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 <asm/uaccess.h>
  37 #include <asm/param.h>
  38 #include <asm/page.h>
  39
  40 #ifndef user_long_t
  41 #define user_long_t long
  42 #endif
  43 #ifndef user_siginfo_t
  44 #define user_siginfo_t siginfo_t
  45 #endif
  46
  47 static int load_elf_binary(struct linux_binprm *bprm);
  48 static int load_elf_library(struct file *);
  49 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
  50                                 int, int, unsigned long);
  51
  52 /*
  53  * If we don't support core dumping, then supply a NULL so we
  54  * don't even try.
  55  */
  56 #ifdef CONFIG_ELF_CORE
  57 static int elf_core_dump(struct coredump_params *cprm);
  58 #else
  59 #define elf_core_dump   NULL
  60 #endif
  61
  62 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
  63 #define ELF_MIN_ALIGN   ELF_EXEC_PAGESIZE
  64 #else
  65 #define ELF_MIN_ALIGN   PAGE_SIZE
  66 #endif
  67
  68 #ifndef ELF_CORE_EFLAGS
  69 #define ELF_CORE_EFLAGS 0
  70 #endif
  71
  72 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
  73 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
  74 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
  75
  76 static struct linux_binfmt elf_format = {
  77         .module         = THIS_MODULE,
  78         .load_binary    = load_elf_binary,
  79         .load_shlib     = load_elf_library,
  80         .core_dump      = elf_core_dump,
  81         .min_coredump   = ELF_EXEC_PAGESIZE,
  82 };
  83
  84 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
  85
  86 static int set_brk(unsigned long start, unsigned long end)
  87 {
  88         start = ELF_PAGEALIGN(start);
  89         end = ELF_PAGEALIGN(end);
  90         if (end > start) {
  91                 unsigned long addr;
  92                 addr = vm_brk(start, end - start);
  93                 if (BAD_ADDR(addr))
  94                         return addr;
  95         }
  96         current->mm->start_brk = current->mm->brk = end;
  97         return 0;
  98 }
  99
 100 /* We need to explicitly zero any fractional pages
 101    after the data section (i.e. bss).  This would
 102    contain the junk from the file that should not
 103    be in memory
 104  */
 105 static int padzero(unsigned long elf_bss)
 106 {
 107         unsigned long nbyte;
 108
 109         nbyte = ELF_PAGEOFFSET(elf_bss);
 110         if (nbyte) {
 111                 nbyte = ELF_MIN_ALIGN - nbyte;
 112                 if (clear_user((void __user *) elf_bss, nbyte))
 113                         return -EFAULT;
 114         }
 115         return 0;
 116 }
 117
 118 /* Let's use some macros to make this stack manipulation a little clearer */
 119 #ifdef CONFIG_STACK_GROWSUP
 120 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
 121 #define STACK_ROUND(sp, items) \
 122         ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
 123 #define STACK_ALLOC(sp, len) ({ \
 124         elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
 125         old_sp; })
 126 #else
 127 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
 128 #define STACK_ROUND(sp, items) \
 129         (((unsigned long) (sp - items)) &~ 15UL)
 130 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
 131 #endif
 132
 133 #ifndef ELF_BASE_PLATFORM
 134 /*
 135  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
 136  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
 137  * will be copied to the user stack in the same manner as AT_PLATFORM.
 138  */
 139 #define ELF_BASE_PLATFORM NULL
 140 #endif
 141
 142 static int
 143 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
 144                 unsigned long load_addr, unsigned long interp_load_addr)
 145 {
 146         unsigned long p = bprm->p;
 147         int argc = bprm->argc;
 148         int envc = bprm->envc;
 149         elf_addr_t __user *argv;
 150         elf_addr_t __user *envp;
 151         elf_addr_t __user *sp;
 152         elf_addr_t __user *u_platform;
 153         elf_addr_t __user *u_base_platform;
 154         elf_addr_t __user *u_rand_bytes;
 155         const char *k_platform = ELF_PLATFORM;
 156         const char *k_base_platform = ELF_BASE_PLATFORM;
 157         unsigned char k_rand_bytes[16];
 158         int items;
 159         elf_addr_t *elf_info;
 160         int ei_index = 0;
 161         const struct cred *cred = current_cred();
 162         struct vm_area_struct *vma;
 163
 164         /*
 165          * In some cases (e.g. Hyper-Threading), we want to avoid L1
 166          * evictions by the processes running on the same package. One
 167          * thing we can do is to shuffle the initial stack for them.
 168          */
 169
 170         p = arch_align_stack(p);
 171
 172         /*
 173          * If this architecture has a platform capability string, copy it
 174          * to userspace.  In some cases (Sparc), this info is impossible
 175          * for userspace to get any other way, in others (i386) it is
 176          * merely difficult.
