fs/binfmt_elf.c

   1 /*
   2  * linux/fs/binfmt_elf.c
   3  *
   4  * These are the functions used to load ELF format executables as used
   5  * on SVr4 machines.  Information on the format may be found in the book
   6  * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
   7  * Tools".
   8  *
   9  * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/kernel.h>
  14 #include <linux/fs.h>
  15 #include <linux/mm.h>
  16 #include <linux/mman.h>
  17 #include <linux/errno.h>
  18 #include <linux/signal.h>
  19 #include <linux/binfmts.h>
  20 #include <linux/string.h>
  21 #include <linux/file.h>
  22 #include <linux/slab.h>
  23 #include <linux/personality.h>
  24 #include <linux/elfcore.h>
  25 #include <linux/init.h>
  26 #include <linux/highuid.h>
  27 #include <linux/compiler.h>
  28 #include <linux/highmem.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/security.h>
  32 #include <linux/random.h>
  33 #include <linux/elf.h>
  34 #include <linux/utsname.h>
  35 #include <linux/coredump.h>
  36 #include <linux/sched.h>
  37 #include <asm/uaccess.h>
  38 #include <asm/param.h>
  39 #include <asm/page.h>
  40
  41 #ifndef user_long_t
  42 #define user_long_t long
  43 #endif
  44 #ifndef user_siginfo_t
  45 #define user_siginfo_t siginfo_t
  46 #endif
  47
  48 static int load_elf_binary(struct linux_binprm *bprm);
  49 static int load_elf_library(struct file *);
  50 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
  51                                 int, int, unsigned long);
  52
  53 /*
  54  * If we don't support core dumping, then supply a NULL so we
  55  * don't even try.
  56  */
  57 #ifdef CONFIG_ELF_CORE
  58 static int elf_core_dump(struct coredump_params *cprm);
  59 #else
  60 #define elf_core_dump   NULL
  61 #endif
  62
  63 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
  64 #define ELF_MIN_ALIGN   ELF_EXEC_PAGESIZE
  65 #else
  66 #define ELF_MIN_ALIGN   PAGE_SIZE
  67 #endif
  68
  69 #ifndef ELF_CORE_EFLAGS
  70 #define ELF_CORE_EFLAGS 0
  71 #endif
  72
  73 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
  74 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
  75 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
  76
  77 static struct linux_binfmt elf_format = {
  78         .module         = THIS_MODULE,
  79         .load_binary    = load_elf_binary,
  80         .load_shlib     = load_elf_library,
  81         .core_dump      = elf_core_dump,
  82         .min_coredump   = ELF_EXEC_PAGESIZE,
  83 };
  84
  85 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
  86
  87 static int set_brk(unsigned long start, unsigned long end)
  88 {
  89         start = ELF_PAGEALIGN(start);
  90         end = ELF_PAGEALIGN(end);
  91         if (end > start) {
  92                 unsigned long addr;
  93                 addr = vm_brk(start, end - start);
  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 little 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 #ifndef ELF_BASE_PLATFORM
 135 /*
 136  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
 137  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
 138  * will be copied to the user stack in the same manner as AT_PLATFORM.
 139  */
 140 #define ELF_BASE_PLATFORM NULL
 141 #endif
 142
 143 static int
 144 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
 145                 unsigned long load_addr, unsigned long interp_load_addr)
 146 {
 147         unsigned long p = bprm->p;
 148         int argc = bprm->argc;
 149         int envc = bprm->envc;
 150         elf_addr_t __user *argv;
 151         elf_addr_t __user *envp;
 152         elf_addr_t __user *sp;
 153         elf_addr_t __user *u_platform;
 154         elf_addr_t __user *u_base_platform;
 155         elf_addr_t __user *u_rand_bytes;
 156         const char *k_platform = ELF_PLATFORM;
 157         const char *k_base_platform = ELF_BASE_PLATFORM;
 158         unsigned char k_rand_bytes[16];
 159         int items;
 160         elf_addr_t *elf_info;
 161         int ei_index = 0;
 162         const struct cred *cred = current_cred();
 163         struct vm_area_struct *vma;
 164
 165         /*
 166          * In some cases (e.g. Hyper-Threading), we want to avoid L1
 167          * evictions by the processes running on the same package. One
 168          * thing we can do is to shuffle the initial stack for them.
 169          */
 170
 171         p = arch_align_stack(p);
 172
 173         /*
 174          * If this architecture has a platform capability string, copy it
 175          * to userspace.  In some cases (Sparc), this info is impossible
 176          * for userspace to get any other way, in others (i386) it is
 177          * merely difficult.
 178          */
 179         u_platform = NULL;
 180         if (k_platform) {
 181                 size_t len = strlen(k_platform) + 1;
 182
 183                 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 184                 if (__copy_to_user(u_platform, k_platform, len))
 185                         return -EFAULT;
 186         }
 187
 188         /*
 189          * If this architecture has a "base" platform capability
 190          * string, copy it to userspace.
 191          */
 192         u_base_platform = NULL;
 193         if (k_base_platform) {
 194                 size_t len = strlen(k_base_platform) + 1;
 195
 196                 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 197                 if (__copy_to_user(u_base_platform, k_base_platform, len))
 198                         return -EFAULT;
 199         }
 200
 201         /*
 202          * Generate 16 random bytes for userspace PRNG seeding.
 203          */
 204         get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
 205         u_rand_bytes = (elf_addr_t __user *)
 206                        STACK_ALLOC(p, sizeof(k_rand_bytes));
 207         if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
 208                 return -EFAULT;
 209
 210         /* Create the ELF interpreter info */
 211         elf_info = (elf_addr_t *)current->mm->saved_auxv;
 212         /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
 213 #define NEW_AUX_ENT(id, val) \
 214         do { \
 215                 elf_info[ei_index++] = id; \
 216                 elf_info[ei_index++] = val; \
 217         } while (0)
 218
 219 #ifdef ARCH_DLINFO
 220         /*
 221          * ARCH_DLINFO must come first so PPC can do its special alignment of
 222          * AUXV.
 223          * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
 224          * ARCH_DLINFO changes
 225          */
 226         ARCH_DLINFO;
 227 #endif
 228         NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
 229         NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
 230         NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
 231         NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
 232         NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
 233         NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
 234         NEW_AUX_ENT(AT_BASE, interp_load_addr);
 235         NEW_AUX_ENT(AT_FLAGS, 0);
 236         NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
 237         NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
 238         NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
 239         NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
 240         NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
 241         NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
 242         NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
 243 #ifdef ELF_HWCAP2
 244         NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
 245 #endif
 246         NEW_AUX_ENT(AT_EXECFN, bprm->exec);
 247         if (k_platform) {
 248                 NEW_AUX_ENT(AT_PLATFORM,
 249                             (elf_addr_t)(unsigned long)u_platform);
 250         }
 251         if (k_base_platform) {
 252                 NEW_AUX_ENT(AT_BASE_PLATFORM,
 253                             (elf_addr_t)(unsigned long)u_base_platform);
 254         }
 255         if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
 256                 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
 257         }
 258 #undef NEW_AUX_ENT
 259         /* AT_NULL is zero; clear the rest too */
 260         memset(&elf_info[ei_index], 0,
 261                sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
 262
 263         /* And advance past the AT_NULL entry.  */
 264         ei_index += 2;
 265
 266         sp = STACK_ADD(p, ei_index);
 267
 268         items = (argc + 1) + (envc + 1) + 1;
 269         bprm->p = STACK_ROUND(sp, items);
 270
 271         /* Point sp at the lowest address on the stack */
 272 #ifdef CONFIG_STACK_GROWSUP
 273         sp = (elf_addr_t __user *)bprm->p - items - ei_index;
 274         bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
 275 #else
 276         sp = (elf_addr_t __user *)bprm->p;
 277 #endif
 278
 279
 280         /*
 281          * Grow the stack manually; some architectures have a limit on how
 282          * far ahead a user-space access may be in order to grow the stack.
 283          */
 284         vma = find_extend_vma(current->mm, bprm->p);
 285         if (!vma)
 286                 return -EFAULT;
 287
 288         /* Now, let's put argc (and argv, envp if appropriate) on the stack */
 289         if (__put_user(argc, sp++))
 290                 return -EFAULT;
 291         argv = sp;
 292         envp = argv + argc + 1;
 293
 294         /* Populate argv and envp */
 295         p = current->mm->arg_end = current->mm->arg_start;
 296         while (argc-- > 0) {
 297                 size_t len;
 298                 if (__put_user((elf_addr_t)p, argv++))
 299                         return -EFAULT;
 300                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 301                 if (!len || len > MAX_ARG_STRLEN)
 302                         return -EINVAL;
 303                 p += len;
 304         }
 305         if (__put_user(0, argv))
 306                 return -EFAULT;
 307         current->mm->arg_end = current->mm->env_start = p;
 308         while (envc-- > 0) {
 309                 size_t len;
 310                 if (__put_user((elf_addr_t)p, envp++))
 311                         return -EFAULT;
 312                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 313                 if (!len || len > MAX_ARG_STRLEN)
 314                         return -EINVAL;
 315                 p += len;
 316         }
 317         if (__put_user(0, envp))
 318                 return -EFAULT;
 319         current->mm->env_end = p;
 320
 321         /* Put the elf_info on the stack in the right place.  */
 322         sp = (elf_addr_t __user *)envp + 1;
 323         if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
 324                 return -EFAULT;
 325         return 0;
 326 }
 327
 328 #ifndef elf_map
 329
 330 static unsigned long elf_map(struct file *filep, unsigned long addr,
 331                 struct elf_phdr *eppnt, int prot, int type,
 332                 unsigned long total_size)
 333 {
 334         unsigned long map_addr;
 335         unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
 336         unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
 337         addr = ELF_PAGESTART(addr);
 338         size = ELF_PAGEALIGN(size);
 339
 340         /* mmap() will return -EINVAL if given a zero size, but a
 341          * segment with zero filesize is perfectly valid */
 342         if (!size)
 343                 return addr;
 344
 345         /*
 346         * total_size is the size of the ELF (interpreter) image.
