);
1168                 if (word == magic.cmp)
1169                         return PAGE_SIZE;
1170         }
1171
1172 #undef  FILTER
1173
1174         return 0;
1175
1176 whole:
1177         return vma->vm_end - vma->vm_start;
1178 }
1179
1180 /* An ELF note in memory */
1181 struct memelfnote
1182 {
1183         const char *name;
1184         int type;
1185         unsigned int datasz;
1186         void *data;
1187 };
1188
1189 static int notesize(struct memelfnote *en)
1190 {
1191         int sz;
1192
1193         sz = sizeof(struct elf_note);
1194         sz += roundup(strlen(en->name) + 1, 4);
1195         sz += roundup(en->datasz, 4);
1196
1197         return sz;
1198 }
1199
1200 #define DUMP_WRITE(addr, nr, foffset)   \
1201         do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1202
1203 static int alignfile(struct file *file, loff_t *foffset)
1204 {
1205         static const char buf[4] = { 0, };
1206         DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1207         return 1;
1208 }
1209
1210 static int writenote(struct memelfnote *men, struct file *file,
1211                         loff_t *foffset)
1212 {
1213         struct elf_note en;
1214         en.n_namesz = strlen(men->name) + 1;
1215         en.n_descsz = men->datasz;
1216         en.n_type = men->type;
1217
1218         DUMP_WRITE(&en, sizeof(en), foffset);
1219         DUMP_WRITE(men->name, en.n_namesz, foffset);
1220         if (!alignfile(file, foffset))
1221                 return 0;
1222         DUMP_WRITE(men->data, men->datasz, foffset);
1223         if (!alignfile(file, foffset))
1224                 return 0;
1225
1226         return 1;
1227 }
1228 #undef DUMP_WRITE
1229
1230 static void fill_elf_header(struct elfhdr *elf, int segs,
1231                             u16 machine, u32 flags, u8 osabi)
1232 {
1233         memset(elf, 0, sizeof(*elf));
1234
1235         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1236         elf->e_ident[EI_CLASS] = ELF_CLASS;
1237         elf->e_ident[EI_DATA] = ELF_DATA;
1238         elf->e_ident[EI_VERSION] = EV_CURRENT;
1239         elf->e_ident[EI_OSABI] = ELF_OSABI;
1240
1241         elf->e_type = ET_CORE;
1242         elf->e_machine = machine;
1243         elf->e_version = EV_CURRENT;
1244         elf->e_phoff = sizeof(struct elfhdr);
1245         elf->e_flags = flags;
1246         elf->e_ehsize = sizeof(struct elfhdr);
1247         elf->e_phentsize = sizeof(struct elf_phdr);
1248         elf->e_phnum = segs;
1249
1250         return;
1251 }
1252
1253 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1254 {
1255         phdr->p_type = PT_NOTE;
1256         phdr->p_offset = offset;
1257         phdr->p_vaddr = 0;
1258         phdr->p_paddr = 0;
1259         phdr->p_filesz = sz;
1260         phdr->p_memsz = 0;
1261         phdr->p_flags = 0;
1262         phdr->p_align = 0;
1263         return;
1264 }
1265
1266 static void fill_note(struct memelfnote *note, const char *name, int type,
1267                 unsigned int sz, void *data)
1268 {
1269         note->name = name;
1270         note->type = type;
1271         note->datasz = sz;
1272         note->data = data;
1273         return;
1274 }
1275
1276 /*
1277  * fill up all the fields in prstatus from the given task struct, except
1278  * registers which need to be filled up separately.
1279  */
1280 static void fill_prstatus(struct elf_prstatus *prstatus,
1281                 struct task_struct *p, long signr)
1282 {
1283         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1284         prstatus->pr_sigpend = p->pending.signal.sig[0];
1285         prstatus->pr_sighold = p->blocked.sig[0];
1286         rcu_read_lock();
1287         prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1288         rcu_read_unlock();
1289         prstatus->pr_pid = task_pid_vnr(p);
1290         prstatus->pr_pgrp = task_pgrp_vnr(p);
1291         prstatus->pr_sid = task_session_vnr(p);
1292         if (thread_group_leader(p)) {
1293                 struct task_cputime cputime;
1294
1295                 /*
1296                  * This is the record for the group leader.  It shows the
1297                  * group-wide total, not its individual thread total.
