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