 347         * The _first_ mmap needs to know the full size, otherwise
 348         * randomization might put this image into an overlapping
 349         * position with the ELF binary image. (since size < total_size)
 350         * So we first map the 'big' image - and unmap the remainder at
 351         * the end. (which unmap is needed for ELF images with holes.)
 352         */
 353         if (total_size) {
 354                 total_size = ELF_PAGEALIGN(total_size);
 355                 map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
 356                 if (!BAD_ADDR(map_addr))
 357                         vm_munmap(map_addr+size, total_size-size);
 358         } else
 359                 map_addr = vm_mmap(filep, addr, size, prot, type, off);
 360
 361         return(map_addr);
 362 }
 363
 364 #endif /* !elf_map */
 365
 366 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
 367 {
 368         int i, first_idx = -1, last_idx = -1;
 369
 370         for (i = 0; i < nr; i++) {
 371                 if (cmds[i].p_type == PT_LOAD) {
 372                         last_idx = i;
 373                         if (first_idx == -1)
 374                                 first_idx = i;
 375                 }
 376         }
 377         if (first_idx == -1)
 378                 return 0;
 379
 380         return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
 381                                 ELF_PAGESTART(cmds[first_idx].p_vaddr);
 382 }
 383
 384
 385 /* This is much more generalized than the library routine read function,
 386    so we keep this separate.  Technically the library read function
 387    is only provided so that we can read a.out libraries that have
 388    an ELF header */
 389
 390 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
 391                 struct file *interpreter, unsigned long *interp_map_addr,
 392                 unsigned long no_base)
 393 {
 394         struct elf_phdr *elf_phdata;
 395         struct elf_phdr *eppnt;
 396         unsigned long load_addr = 0;
 397         int load_addr_set = 0;
 398         unsigned long last_bss = 0, elf_bss = 0;
 399         unsigned long error = ~0UL;
 400         unsigned long total_size;
 401         int retval, i, size;
 402
 403         /* First of all, some simple consistency checks */
 404         if (interp_elf_ex->e_type != ET_EXEC &&
 405             interp_elf_ex->e_type != ET_DYN)
 406                 goto out;
 407         if (!elf_check_arch(interp_elf_ex))
 408                 goto out;
 409         if (!interpreter->f_op || !interpreter->f_op->mmap)
 410                 goto out;
 411
 412         /*
 413          * If the size of this structure has changed, then punt, since
 414          * we will be doing the wrong thing.
 415          */
 416         if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
 417                 goto out;
 418         if (interp_elf_ex->e_phnum < 1 ||
 419                 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
 420                 goto out;
 421
 422         /* Now read in all of the header information */
 423         size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
 424         if (size > ELF_MIN_ALIGN)
 425                 goto out;
 426         elf_phdata = kmalloc(size, GFP_KERNEL);
 427         if (!elf_phdata)
 428                 goto out;
 429
 430         retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
 431                              (char *)elf_phdata, size);
 432         error = -EIO;
 433         if (retval != size) {
 434                 if (retval < 0)
 435                         error = retval;
 436                 goto out_close;
 437         }
 438
 439         total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
 440         if (!total_size) {
 441                 error = -EINVAL;
 442                 goto out_close;
 443         }
 444
 445         eppnt = elf_phdata;
 446         for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
 447                 if (eppnt->p_type == PT_LOAD) {
 448                         int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
 449                         int elf_prot = 0;
 450                         unsigned long vaddr = 0;
 451                         unsigned long k, map_addr;
 452
 453                         if (eppnt->p_flags & PF_R)
 454                                 elf_prot = PROT_READ;
 455                         if (eppnt->p_flags & PF_W)
 456                                 elf_prot |= PROT_WRITE;
 457                         if (eppnt->p_flags & PF_X)
 458                                 elf_prot |= PROT_EXEC;
 459                         vaddr = eppnt->p_vaddr;
 460                         if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
 461                                 elf_type |= MAP_FIXED;
 462                         else if (no_base && interp_elf_ex->e_type == ET_DYN)
 463                                 load_addr = -vaddr;
 464
 465                         map_addr = elf_map(interpreter, load_addr + vaddr,
 466                                         eppnt, elf_prot, elf_type, total_size);
 467                         total_size = 0;
 468                         if (!*interp_map_addr)
 469                                 *interp_map_addr = map_addr;
 470                         error = map_addr;
 471                         if (BAD_ADDR(map_addr))
 472                                 goto out_close;
 473
 474                         if (!load_addr_set &&
 475                             interp_elf_ex->e_type == ET_DYN) {
 476                                 load_addr = map_addr - ELF_PAGESTART(vaddr);
 477                                 load_addr_set = 1;
 478                         }
 479
 480                         /*
 481                          * Check to see if the section's size will overflow the
 482                          * allowed task size. Note that p_filesz must always be
 483                          * <= p_memsize so it's only necessary to check p_memsz.
 484                          */
 485                         k = load_addr + eppnt->p_vaddr;
 486                         if (BAD_ADDR(k) ||
 487                             eppnt->p_filesz > eppnt->p_memsz ||
 488                             eppnt->p_memsz > TASK_SIZE ||
 489                             TASK_SIZE - eppnt->p_memsz < k) {
 490                                 error = -ENOMEM;
 491                                 goto out_close;
 492                         }
 493
 494                         /*
 495                          * Find the end of the file mapping for this phdr, and
 496                          * keep track of the largest address we see for this.
 497                          */
 498                         k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
 499                         if (k > elf_bss)
 500                                 elf_bss = k;
 501
 502                         /*
 503                          * Do the same thing for the memory mapping - between
 504                          * elf_bss and last_bss is the bss section.
 505                          */
 506                         k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
 507                         if (k > last_bss)
 508                                 last_bss = k;
 509                 }
 510         }
 511
 512         if (last_bss > elf_bss) {
 513                 /*
 514                  * Now fill out the bss section.  First pad the last page up
 515                  * to the page boundary, and then perform a mmap to make sure
 516                  * that there are zero-mapped pages up to and including the
 517                  * last bss page.
 518                  */
 519                 if (padzero(elf_bss)) {
 520                         error = -EFAULT;
 521                         goto out_close;
 522                 }
 523
 524                 /* What we have mapped so far */
 525                 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
 526
 527                 /* Map the last of the bss segment */
 528                 error = vm_brk(elf_bss, last_bss - elf_bss);
 529                 if (BAD_ADDR(error))
 530                         goto out_close;
 531         }
 532
 533         error = load_addr;
 534
 535 out_close:
 536         kfree(elf_phdata);
 537 out:
 538         return error;
 539 }
 540
 541 /*
 542  * These are the functions used to load ELF style executables and shared
 543  * libraries.  There is no binary dependent code anywhere else.
