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