 177          */
 178         u_platform = NULL;
 179         if (k_platform) {
 180                 size_t len = strlen(k_platform) + 1;
 181
 182                 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 183                 if (__copy_to_user(u_platform, k_platform, len))
 184                         return -EFAULT;
 185         }
 186
 187         /*
 188          * If this architecture has a "base" platform capability
 189          * string, copy it to userspace.
 190          */
 191         u_base_platform = NULL;
 192         if (k_base_platform) {
 193                 size_t len = strlen(k_base_platform) + 1;
 194
 195                 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
 196                 if (__copy_to_user(u_base_platform, k_base_platform, len))
 197                         return -EFAULT;
 198         }
 199
 200         /*
 201          * Generate 16 random bytes for userspace PRNG seeding.
 202          */
 203         get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
 204         u_rand_bytes = (elf_addr_t __user *)
 205                        STACK_ALLOC(p, sizeof(k_rand_bytes));
 206         if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
 207                 return -EFAULT;
 208
 209         /* Create the ELF interpreter info */
 210         elf_info = (elf_addr_t *)current->mm->saved_auxv;
 211         /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
 212 #define NEW_AUX_ENT(id, val) \
 213         do { \
 214                 elf_info[ei_index++] = id; \
 215                 elf_info[ei_index++] = val; \
 216         } while (0)
 217
 218 #ifdef ARCH_DLINFO
 219         /*
 220          * ARCH_DLINFO must come first so PPC can do its special alignment of
 221          * AUXV.
 222          * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
 223          * ARCH_DLINFO changes
 224          */
 225         ARCH_DLINFO;
 226 #endif
 227         NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
 228         NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
 229         NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
 230         NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
 231         NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
 232         NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
 233         NEW_AUX_ENT(AT_BASE, interp_load_addr);
 234         NEW_AUX_ENT(AT_FLAGS, 0);
 235         NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
 236         NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
 237         NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
 238         NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
 239         NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
 240         NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
 241         NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
 242         NEW_AUX_ENT(AT_EXECFN, bprm->exec);
 243         if (k_platform) {
 244                 NEW_AUX_ENT(AT_PLATFORM,
 245                             (elf_addr_t)(unsigned long)u_platform);
 246         }
 247         if (k_base_platform) {
 248                 NEW_AUX_ENT(AT_BASE_PLATFORM,
 249                             (elf_addr_t)(unsigned long)u_base_platform);
 250         }
 251         if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
 252                 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
 253         }
 254 #undef NEW_AUX_ENT
 255         /* AT_NULL is zero; clear the rest too */
 256         memset(&elf_info[ei_index], 0,
 257                sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
 258
 259         /* And advance past the AT_NULL entry.  */
 260         ei_index += 2;
 261
 262         sp = STACK_ADD(p, ei_index);
 263
 264         items = (argc + 1) + (envc + 1) + 1;
 265         bprm->p = STACK_ROUND(sp, items);
 266
 267         /* Point sp at the lowest address on the stack */
 268 #ifdef CONFIG_STACK_GROWSUP
 269         sp = (elf_addr_t __user *)bprm->p - items - ei_index;
 270         bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
 271 #else
 272         sp = (elf_addr_t __user *)bprm->p;
 273 #endif
 274
 275
 276         /*
 277          * Grow the stack manually; some architectures have a limit on how
 278          * far ahead a user-space access may be in order to grow the stack.
 279          */
 280         vma = find_extend_vma(current->mm, bprm->p);
 281         if (!vma)
 282                 return -EFAULT;
 283
 284         /* Now, let's put argc (and argv, envp if appropriate) on the stack */
 285         if (__put_user(argc, sp++))
 286                 return -EFAULT;
 287         argv = sp;
 288         envp = argv + argc + 1;
 289
 290         /* Populate argv and envp */
 291         p = current->mm->arg_end = current->mm->arg_start;
 292         while (argc-- > 0) {
 293                 size_t len;
 294                 if (__put_user((elf_addr_t)p, argv++))
 295                         return -EFAULT;
 296                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 297                 if (!len || len > MAX_ARG_STRLEN)
 298                         return -EINVAL;
 299                 p += len;
 300         }
 301         if (__put_user(0, argv))
 302                 return -EFAULT;
 303         current->mm->arg_end = current->mm->env_start = p;
 304         while (envc-- > 0) {
 305                 size_t len;
 306                 if (__put_user((elf_addr_t)p, envp++))
 307                         return -EFAULT;
 308                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
 309                 if (!len || len > MAX_ARG_STRLEN)
 310                         return -EINVAL;
 311                 p += len;
 312         }
 313         if (__put_user(0, envp))
 314                 return -EFAULT;
 315         current->mm->env_end = p;
 316
 317         /* Put the elf_info on the stack in the right place.  */
 318         sp = (elf_addr_t __user *)envp + 1;
 319         if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
 320                 return -EFAULT;
 321         return 0;
 322 }
 323
 324 #ifndef elf_map
 325
 326 static unsigned long elf_map(struct file *filep, unsigned long addr,
 327                 struct elf_phdr *eppnt, int prot, int type,
 328                 unsigned long total_size)
 329 {
 330         unsigned long map_addr;
 331         unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
 332         unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
 333         addr = ELF_PAGESTART(addr);
 334         size = ELF_PAGEALIGN(size);
 335
 336         /* mmap() will return -EINVAL if given a zero size, but a
 337          * segment with zero filesize is perfectly valid */
 338         if (!size)
 339                 return addr;
 340
 341         /*
 342         * total_size is the size of the ELF (interpreter) image.