 544  */
 545
 546 #define INTERPRETER_NONE 0
 547 #define INTERPRETER_ELF 2
 548
 549 #ifndef STACK_RND_MASK
 550 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))     /* 8MB of VA */
 551 #endif
 552
 553 static unsigned long randomize_stack_top(unsigned long stack_top)
 554 {
 555         unsigned int random_variable = 0;
 556
 557         if ((current->flags & PF_RANDOMIZE) &&
 558                 !(current->personality & ADDR_NO_RANDOMIZE)) {
 559                 random_variable = get_random_int() & STACK_RND_MASK;
 560                 random_variable <<= PAGE_SHIFT;
 561         }
 562 #ifdef CONFIG_STACK_GROWSUP
 563         return PAGE_ALIGN(stack_top) + random_variable;
 564 #else
 565         return PAGE_ALIGN(stack_top) - random_variable;
 566 #endif
 567 }
 568
 569 static int load_elf_binary(struct linux_binprm *bprm)
 570 {
 571         struct file *interpreter = NULL; /* to shut gcc up */
 572         unsigned long load_addr = 0, load_bias = 0;
 573         int load_addr_set = 0;
 574         char * elf_interpreter = NULL;
 575         unsigned long error;
 576         struct elf_phdr *elf_ppnt, *elf_phdata;
 577         unsigned long elf_bss, elf_brk;
 578         int retval, i;
 579         unsigned int size;
 580         unsigned long elf_entry;
 581         unsigned long interp_load_addr = 0;
 582         unsigned long start_code, end_code, start_data, end_data;
 583         unsigned long reloc_func_desc __maybe_unused = 0;
 584         int executable_stack = EXSTACK_DEFAULT;
 585         unsigned long def_flags = 0;
 586         struct pt_regs *regs = current_pt_regs();
 587         struct {
 588                 struct elfhdr elf_ex;
 589                 struct elfhdr interp_elf_ex;
 590         } *loc;
 591
 592         loc = kmalloc(sizeof(*loc), GFP_KERNEL);
 593         if (!loc) {
 594                 retval = -ENOMEM;
 595                 goto out_ret;
 596         }
 597
 598         /* Get the exec-header */
 599         loc->elf_ex = *((struct elfhdr *)bprm->buf);
 600
 601         retval = -ENOEXEC;
 602         /* First of all, some simple consistency checks */
 603         if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 604                 goto out;
 605
 606         if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
 607                 goto out;
 608         if (!elf_check_arch(&loc->elf_ex))
 609                 goto out;
 610         if (!bprm->file->f_op || !bprm->file->f_op->mmap)
 611                 goto out;
 612
 613         /* Now read in all of the header information */
 614         if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
 615                 goto out;
 616         if (loc->elf_ex.e_phnum < 1 ||
 617                 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
 618                 goto out;
 619         size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
 620         retval = -ENOMEM;
 621         elf_phdata = kmalloc(size, GFP_KERNEL);
 622         if (!elf_phdata)
 623                 goto out;
 624
 625         retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
 626                              (char *)elf_phdata, size);
 627         if (retval != size) {
 628                 if (retval >= 0)
 629                         retval = -EIO;
 630                 goto out_free_ph;
 631         }
 632
 633         elf_ppnt = elf_phdata;
 634         elf_bss = 0;
 635         elf_brk = 0;
 636
 637         start_code = ~0UL;
 638         end_code = 0;
 639         start_data = 0;
 640         end_data = 0;
 641
 642         for (i = 0; i < loc->elf_ex.e_phnum; i++) {
 643                 if (elf_ppnt->p_type == PT_INTERP) {
 644                         /* This is the program interpreter used for
 645                          * shared libraries - for now assume that this
 646                          * is an a.out format binary
 647                          */
 648                         retval = -ENOEXEC;
 649                         if (elf_ppnt->p_filesz > PATH_MAX ||
 650                             elf_ppnt->p_filesz < 2)
 651                                 goto out_free_ph;
 652
 653                         retval = -ENOMEM;
 654                         elf_interpreter = kmalloc(elf_ppnt->p_filesz,
 655                                                   GFP_KERNEL);
 656                         if (!elf_interpreter)
 657                                 goto out_free_ph;
 658
 659                         retval = kernel_read(bprm->file, elf_ppnt->p_offset,
 660                                              elf_interpreter,
 661                                              elf_ppnt->p_filesz);
 662                         if (retval != elf_ppnt->p_filesz) {
 663                                 if (retval >= 0)
 664                                         retval = -EIO;
 665                                 goto out_free_interp;
 666                         }
 667                         /* make sure path is NULL terminated */
 668                         retval = -ENOEXEC;
 669                         if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
 670                                 goto out_free_interp;
 671
 672                         interpreter = open_exec(elf_interpreter);
 673                         retval = PTR_ERR(interpreter);
 674                         if (IS_ERR(interpreter))
 675                                 goto out_free_interp;
 676
 677                         /*
 678                          * If the binary is not readable then enforce
 679                          * mm->dumpable = 0 regardless of the interpreter's
 680                          * permissions.
 681                          */
 682                         would_dump(bprm, interpreter);
 683
 684                         retval = kernel_read(interpreter, 0, bprm->buf,
 685                                              BINPRM_BUF_SIZE);
 686                         if (retval != BINPRM_BUF_SIZE) {
 687                                 if (retval >= 0)
 688                                         retval = -EIO;
 689                                 goto out_free_dentry;
 690                         }
 691
 692                         /* Get the exec headers */
 693                         loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
 694                         break;
 695                 }
 696                 elf_ppnt++;
 697         }
 698
 699         elf_ppnt = elf_phdata;
 700         for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
 701                 if (elf_ppnt->p_type == PT_GNU_STACK) {
 702                         if (elf_ppnt->p_flags & PF_X)
 703                                 executable_stack = EXSTACK_ENABLE_X;
 704                         else
 705                                 executable_stack = EXSTACK_DISABLE_X;
 706                         break;
 707                 }
 708
 709         /* Some simple consistency checks for the interpreter */
 710         if (elf_interpreter) {
 711                 retval = -ELIBBAD;
 712                 /* Not an ELF interpreter */
 713                 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 714                         goto out_free_dentry;
 715                 /* Verify the interpreter has a valid arch */
 716                 if (!elf_check_arch(&loc->interp_elf_ex))
 717                         goto out_free_dentry;
 718         }
 719
 720         /* Flush all traces of the currently running executable */
 721         retval = flush_old_exec(bprm);
 722         if (retval)
 723                 goto out_free_dentry;
 724
 725         /* OK, This is the point of no return */
 726         current->mm->def_flags = def_flags;
 727
 728         /* Do this immediately, since STACK_TOP as used in setup_arg_pages
 729            may depend on the personality.  */
 730         SET_PERSONALITY(loc->elf_ex);
 731         if (elf_read_implies_exec(loc->elf_ex, executable_stack))
 732                 current->personality |= READ_IMPLIES_EXEC;
 733
 734         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 735                 current->flags |= PF_RANDOMIZE;
 736
 737         setup_new_exec(bprm);
 738
 739         /* Do this so that we can load the interpreter, if need be.  We will
 740            change some of these later */
 741         retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
 742                                  executable_stack);
 743         if (retval < 0) {
 744                 send_sig(SIGKILL, current, 0);
 745                 goto out_free_dentry;
 746         }
 747
 748         current->mm->start_stack = bprm->p;
 749
 750         /* Now we do a little grungy work by mmapping the ELF image into
 751            the correct location in memory. */
 752         for(i = 0, elf_ppnt = elf_phdata;
 753             i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
 754                 int elf_prot = 0, elf_flags;
 755                 unsigned long k, vaddr;
 756
 757                 if (elf_ppnt->p_type != PT_LOAD)
 758                         continue;
 759
 760                 if (unlikely (elf_brk > elf_bss)) {
 761                         unsigned long nbyte;
 762
 763                         /* There was a PT_LOAD segment with p_memsz > p_filesz
 764                            before this one. Map anonymous pages, if needed,
 765                            and clear the area.  */
 766                         retval = set_brk(elf_bss + load_bias,
 767                                          elf_brk + load_bias);
 768                         if (retval) {
 769                                 send_sig(SIGKILL, current, 0);
 770                                 goto out_free_dentry;
 771                         }
 772                         nbyte = ELF_PAGEOFFSET(elf_bss);
 773                         if (nbyte) {
 774                                 nbyte = ELF_MIN_ALIGN - nbyte;
 775                                 if (nbyte > elf_brk - elf_bss)
 776                                         nbyte = elf_brk - elf_bss;
 777                                 if (clear_user((void __user *)elf_bss +
 778                                                         load_bias, nbyte)) {
 779                                         /*
 780                                          * This bss-zeroing can fail if the ELF
 781                                          * file specifies odd protections. So
 782                                          * we don't check the return value
 783                                          */
 784                                 }
 785                         }
 786                 }
 787
 788                 if (elf_ppnt->p_flags & PF_R)
 789                         elf_prot |= PROT_READ;
 790                 if (elf_ppnt->p_flags & PF_W)
 791                         elf_prot |= PROT_WRITE;
 792                 if (elf_ppnt->p_flags & PF_X)
 793                         elf_prot |= PROT_EXEC;
 794
 795                 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
 796
 797                 vaddr = elf_ppnt->p_vaddr;
 798                 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
 799                         elf_flags |= MAP_FIXED;
 800                 } else if (loc->elf_ex.e_type == ET_DYN) {
 801                         /* Try and get dynamic programs out of the way of the
 802                          * default mmap base, as well as whatever program they
 803                          * might try to exec.  This is because the brk will
 804                          * follow the loader, and is not movable.  */
 805 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
 806                         /* Memory randomization might have been switched off
 807                          * in runtime via sysctl or explicit setting of
 808                          * personality flags.
 809                          * If that is the case, retain the original non-zero
 810                          * load_bias value in order to establish proper
 811                          * non-randomized mappings.
 812                          */
 813                         if (current->flags & PF_RANDOMIZE)
 814                                 load_bias = 0;
 815                         else
 816                                 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 817 #else
 818                         load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 819 #endif
 820                 }
 821
 822                 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
 823                                 elf_prot, elf_flags, 0);
 824                 if (BAD_ADDR(error)) {
 825                         send_sig(SIGKILL, current, 0);
 826                         retval = IS_ERR((void *)error) ?
