fs/binfmt_elf.c

   1 /*
   2  * linux/fs/binfmt_elf.c
   3  *
   4  * These are the functions used to load ELF format executables as used
   5  * on SVr4 machines.  Information on the format may be found in the book
   6  * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
   7  * Tools".
   8  *
   9  * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/kernel.h>
  14 #include <linux/fs.h>
  15 #include <linux/stat.h>
  16 #include <linux/time.h>
  17 #include <linux/mm.h>
  18 #include <linux/mman.h>
  19 #include <linux/a.out.h>
  20 #include <linux/errno.h>
  21 #include <linux/signal.h>
  22 #include <linux/binfmts.h>
  23 #include <linux/string.h>
  24 #include <linux/file.h>
  25 #include <linux/fcntl.h>
  26 #include <linux/ptrace.h>
  27 #include <linux/slab.h>
  28 #include <linux/shm.h>
  29 #include <linux/personality.h>
  30 #include <linux/elfcore.h>
  31 #include <linux/init.h>
  32 #include <linux/highuid.h>
  33 #include <linux/smp.h>
  34 #include <linux/compiler.h>
  35 #include <linux/highmem.h>
  36 #include <linux/pagemap.h>
  37 #include <linux/security.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/random.h>
  40 #include <linux/elf.h>
  41 #include <linux/utsname.h>
  42 #include <asm/uaccess.h>
  43 #include <asm/param.h>
  44 #include <asm/page.h>
  45
  46 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
  47 static int load_elf_library(struct file *);
  48 static unsigned long elf_map (struct file *, unsigned long, struct elf_phdr *, int, int);
  49
  50 /*
  51  * If we don't support core dumping, then supply a NULL so we
  52  * don't even try.
  53  */
  54 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
  55 static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file);
  56 #else
  57 #define elf_core_dump   NULL
  58 #endif
  59
  60 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
  61 #define ELF_MIN_ALIGN   ELF_EXEC_PAGESIZE
  62 #else
  63 #define ELF_MIN_ALIGN   PAGE_SIZE
  64 #endif
  65
  66 #ifndef ELF_CORE_EFLAGS
  67 #define ELF_CORE_EFLAGS 0
  68 #endif
  69
  70 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
  71 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
  72 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
  73
  74 static struct linux_binfmt elf_format = {
  75                 .module         = THIS_MODULE,
  76                 .load_binary    = load_elf_binary,
  77                 .load_shlib     = load_elf_library,
  78                 .core_dump      = elf_core_dump,
  79                 .min_coredump   = ELF_EXEC_PAGESIZE,
  80                 .hasvdso        = 1
  81 };
  82
  83 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
  84
  85 static int set_brk(unsigned long start, unsigned long end)
  86 {
  87         start = ELF_PAGEALIGN(start);
  88         end = ELF_PAGEALIGN(end);
  89         if (end > start) {
  90                 unsigned long addr;
  91                 down_write(&current->mm->mmap_sem);
  92                 addr = do_brk(start, end - start);
  93                 up_write(&current->mm->mmap_sem);
  94                 if (BAD_ADDR(addr))
  95                         return addr;
  96         }
  97         current->mm->start_brk = current->mm->brk = end;
  98         return 0;
  99 }
 100
 101 /* We need to explicitly zero any fractional pages
 102    after the data section (i.e. bss).  This would
 103    contain the junk from the file that should not
 104    be in memory
 105  */
 106 static int padzero(unsigned long elf_bss)
 107 {
 108         unsigned long nbyte;
 109
 110         nbyte = ELF_PAGEOFFSET(elf_bss);
 111         if (nbyte) {
 112                 nbyte = ELF_MIN_ALIGN - nbyte;
 113                 if (clear_user((void __user *) elf_bss, nbyte))
 114                         return -EFAULT;
 115         }
 116         return 0;
 117 }
 118
 119 /* Let's use some macros to make this stack manipulation a litle clearer */
 120 #ifdef CONFIG_STACK_GROWSUP
 121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
 122 #define STACK_ROUND(sp, items) \
 123         ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
 124 #define STACK_ALLOC(sp, len) ({ \
 125         elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
 126         old_sp; })
 127 #else
 128 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
 129 #define STACK_ROUND(sp, items) \
 130         (((unsigned long) (sp - items)) &~ 15UL)
 131 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
 132 #endif
 133
 134 static int
 135 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
 136                 int interp_aout, unsigned long load_addr,
 137                 unsigned long interp_load_addr)
 138 {
 139         unsigned long p = bprm->p;
 140         int argc = bprm->argc;
 141         int envc = bprm->envc;
 142         elf_addr_t __user *argv;
 143         elf_addr_t __user *envp;
 144         elf_addr_t __user *sp;
 145         elf_addr_t __user *u_platform;
 146         const char *k_platform = ELF_PLATFORM;
 147         int items;
 148         elf_addr_t *elf_info;
 149         int ei_index = 0;
 150         struct task_struct *tsk = current;
 151         struct vm_area_struct *vma;
 152
 153         /*
 154          * If this architecture has a platform capability string, copy it
 155          * to userspace.  In some cases (Sparc), this info is impossible
 156          * for userspace to get any other way, in others (i386) it is
 157          * merely difficult.
 158          */
 159         u_platform = NULL;
 160         if (k_platform) {
 161                 size_t len = strlen(k_platform) + 1;
 162
 163                 /*
 164                  * In some cases (e.g. Hyper-Threading), we want to avoid L1
 165                  * evictions by the processes running on the same package. One
 166                  * thing we can do is to shuffle the initial stack for them.
 167                  */
 168
 169                 p = arch_align_stack(p);
 170
 171                 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 172                 if (__copy_to_user(u_platform, k_platform, len))
 173                         return -EFAULT;
 174         }
 175
 176         /* Create the ELF interpreter info */
 177         elf_info = (elf_addr_t *)current->mm->saved_auxv;
 178 #define NEW_AUX_ENT(id, val) \
 179         do { \
 180                 elf_info[ei_index++] = id; \
 181                 elf_info[ei_index++] = val; \
 182         } while (0)
 183
 184 #ifdef ARCH_DLINFO
 185         /*
 186          * ARCH_DLINFO must come first so PPC can do its special alignment of
 187          * AUXV.
 188          */
 189         ARCH_DLINFO;
 190 #endif
 191         NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
 192         NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
 193         NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
 194         NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
 195         NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
 196         NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
 197         NEW_AUX_ENT(AT_BASE, interp_load_addr);
 198         NEW_AUX_ENT(AT_FLAGS, 0);
 199         NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
 200         NEW_AUX_ENT(AT_UID, tsk->uid);
 201         NEW_AUX_ENT(AT_EUID, tsk->euid);
 202         NEW_AUX_ENT(AT_GID, tsk->gid);
 203         NEW_AUX_ENT(AT_EGID, tsk->egid);
 204         NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
 205         if (k_platform) {
 206                 NEW_AUX_ENT(AT_PLATFORM,
 207                             (elf_addr_t)(unsigned long)u_platform);
 208         }
 209         if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
 210                 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
 211         }
 212 #undef NEW_AUX_ENT
 213         /* AT_NULL is zero; clear the rest too */
 214         memset(&elf_info[ei_index], 0,
 215                sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
 216
 217         /* And advance past the AT_NULL entry.  */
 218         ei_index += 2;
 219
 220         sp = STACK_ADD(p, ei_index);
 221
 222         items = (argc + 1) + (envc + 1);
 223         if (interp_aout) {
 224                 items += 3; /* a.out interpreters require argv & envp too */
 225         } else {
 226                 items += 1; /* ELF interpreters only put argc on the stack */
 227         }
 228         bprm->p = STACK_ROUND(sp, items);
 229
 230         /* Point sp at the lowest address on the stack */
 231 #ifdef CONFIG_STACK_GROWSUP
 232         sp = (elf_addr_t __user *)bprm->p - items - ei_index;
 233         bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
 234 #else
 235         sp = (elf_addr_t __user *)bprm->p;
 236 #endif
 237
 238
 239         /*
 240          * Grow the stack manually; some architectures have a limit on how
 241          * far ahead a user-space access may be in order to grow the stack.