1298                  */
1299                 thread_group_cputime(p, &cputime);
1300                 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1301                 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1302         } else {
1303                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1304                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1305         }
1306         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1307         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1308 }
1309
1310 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1311                        struct mm_struct *mm)
1312 {
1313         const struct cred *cred;
1314         unsigned int i, len;
1315
1316         /* first copy the parameters from user space */
1317         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1318
1319         len = mm->arg_end - mm->arg_start;
1320         if (len >= ELF_PRARGSZ)
1321                 len = ELF_PRARGSZ-1;
1322         if (copy_from_user(&psinfo->pr_psargs,
1323                            (const char __user *)mm->arg_start, len))
1324                 return -EFAULT;
1325         for(i = 0; i < len; i++)
1326                 if (psinfo->pr_psargs[i] == 0)
1327                         psinfo->pr_psargs[i] = ' ';
1328         psinfo->pr_psargs[len] = 0;
1329
1330         rcu_read_lock();
1331         psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1332         rcu_read_unlock();
1333         psinfo->pr_pid = task_pid_vnr(p);
1334         psinfo->pr_pgrp = task_pgrp_vnr(p);
1335         psinfo->pr_sid = task_session_vnr(p);
1336
1337         i = p->state ? ffz(~p->state) + 1 : 0;
1338         psinfo->pr_state = i;
1339         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1340         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1341         psinfo->pr_nice = task_nice(p);
1342         psinfo->pr_flag = p->flags;
1343         rcu_read_lock();
1344         cred = __task_cred(p);
1345         SET_UID(psinfo->pr_uid, cred->uid);
1346         SET_GID(psinfo->pr_gid, cred->gid);
1347         rcu_read_unlock();
1348         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1349
1350         return 0;
1351 }
1352
1353 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1354 {
1355         elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1356         int i = 0;
1357         do
1358                 i += 2;
1359         while (auxv[i - 2] != AT_NULL);
1360         fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1361 }
1362
1363 #ifdef CORE_DUMP_USE_REGSET
1364 #include <linux/regset.h>
1365
1366 struct elf_thread_core_info {
1367         struct elf_thread_core_info *next;
1368         struct task_struct *task;
1369         struct elf_prstatus prstatus;
1370         struct memelfnote notes[0];
1371 };
1372
1373 struct elf_note_info {
1374         struct elf_thread_core_info *thread;
1375         struct memelfnote psinfo;
1376         struct memelfnote auxv;
1377         size_t size;
1378         int thread_notes;
1379 };
1380
1381 /*
1382  * When a regset has a writeback hook, we call it on each thread before
1383  * dumping user memory.  On register window machines, this makes sure the
1384  * user memory backing the register data is up to date before we read it.
1385  */
1386 static void do_thread_regset_writeback(struct task_struct *task,
1387                                        const struct user_regset *regset)
1388 {
1389         if (regset->writeback)
1390                 regset->writeback(task, regset, 1);
1391 }
1392
1393 static int fill_thread_core_info(struct elf_thread_core_info *t,
1394                                  const struct user_regset_view *view,
1395                                  long signr, size_t *total)
1396 {
1397         unsigned int i;
1398
1399         /*
1400          * NT_PRSTATUS is the one special case, because the regset data
1401          * goes into the pr_reg field inside the note contents, rather
1402          * than being the whole note contents.  We fill the reset in here.
1403          * We assume that regset 0 is NT_PRSTATUS.
1404          */
1405         fill_prstatus(&t->prstatus, t->task, signr);
1406         (void) view->regsets[0].get(t->task, &view->regsets[0],
1407                                     0, sizeof(t->prstatus.pr_reg),
1408                                     &t->prstatus.pr_reg, NULL);
1409
1410         fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1411                   sizeof(t->prstatus), &t->prstatus);
1412         *total += notesize(&t->notes[0]);
1413
1414         do_thread_regset_writeback(t->task, &view->regsets[0]);
1415
1416         /*
1417          * Each other regset might generate a note too.  For each regset
1418          * that has no core_note_type or is inactive, we leave t->notes[i]
1419          * all zero and we'll know to skip writing it later.