 827                                 PTR_ERR((void*)error) : -EINVAL;
 828                         goto out_free_dentry;
 829                 }
 830
 831                 if (!load_addr_set) {
 832                         load_addr_set = 1;
 833                         load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
 834                         if (loc->elf_ex.e_type == ET_DYN) {
 835                                 load_bias += error -
 836                                              ELF_PAGESTART(load_bias + vaddr);
 837                                 load_addr += load_bias;
 838                                 reloc_func_desc = load_bias;
 839                         }
 840                 }
 841                 k = elf_ppnt->p_vaddr;
 842                 if (k < start_code)
 843                         start_code = k;
 844                 if (start_data < k)
 845                         start_data = k;
 846
 847                 /*
 848                  * Check to see if the section's size will overflow the
 849                  * allowed task size. Note that p_filesz must always be
 850                  * <= p_memsz so it is only necessary to check p_memsz.
 851                  */
 852                 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
 853                     elf_ppnt->p_memsz > TASK_SIZE ||
 854                     TASK_SIZE - elf_ppnt->p_memsz < k) {
 855                         /* set_brk can never work. Avoid overflows. */
 856                         send_sig(SIGKILL, current, 0);
 857                         retval = -EINVAL;
 858                         goto out_free_dentry;
 859                 }
 860
 861                 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
 862
 863                 if (k > elf_bss)
 864                         elf_bss = k;
 865                 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
 866                         end_code = k;
 867                 if (end_data < k)
 868                         end_data = k;
 869                 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
 870                 if (k > elf_brk)
 871                         elf_brk = k;
 872         }
 873
 874         loc->elf_ex.e_entry += load_bias;
 875         elf_bss += load_bias;
 876         elf_brk += load_bias;
 877         start_code += load_bias;
 878         end_code += load_bias;
 879         start_data += load_bias;
 880         end_data += load_bias;
 881
 882         /* Calling set_brk effectively mmaps the pages that we need
 883          * for the bss and break sections.  We must do this before
 884          * mapping in the interpreter, to make sure it doesn't wind
 885          * up getting placed where the bss needs to go.
 886          */
 887         retval = set_brk(elf_bss, elf_brk);
 888         if (retval) {
 889                 send_sig(SIGKILL, current, 0);
 890                 goto out_free_dentry;
 891         }
 892         if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
 893                 send_sig(SIGSEGV, current, 0);
 894                 retval = -EFAULT; /* Nobody gets to see this, but.. */
 895                 goto out_free_dentry;
 896         }
 897
 898         if (elf_interpreter) {
 899                 unsigned long interp_map_addr = 0;
 900
 901                 elf_entry = load_elf_interp(&loc->interp_elf_ex,
 902                                             interpreter,
 903                                             &interp_map_addr,
 904                                             load_bias);
 905                 if (!IS_ERR((void *)elf_entry)) {
 906                         /*
 907                          * load_elf_interp() returns relocation
 908                          * adjustment
 909                          */
 910                         interp_load_addr = elf_entry;
 911                         elf_entry += loc->interp_elf_ex.e_entry;
 912                 }
 913                 if (BAD_ADDR(elf_entry)) {
 914                         force_sig(SIGSEGV, current);
 915                         retval = IS_ERR((void *)elf_entry) ?
 916                                         (int)elf_entry : -EINVAL;
 917                         goto out_free_dentry;
 918                 }
 919                 reloc_func_desc = interp_load_addr;
 920
 921                 allow_write_access(interpreter);
 922                 fput(interpreter);
 923                 kfree(elf_interpreter);
 924         } else {
 925                 elf_entry = loc->elf_ex.e_entry;
 926                 if (BAD_ADDR(elf_entry)) {
 927                         force_sig(SIGSEGV, current);
 928                         retval = -EINVAL;
 929                         goto out_free_dentry;
 930                 }
 931         }
 932
 933         kfree(elf_phdata);
 934
 935         set_binfmt(&elf_format);
 936
 937 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
 938         retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
 939         if (retval < 0) {
 940                 send_sig(SIGKILL, current, 0);
 941                 goto out;
 942         }
 943 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
 944
 945         install_exec_creds(bprm);
 946         retval = create_elf_tables(bprm, &loc->elf_ex,
 947                           load_addr, interp_load_addr);
 948         if (retval < 0) {
 949                 send_sig(SIGKILL, current, 0);
 950                 goto out;
 951         }
 952         /* N.B. passed_fileno might not be initialized? */
 953         current->mm->end_code = end_code;
 954         current->mm->start_code = start_code;
 955         current->mm->start_data = start_data;
 956         current->mm->end_data = end_data;
 957         current->mm->start_stack = bprm->p;
 958
 959 #ifdef arch_randomize_brk
 960         if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
 961                 current->mm->brk = current->mm->start_brk =
 962                         arch_randomize_brk(current->mm);
 963 #ifdef CONFIG_COMPAT_BRK
 964                 current->brk_randomized = 1;
 965 #endif
 966         }
 967 #endif
 968
 969         if (current->personality & MMAP_PAGE_ZERO) {
 970                 /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
 971                    and some applications "depend" upon this behavior.
 972                    Since we do not have the power to recompile these, we
 973                    emulate the SVr4 behavior. Sigh. */
 974                 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
 975                                 MAP_FIXED | MAP_PRIVATE, 0);
 976         }
 977
 978 #ifdef ELF_PLAT_INIT
 979         /*
 980          * The ABI may specify that certain registers be set up in special
 981          * ways (on i386 %edx is the address of a DT_FINI function, for
 982          * example.  In addition, it may also specify (eg, PowerPC64 ELF)
 983          * that the e_entry field is the address of the function descriptor
 984          * for the startup routine, rather than the address of the startup
 985          * routine itself.  This macro performs whatever initialization to
 986          * the regs structure is required as well as any relocations to the
 987          * function descriptor entries when executing dynamically links apps.
 988          */
 989         ELF_PLAT_INIT(regs, reloc_func_desc);
 990 #endif
 991
 992         start_thread(regs, elf_entry, bprm->p);
 993         retval = 0;
 994 out:
 995         kfree(loc);
 996 out_ret:
 997         return retval;
 998
 999         /* error cleanup */
1000 out_free_dentry:
1001         allow_write_access(interpreter);
1002         if (interpreter)
1003                 fput(interpreter);
1004 out_free_interp:
1005         kfree(elf_interpreter);
1006 out_free_ph:
1007         kfree(elf_phdata);
1008         goto out;
1009 }
1010
1011 /* This is really simpleminded and specialized - we are loading an
1012    a.out library that is given an ELF header. */
1013 static int load_elf_library(struct file *file)
1014 {
1015         struct elf_phdr *elf_phdata;
1016         struct elf_phdr *eppnt;
1017         unsigned long elf_bss, bss, len;
1018         int retval, error, i, j;
1019         struct elfhdr elf_ex;
1020
1021         error = -ENOEXEC;
1022         retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1023         if (retval != sizeof(elf_ex))
1024                 goto out;
1025
1026         if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1027                 goto out;
1028
1029         /* First of all, some simple consistency checks */
1030         if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1031             !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1032                 goto out;
1033
1034         /* Now read in all of the header information */
1035
1036         j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1037         /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1038
1039         error = -ENOMEM;
1040         elf_phdata = kmalloc(j, GFP_KERNEL);
1041         if (!elf_phdata)
1042                 goto out;
1043
1044         eppnt = elf_phdata;
1045         error = -ENOEXEC;
1046         retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1047         if (retval != j)
1048                 goto out_free_ph;
1049
1050         for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1051                 if ((eppnt + i)->p_type == PT_LOAD)
1052                         j++;
1053         if (j != 1)
1054                 goto out_free_ph;
1055
1056         while (eppnt->p_type != PT_LOAD)
1057                 eppnt++;
1058
1059         /* Now use mmap to map the library into memory. */
1060         error = vm_mmap(file,
1061                         ELF_PAGESTART(eppnt->p_vaddr),
1062                         (eppnt->p_filesz +
1063                          ELF_PAGEOFFSET(eppnt->p_vaddr)),
1064                         PROT_READ | PROT_WRITE | PROT_EXEC,
1065                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1066                         (eppnt->p_offset -
1067                          ELF_PAGEOFFSET(eppnt->p_vaddr)));
1068         if (error != ELF_PAGESTART(eppnt->p_vaddr))
1069                 goto out_free_ph;
1070
1071         elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1072         if (padzero(elf_bss)) {
1073                 error = -EFAULT;
1074                 goto out_free_ph;
1075         }
1076
1077         len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1078                             ELF_MIN_ALIGN - 1);
1079         bss = eppnt->p_memsz + eppnt->p_vaddr;
1080         if (bss > len)
1081                 vm_brk(len, bss - len);
1082         error = 0;
1083
1084 out_free_ph:
1085         kfree(elf_phdata);
1086 out:
1087         return error;
1088 }
1089
1090 #ifdef CONFIG_ELF_CORE
1091 /*
1092  * ELF core dumper
1093  *
1094  * Modelled on fs/exec.c:aout_core_dump()
1095  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1096  */
1097
1098 /*
1099  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1100  * that are useful for post-mortem analysis are included in every core dump.