 242          */
 243         vma = find_extend_vma(current->mm, bprm->p);
 244         if (!vma)
 245                 return -EFAULT;
 246
 247         /* Now, let's put argc (and argv, envp if appropriate) on the stack */
 248         if (__put_user(argc, sp++))
 249                 return -EFAULT;
 250         if (interp_aout) {
 251                 argv = sp + 2;
 252                 envp = argv + argc + 1;
 253                 if (__put_user((elf_addr_t)(unsigned long)argv, sp++) ||
 254                     __put_user((elf_addr_t)(unsigned long)envp, sp++))
 255                         return -EFAULT;
 256         } else {
 257                 argv = sp;
 258                 envp = argv + argc + 1;
 259         }
 260
 261         /* Populate argv and envp */
 262         p = current->mm->arg_end = current->mm->arg_start;
 263         while (argc-- > 0) {
 264                 size_t len;
 265                 if (__put_user((elf_addr_t)p, argv++))
 266                         return -EFAULT;
 267                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 268                 if (!len || len > MAX_ARG_STRLEN)
 269                         return 0;
 270                 p += len;
 271         }
 272         if (__put_user(0, argv))
 273                 return -EFAULT;
 274         current->mm->arg_end = current->mm->env_start = p;
 275         while (envc-- > 0) {
 276                 size_t len;
 277                 if (__put_user((elf_addr_t)p, envp++))
 278                         return -EFAULT;
 279                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 280                 if (!len || len > MAX_ARG_STRLEN)
 281                         return 0;
 282                 p += len;
 283         }
 284         if (__put_user(0, envp))
 285                 return -EFAULT;
 286         current->mm->env_end = p;
 287
 288         /* Put the elf_info on the stack in the right place.  */
 289         sp = (elf_addr_t __user *)envp + 1;
 290         if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
 291                 return -EFAULT;
 292         return 0;
 293 }
 294
 295 #ifndef elf_map
 296
 297 static unsigned long elf_map(struct file *filep, unsigned long addr,
 298                 struct elf_phdr *eppnt, int prot, int type)
 299 {
 300         unsigned long map_addr;
 301         unsigned long pageoffset = ELF_PAGEOFFSET(eppnt->p_vaddr);
 302
 303         down_write(&current->mm->mmap_sem);
 304         /* mmap() will return -EINVAL if given a zero size, but a
 305          * segment with zero filesize is perfectly valid */
 306         if (eppnt->p_filesz + pageoffset)
 307                 map_addr = do_mmap(filep, ELF_PAGESTART(addr),
 308                                    eppnt->p_filesz + pageoffset, prot, type,
 309                                    eppnt->p_offset - pageoffset);
 310         else
 311                 map_addr = ELF_PAGESTART(addr);
 312         up_write(&current->mm->mmap_sem);
 313         return(map_addr);
 314 }
 315
 316 #endif /* !elf_map */
 317
 318 /* This is much more generalized than the library routine read function,
 319    so we keep this separate.  Technically the library read function
 320    is only provided so that we can read a.out libraries that have
 321    an ELF header */
 322
 323 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
 324                 struct file *interpreter, unsigned long *interp_load_addr)
 325 {
 326         struct elf_phdr *elf_phdata;
 327         struct elf_phdr *eppnt;
 328         unsigned long load_addr = 0;
 329         int load_addr_set = 0;
 330         unsigned long last_bss = 0, elf_bss = 0;
 331         unsigned long error = ~0UL;
 332         int retval, i, size;
 333
 334         /* First of all, some simple consistency checks */
 335         if (interp_elf_ex->e_type != ET_EXEC &&
 336             interp_elf_ex->e_type != ET_DYN)
 337                 goto out;
 338         if (!elf_check_arch(interp_elf_ex))
 339                 goto out;
 340         if (!interpreter->f_op || !interpreter->f_op->mmap)
 341                 goto out;
 342
 343         /*
 344          * If the size of this structure has changed, then punt, since
 345          * we will be doing the wrong thing.
 346          */
 347         if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
 348                 goto out;
 349         if (interp_elf_ex->e_phnum < 1 ||
 350                 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
 351                 goto out;
 352
 353         /* Now read in all of the header information */
 354         size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
 355         if (size > ELF_MIN_ALIGN)
 356                 goto out;
 357         elf_phdata = kmalloc(size, GFP_KERNEL);
 358         if (!elf_phdata)
 359                 goto out;
 360
 361         retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
 362                              (char *)elf_phdata,size);
 363         error = -EIO;
 364         if (retval != size) {
 365                 if (retval < 0)
 366                         error = retval;
 367                 goto out_close;
 368         }
 369
 370         eppnt = elf_phdata;
 371         for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
 372                 if (eppnt->p_type == PT_LOAD) {
 373                         int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
 374                         int elf_prot = 0;
 375                         unsigned long vaddr = 0;
 376                         unsigned long k, map_addr;
 377
 378                         if (eppnt->p_flags & PF_R)
 379                                 elf_prot = PROT_READ;
 380                         if (eppnt->p_flags & PF_W)
 381                                 elf_prot |= PROT_WRITE;
 382                         if (eppnt->p_flags & PF_X)
 383                                 elf_prot |= PROT_EXEC;
 384                         vaddr = eppnt->p_vaddr;
 385                         if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
 386                                 elf_type |= MAP_FIXED;
 387
 388                         map_addr = elf_map(interpreter, load_addr + vaddr,
 389                                            eppnt, elf_prot, elf_type);
 390                         error = map_addr;
 391                         if (BAD_ADDR(map_addr))
 392                                 goto out_close;
 393
 394                         if (!load_addr_set &&
 395                             interp_elf_ex->e_type == ET_DYN) {
 396                                 load_addr = map_addr - ELF_PAGESTART(vaddr);
 397                                 load_addr_set = 1;
 398                         }
 399
 400                         /*
 401                          * Check to see if the section's size will overflow the
 402                          * allowed task size. Note that p_filesz must always be
 403                          * <= p_memsize so it's only necessary to check p_memsz.
 404                          */
 405                         k = load_addr + eppnt->p_vaddr;
 406                         if (BAD_ADDR(k) ||
 407                             eppnt->p_filesz > eppnt->p_memsz ||
 408                             eppnt->p_memsz > TASK_SIZE ||
 409                             TASK_SIZE - eppnt->p_memsz < k) {
 410                                 error = -ENOMEM;
 411                                 goto out_close;
 412                         }
 413
 414                         /*
 415                          * Find the end of the file mapping for this phdr, and
 416                          * keep track of the largest address we see for this.
 417                          */
 418                         k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
 419                         if (k > elf_bss)
 420                                 elf_bss = k;
 421
 422                         /*
 423                          * Do the same thing for the memory mapping - between
 424                          * elf_bss and last_bss is the bss section.
 425                          */
 426                         k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
 427                         if (k > last_bss)
 428                                 last_bss = k;
 429                 }
 430         }
 431
 432         /*
 433          * Now fill out the bss section.  First pad the last page up
 434          * to the page boundary, and then perform a mmap to make sure
 435          * that there are zero-mapped pages up to and including the
 436          * last bss page.