1420          */
1421         for (i = 1; i < view->n; ++i) {
1422                 const struct user_regset *regset = &view->regsets[i];
1423                 do_thread_regset_writeback(t->task, regset);
1424                 if (regset->core_note_type && regset->get &&
1425                     (!regset->active || regset->active(t->task, regset))) {
1426                         int ret;
1427                         size_t size = regset->n * regset->size;
1428                         void *data = kmalloc(size, GFP_KERNEL);
1429                         if (unlikely(!data))
1430                                 return 0;
1431                         ret = regset->get(t->task, regset,
1432                                           0, size, data, NULL);
1433                         if (unlikely(ret))
1434                                 kfree(data);
1435                         else {
1436                                 if (regset->core_note_type != NT_PRFPREG)
1437                                         fill_note(&t->notes[i], "LINUX",
1438                                                   regset->core_note_type,
1439                                                   size, data);
1440                                 else {
1441                                         t->prstatus.pr_fpvalid = 1;
1442                                         fill_note(&t->notes[i], "CORE",
1443                                                   NT_PRFPREG, size, data);
1444                                 }
1445                                 *total += notesize(&t->notes[i]);
1446                         }
1447                 }
1448         }
1449
1450         return 1;
1451 }
1452
1453 static int fill_note_info(struct elfhdr *elf, int phdrs,
1454                           struct elf_note_info *info,
1455                           long signr, struct pt_regs *regs)
1456 {
1457         struct task_struct *dump_task = current;
1458         const struct user_regset_view *view = task_user_regset_view(dump_task);
1459         struct elf_thread_core_info *t;
1460         struct elf_prpsinfo *psinfo;
1461         struct core_thread *ct;
1462         unsigned int i;
1463
1464         info->size = 0;
1465         info->thread = NULL;
1466
1467         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1468         if (psinfo == NULL)
1469                 return 0;
1470
1471         fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1472
1473         /*
1474          * Figure out how many notes we're going to need for each thread.
1475          */
1476         info->thread_notes = 0;
1477         for (i = 0; i < view->n; ++i)
1478                 if (view->regsets[i].core_note_type != 0)
1479                         ++info->thread_notes;
1480
1481         /*
1482          * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1483          * since it is our one special case.
1484          */
1485         if (unlikely(info->thread_notes == 0) ||
1486             unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1487                 WARN_ON(1);
1488                 return 0;
1489         }
1490
1491         /*
1492          * Initialize the ELF file header.
1493          */
1494         fill_elf_header(elf, phdrs,
1495                         view->e_machine, view->e_flags, view->ei_osabi);
1496
1497         /*
1498          * Allocate a structure for each thread.
1499          */
1500         for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1501                 t = kzalloc(offsetof(struct elf_thread_core_info,
1502                                      notes[info->thread_notes]),
1503                             GFP_KERNEL);
1504                 if (unlikely(!t))
1505                         return 0;
1506
1507                 t->task = ct->task;
1508                 if (ct->task == dump_task || !info->thread) {
1509                         t->next = info->thread;
1510                         info->thread = t;
1511                 } else {
1512                         /*
1513                          * Make sure to keep the original task at
1514                          * the head of the list.
1515                          */
1516                         t->next = info->thread->next;
1517                         info->thread->next = t;
1518                 }
1519         }
1520
1521         /*
1522          * Now fill in each thread's information.
1523          */
1524         for (t = info->thread; t != NULL; t = t->next)
1525                 if (!fill_thread_core_info(t, view, signr, &info->size))
1526                         return 0;
1527
1528         /*
1529          * Fill in the two process-wide notes.
1530          */
1531         fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1532         info->size += notesize(&info->psinfo);
1533
1534         fill_auxv_note(&info->auxv, current->mm);
1535         info->size += notesize(&info->auxv);
1536
1537         return 1;
1538 }
1539
1540 static size_t get_note_info_size(struct elf_note_info *info)
1541 {
1542         return info->size;
1543 }
1544
1545 /*
1546  * Write all the notes for each thread.  When writing the first thread, the
1547  * process-wide notes are interleaved after the first thread-specific note.