1101  * In that way we ensure that the core dump is fully interpretable later
1102  * without matching up the same kernel and hardware config to see what PC values
1103  * meant. These special mappings include - vDSO, vsyscall, and other
1104  * architecture specific mappings
1105  */
1106 static bool always_dump_vma(struct vm_area_struct *vma)
1107 {
1108         /* Any vsyscall mappings? */
1109         if (vma == get_gate_vma(vma->vm_mm))
1110                 return true;
1111         /*
1112          * arch_vma_name() returns non-NULL for special architecture mappings,
1113          * such as vDSO sections.
1114          */
1115         if (arch_vma_name(vma))
1116                 return true;
1117
1118         return false;
1119 }
1120
1121 /*
1122  * Decide what to dump of a segment, part, all or none.
1123  */
1124 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1125                                    unsigned long mm_flags)
1126 {
1127 #define FILTER(type)    (mm_flags & (1UL << MMF_DUMP_##type))
1128
1129         /* always dump the vdso and vsyscall sections */
1130         if (always_dump_vma(vma))
1131                 goto whole;
1132
1133         if (vma->vm_flags & VM_DONTDUMP)
1134                 return 0;
1135
1136         /* Hugetlb memory check */
1137         if (vma->vm_flags & VM_HUGETLB) {
1138                 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1139                         goto whole;
1140                 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1141                         goto whole;
1142                 return 0;
1143         }
1144
1145         /* Do not dump I/O mapped devices or special mappings */
1146         if (vma->vm_flags & VM_IO)
1147                 return 0;
1148
1149         /* By default, dump shared memory if mapped from an anonymous file. */
1150         if (vma->vm_flags & VM_SHARED) {
1151                 if (file_inode(vma->vm_file)->i_nlink == 0 ?
1152                     FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1153                         goto whole;
1154                 return 0;
1155         }
1156
1157         /* Dump segments that have been written to.  */
1158         if (vma->anon_vma && FILTER(ANON_PRIVATE))
1159                 goto whole;
1160         if (vma->vm_file == NULL)
1161                 return 0;
1162
1163         if (FILTER(MAPPED_PRIVATE))
1164                 goto whole;
1165
1166         /*
1167          * If this looks like the beginning of a DSO or executable mapping,
1168          * check for an ELF header.  If we find one, dump the first page to
1169          * aid in determining what was mapped here.
1170          */
1171         if (FILTER(ELF_HEADERS) &&
1172             vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1173                 u32 __user *header = (u32 __user *) vma->vm_start;
1174                 u32 word;
1175                 mm_segment_t fs = get_fs();
1176                 /*
1177                  * Doing it this way gets the constant folded by GCC.
1178                  */
1179                 union {
1180                         u32 cmp;
1181                         char elfmag[SELFMAG];
1182                 } magic;
1183                 BUILD_BUG_ON(SELFMAG != sizeof word);
1184                 magic.elfmag[EI_MAG0] = ELFMAG0;
1185                 magic.elfmag[EI_MAG1] = ELFMAG1;
1186                 magic.elfmag[EI_MAG2] = ELFMAG2;
1187                 magic.elfmag[EI_MAG3] = ELFMAG3;
1188                 /*
1189                  * Switch to the user "segment" for get_user(),
1190                  * then put back what elf_core_dump() had in place.
1191                  */
1192                 set_fs(USER_DS);
1193                 if (unlikely(get_user(word, header)))
1194                         word = 0;
1195                 set_fs(fs);
1196                 if (word == magic.cmp)
1197                         return PAGE_SIZE;
1198         }
1199
1200 #undef  FILTER
1201
1202         return 0;
1203
1204 whole:
1205         return vma->vm_end - vma->vm_start;
1206 }
1207
1208 /* An ELF note in memory */
1209 struct memelfnote
1210 {
1211         const char *name;
1212         int type;
1213         unsigned int datasz;
1214         void *data;
1215 };
1216
1217 static int notesize(struct memelfnote *en)
1218 {
1219         int sz;
1220
1221         sz = sizeof(struct elf_note);
1222         sz += roundup(strlen(en->name) + 1, 4);
1223         sz += roundup(en->datasz, 4);
1224
1225         return sz;
1226 }
1227
1228 #define DUMP_WRITE(addr, nr, foffset)   \
1229         do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1230
1231 static int alignfile(struct file *file, loff_t *foffset)
1232 {
1233         static const char buf[4] = { 0, };
1234         DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1235         return 1;
1236 }
1237
1238 static int writenote(struct memelfnote *men, struct file *file,
1239                         loff_t *foffset)
1240 {
1241         struct elf_note en;
1242         en.n_namesz = strlen(men->name) + 1;
1243         en.n_descsz = men->datasz;
1244         en.n_type = men->type;
1245
1246         DUMP_WRITE(&en, sizeof(en), foffset);
1247         DUMP_WRITE(men->name, en.n_namesz, foffset);
1248         if (!alignfile(file, foffset))
1249                 return 0;
1250         DUMP_WRITE(men->data, men->datasz, foffset);
1251         if (!alignfile(file, foffset))
1252                 return 0;
1253
1254         return 1;
1255 }
1256 #undef DUMP_WRITE
1257
1258 static void fill_elf_header(struct elfhdr *elf, int segs,
1259                             u16 machine, u32 flags)
1260 {
1261         memset(elf, 0, sizeof(*elf));
1262
1263         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1264         elf->e_ident[EI_CLASS] = ELF_CLASS;
1265         elf->e_ident[EI_DATA] = ELF_DATA;
1266         elf->e_ident[EI_VERSION] = EV_CURRENT;
1267         elf->e_ident[EI_OSABI] = ELF_OSABI;
1268
1269         elf->e_type = ET_CORE;
1270         elf->e_machine = machine;
1271         elf->e_version = EV_CURRENT;
1272         elf->e_phoff = sizeof(struct elfhdr);
1273         elf->e_flags = flags;
1274         elf->e_ehsize = sizeof(struct elfhdr);
1275         elf->e_phentsize = sizeof(struct elf_phdr);
1276         elf->e_phnum = segs;
1277
1278         return;
1279 }
1280
1281 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1282 {
1283         phdr->p_type = PT_NOTE;
1284         phdr->p_offset = offset;
1285         phdr->p_vaddr = 0;
1286         phdr->p_paddr = 0;
1287         phdr->p_filesz = sz;
1288         phdr->p_memsz = 0;
1289         phdr->p_flags = 0;
1290         phdr->p_align = 0;
1291         return;
1292 }
1293
1294 static void fill_note(struct memelfnote *note, const char *name, int type,
1295                 unsigned int sz, void *data)
1296 {
1297         note->name = name;
1298         note->type = type;
1299         note->datasz = sz;
1300         note->data = data;
1301         return;
1302 }
1303
1304 /*
1305  * fill up all the fields in prstatus from the given task struct, except
1306  * registers which need to be filled up separately.
1307  */
1308 static void fill_prstatus(struct elf_prstatus *prstatus,
1309                 struct task_struct *p, long signr)
1310 {
1311         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1312         prstatus->pr_sigpend = p->pending.signal.sig[0];
1313         prstatus->pr_sighold = p->blocked.sig[0];
1314         rcu_read_lock();
1315         prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1316         rcu_read_unlock();
1317         prstatus->pr_pid = task_pid_vnr(p);
1318         prstatus->pr_pgrp = task_pgrp_vnr(p);
1319         prstatus->pr_sid = task_session_vnr(p);
1320         if (thread_group_leader(p)) {
1321                 struct task_cputime cputime;
1322
1323                 /*
1324                  * This is the record for the group leader.  It shows the
1325                  * group-wide total, not its individual thread total.