 437          */
 438         if (padzero(elf_bss)) {
 439                 error = -EFAULT;
 440                 goto out_close;
 441         }
 442
 443         /* What we have mapped so far */
 444         elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
 445
 446         /* Map the last of the bss segment */
 447         if (last_bss > elf_bss) {
 448                 down_write(&current->mm->mmap_sem);
 449                 error = do_brk(elf_bss, last_bss - elf_bss);
 450                 up_write(&current->mm->mmap_sem);
 451                 if (BAD_ADDR(error))
 452                         goto out_close;
 453         }
 454
 455         *interp_load_addr = load_addr;
 456         error = ((unsigned long)interp_elf_ex->e_entry) + load_addr;
 457
 458 out_close:
 459         kfree(elf_phdata);
 460 out:
 461         return error;
 462 }
 463
 464 static unsigned long load_aout_interp(struct exec *interp_ex,
 465                 struct file *interpreter)
 466 {
 467         unsigned long text_data, elf_entry = ~0UL;
 468         char __user * addr;
 469         loff_t offset;
 470
 471         current->mm->end_code = interp_ex->a_text;
 472         text_data = interp_ex->a_text + interp_ex->a_data;
 473         current->mm->end_data = text_data;
 474         current->mm->brk = interp_ex->a_bss + text_data;
 475
 476         switch (N_MAGIC(*interp_ex)) {
 477         case OMAGIC:
 478                 offset = 32;
 479                 addr = (char __user *)0;
 480                 break;
 481         case ZMAGIC:
 482         case QMAGIC:
 483                 offset = N_TXTOFF(*interp_ex);
 484                 addr = (char __user *)N_TXTADDR(*interp_ex);
 485                 break;
 486         default:
 487                 goto out;
 488         }
 489
 490         down_write(&current->mm->mmap_sem);
 491         do_brk(0, text_data);
 492         up_write(&current->mm->mmap_sem);
 493         if (!interpreter->f_op || !interpreter->f_op->read)
 494                 goto out;
 495         if (interpreter->f_op->read(interpreter, addr, text_data, &offset) < 0)
 496                 goto out;
 497         flush_icache_range((unsigned long)addr,
 498                            (unsigned long)addr + text_data);
 499
 500         down_write(&current->mm->mmap_sem);
 501         do_brk(ELF_PAGESTART(text_data + ELF_MIN_ALIGN - 1),
 502                 interp_ex->a_bss);
 503         up_write(&current->mm->mmap_sem);
 504         elf_entry = interp_ex->a_entry;
 505
 506 out:
 507         return elf_entry;
 508 }
 509
 510 /*
 511  * These are the functions used to load ELF style executables and shared
 512  * libraries.  There is no binary dependent code anywhere else.
 513  */
 514
 515 #define INTERPRETER_NONE 0
 516 #define INTERPRETER_AOUT 1
 517 #define INTERPRETER_ELF 2
 518
 519 #ifndef STACK_RND_MASK
 520 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))     /* 8MB of VA */
 521 #endif
 522
 523 static unsigned long randomize_stack_top(unsigned long stack_top)
 524 {
 525         unsigned int random_variable = 0;
 526
 527         if ((current->flags & PF_RANDOMIZE) &&
 528                 !(current->personality & ADDR_NO_RANDOMIZE)) {
 529                 random_variable = get_random_int() & STACK_RND_MASK;
 530                 random_variable <<= PAGE_SHIFT;
 531         }
 532 #ifdef CONFIG_STACK_GROWSUP
 533         return PAGE_ALIGN(stack_top) + random_variable;
 534 #else
 535         return PAGE_ALIGN(stack_top) - random_variable;
 536 #endif
 537 }
 538
 539 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
 540 {
 541         struct file *interpreter = NULL; /* to shut gcc up */
 542         unsigned long load_addr = 0, load_bias = 0;
 543         int load_addr_set = 0;
 544         char * elf_interpreter = NULL;
 545         unsigned int interpreter_type = INTERPRETER_NONE;
 546         unsigned char ibcs2_interpreter = 0;
 547         unsigned long error;
 548         struct elf_phdr *elf_ppnt, *elf_phdata;
 549         unsigned long elf_bss, elf_brk;
 550         int elf_exec_fileno;
 551         int retval, i;
 552         unsigned int size;
 553         unsigned long elf_entry, interp_load_addr = 0;
 554         unsigned long start_code, end_code, start_data, end_data;
 555         unsigned long reloc_func_desc = 0;
 556         char passed_fileno[6];
 557         struct files_struct *files;
 558         int executable_stack = EXSTACK_DEFAULT;
 559         unsigned long def_flags = 0;
 560         struct {
 561                 struct elfhdr elf_ex;
 562                 struct elfhdr interp_elf_ex;
 563                 struct exec interp_ex;
 564         } *loc;
 565
 566         loc = kmalloc(sizeof(*loc), GFP_KERNEL);
 567         if (!loc) {
 568                 retval = -ENOMEM;
 569                 goto out_ret;
 570         }
 571
 572         /* Get the exec-header */
 573         loc->elf_ex = *((struct elfhdr *)bprm->buf);
 574
 575         retval = -ENOEXEC;
 576         /* First of all, some simple consistency checks */
 577         if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 578                 goto out;
 579
 580         if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
 581                 goto out;
 582         if (!elf_check_arch(&loc->elf_ex))
 583                 goto out;
 584         if (!bprm->file->f_op||!bprm->file->f_op->mmap)
 585                 goto out;
 586
 587         /* Now read in all of the header information */
 588         if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
 589                 goto out;
 590         if (loc->elf_ex.e_phnum < 1 ||
 591                 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
 592                 goto out;
 593         size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
 594         retval = -ENOMEM;
 595         elf_phdata = kmalloc(size, GFP_KERNEL);
 596         if (!elf_phdata)
 597                 goto out;
 598
 599         retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
 600                              (char *)elf_phdata, size);
 601         if (retval != size) {
 602                 if (retval >= 0)
 603                         retval = -EIO;
 604                 goto out_free_ph;
 605         }
 606
 607         files = current->files; /* Refcounted so ok */
 608         retval = unshare_files();
 609         if (retval < 0)
 610                 goto out_free_ph;
 611         if (files == current->files) {
 612                 put_files_struct(files);
 613                 files = NULL;
 614         }
 615
 616         /* exec will make our files private anyway, but for the a.out
 617            loader stuff we need to do it earlier */
 618         retval = get_unused_fd();
 619         if (retval < 0)
 620                 goto out_free_fh;
 621         get_file(bprm->file);
 622         fd_install(elf_exec_fileno = retval, bprm->file);
 623
 624         elf_ppnt = elf_phdata;
 625         elf_bss = 0;
 626         elf_brk = 0;
 627
 628         start_code = ~0UL;
 629         end_code = 0;
 630         start_data = 0;
 631         end_data = 0;
 632
 633         for (i = 0; i < loc->elf_ex.e_phnum; i++) {
 634                 if (elf_ppnt->p_type == PT_INTERP) {
 635                         /* This is the program interpreter used for
 636                          * shared libraries - for now assume that this
 637                          * is an a.out format binary
 638                          */
 639                         retval = -ENOEXEC;
 640                         if (elf_ppnt->p_filesz > PATH_MAX ||
 641                             elf_ppnt->p_filesz < 2)
 642                                 goto out_free_file;
 643
 644                         retval = -ENOMEM;
 645                         elf_interpreter = kmalloc(elf_ppnt->p_filesz,
 646                                                   GFP_KERNEL);
 647                         if (!elf_interpreter)
 648                                 goto out_free_file;
 649
 650                         retval = kernel_read(bprm->file, elf_ppnt->p_offset,
 651                                              elf_interpreter,
 652                                              elf_ppnt->p_filesz);
 653                         if (retval != elf_ppnt->p_filesz) {
 654                                 if (retval >= 0)
 655                                         retval = -EIO;
 656                                 goto out_free_interp;
 657                         }
 658                         /* make sure path is NULL terminated */
 659                         retval = -ENOEXEC;
 660                         if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
 661                                 goto out_free_interp;
 662
 663                         /* If the program interpreter is one of these two,
 664                          * then assume an iBCS2 image. Otherwise assume
 665                          * a native linux image.
 666                          */
 667                         if (strcmp(elf_interpreter,"/usr/lib/libc.so.1") == 0 ||
 668                             strcmp(elf_interpreter,"/usr/lib/ld.so.1") == 0)
 669                                 ibcs2_interpreter = 1;
 670
 671                         /*
 672                          * The early SET_PERSONALITY here is so that the lookup
 673                          * for the interpreter happens in the namespace of the
 674                          * to-be-execed image.  SET_PERSONALITY can select an
 675                          * alternate root.