1548  */
1549 static int write_note_info(struct elf_note_info *info,
1550                            struct file *file, loff_t *foffset)
1551 {
1552         bool first = 1;
1553         struct elf_thread_core_info *t = info->thread;
1554
1555         do {
1556                 int i;
1557
1558                 if (!writenote(&t->notes[0], file, foffset))
1559                         return 0;
1560
1561                 if (first && !writenote(&info->psinfo, file, foffset))
1562                         return 0;
1563                 if (first && !writenote(&info->auxv, file, foffset))
1564                         return 0;
1565
1566                 for (i = 1; i < info->thread_notes; ++i)
1567                         if (t->notes[i].data &&
1568                             !writenote(&t->notes[i], file, foffset))
1569                                 return 0;
1570
1571                 first = 0;
1572                 t = t->next;
1573         } while (t);
1574
1575         return 1;
1576 }
1577
1578 static void free_note_info(struct elf_note_info *info)
1579 {
1580         struct elf_thread_core_info *threads = info->thread;
1581         while (threads) {
1582                 unsigned int i;
1583                 struct elf_thread_core_info *t = threads;
1584                 threads = t->next;
1585                 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1586                 for (i = 1; i < info->thread_notes; ++i)
1587                         kfree(t->notes[i].data);
1588                 kfree(t);
1589         }
1590         kfree(info->psinfo.data);
1591 }
1592
1593 #else
1594
1595 /* Here is the structure in which status of each thread is captured. */
1596 struct elf_thread_status
1597 {
1598         struct list_head list;
1599         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1600         elf_fpregset_t fpu;             /* NT_PRFPREG */
1601         struct task_struct *thread;
1602 #ifdef ELF_CORE_COPY_XFPREGS
1603         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1604 #endif
1605         struct memelfnote notes[3];
1606         int num_notes;
1607 };
1608
1609 /*
1610  * In order to add the specific thread information for the elf file format,
1611  * we need to keep a linked list of every threads pr_status and then create
1612  * a single section for them in the final core file.
1613  */
1614 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1615 {
1616         int sz = 0;
1617         struct task_struct *p = t->thread;
1618         t->num_notes = 0;
1619
1620         fill_prstatus(&t->prstatus, p, signr);
1621         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1622
1623         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1624                   &(t->prstatus));
1625         t->num_notes++;
1626         sz += notesize(&t->notes[0]);
1627
1628         if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1629                                                                 &t->fpu))) {
1630                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1631                           &(t->fpu));
1632                 t->num_notes++;
1633                 sz += notesize(&t->notes[1]);
1634         }
1635
1636 #ifdef ELF_CORE_COPY_XFPREGS
1637         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1638                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1639                           sizeof(t->xfpu), &t->xfpu);
1640                 t->num_notes++;
1641                 sz += notesize(&t->notes[2]);
1642         }
1643 #endif
1644         return sz;
1645 }
1646
1647 struct elf_note_info {
1648         struct memelfnote *notes;
1649         struct elf_prstatus *prstatus;  /* NT_PRSTATUS */
1650         struct elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
1651         struct list_head thread_list;
1652         elf_fpregset_t *fpu;
1653 #ifdef ELF_CORE_COPY_XFPREGS
1654         elf_fpxregset_t *xfpu;
1655 #endif
1656         int thread_status_size;
1657         int numnote;
1658 };
1659
1660 static int elf_note_info_init(struct elf_note_info *info)
1661 {
1662         memset(info, 0, sizeof(*info));
1663         INIT_LIST_HEAD(&info->thread_list);
1664
1665         /* Allocate space for six ELF notes */
1666         info->notes = kmalloc(6 * sizeof(struct memelfnote), GFP_KERNEL);
1667         if (!info->notes)
1668                 return 0;
1669         info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1670         if (!info->psinfo)
1671                 goto notes_free;
1672         info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1673         if (!info->prstatus)
1674                 goto psinfo_free;
1675         info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1676         if (!info->fpu)
1677                 goto prstatus_free;
1678 #ifdef ELF_CORE_COPY_XFPREGS
1679         info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1680         if (!info->xfpu)
1681                 goto fpu_free;
1682 #endif
1683         return 1;
1684 #ifdef ELF_CORE_COPY_XFPREGS
1685  fpu_free:
1686         kfree(info->fpu);
1687 #endif
1688  prstatus_free:
1689         kfree(info->prstatus);
1690  psinfo_free:
1691         kfree(info->psinfo);