1326                  */
1327                 thread_group_cputime(p, &cputime);
1328                 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1329                 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1330         } else {
1331                 cputime_t utime, stime;
1332
1333                 task_cputime(p, &utime, &stime);
1334                 cputime_to_timeval(utime, &prstatus->pr_utime);
1335                 cputime_to_timeval(stime, &prstatus->pr_stime);
1336         }
1337         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1338         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1339 }
1340
1341 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1342                        struct mm_struct *mm)
1343 {
1344         const struct cred *cred;
1345         unsigned int i, len;
1346
1347         /* first copy the parameters from user space */
1348         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1349
1350         len = mm->arg_end - mm->arg_start;
1351         if (len >= ELF_PRARGSZ)
1352                 len = ELF_PRARGSZ-1;
1353         if (copy_from_user(&psinfo->pr_psargs,
1354                            (const char __user *)mm->arg_start, len))
1355                 return -EFAULT;
1356         for(i = 0; i < len; i++)
1357                 if (psinfo->pr_psargs[i] == 0)
1358                         psinfo->pr_psargs[i] = ' ';
1359         psinfo->pr_psargs[len] = 0;
1360
1361         rcu_read_lock();
1362         psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1363         rcu_read_unlock();
1364         psinfo->pr_pid = task_pid_vnr(p);
1365         psinfo->pr_pgrp = task_pgrp_vnr(p);
1366         psinfo->pr_sid = task_session_vnr(p);
1367
1368         i = p->state ? ffz(~p->state) + 1 : 0;
1369         psinfo->pr_state = i;
1370         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1371         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1372         psinfo->pr_nice = task_nice(p);
1373         psinfo->pr_flag = p->flags;
1374         rcu_read_lock();
1375         cred = __task_cred(p);
1376         SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1377         SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1378         rcu_read_unlock();
1379         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1380
1381         return 0;
1382 }
1383
1384 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1385 {
1386         elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1387         int i = 0;
1388         do
1389                 i += 2;
1390         while (auxv[i - 2] != AT_NULL);
1391         fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1392 }
1393
1394 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1395                 siginfo_t *siginfo)
1396 {
1397         mm_segment_t old_fs = get_fs();
1398         set_fs(KERNEL_DS);
1399         copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1400         set_fs(old_fs);
1401         fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1402 }
1403
1404 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1405 /*
1406  * Format of NT_FILE note:
1407  *
1408  * long count     -- how many files are mapped
1409  * long page_size -- units for file_ofs
1410  * array of [COUNT] elements of
1411  *   long start
1412  *   long end
1413  *   long file_ofs
1414  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1415  */
1416 static int fill_files_note(struct memelfnote *note)
1417 {
1418         struct vm_area_struct *vma;
1419         unsigned count, size, names_ofs, remaining, n;
1420         user_long_t *data;
1421         user_long_t *start_end_ofs;
1422         char *name_base, *name_curpos;
1423
1424         /* *Estimated* file count and total data size needed */
1425         count = current->mm->map_count;
1426         size = count * 64;
1427
1428         names_ofs = (2 + 3 * count) * sizeof(data[0]);
1429  alloc:
1430         if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1431                 return -EINVAL;
1432         size = round_up(size, PAGE_SIZE);
1433         data = vmalloc(size);
1434         if (!data)
1435                 return -ENOMEM;
1436
1437         start_end_ofs = data + 2;
1438         name_base = name_curpos = ((char *)data) + names_ofs;
1439         remaining = size - names_ofs;
1440         count = 0;
1441         for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1442                 struct file *file;
1443                 const char *filename;
1444
1445                 file = vma->vm_file;
1446                 if (!file)
1447                         continue;
1448                 filename = d_path(&file->f_path, name_curpos, remaining);
1449                 if (IS_ERR(filename)) {
1450                         if (PTR_ERR(filename) == -ENAMETOOLONG) {
1451                                 vfree(data);
1452                                 size = size * 5 / 4;
1453                                 goto alloc;
1454                         }
1455                         continue;
1456                 }
1457
1458                 /* d_path() fills at the end, move name down */
1459                 /* n = strlen(filename) + 1: */
1460                 n = (name_curpos + remaining) - filename;
1461                 remaining = filename - name_curpos;
1462                 memmove(name_curpos, filename, n);
1463                 name_curpos += n;
1464
1465                 *start_end_ofs++ = vma->vm_start;
1466                 *start_end_ofs++ = vma->vm_end;
1467                 *start_end_ofs++ = vma->vm_pgoff;
1468                 count++;
1469         }
1470
1471         /* Now we know exact count of files, can store it */
1472         data[0] = count;
1473         data[1] = PAGE_SIZE;
1474         /*
1475          * Count usually is less than current->mm->map_count,
1476          * we need to move filenames down.
1477          */
1478         n = current->mm->map_count - count;
1479         if (n != 0) {
1480                 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1481                 memmove(name_base - shift_bytes, name_base,
1482                         name_curpos - name_base);
1483                 name_curpos -= shift_bytes;
1484         }
1485
1486         size = name_curpos - (char *)data;
1487         fill_note(note, "CORE", NT_FILE, size, data);
1488         return 0;
1489 }
1490
1491 #ifdef CORE_DUMP_USE_REGSET
1492 #include <linux/regset.h>
1493
1494 struct elf_thread_core_info {
1495         struct elf_thread_core_info *next;
1496         struct task_struct *task;
1497         struct elf_prstatus prstatus;
1498         struct memelfnote notes[0];
1499 };
1500
1501 struct elf_note_info {
1502         struct elf_thread_core_info *thread;
1503         struct memelfnote psinfo;
1504         struct memelfnote signote;
1505         struct memelfnote auxv;
1506         struct memelfnote files;
1507         user_siginfo_t csigdata;
1508         size_t size;
1509         int thread_notes;
1510 };
1511
1512 /*
1513  * When a regset has a writeback hook, we call it on each thread before
1514  * dumping user memory.  On register window machines, this makes sure the
1515  * user memory backing the register data is up to date before we read it.
1516  */
1517 static void do_thread_regset_writeback(struct task_struct *task,
1518                                        const struct user_regset *regset)
1519 {
1520         if (regset->writeback)
1521                 regset->writeback(task, regset, 1);
1522 }
1523
1524 #ifndef PR_REG_SIZE
1525 #define PR_REG_SIZE(S) sizeof(S)
1526 #endif
1527
1528 #ifndef PRSTATUS_SIZE
1529 #define PRSTATUS_SIZE(S) sizeof(S)
1530 #endif
1531
1532 #ifndef PR_REG_PTR
1533 #define PR_REG_PTR(S) (&((S)->pr_reg))
1534 #endif
1535
1536 #ifndef SET_PR_FPVALID
1537 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1538 #endif
1539
1540 static int fill_thread_core_info(struct elf_thread_core_info *t,
1541                                  const struct user_regset_view *view,
1542                                  long signr, size_t *total)
1543 {
1544         unsigned int i;
1545
1546         /*
1547          * NT_PRSTATUS is the one special case, because the regset data
1548          * goes into the pr_reg field inside the note contents, rather
1549          * than being the whole note contents.  We fill the reset in here.
1550          * We assume that regset 0 is NT_PRSTATUS.
1551          */
1552         fill_prstatus(&t->prstatus, t->task, signr);
1553         (void) view->regsets[0].get(t->task, &view->regsets[0],
1554                                     0, PR_REG_SIZE(t->prstatus.pr_reg),
1555                                     PR_REG_PTR(&t->prstatus), NULL);
1556
1557         fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1558                   PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1559         *total += notesize(&t->notes[0]);
1560
1561         do_thread_regset_writeback(t->task, &view->regsets[0]);
1562
1563         /*
1564          * Each other regset might generate a note too.  For each regset
1565          * that has no core_note_type or is inactive, we leave t->notes[i]
1566          * all zero and we'll know to skip writing it later.
1567          */
1568         for (i = 1; i < view->n; ++i) {
1569                 const struct user_regset *regset = &view->regsets[i];
1570                 do_thread_regset_writeback(t->task, regset);
1571                 if (regset->core_note_type && regset->get &&
1572                     (!regset->active || regset->active(t->task, regset))) {
1573                         int ret;
1574                         size_t size = regset->n * regset->size;
1575                         void *data = kmalloc(size, GFP_KERNEL);
1576                         if (unlikely(!data))
1577                                 return 0;
1578                         ret = regset->get(t->task, regset,
1579                                           0, size, data, NULL);
1580                         if (unlikely(ret))
1581                                 kfree(data);
1582                         else {
1583                                 if (regset->core_note_type != NT_PRFPREG)
1584                                         fill_note(&t->notes[i], "LINUX",
1585                                                   regset->core_note_type,
1586                                                   size, data);
1587                                 else {
1588                                         SET_PR_FPVALID(&t->prstatus, 1);
1589                                         fill_note(&t->notes[i], "CORE",
1590                                                   NT_PRFPREG, size, data);
1591                                 }
1592                                 *total += notesize(&t->notes[i]);
1593                         }
1594                 }
1595         }
1596
1597         return 1;
1598 }
1599
1600 static int fill_note_info(struct elfhdr *elf, int phdrs,
1601                           struct elf_note_info *info,
1602                           siginfo_t *siginfo, struct pt_regs *regs)
1603 {
1604         struct task_struct *dump_task = current;
1605         const struct user_regset_view *view = task_user_regset_view(dump_task);
1606         struct elf_thread_core_info *t;
1607         struct elf_prpsinfo *psinfo;
1608         struct core_thread *ct;
1609         unsigned int i;
1610
1611         info->size = 0;
1612         info->thread = NULL;
1613
1614         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1615         if (psinfo == NULL) {
1616                 info->psinfo.data = NULL; /* So we don't free this wrongly */
1617                 return 0;
1618         }
1619
1620         fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1621
1622         /*
1623          * Figure out how many notes we're going to need for each thread.
1624          */
1625         info->thread_notes = 0;
1626         for (i = 0; i < view->n; ++i)
1627                 if (view->regsets[i].core_note_type != 0)
1628                         ++info->thread_notes;
1629
1630         /*
1631          * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1632          * since it is our one special case.
1633          */
1634         if (unlikely(info->thread_notes == 0) ||
1635             unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1636                 WARN_ON(1);
1637                 return 0;
1638         }
1639
1640         /*
1641          * Initialize the ELF file header.
1642          */
1643         fill_elf_header(elf, phdrs,
1644                         view->e_machine, view->e_flags);
1645
1646         /*
1647          * Allocate a structure for each thread.
1648          */
1649         for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1650                 t = kzalloc(offsetof(struct elf_thread_core_info,
1651                                      notes[info->thread_notes]),
1652                             GFP_KERNEL);
1653                 if (unlikely(!t))
1654                         return 0;
1655
1656                 t->task = ct->task;
1657                 if (ct->task == dump_task || !info->thread) {
1658                         t->next = info->thread;
1659                         info->thread = t;
1660                 } else {
1661                         /*
1662                          * Make sure to keep the original task at
1663                          * the head of the list.