 676                          *
 677                          * However, SET_PERSONALITY is NOT allowed to switch
 678                          * this task into the new images's memory mapping
 679                          * policy - that is, TASK_SIZE must still evaluate to
 680                          * that which is appropriate to the execing application.
 681                          * This is because exit_mmap() needs to have TASK_SIZE
 682                          * evaluate to the size of the old image.
 683                          *
 684                          * So if (say) a 64-bit application is execing a 32-bit
 685                          * application it is the architecture's responsibility
 686                          * to defer changing the value of TASK_SIZE until the
 687                          * switch really is going to happen - do this in
 688                          * flush_thread().      - akpm
 689                          */
 690                         SET_PERSONALITY(loc->elf_ex, ibcs2_interpreter);
 691
 692                         interpreter = open_exec(elf_interpreter);
 693                         retval = PTR_ERR(interpreter);
 694                         if (IS_ERR(interpreter))
 695                                 goto out_free_interp;
 696
 697                         /*
 698                          * If the binary is not readable then enforce
 699                          * mm->dumpable = 0 regardless of the interpreter's
 700                          * permissions.
 701                          */
 702                         if (file_permission(interpreter, MAY_READ) < 0)
 703                                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
 704
 705                         retval = kernel_read(interpreter, 0, bprm->buf,
 706                                              BINPRM_BUF_SIZE);
 707                         if (retval != BINPRM_BUF_SIZE) {
 708                                 if (retval >= 0)
 709                                         retval = -EIO;
 710                                 goto out_free_dentry;
 711                         }
 712
 713                         /* Get the exec headers */
 714                         loc->interp_ex = *((struct exec *)bprm->buf);
 715                         loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
 716                         break;
 717                 }
 718                 elf_ppnt++;
 719         }
 720
 721         elf_ppnt = elf_phdata;
 722         for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
 723                 if (elf_ppnt->p_type == PT_GNU_STACK) {
 724                         if (elf_ppnt->p_flags & PF_X)
 725                                 executable_stack = EXSTACK_ENABLE_X;
 726                         else
 727                                 executable_stack = EXSTACK_DISABLE_X;
 728                         break;
 729                 }
 730
 731         /* Some simple consistency checks for the interpreter */
 732         if (elf_interpreter) {
 733                 interpreter_type = INTERPRETER_ELF | INTERPRETER_AOUT;
 734
 735                 /* Now figure out which format our binary is */
 736                 if ((N_MAGIC(loc->interp_ex) != OMAGIC) &&
 737                     (N_MAGIC(loc->interp_ex) != ZMAGIC) &&
 738                     (N_MAGIC(loc->interp_ex) != QMAGIC))
 739                         interpreter_type = INTERPRETER_ELF;
 740
 741                 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 742                         interpreter_type &= ~INTERPRETER_ELF;
 743
 744                 retval = -ELIBBAD;
 745                 if (!interpreter_type)
 746                         goto out_free_dentry;
 747
 748                 /* Make sure only one type was selected */
 749                 if ((interpreter_type & INTERPRETER_ELF) &&
 750                      interpreter_type != INTERPRETER_ELF) {
 751                         // FIXME - ratelimit this before re-enabling
 752                         // printk(KERN_WARNING "ELF: Ambiguous type, using ELF\n");
 753                         interpreter_type = INTERPRETER_ELF;
 754                 }
 755                 /* Verify the interpreter has a valid arch */
 756                 if ((interpreter_type == INTERPRETER_ELF) &&
 757                     !elf_check_arch(&loc->interp_elf_ex))
 758                         goto out_free_dentry;
 759         } else {
 760                 /* Executables without an interpreter also need a personality  */
 761                 SET_PERSONALITY(loc->elf_ex, ibcs2_interpreter);
 762         }
 763
 764         /* OK, we are done with that, now set up the arg stuff,
 765            and then start this sucker up */
 766         if ((!bprm->sh_bang) && (interpreter_type == INTERPRETER_AOUT)) {
 767                 char *passed_p = passed_fileno;
 768                 sprintf(passed_fileno, "%d", elf_exec_fileno);
 769
 770                 if (elf_interpreter) {
 771                         retval = copy_strings_kernel(1, &passed_p, bprm);
 772                         if (retval)
 773                                 goto out_free_dentry;
 774                         bprm->argc++;
 775                 }
 776         }
 777
 778         /* Flush all traces of the currently running executable */
 779         retval = flush_old_exec(bprm);
 780         if (retval)
 781                 goto out_free_dentry;
 782
 783         /* Discard our unneeded old files struct */
 784         if (files) {
 785                 put_files_struct(files);
 786                 files = NULL;
 787         }
 788
 789         /* OK, This is the point of no return */
 790         current->flags &= ~PF_FORKNOEXEC;
 791         current->mm->def_flags = def_flags;
 792
 793         /* Do this immediately, since STACK_TOP as used in setup_arg_pages
 794            may depend on the personality.  */
 795         SET_PERSONALITY(loc->elf_ex, ibcs2_interpreter);
 796         if (elf_read_implies_exec(loc->elf_ex, executable_stack))
 797                 current->personality |= READ_IMPLIES_EXEC;
 798
 799         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 800                 current->flags |= PF_RANDOMIZE;
 801         arch_pick_mmap_layout(current->mm);
 802
 803         /* Do this so that we can load the interpreter, if need be.  We will
 804            change some of these later */
 805         current->mm->free_area_cache = current->mm->mmap_base;
 806         current->mm->cached_hole_size = 0;
 807         retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
 808                                  executable_stack);
 809         if (retval < 0) {
 810                 send_sig(SIGKILL, current, 0);
 811                 goto out_free_dentry;
 812         }
 813
 814         current->mm->start_stack = bprm->p;
 815
 816         /* Now we do a little grungy work by mmaping the ELF image into
 817            the correct location in memory.  At this point, we assume that
 818            the image should be loaded at fixed address, not at a variable
 819            address. */
 820         for(i = 0, elf_ppnt = elf_phdata;
 821             i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
 822                 int elf_prot = 0, elf_flags;
 823                 unsigned long k, vaddr;
 824
 825                 if (elf_ppnt->p_type != PT_LOAD)
 826                         continue;
 827
 828                 if (unlikely (elf_brk > elf_bss)) {
 829                         unsigned long nbyte;
 830
 831                         /* There was a PT_LOAD segment with p_memsz > p_filesz
 832                            before this one. Map anonymous pages, if needed,
 833                            and clear the area.  */
 834                         retval = set_brk (elf_bss + load_bias,
 835                                           elf_brk + load_bias);
 836                         if (retval) {
 837                                 send_sig(SIGKILL, current, 0);
 838                                 goto out_free_dentry;
 839                         }
 840                         nbyte = ELF_PAGEOFFSET(elf_bss);
 841                         if (nbyte) {
 842                                 nbyte = ELF_MIN_ALIGN - nbyte;
 843                                 if (nbyte > elf_brk - elf_bss)
 844                                         nbyte = elf_brk - elf_bss;
 845                                 if (clear_user((void __user *)elf_bss +
 846                                                         load_bias, nbyte)) {
 847                                         /*
 848                                          * This bss-zeroing can fail if the ELF
 849                                          * file specifies odd protections. So
 850                                          * we don't check the return value
 851                                          */
 852                                 }
 853                         }
 854                 }
 855
 856                 if (elf_ppnt->p_flags & PF_R)
 857                         elf_prot |= PROT_READ;
 858                 if (elf_ppnt->p_flags & PF_W)
 859                         elf_prot |= PROT_WRITE;
 860                 if (elf_ppnt->p_flags & PF_X)
 861                         elf_prot |= PROT_EXEC;
 862
 863                 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
 864
 865                 vaddr = elf_ppnt->p_vaddr;
 866                 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
 867                         elf_flags |= MAP_FIXED;
 868                 } else if (loc->elf_ex.e_type == ET_DYN) {
 869                         /* Try and get dynamic programs out of the way of the
 870                          * default mmap base, as well as whatever program they
 871                          * might try to exec.  This is because the brk will
 872                          * follow the loader, and is not movable.  */
 873                         load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 874                 }
 875
 876                 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
 877                                 elf_prot, elf_flags);
 878                 if (BAD_ADDR(error)) {
 879                         send_sig(SIGKILL, current, 0);
 880                         retval = IS_ERR((void *)error) ?