1692  notes_free:
1693         kfree(info->notes);
1694         return 0;
1695 }
1696
1697 static int fill_note_info(struct elfhdr *elf, int phdrs,
1698                           struct elf_note_info *info,
1699                           long signr, struct pt_regs *regs)
1700 {
1701         struct list_head *t;
1702
1703         if (!elf_note_info_init(info))
1704                 return 0;
1705
1706         if (signr) {
1707                 struct core_thread *ct;
1708                 struct elf_thread_status *ets;
1709
1710                 for (ct = current->mm->core_state->dumper.next;
1711                                                 ct; ct = ct->next) {
1712                         ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1713                         if (!ets)
1714                                 return 0;
1715
1716                         ets->thread = ct->task;
1717                         list_add(&ets->list, &info->thread_list);
1718                 }
1719
1720                 list_for_each(t, &info->thread_list) {
1721                         int sz;
1722
1723                         ets = list_entry(t, struct elf_thread_status, list);
1724                         sz = elf_dump_thread_status(signr, ets);
1725                         info->thread_status_size += sz;
1726                 }
1727         }
1728         /* now collect the dump for the current */
1729         memset(info->prstatus, 0, sizeof(*info->prstatus));
1730         fill_prstatus(info->prstatus, current, signr);
1731         elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1732
1733         /* Set up header */
1734         fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1735
1736         /*
1737          * Set up the notes in similar form to SVR4 core dumps made
1738          * with info from their /proc.
1739          */
1740
1741         fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1742                   sizeof(*info->prstatus), info->prstatus);
1743         fill_psinfo(info->psinfo, current->group_leader, current->mm);
1744         fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1745                   sizeof(*info->psinfo), info->psinfo);
1746
1747         info->numnote = 2;
1748
1749         fill_auxv_note(&info->notes[info->numnote++], current->mm);
1750
1751         /* Try to dump the FPU. */
1752         info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1753                                                                info->fpu);
1754         if (info->prstatus->pr_fpvalid)
1755                 fill_note(info->notes + info->numnote++,
1756                           "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1757 #ifdef ELF_CORE_COPY_XFPREGS
1758         if (elf_core_copy_task_xfpregs(current, info->xfpu))
1759                 fill_note(info->notes + info->numnote++,
1760                           "LINUX", ELF_CORE_XFPREG_TYPE,
1761                           sizeof(*info->xfpu), info->xfpu);
1762 #endif
1763
1764         return 1;
1765 }
1766
1767 static size_t get_note_info_size(struct elf_note_info *info)
1768 {
1769         int sz = 0;
1770         int i;
1771
1772         for (i = 0; i < info->numnote; i++)
1773                 sz += notesize(info->notes + i);
1774
1775         sz += info->thread_status_size;
1776
1777         return sz;
1778 }
1779
1780 static int write_note_info(struct elf_note_info *info,
1781                            struct file *file, loff_t *foffset)
1782 {
1783         int i;
1784         struct list_head *t;
1785
1786         for (i = 0; i < info->numnote; i++)
1787                 if (!writenote(info->notes + i, file, foffset))
1788                         return 0;
1789
1790         /* write out the thread status notes section */
1791         list_for_each(t, &info->thread_list) {
1792                 struct elf_thread_status *tmp =
1793                                 list_entry(t, struct elf_thread_status, list);
1794
1795                 for (i = 0; i < tmp->num_notes; i++)
1796                         if (!writenote(&tmp->notes[i], file, foffset))
1797                                 return 0;
1798         }
1799
1800         return 1;
1801 }
1802
1803 static void free_note_info(struct elf_note_info *info)
1804 {
1805         while (!list_empty(&info->thread_list)) {
1806                 struct list_head *tmp = info->thread_list.next;
1807                 list_del(tmp);
1808                 kfree(list_entry(tmp, struct elf_thread_status, list));
1809         }
1810
1811         kfree(info->prstatus);
1812         kfree(info->psinfo);
1813         kfree(info->notes);
1814         kfree(info->fpu);
1815 #ifdef ELF_CORE_COPY_XFPREGS
1816         kfree(info->xfpu);
1817 #endif
1818 }
1819
1820 #endif
1821
1822 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1823                                         struct vm_area_struct *gate_vma)
1824 {
1825         struct vm_area_struct *ret = tsk->mm->mmap;
1826
1827         if (ret)
1828                 return ret;
1829         return gate_vma;
1830 }
1831 /*
1832  * Helper function for iterating across a vma list.  It ensures that the caller
1833  * will visit `gate_vma' prior to terminating the search.