1664                          */
1665                         t->next = info->thread->next;
1666                         info->thread->next = t;
1667                 }
1668         }
1669
1670         /*
1671          * Now fill in each thread's information.
1672          */
1673         for (t = info->thread; t != NULL; t = t->next)
1674                 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1675                         return 0;
1676
1677         /*
1678          * Fill in the two process-wide notes.
1679          */
1680         fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1681         info->size += notesize(&info->psinfo);
1682
1683         fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1684         info->size += notesize(&info->signote);
1685
1686         fill_auxv_note(&info->auxv, current->mm);
1687         info->size += notesize(&info->auxv);
1688
1689         if (fill_files_note(&info->files) == 0)
1690                 info->size += notesize(&info->files);
1691
1692         return 1;
1693 }
1694
1695 static size_t get_note_info_size(struct elf_note_info *info)
1696 {
1697         return info->size;
1698 }
1699
1700 /*
1701  * Write all the notes for each thread.  When writing the first thread, the
1702  * process-wide notes are interleaved after the first thread-specific note.
1703  */
1704 static int write_note_info(struct elf_note_info *info,
1705                            struct file *file, loff_t *foffset)
1706 {
1707         bool first = 1;
1708         struct elf_thread_core_info *t = info->thread;
1709
1710         do {
1711                 int i;
1712
1713                 if (!writenote(&t->notes[0], file, foffset))
1714                         return 0;
1715
1716                 if (first && !writenote(&info->psinfo, file, foffset))
1717                         return 0;
1718                 if (first && !writenote(&info->signote, file, foffset))
1719                         return 0;
1720                 if (first && !writenote(&info->auxv, file, foffset))
1721                         return 0;
1722                 if (first && info->files.data &&
1723                                 !writenote(&info->files, file, foffset))
1724                         return 0;
1725
1726                 for (i = 1; i < info->thread_notes; ++i)
1727                         if (t->notes[i].data &&
1728                             !writenote(&t->notes[i], file, foffset))
1729                                 return 0;
1730
1731                 first = 0;
1732                 t = t->next;
1733         } while (t);
1734
1735         return 1;
1736 }
1737
1738 static void free_note_info(struct elf_note_info *info)
1739 {
1740         struct elf_thread_core_info *threads = info->thread;
1741         while (threads) {
1742                 unsigned int i;
1743                 struct elf_thread_core_info *t = threads;
1744                 threads = t->next;
1745                 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1746                 for (i = 1; i < info->thread_notes; ++i)
1747                         kfree(t->notes[i].data);
1748                 kfree(t);
1749         }
1750         kfree(info->psinfo.data);
1751         vfree(info->files.data);
1752 }
1753
1754 #else
1755
1756 /* Here is the structure in which status of each thread is captured. */
1757 struct elf_thread_status
1758 {
1759         struct list_head list;
1760         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1761         elf_fpregset_t fpu;             /* NT_PRFPREG */
1762         struct task_struct *thread;
1763 #ifdef ELF_CORE_COPY_XFPREGS
1764         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1765 #endif
1766         struct memelfnote notes[3];
1767         int num_notes;
1768 };
1769
1770 /*
1771  * In order to add the specific thread information for the elf file format,
1772  * we need to keep a linked list of every threads pr_status and then create
1773  * a single section for them in the final core file.
1774  */
1775 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1776 {
1777         int sz = 0;
1778         struct task_struct *p = t->thread;
1779         t->num_notes = 0;
1780
1781         fill_prstatus(&t->prstatus, p, signr);
1782         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1783
1784         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1785                   &(t->prstatus));
1786         t->num_notes++;
1787         sz += notesize(&t->notes[0]);
1788
1789         if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1790                                                                 &t->fpu))) {
1791                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1792                           &(t->fpu));
1793                 t->num_notes++;
1794                 sz += notesize(&t->notes[1]);
1795         }
1796
1797 #ifdef ELF_CORE_COPY_XFPREGS
1798         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1799                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1800                           sizeof(t->xfpu), &t->xfpu);
1801                 t->num_notes++;
1802                 sz += notesize(&t->notes[2]);
1803         }
1804 #endif
1805         return sz;
1806 }
1807
1808 struct elf_note_info {
1809         struct memelfnote *notes;
1810         struct memelfnote *notes_files;
1811         struct elf_prstatus *prstatus;  /* NT_PRSTATUS */
1812         struct elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
1813         struct list_head thread_list;
1814         elf_fpregset_t *fpu;
1815 #ifdef ELF_CORE_COPY_XFPREGS
1816         elf_fpxregset_t *xfpu;
1817 #endif
1818         user_siginfo_t csigdata;
1819         int thread_status_size;
1820         int numnote;
1821 };
1822
1823 static int elf_note_info_init(struct elf_note_info *info)
1824 {
1825         memset(info, 0, sizeof(*info));
1826         INIT_LIST_HEAD(&info->thread_list);
1827
1828         /* Allocate space for ELF notes */
1829         info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1830         if (!info->notes)
1831                 return 0;
1832         info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1833         if (!info->psinfo)
1834                 return 0;
1835         info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1836         if (!info->prstatus)
1837                 return 0;
1838         info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1839         if (!info->fpu)
1840                 return 0;
1841 #ifdef ELF_CORE_COPY_XFPREGS
1842         info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1843         if (!info->xfpu)
1844                 return 0;
1845 #endif
1846         return 1;
1847 }
1848
1849 static int fill_note_info(struct elfhdr *elf, int phdrs,
1850                           struct elf_note_info *info,
1851                           siginfo_t *siginfo, struct pt_regs *regs)
1852 {
1853         struct list_head *t;
1854
1855         if (!elf_note_info_init(info))
1856                 return 0;
1857
1858         if (siginfo->si_signo) {
1859                 struct core_thread *ct;
1860                 struct elf_thread_status *ets;
1861
1862                 for (ct = current->mm->core_state->dumper.next;
1863                                                 ct; ct = ct->next) {
1864                         ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1865                         if (!ets)
1866                                 return 0;
1867
1868                         ets->thread = ct->task;
1869                         list_add(&ets->list, &info->thread_list);
1870                 }
1871
1872                 list_for_each(t, &info->thread_list) {
1873                         int sz;
1874
1875                         ets = list_entry(t, struct elf_thread_status, list);
1876                         sz = elf_dump_thread_status(siginfo->si_signo, ets);
1877                         info->thread_status_size += sz;
1878                 }
1879         }
1880         /* now collect the dump for the current */
1881         memset(info->prstatus, 0, sizeof(*info->prstatus));
1882         fill_prstatus(info->prstatus, current, siginfo->si_signo);
1883         elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1884
1885         /* Set up header */
1886         fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
1887
1888         /*
1889          * Set up the notes in similar form to SVR4 core dumps made
1890          * with info from their /proc.
1891          */
1892
1893         fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1894                   sizeof(*info->prstatus), info->prstatus);
1895         fill_psinfo(info->psinfo, current->group_leader, current->mm);
1896         fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1897                   sizeof(*info->psinfo), info->psinfo);
1898
1899         fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1900         fill_auxv_note(info->notes + 3, current->mm);
1901         info->numnote = 4;
1902
1903         if (fill_files_note(info->notes + info->numnote) == 0) {
1904                 info->notes_files = info->notes + info->numnote;
1905                 info->numnote++;
1906         }
1907
1908         /* Try to dump the FPU. */
1909         info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1910                                                                info->fpu);
1911         if (info->prstatus->pr_fpvalid)
1912                 fill_note(info->notes + info->numnote++,
1913                           "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1914 #ifdef ELF_CORE_COPY_XFPREGS
1915         if (elf_core_copy_task_xfpregs(current, info->xfpu))
1916                 fill_note(info->notes + info->numnote++,
1917                           "LINUX", ELF_CORE_XFPREG_TYPE,
1918                           sizeof(*info->xfpu), info->xfpu);
1919 #endif
1920
1921         return 1;
1922 }
1923
1924 static size_t get_note_info_size(struct elf_note_info *info)
1925 {
1926         int sz = 0;
1927         int i;
1928
1929         for (i = 0; i < info->numnote; i++)
1930                 sz += notesize(info->notes + i);
1931
1932         sz += info->thread_status_size;
1933
1934         return sz;
1935 }
1936
1937 static int write_note_info(struct elf_note_info *info,
1938                            struct file *file, loff_t *foffset)
1939 {
1940         int i;
1941         struct list_head *t;
1942
1943         for (i = 0; i < info->numnote; i++)
1944                 if (!writenote(info->notes + i, file, foffset))
1945                         return 0;
1946
1947         /* write out the thread status notes section */
1948         list_for_each(t, &info->thread_list) {
1949                 struct elf_thread_status *tmp =
1950                                 list_entry(t, struct elf_thread_status, list);
1951
1952                 for (i = 0; i < tmp->num_notes; i++)
1953                         if (!writenote(&tmp->notes[i], file, foffset))
1954                                 return 0;
1955         }
1956
1957         return 1;
1958 }
1959
1960 static void free_note_info(struct elf_note_info *info)
1961 {
1962         while (!list_empty(&info->thread_list)) {
1963                 struct list_head *tmp = info->thread_list.next;
1964                 list_del(tmp);
1965                 kfree(list_entry(tmp, struct elf_thread_status, list));
1966         }
1967
1968         /* Free data possibly allocated by fill_files_note(): */
1969         if (info->notes_files)
1970                 vfree(info->notes_files->data);
1971
1972         kfree(info->prstatus);
1973         kfree(info->psinfo);
1974         kfree(info->notes);
1975         kfree(info->fpu);
1976 #ifdef ELF_CORE_COPY_XFPREGS
1977         kfree(info->xfpu);
1978 #endif
1979 }
1980
1981 #endif
1982
1983 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1984                                         struct vm_area_struct *gate_vma)
1985 {
1986         struct vm_area_struct *ret = tsk->mm->mmap;
1987
1988         if (ret)
1989                 return ret;
1990         return gate_vma;
1991 }
1992 /*
1993  * Helper function for iterating across a vma list.  It ensures that the caller
1994  * will visit `gate_vma' prior to terminating the search.