 881                                 PTR_ERR((void*)error) : -EINVAL;
 882                         goto out_free_dentry;
 883                 }
 884
 885                 if (!load_addr_set) {
 886                         load_addr_set = 1;
 887                         load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
 888                         if (loc->elf_ex.e_type == ET_DYN) {
 889                                 load_bias += error -
 890                                              ELF_PAGESTART(load_bias + vaddr);
 891                                 load_addr += load_bias;
 892                                 reloc_func_desc = load_bias;
 893                         }
 894                 }
 895                 k = elf_ppnt->p_vaddr;
 896                 if (k < start_code)
 897                         start_code = k;
 898                 if (start_data < k)
 899                         start_data = k;
 900
 901                 /*
 902                  * Check to see if the section's size will overflow the
 903                  * allowed task size. Note that p_filesz must always be
 904                  * <= p_memsz so it is only necessary to check p_memsz.
 905                  */
 906                 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
 907                     elf_ppnt->p_memsz > TASK_SIZE ||
 908                     TASK_SIZE - elf_ppnt->p_memsz < k) {
 909                         /* set_brk can never work. Avoid overflows. */
 910                         send_sig(SIGKILL, current, 0);
 911                         retval = -EINVAL;
 912                         goto out_free_dentry;
 913                 }
 914
 915                 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
 916
 917                 if (k > elf_bss)
 918                         elf_bss = k;
 919                 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
 920                         end_code = k;
 921                 if (end_data < k)
 922                         end_data = k;
 923                 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
 924                 if (k > elf_brk)
 925                         elf_brk = k;
 926         }
 927
 928         loc->elf_ex.e_entry += load_bias;
 929         elf_bss += load_bias;
 930         elf_brk += load_bias;
 931         start_code += load_bias;
 932         end_code += load_bias;
 933         start_data += load_bias;
 934         end_data += load_bias;
 935
 936         /* Calling set_brk effectively mmaps the pages that we need
 937          * for the bss and break sections.  We must do this before
 938          * mapping in the interpreter, to make sure it doesn't wind
 939          * up getting placed where the bss needs to go.
 940          */
 941         retval = set_brk(elf_bss, elf_brk);
 942         if (retval) {
 943                 send_sig(SIGKILL, current, 0);
 944                 goto out_free_dentry;
 945         }
 946         if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
 947                 send_sig(SIGSEGV, current, 0);
 948                 retval = -EFAULT; /* Nobody gets to see this, but.. */
 949                 goto out_free_dentry;
 950         }
 951
 952         if (elf_interpreter) {
 953                 if (interpreter_type == INTERPRETER_AOUT)
 954                         elf_entry = load_aout_interp(&loc->interp_ex,
 955                                                      interpreter);
 956                 else
 957                         elf_entry = load_elf_interp(&loc->interp_elf_ex,
 958                                                     interpreter,
 959                                                     &interp_load_addr);
 960                 if (BAD_ADDR(elf_entry)) {
 961                         force_sig(SIGSEGV, current);
 962                         retval = IS_ERR((void *)elf_entry) ?
 963                                         (int)elf_entry : -EINVAL;
 964                         goto out_free_dentry;
 965                 }
 966                 reloc_func_desc = interp_load_addr;
 967
 968                 allow_write_access(interpreter);
 969                 fput(interpreter);
 970                 kfree(elf_interpreter);
 971         } else {
 972                 elf_entry = loc->elf_ex.e_entry;
 973                 if (BAD_ADDR(elf_entry)) {
 974                         force_sig(SIGSEGV, current);
 975                         retval = -EINVAL;
 976                         goto out_free_dentry;
 977                 }
 978         }
 979
 980         kfree(elf_phdata);
 981
 982         if (interpreter_type != INTERPRETER_AOUT)
 983                 sys_close(elf_exec_fileno);
 984
 985         set_binfmt(&elf_format);
 986
 987 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
 988         retval = arch_setup_additional_pages(bprm, executable_stack);
 989         if (retval < 0) {
 990                 send_sig(SIGKILL, current, 0);
 991                 goto out;
 992         }
 993 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
 994
 995         compute_creds(bprm);
 996         current->flags &= ~PF_FORKNOEXEC;
 997         retval = create_elf_tables(bprm, &loc->elf_ex,
 998                           (interpreter_type == INTERPRETER_AOUT),
 999                           load_addr, interp_load_addr);
1000         if (retval < 0) {
1001                 send_sig(SIGKILL, current, 0);
1002                 goto out;
1003         }
1004         /* N.B. passed_fileno might not be initialized? */
1005         if (interpreter_type == INTERPRETER_AOUT)
1006                 current->mm->arg_start += strlen(passed_fileno) + 1;
1007         current->mm->end_code = end_code;
1008         current->mm->start_code = start_code;
1009         current->mm->start_data = start_data;
1010         current->mm->end_data = end_data;
1011         current->mm->start_stack = bprm->p;
1012
1013         if (current->personality & MMAP_PAGE_ZERO) {
1014                 /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
1015                    and some applications "depend" upon this behavior.
1016                    Since we do not have the power to recompile these, we
1017                    emulate the SVr4 behavior. Sigh. */
1018                 down_write(&current->mm->mmap_sem);
1019                 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1020                                 MAP_FIXED | MAP_PRIVATE, 0);
1021                 up_write(&current->mm->mmap_sem);
1022         }
1023
1024 #ifdef ELF_PLAT_INIT
1025         /*
1026          * The ABI may specify that certain registers be set up in special
1027          * ways (on i386 %edx is the address of a DT_FINI function, for
1028          * example.  In addition, it may also specify (eg, PowerPC64 ELF)
1029          * that the e_entry field is the address of the function descriptor
1030          * for the startup routine, rather than the address of the startup
1031          * routine itself.  This macro performs whatever initialization to
1032          * the regs structure is required as well as any relocations to the
1033          * function descriptor entries when executing dynamically links apps.