1834  */
1835 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1836                                         struct vm_area_struct *gate_vma)
1837 {
1838         struct vm_area_struct *ret;
1839
1840         ret = this_vma->vm_next;
1841         if (ret)
1842                 return ret;
1843         if (this_vma == gate_vma)
1844                 return NULL;
1845         return gate_vma;
1846 }
1847
1848 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1849                              elf_addr_t e_shoff, int segs)
1850 {
1851         elf->e_shoff = e_shoff;
1852         elf->e_shentsize = sizeof(*shdr4extnum);
1853         elf->e_shnum = 1;
1854         elf->e_shstrndx = SHN_UNDEF;
1855
1856         memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1857
1858         shdr4extnum->sh_type = SHT_NULL;
1859         shdr4extnum->sh_size = elf->e_shnum;
1860         shdr4extnum->sh_link = elf->e_shstrndx;
1861         shdr4extnum->sh_info = segs;
1862 }
1863
1864 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
1865                                      unsigned long mm_flags)
1866 {
1867         struct vm_area_struct *vma;
1868         size_t size = 0;
1869
1870         for (vma = first_vma(current, gate_vma); vma != NULL;
1871              vma = next_vma(vma, gate_vma))
1872                 size += vma_dump_size(vma, mm_flags);
1873         return size;
1874 }
1875
1876 /*
1877  * Actual dumper
1878  *
1879  * This is a two-pass process; first we find the offsets of the bits,
1880  * and then they are actually written out.  If we run out of core limit
1881  * we just truncate.
1882  */
1883 static int elf_core_dump(struct coredump_params *cprm)
1884 {
1885         int has_dumped = 0;
1886         mm_segment_t fs;
1887         int segs;
1888         size_t size = 0;
1889         struct vm_area_struct *vma, *gate_vma;
1890         struct elfhdr *elf = NULL;
1891         loff_t offset = 0, dataoff, foffset;
1892         struct elf_note_info info;
1893         struct elf_phdr *phdr4note = NULL;
1894         struct elf_shdr *shdr4extnum = NULL;
1895         Elf_Half e_phnum;
1896         elf_addr_t e_shoff;
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 out;
1914         /*
1915          * The number of segs are recored into ELF header as 16bit value.
1916          * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
1917          */
1918         segs = current->mm->map_count;
1919         segs += elf_core_extra_phdrs();
1920
1921         gate_vma = get_gate_vma(current->mm);
1922         if (gate_vma != NULL)
1923                 segs++;
1924
1925         /* for notes section */
1926         segs++;
1927
1928         /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1929          * this, kernel supports extended numbering. Have a look at
1930          * include/linux/elf.h for further information. */
1931         e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1932
1933         /*
1934          * Collect all the non-memory information about the process for the
1935          * notes.  This also sets up the file header.