1995  */
1996 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1997                                         struct vm_area_struct *gate_vma)
1998 {
1999         struct vm_area_struct *ret;
2000
2001         ret = this_vma->vm_next;
2002         if (ret)
2003                 return ret;
2004         if (this_vma == gate_vma)
2005                 return NULL;
2006         return gate_vma;
2007 }
2008
2009 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
2010                              elf_addr_t e_shoff, int segs)
2011 {
2012         elf->e_shoff = e_shoff;
2013         elf->e_shentsize = sizeof(*shdr4extnum);
2014         elf->e_shnum = 1;
2015         elf->e_shstrndx = SHN_UNDEF;
2016
2017         memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2018
2019         shdr4extnum->sh_type = SHT_NULL;
2020         shdr4extnum->sh_size = elf->e_shnum;
2021         shdr4extnum->sh_link = elf->e_shstrndx;
2022         shdr4extnum->sh_info = segs;
2023 }
2024
2025 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
2026                                      unsigned long mm_flags)
2027 {
2028         struct vm_area_struct *vma;
2029         size_t size = 0;
2030
2031         for (vma = first_vma(current, gate_vma); vma != NULL;
2032              vma = next_vma(vma, gate_vma))
2033                 size += vma_dump_size(vma, mm_flags);
2034         return size;
2035 }
2036
2037 /*
2038  * Actual dumper
2039  *
2040  * This is a two-pass process; first we find the offsets of the bits,
2041  * and then they are actually written out.  If we run out of core limit
2042  * we just truncate.
2043  */
2044 static int elf_core_dump(struct coredump_params *cprm)
2045 {
2046         int has_dumped = 0;
2047         mm_segment_t fs;
2048         int segs;
2049         size_t size = 0;
2050         struct vm_area_struct *vma, *gate_vma;
2051         struct elfhdr *elf = NULL;
2052         loff_t offset = 0, dataoff, foffset;
2053         struct elf_note_info info = { };
2054         struct elf_phdr *phdr4note = NULL;
2055         struct elf_shdr *shdr4extnum = NULL;
2056         Elf_Half e_phnum;
2057         elf_addr_t e_shoff;
2058
2059         /*
2060          * We no longer stop all VM operations.
2061          *
2062          * This is because those proceses that could possibly change map_count
2063          * or the mmap / vma pages are now blocked in do_exit on current
2064          * finishing this core dump.
2065          *
2066          * Only ptrace can touch these memory addresses, but it doesn't change
2067          * the map_count or the pages allocated. So no possibility of crashing
2068          * exists while dumping the mm->vm_next areas to the core file.
2069          */
2070
2071         /* alloc memory for large data structures: too large to be on stack */
2072         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2073         if (!elf)
2074                 goto out;
2075         /*
2076          * The number of segs are recored into ELF header as 16bit value.
2077          * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2078          */
2079         segs = current->mm->map_count;
2080         segs += elf_core_extra_phdrs();
2081
2082         gate_vma = get_gate_vma(current->mm);
2083         if (gate_vma != NULL)
2084                 segs++;
2085
2086         /* for notes section */
2087         segs++;
2088
2089         /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2090          * this, kernel supports extended numbering. Have a look at
2091          * include/linux/elf.h for further information. */
2092         e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2093
2094         /*
2095          * Collect all the non-memory information about the process for the
2096          * notes.  This also sets up the file header.
2097          */
2098         if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2099                 goto cleanup;
2100
2101         has_dumped = 1;
2102
2103         fs = get_fs();
2104         set_fs(KERNEL_DS);
2105
2106         offset += sizeof(*elf);                         /* Elf header */
2107         offset += segs * sizeof(struct elf_phdr);       /* Program headers */
2108         foffset = offset;
2109
2110         /* Write notes phdr entry */
2111         {
2112                 size_t sz = get_note_info_size(&info);
2113
2114                 sz += elf_coredump_extra_notes_size();
2115
2116                 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2117                 if (!phdr4note)
2118                         goto end_coredump;
2119
2120                 fill_elf_note_phdr(phdr4note, sz, offset);
2121                 offset += sz;
2122         }
2123
2124         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2125
2126         offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2127         offset += elf_core_extra_data_size();
2128         e_shoff = offset;
2129
2130         if (e_phnum == PN_XNUM) {
2131                 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2132                 if (!shdr4extnum)
2133                         goto end_coredump;
2134                 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2135         }
2136
2137         offset = dataoff;
2138
2139         size += sizeof(*elf);
2140         if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
2141                 goto end_coredump;
2142
2143         size += sizeof(*phdr4note);
2144         if (size > cprm->limit
2145             || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
2146                 goto end_coredump;
2147
2148         /* Write program headers for segments dump */
2149         for (vma = first_vma(current, gate_vma); vma != NULL;
2150                         vma = next_vma(vma, gate_vma)) {
2151                 struct elf_phdr phdr;
2152
2153                 phdr.p_type = PT_LOAD;
2154                 phdr.p_offset = offset;
2155                 phdr.p_vaddr = vma->vm_start;
2156                 phdr.p_paddr = 0;
2157                 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2158                 phdr.p_memsz = vma->vm_end - vma->vm_start;
2159                 offset += phdr.p_filesz;
2160                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2161                 if (vma->vm_flags & VM_WRITE)
2162                         phdr.p_flags |= PF_W;
2163                 if (vma->vm_flags & VM_EXEC)
2164                         phdr.p_flags |= PF_X;
2165                 phdr.p_align = ELF_EXEC_PAGESIZE;
2166
2167                 size += sizeof(phdr);
2168                 if (size > cprm->limit
2169                     || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2170                         goto end_coredump;
2171         }
2172
2173         if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2174                 goto end_coredump;
2175
2176         /* write out the notes section */
2177         if (!write_note_info(&info, cprm->file, &foffset))
2178                 goto end_coredump;
2179
2180         if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2181                 goto end_coredump;
2182
2183         /* Align to page */
2184         if (!dump_seek(cprm->file, dataoff - foffset))
2185                 goto end_coredump;
2186
2187         for (vma = first_vma(current, gate_vma); vma != NULL;
2188                         vma = next_vma(vma, gate_vma)) {
2189                 unsigned long addr;
2190                 unsigned long end;
2191
2192                 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2193
2194                 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2195                         struct page *page;
2196                         int stop;
2197
2198                         page = get_dump_page(addr);
2199                         if (page) {
2200                                 void *kaddr = kmap(page);
2201                                 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2202                                         !dump_write(cprm->file, kaddr,
2203                                                     PAGE_SIZE);
2204                                 kunmap(page);
2205                                 page_cache_release(page);
2206                         } else
2207                                 stop = !dump_seek(cprm->file, PAGE_SIZE);
2208                         if (stop)
2209                                 goto end_coredump;
2210                 }
2211         }
2212
2213         if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2214                 goto end_coredump;
2215
2216         if (e_phnum == PN_XNUM) {
2217                 size += sizeof(*shdr4extnum);
2218                 if (size > cprm->limit
2219                     || !dump_write(cprm->file, shdr4extnum,
2220                                    sizeof(*shdr4extnum)))
2221                         goto end_coredump;
2222         }
2223
2224 end_coredump:
2225         set_fs(fs);
2226
2227 cleanup:
2228         free_note_info(&info);
2229         kfree(shdr4extnum);
2230         kfree(phdr4note);
2231         kfree(elf);
2232 out:
2233         return has_dumped;
2234 }
2235
2236 #endif          /* CONFIG_ELF_CORE */
2237
2238 static int __init init_elf_binfmt(void)
2239 {
2240         register_binfmt(&elf_format);
2241         return 0;
2242 }
2243
2244 static void __exit exit_elf_binfmt(void)
2245 {
2246         /* Remove the COFF and ELF loaders. */
2247         unregister_binfmt(&elf_format);
2248 }
2249
2250 core_initcall(init_elf_binfmt);
2251 module_exit(exit_elf_binfmt);
2252 MODULE_LICENSE("GPL");