1034          */
1035         ELF_PLAT_INIT(regs, reloc_func_desc);
1036 #endif
1037
1038         start_thread(regs, elf_entry, bprm->p);
1039         if (unlikely(current->ptrace & PT_PTRACED)) {
1040                 if (current->ptrace & PT_TRACE_EXEC)
1041                         ptrace_notify ((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
1042                 else
1043                         send_sig(SIGTRAP, current, 0);
1044         }
1045         retval = 0;
1046 out:
1047         kfree(loc);
1048 out_ret:
1049         return retval;
1050
1051         /* error cleanup */
1052 out_free_dentry:
1053         allow_write_access(interpreter);
1054         if (interpreter)
1055                 fput(interpreter);
1056 out_free_interp:
1057         kfree(elf_interpreter);
1058 out_free_file:
1059         sys_close(elf_exec_fileno);
1060 out_free_fh:
1061         if (files)
1062                 reset_files_struct(current, files);
1063 out_free_ph:
1064         kfree(elf_phdata);
1065         goto out;
1066 }
1067
1068 /* This is really simpleminded and specialized - we are loading an
1069    a.out library that is given an ELF header. */
1070 static int load_elf_library(struct file *file)
1071 {
1072         struct elf_phdr *elf_phdata;
1073         struct elf_phdr *eppnt;
1074         unsigned long elf_bss, bss, len;
1075         int retval, error, i, j;
1076         struct elfhdr elf_ex;
1077
1078         error = -ENOEXEC;
1079         retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1080         if (retval != sizeof(elf_ex))
1081                 goto out;
1082
1083         if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1084                 goto out;
1085
1086         /* First of all, some simple consistency checks */
1087         if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1088             !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1089                 goto out;
1090
1091         /* Now read in all of the header information */
1092
1093         j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1094         /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1095
1096         error = -ENOMEM;
1097         elf_phdata = kmalloc(j, GFP_KERNEL);
1098         if (!elf_phdata)
1099                 goto out;
1100
1101         eppnt = elf_phdata;
1102         error = -ENOEXEC;
1103         retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1104         if (retval != j)
1105                 goto out_free_ph;
1106
1107         for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1108                 if ((eppnt + i)->p_type == PT_LOAD)
1109                         j++;
1110         if (j != 1)
1111                 goto out_free_ph;
1112
1113         while (eppnt->p_type != PT_LOAD)
1114                 eppnt++;
1115
1116         /* Now use mmap to map the library into memory. */
1117         down_write(&current->mm->mmap_sem);
1118         error = do_mmap(file,
1119                         ELF_PAGESTART(eppnt->p_vaddr),
1120                         (eppnt->p_filesz +
1121                          ELF_PAGEOFFSET(eppnt->p_vaddr)),
1122                         PROT_READ | PROT_WRITE | PROT_EXEC,
1123                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1124                         (eppnt->p_offset -
1125                          ELF_PAGEOFFSET(eppnt->p_vaddr)));
1126         up_write(&current->mm->mmap_sem);
1127         if (error != ELF_PAGESTART(eppnt->p_vaddr))
1128                 goto out_free_ph;
1129
1130         elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1131         if (padzero(elf_bss)) {
1132                 error = -EFAULT;
1133                 goto out_free_ph;
1134         }
1135
1136         len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1137                             ELF_MIN_ALIGN - 1);
1138         bss = eppnt->p_memsz + eppnt->p_vaddr;
1139         if (bss > len) {
1140                 down_write(&current->mm->mmap_sem);
1141                 do_brk(len, bss - len);
1142                 up_write(&current->mm->mmap_sem);
1143         }
1144         error = 0;
1145
1146 out_free_ph:
1147         kfree(elf_phdata);
1148 out:
1149         return error;
1150 }
1151
1152 /*
1153  * Note that some platforms still use traditional core dumps and not
1154  * the ELF core dump.  Each platform can select it as appropriate.
1155  */
1156 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1157
1158 /*
1159  * ELF core dumper
1160  *
1161  * Modelled on fs/exec.c:aout_core_dump()
1162  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1163  */
1164 /*
1165  * These are the only things you should do on a core-file: use only these
1166  * functions to write out all the necessary info.
1167  */
1168 static int dump_write(struct file *file, const void *addr, int nr)
1169 {
1170         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1171 }
1172
1173 static int dump_seek(struct file *file, loff_t off)
1174 {
1175         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
1176                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
1177                         return 0;
1178         } else {
1179                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
1180                 if (!buf)
1181                         return 0;
1182                 while (off > 0) {
1183                         unsigned long n = off;
1184                         if (n > PAGE_SIZE)
1185                                 n = PAGE_SIZE;
1186                         if (!dump_write(file, buf, n))
1187                                 return 0;
1188                         off -= n;
1189                 }
1190                 free_page((unsigned long)buf);
1191         }
1192         return 1;
1193 }
1194
1195 /*
1196  * Decide whether a segment is worth dumping; default is yes to be
1197  * sure (missing info is worse than too much; etc).
1198  * Personally I'd include everything, and use the coredump limit...
1199  *
1200  * I think we should skip something. But I am not sure how. H.J.
1201  */
1202 static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
1203 {
1204         /* The vma can be set up to tell us the answer directly.  */
1205         if (vma->vm_flags & VM_ALWAYSDUMP)
1206                 return 1;
1207
1208         /* Do not dump I/O mapped devices or special mappings */
1209         if (vma->vm_flags & (VM_IO | VM_RESERVED))
1210                 return 0;
1211
1212         /* By default, dump shared memory if mapped from an anonymous file. */
1213         if (vma->vm_flags & VM_SHARED) {
1214                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0)
1215                         return test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
1216                 else
1217                         return test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
1218         }
1219
1220         /* By default, if it hasn't been written to, don't write it out. */
1221         if (!vma->anon_vma)
1222                 return test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
1223
1224         return test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
1225 }
1226
1227 /* An ELF note in memory */
1228 struct memelfnote
1229 {
1230         const char *name;
1231         int type;
1232         unsigned int datasz;
1233         void *data;
1234 };
1235
1236 static int notesize(struct memelfnote *en)
1237 {
1238         int sz;
1239
1240         sz = sizeof(struct elf_note);
1241         sz += roundup(strlen(en->name) + 1, 4);
1242         sz += roundup(en->datasz, 4);
1243
1244         return sz;
1245 }
1246
1247 #define DUMP_WRITE(addr, nr, foffset)   \
1248         do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1249
1250 static int alignfile(struct file *file, loff_t *foffset)
1251 {
1252         static const char buf[4] = { 0, };
1253         DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1254         return 1;
1255 }
1256
1257 static int writenote(struct memelfnote *men, struct file *file,
1258                         loff_t *foffset)
1259 {
1260         struct elf_note en;
1261         en.n_namesz = strlen(men->name) + 1;
1262         en.n_descsz = men->datasz;
1263         en.n_type = men->type;
1264
1265         DUMP_WRITE(&en, sizeof(en), foffset);
1266         DUMP_WRITE(men->name, en.n_namesz, foffset);
1267         if (!alignfile(file, foffset))
1268                 return 0;
1269         DUMP_WRITE(men->data, men->datasz, foffset);
1270         if (!alignfile(file, foffset))
1271                 return 0;
1272
1273         return 1;
1274 }
1275 #undef DUMP_WRITE
1276
1277 #define DUMP_WRITE(addr, nr)    \
1278         if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1279                 goto end_coredump;
1280 #define DUMP_SEEK(off)  \
1281         if (!dump_seek(file, (off))) \
1282                 goto end_coredump;
1283
1284 static void fill_elf_header(struct elfhdr *elf, int segs)
1285 {
1286         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1287         elf->e_ident[EI_CLASS] = ELF_CLASS;
1288         elf->e_ident[EI_DATA] = ELF_DATA;
1289         elf->e_ident[EI_VERSION] = EV_CURRENT;
1290         elf->e_ident[EI_OSABI] = ELF_OSABI;
1291         memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1292
1293         elf->e_type = ET_CORE;
1294         elf->e_machine = ELF_ARCH;
1295         elf->e_version = EV_CURRENT;
1296         elf->e_entry = 0;
1297         elf->e_phoff = sizeof(struct elfhdr);
1298         elf->e_shoff = 0;
1299         elf->e_flags = ELF_CORE_EFLAGS;
1300         elf->e_ehsize = sizeof(struct elfhdr);
1301         elf->e_phentsize = sizeof(struct elf_phdr);
1302         elf->e_phnum = segs;
1303         elf->e_shentsize = 0;
1304         elf->e_shnum = 0;
1305         elf->e_shstrndx = 0;
1306         return;
1307 }
1308
1309 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1310 {
1311         phdr->p_type = PT_NOTE;
1312         phdr->p_offset = offset;
1313         phdr->p_vaddr = 0;
1314         phdr->p_paddr = 0;
1315         phdr->p_filesz = sz;
1316         phdr->p_memsz = 0;
1317         phdr->p_flags = 0;
1318         phdr->p_align = 0;
1319         return;
1320 }
1321
1322 static void fill_note(struct memelfnote *note, const char *name, int type,
1323                 unsigned int sz, void *data)
1324 {
1325         note->name = name;
1326         note->type = type;
1327         note->datasz = sz;
1328         note->data = data;
1329         return;
1330 }
1331
1332 /*
1333  * fill up all the fields in prstatus from the given task struct, except
1334  * registers which need to be filled up separately.