1936          */
1937         if (!fill_note_info(elf, e_phnum, &info, cprm->signr, cprm->regs))
1938                 goto cleanup;
1939
1940         has_dumped = 1;
1941         current->flags |= PF_DUMPCORE;
1942
1943         fs = get_fs();
1944         set_fs(KERNEL_DS);
1945
1946         offset += sizeof(*elf);                         /* Elf header */
1947         offset += segs * sizeof(struct elf_phdr);       /* Program headers */
1948         foffset = offset;
1949
1950         /* Write notes phdr entry */
1951         {
1952                 size_t sz = get_note_info_size(&info);
1953
1954                 sz += elf_coredump_extra_notes_size();
1955
1956                 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1957                 if (!phdr4note)
1958                         goto end_coredump;
1959
1960                 fill_elf_note_phdr(phdr4note, sz, offset);
1961                 offset += sz;
1962         }
1963
1964         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1965
1966         offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
1967         offset += elf_core_extra_data_size();
1968         e_shoff = offset;
1969
1970         if (e_phnum == PN_XNUM) {
1971                 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1972                 if (!shdr4extnum)
1973                         goto end_coredump;
1974                 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1975         }
1976
1977         offset = dataoff;
1978
1979         size += sizeof(*elf);
1980         if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
1981                 goto end_coredump;
1982
1983         size += sizeof(*phdr4note);
1984         if (size > cprm->limit
1985             || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
1986                 goto end_coredump;
1987
1988         /* Write program headers for segments dump */
1989         for (vma = first_vma(current, gate_vma); vma != NULL;
1990                         vma = next_vma(vma, gate_vma)) {
1991                 struct elf_phdr phdr;
1992
1993                 phdr.p_type = PT_LOAD;
1994                 phdr.p_offset = offset;
1995                 phdr.p_vaddr = vma->vm_start;
1996                 phdr.p_paddr = 0;
1997                 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
1998                 phdr.p_memsz = vma->vm_end - vma->vm_start;
1999                 offset += phdr.p_filesz;
2000                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2001                 if (vma->vm_flags & VM_WRITE)
2002                         phdr.p_flags |= PF_W;
2003                 if (vma->vm_flags & VM_EXEC)
2004                         phdr.p_flags |= PF_X;
2005                 phdr.p_align = ELF_EXEC_PAGESIZE;
2006
2007                 size += sizeof(phdr);
2008                 if (size > cprm->limit
2009                     || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2010                         goto end_coredump;
2011         }
2012
2013         if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2014                 goto end_coredump;
2015
2016         /* write out the notes section */
2017         if (!write_note_info(&info, cprm->file, &foffset))
2018                 goto end_coredump;
2019
2020         if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2021                 goto end_coredump;
2022
2023         /* Align to page */
2024         if (!dump_seek(cprm->file, dataoff - foffset))
2025                 goto end_coredump;
2026
2027         for (vma = first_vma(current, gate_vma); vma != NULL;
2028                         vma = next_vma(vma, gate_vma)) {
2029                 unsigned long addr;
2030                 unsigned long end;
2031
2032                 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2033
2034                 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2035                         struct page *page;
2036                         int stop;
2037
2038                         page = get_dump_page(addr);
2039                         if (page) {
2040                                 void *kaddr = kmap(page);
2041                                 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2042                                         !dump_write(cprm->file, kaddr,
2043                                                     PAGE_SIZE);
2044                                 kunmap(page);
2045                                 page_cache_release(page);
2046                         } else
2047                                 stop = !dump_seek(cprm->file, PAGE_SIZE);
2048                         if (stop)
2049                                 goto end_coredump;
2050                 }
2051         }
2052
2053         if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2054                 goto end_coredump;
2055
2056         if (e_phnum == PN_XNUM) {
2057                 size += sizeof(*shdr4extnum);
2058                 if (size > cprm->limit
2059                     || !dump_write(cprm->file, shdr4extnum,
2060                                    sizeof(*shdr4extnum)))
2061                         goto end_coredump;
2062         }
2063
2064 end_coredump:
2065         set_fs(fs);
2066
2067 cleanup:
2068         free_note_info(&info);
2069         kfree(shdr4extnum);
2070         kfree(phdr4note);
2071         kfree(elf);
2072 out:
2073         return has_dumped;
2074 }
2075
2076 #endif          /* CONFIG_ELF_CORE */
2077
2078 static int __init init_elf_binfmt(void)
2079 {
2080         return register_binfmt(&elf_format);
2081 }
2082
2083 static void __exit exit_elf_binfmt(void)
2084 {
2085         /* Remove the COFF and ELF loaders. */
2086         unregister_binfmt(&elf_format);
2087 }
2088
2089 core_initcall(init_elf_binfmt);
2090 module_exit(exit_elf_binfmt);
2091 MODULE_LICENSE("GPL");