1335  */
1336 static void fill_prstatus(struct elf_prstatus *prstatus,
1337                 struct task_struct *p, long signr)
1338 {
1339         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1340         prstatus->pr_sigpend = p->pending.signal.sig[0];
1341         prstatus->pr_sighold = p->blocked.sig[0];
1342         prstatus->pr_pid = p->pid;
1343         prstatus->pr_ppid = p->parent->pid;
1344         prstatus->pr_pgrp = process_group(p);
1345         prstatus->pr_sid = process_session(p);
1346         if (thread_group_leader(p)) {
1347                 /*
1348                  * This is the record for the group leader.  Add in the
1349                  * cumulative times of previous dead threads.  This total
1350                  * won't include the time of each live thread whose state
1351                  * is included in the core dump.  The final total reported
1352                  * to our parent process when it calls wait4 will include
1353                  * those sums as well as the little bit more time it takes
1354                  * this and each other thread to finish dying after the
1355                  * core dump synchronization phase.
1356                  */
1357                 cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1358                                    &prstatus->pr_utime);
1359                 cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1360                                    &prstatus->pr_stime);
1361         } else {
1362                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1363                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1364         }
1365         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1366         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1367 }
1368
1369 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1370                        struct mm_struct *mm)
1371 {
1372         unsigned int i, len;
1373
1374         /* first copy the parameters from user space */
1375         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1376
1377         len = mm->arg_end - mm->arg_start;
1378         if (len >= ELF_PRARGSZ)
1379                 len = ELF_PRARGSZ-1;
1380         if (copy_from_user(&psinfo->pr_psargs,
1381                            (const char __user *)mm->arg_start, len))
1382                 return -EFAULT;
1383         for(i = 0; i < len; i++)
1384                 if (psinfo->pr_psargs[i] == 0)
1385                         psinfo->pr_psargs[i] = ' ';
1386         psinfo->pr_psargs[len] = 0;
1387
1388         psinfo->pr_pid = p->pid;
1389         psinfo->pr_ppid = p->parent->pid;
1390         psinfo->pr_pgrp = process_group(p);
1391         psinfo->pr_sid = process_session(p);
1392
1393         i = p->state ? ffz(~p->state) + 1 : 0;
1394         psinfo->pr_state = i;
1395         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1396         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1397         psinfo->pr_nice = task_nice(p);
1398         psinfo->pr_flag = p->flags;
1399         SET_UID(psinfo->pr_uid, p->uid);
1400         SET_GID(psinfo->pr_gid, p->gid);
1401         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1402
1403         return 0;
1404 }
1405
1406 /* Here is the structure in which status of each thread is captured. */
1407 struct elf_thread_status
1408 {
1409         struct list_head list;
1410         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1411         elf_fpregset_t fpu;             /* NT_PRFPREG */
1412         struct task_struct *thread;
1413 #ifdef ELF_CORE_COPY_XFPREGS
1414         elf_fpxregset_t xfpu;           /* NT_PRXFPREG */
1415 #endif
1416         struct memelfnote notes[3];
1417         int num_notes;
1418 };
1419
1420 /*
1421  * In order to add the specific thread information for the elf file format,
1422  * we need to keep a linked list of every threads pr_status and then create
1423  * a single section for them in the final core file.
1424  */
1425 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1426 {
1427         int sz = 0;
1428         struct task_struct *p = t->thread;
1429         t->num_notes = 0;
1430
1431         fill_prstatus(&t->prstatus, p, signr);
1432         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1433
1434         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1435                   &(t->prstatus));
1436         t->num_notes++;
1437         sz += notesize(&t->notes[0]);
1438
1439         if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1440                                                                 &t->fpu))) {
1441                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1442                           &(t->fpu));
1443                 t->num_notes++;
1444                 sz += notesize(&t->notes[1]);
1445         }
1446
1447 #ifdef ELF_CORE_COPY_XFPREGS
1448         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1449                 fill_note(&t->notes[2], "LINUX", NT_PRXFPREG, sizeof(t->xfpu),
1450                           &t->xfpu);
1451                 t->num_notes++;
1452                 sz += notesize(&t->notes[2]);
1453         }
1454 #endif
1455         return sz;
1456 }
1457
1458 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1459                                         struct vm_area_struct *gate_vma)
1460 {
1461         struct vm_area_struct *ret = tsk->mm->mmap;
1462
1463         if (ret)
1464                 return ret;
1465         return gate_vma;
1466 }
1467 /*
1468  * Helper function for iterating across a vma list.  It ensures that the caller
1469  * will visit `gate_vma' prior to terminating the search.
1470  */
1471 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1472                                         struct vm_area_struct *gate_vma)
1473 {
1474         struct vm_area_struct *ret;
1475
1476         ret = this_vma->vm_next;
1477         if (ret)
1478                 return ret;
1479         if (this_vma == gate_vma)
1480                 return NULL;
1481         return gate_vma;
1482 }
1483
1484 /*
1485  * Actual dumper
1486  *
1487  * This is a two-pass process; first we find the offsets of the bits,
1488  * and then they are actually written out.  If we run out of core limit
1489  * we just truncate.
1490  */
1491 static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file)
1492 {
1493 #define NUM_NOTES       6
1494         int has_dumped = 0;
1495         mm_segment_t fs;
1496         int segs;
1497         size_t size = 0;
1498         int i;
1499         struct vm_area_struct *vma, *gate_vma;
1500         struct elfhdr *elf = NULL;
1501         loff_t offset = 0, dataoff, foffset;
1502         unsigned long limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1503         int numnote;
1504         struct memelfnote *notes = NULL;
1505         struct elf_prstatus *prstatus = NULL;   /* NT_PRSTATUS */
1506         struct elf_prpsinfo *psinfo = NULL;     /* NT_PRPSINFO */
1507         struct task_struct *g, *p;
1508         LIST_HEAD(thread_list);
1509         struct list_head *t;
1510         elf_fpregset_t *fpu = NULL;
1511 #ifdef ELF_CORE_COPY_XFPREGS
1512         elf_fpxregset_t *xfpu = NULL;
1513 #endif
1514         int thread_status_size = 0;
1515         elf_addr_t *auxv;
1516         unsigned long mm_flags;
1517
1518         /*
1519          * We no longer stop all VM operations.
1520          *
1521          * This is because those proceses that could possibly change map_count
1522          * or the mmap / vma pages are now blocked in do_exit on current
1523          * finishing this core dump.
1524          *
1525          * Only ptrace can touch these memory addresses, but it doesn't change
1526          * the map_count or the pages allocated. So no possibility of crashing
1527          * exists while dumping the mm->vm_next areas to the core file.
1528          */
1529
1530         /* alloc memory for large data structures: too large to be on stack */
1531         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1532         if (!elf)
1533                 goto cleanup;
1534         prstatus = kmalloc(sizeof(*prstatus), GFP_KERNEL);
1535         if (!prstatus)
1536                 goto cleanup;
1537         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1538         if (!psinfo)
1539                 goto cleanup;
1540         notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1541         if (!notes)
1542                 goto cleanup;
1543         fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1544         if (!fpu)
1545                 goto cleanup;
1546 #ifdef ELF_CORE_COPY_XFPREGS
1547         xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1548         if (!xfpu)
1549                 goto cleanup;
1550 #endif
1551
1552         if (signr) {
1553                 struct elf_thread_status *tmp;
1554                 rcu_read_lock();
1555                 do_each_thread(g,p)
1556                         if (current->mm == p->mm && current != p) {
1557                                 tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC);
1558                                 if (!tmp) {
1559                                         rcu_read_unlock();
1560                                         goto cleanup;
1561                                 }
1562                                 tmp->thread = p;
1563                                 list_add(&tmp->list, &thread_list);
1564                         }
1565                 while_each_thread(g,p);
1566                 rcu_read_unlock();
1567                 list_for_each(t, &thread_list) {
1568                         struct elf_thread_status *tmp;
1569                         int sz;
1570
1571                         tmp = list_entry(t, struct elf_thread_status, list);
1572                         sz = elf_dump_thread_status(signr, tmp);
1573                         thread_status_size += sz;
1574                 }
1575         }
1576         /* now collect the dump for the current */
1577         memset(prstatus, 0, sizeof(*prstatus));
1578         fill_prstatus(prstatus, current, signr);
1579         elf_core_copy_regs(&prstatus->pr_reg, regs);
1580
1581         segs = current->mm->map_count;
1582 #ifdef ELF_CORE_EXTRA_PHDRS
1583         segs += ELF_CORE_EXTRA_PHDRS;
1584 #endif
1585
1586         gate_vma = get_gate_vma(current);
1587         if (gate_vma != NULL)
1588                 segs++;
1589
1590         /* Set up header */
1591         fill_elf_header(elf, segs + 1); /* including notes section */
1592
1593         has_dumped = 1;
1594         current->flags |= PF_DUMPCORE;
1595
1596         /*
1597          * Set up the notes in similar form to SVR4 core dumps made
1598          * with info from their /proc.
1599          */
1600
1601         fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1602         fill_psinfo(psinfo, current->group_leader, current->mm);
1603         fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1604
1605         numnote = 2;
1606
1607         auxv = (elf_addr_t *)current->mm->saved_auxv;
1608
1609         i = 0;
1610         do
1611                 i += 2;
1612         while (auxv[i - 2] != AT_NULL);
1613         fill_note(&notes[numnote++], "CORE", NT_AUXV,
1614                   i * sizeof(elf_addr_t), auxv);
1615
1616         /* Try to dump the FPU. */
1617         if ((prstatus->pr_fpvalid =
1618              elf_core_copy_task_fpregs(current, regs, fpu)))
1619                 fill_note(notes + numnote++,
1620                           "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1621 #ifdef ELF_CORE_COPY_XFPREGS
1622         if (elf_core_copy_task_xfpregs(current, xfpu))
1623                 fill_note(notes + numnote++,
1624                           "LINUX", NT_PRXFPREG, sizeof(*xfpu), xfpu);
1625 #endif
1626
1627         fs = get_fs();
1628         set_fs(KERNEL_DS);
1629
1630         DUMP_WRITE(elf, sizeof(*elf));
1631         offset += sizeof(*elf);                         /* Elf header */
1632         offset += (segs + 1) * sizeof(struct elf_phdr); /* Program headers */
1633         foffset = offset;
1634
1635         /* Write notes phdr entry */
1636         {
1637                 struct elf_phdr phdr;
1638                 int sz = 0;
1639
1640                 for (i = 0; i < numnote; i++)
1641                         sz += notesize(notes + i);
1642
1643                 sz += thread_status_size;
1644
1645                 sz += elf_coredump_extra_notes_size();
1646
1647                 fill_elf_note_phdr(&phdr, sz, offset);
1648                 offset += sz;
1649                 DUMP_WRITE(&phdr, sizeof(phdr));
1650         }
1651
1652         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1653
1654         /*
1655          * We must use the same mm->flags while dumping core to avoid
1656          * inconsistency between the program headers and bodies, otherwise an
1657          * unusable core file can be generated.
1658          */
1659         mm_flags = current->mm->flags;
1660
1661         /* Write program headers for segments dump */
1662         for (vma = first_vma(current, gate_vma); vma != NULL;
1663                         vma = next_vma(vma, gate_vma)) {
1664                 struct elf_phdr phdr;
1665                 size_t sz;
1666
1667                 sz = vma->vm_end - vma->vm_start;
1668
1669                 phdr.p_type = PT_LOAD;
1670                 phdr.p_offset = offset;
1671                 phdr.p_vaddr = vma->vm_start;
1672                 phdr.p_paddr = 0;
1673                 phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
1674                 phdr.p_memsz = sz;
1675                 offset += phdr.p_filesz;
1676                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1677                 if (vma->vm_flags & VM_WRITE)
1678                         phdr.p_flags |= PF_W;
1679                 if (vma->vm_flags & VM_EXEC)
1680                         phdr.p_flags |= PF_X;
1681                 phdr.p_align = ELF_EXEC_PAGESIZE;
1682
1683                 DUMP_WRITE(&phdr, sizeof(phdr));
1684         }
1685
1686 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1687         ELF_CORE_WRITE_EXTRA_PHDRS;
1688 #endif
1689
1690         /* write out the notes section */
1691         for (i = 0; i < numnote; i++)
1692                 if (!writenote(notes + i, file, &foffset))
1693                         goto end_coredump;
1694
1695         if (elf_coredump_extra_notes_write(file, &foffset))
1696                 goto end_coredump;
1697
1698         /* write out the thread status notes section */
1699         list_for_each(t, &thread_list) {
1700                 struct elf_thread_status *tmp =
1701                                 list_entry(t, struct elf_thread_status, list);
1702
1703                 for (i = 0; i < tmp->num_notes; i++)
1704                         if (!writenote(&tmp->notes[i], file, &foffset))
1705                                 goto end_coredump;
1706         }
1707
1708         /* Align to page */
1709         DUMP_SEEK(dataoff - foffset);
1710
1711         for (vma = first_vma(current, gate_vma); vma != NULL;
1712                         vma = next_vma(vma, gate_vma)) {
1713                 unsigned long addr;
1714
1715                 if (!maydump(vma, mm_flags))
1716                         continue;
1717
1718                 for (addr = vma->vm_start;
1719                      addr < vma->vm_end;
1720                      addr += PAGE_SIZE) {
1721                         struct page *page;
1722                         struct vm_area_struct *vma;
1723
1724                         if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1725                                                 &page, &vma) <= 0) {
1726                                 DUMP_SEEK(PAGE_SIZE);
1727                         } else {
1728                                 if (page == ZERO_PAGE(addr)) {
1729                                         if (!dump_seek(file, PAGE_SIZE)) {
1730                                                 page_cache_release(page);
1731                                                 goto end_coredump;
1732                                         }
1733                                 } else {
1734                                         void *kaddr;
1735                                         flush_cache_page(vma, addr,
1736                                                          page_to_pfn(page));
1737                                         kaddr = kmap(page);
1738                                         if ((size += PAGE_SIZE) > limit ||
1739                                             !dump_write(file, kaddr,
1740                                             PAGE_SIZE)) {
1741                                                 kunmap(page);
1742                                                 page_cache_release(page);
1743                                                 goto end_coredump;
1744                                         }
1745                                         kunmap(page);
1746                                 }
1747                                 page_cache_release(page);
1748                         }
1749                 }
1750         }
1751
1752 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1753         ELF_CORE_WRITE_EXTRA_DATA;
1754 #endif
1755
1756 end_coredump:
1757         set_fs(fs);
1758
1759 cleanup:
1760         while (!list_empty(&thread_list)) {
1761                 struct list_head *tmp = thread_list.next;
1762                 list_del(tmp);
1763                 kfree(list_entry(tmp, struct elf_thread_status, list));
1764         }
1765
1766         kfree(elf);
1767         kfree(prstatus);
1768         kfree(psinfo);
1769         kfree(notes);
1770         kfree(fpu);
1771 #ifdef ELF_CORE_COPY_XFPREGS
1772         kfree(xfpu);
1773 #endif
1774         return has_dumped;
1775 #undef NUM_NOTES
1776 }
1777
1778 #endif          /* USE_ELF_CORE_DUMP */
1779
1780 static int __init init_elf_binfmt(void)
1781 {
1782         return register_binfmt(&elf_format);
1783 }
1784
1785 static void __exit exit_elf_binfmt(void)
1786 {
1787         /* Remove the COFF and ELF loaders. */
1788         unregister_binfmt(&elf_format);
1789 }
1790
1791 core_initcall(init_elf_binfmt);
1792 module_exit(exit_elf_binfmt);
1793 MODULE_LICENSE("GPL");