 343         * The _first_ mmap needs to know the full size, otherwise
 344         * randomization might put this image into an overlapping
 345         * position with the ELF binary image. (since size < total_size)
 346         * So we first map the 'big' image - and unmap the remainder at
 347         * the end. (which unmap is needed for ELF images with holes.)
 348         */
 349         if (total_size) {
 350                 total_size = ELF_PAGEALIGN(total_size);
 351                 map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
 352                 if (!BAD_ADDR(map_addr))
 353                         vm_munmap(map_addr+size, total_size-size);
 354         } else
 355                 map_addr = vm_mmap(filep, addr, size, prot, type, off);
 356
 357         return(map_addr);
 358 }
 359
 360 #endif /* !elf_map */
 361
 362 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
 363 {
 364         int i, first_idx = -1, last_idx = -1;
 365
 366         for (i = 0; i < nr; i++) {
 367                 if (cmds[i].p_type == PT_LOAD) {
 368                         last_idx = i;
 369                         if (first_idx == -1)
 370                                 first_idx = i;
 371                 }
 372         }
 373         if (first_idx == -1)
 374                 return 0;
 375
 376         return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
 377                                 ELF_PAGESTART(cmds[first_idx].p_vaddr);
 378 }
 379
 380
 381 /* This is much more generalized than the library routine read function,
 382    so we keep this separate.  Technically the library read function
 383    is only provided so that we can read a.out libraries that have
 384    an ELF header */
 385
 386 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
 387                 struct file *interpreter, unsigned long *interp_map_addr,
 388                 unsigned long no_base)
 389 {
 390         struct elf_phdr *elf_phdata;
 391         struct elf_phdr *eppnt;
 392         unsigned long load_addr = 0;
 393         int load_addr_set = 0;
 394         unsigned long last_bss = 0, elf_bss = 0;
 395         unsigned long error = ~0UL;
 396         unsigned long total_size;
 397         int retval, i, size;
 398
 399         /* First of all, some simple consistency checks */
 400         if (interp_elf_ex->e_type != ET_EXEC &&
 401             interp_elf_ex->e_type != ET_DYN)
 402                 goto out;
 403         if (!elf_check_arch(interp_elf_ex))
 404                 goto out;
 405         if (!interpreter->f_op || !interpreter->f_op->mmap)
 406                 goto out;
 407
 408         /*
 409          * If the size of this structure has changed, then punt, since
 410          * we will be doing the wrong thing.
 411          */
 412         if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
 413                 goto out;
 414         if (interp_elf_ex->e_phnum < 1 ||
 415                 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
 416                 goto out;
 417
 418         /* Now read in all of the header information */
 419         size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
 420         if (size > ELF_MIN_ALIGN)
 421                 goto out;
 422         elf_phdata = kmalloc(size, GFP_KERNEL);
 423         if (!elf_phdata)
 424                 goto out;
 425
 426         retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
 427                              (char *)elf_phdata, size);
 428         error = -EIO;
 429         if (retval != size) {
 430                 if (retval < 0)
 431                         error = retval;
 432                 goto out_close;
 433         }
 434
 435         total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
 436         if (!total_size) {
 437                 error = -EINVAL;
 438                 goto out_close;
 439         }
 440
 441         eppnt = elf_phdata;
 442         for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
 443                 if (eppnt->p_type == PT_LOAD) {
 444                         int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
 445                         int elf_prot = 0;
 446                         unsigned long vaddr = 0;
 447                         unsigned long k, map_addr;
 448
 449                         if (eppnt->p_flags & PF_R)
 450                                 elf_prot = PROT_READ;
 451                         if (eppnt->p_flags & PF_W)
 452                                 elf_prot |= PROT_WRITE;
 453                         if (eppnt->p_flags & PF_X)
 454                                 elf_prot |= PROT_EXEC;
 455                         vaddr = eppnt->p_vaddr;
 456                         if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
 457                                 elf_type |= MAP_FIXED;
 458                         else if (no_base && interp_elf_ex->e_type == ET_DYN)
 459                                 load_addr = -vaddr;
 460
 461                         map_addr = elf_map(interpreter, load_addr + vaddr,
 462                                         eppnt, elf_prot, elf_type, total_size);
 463                         total_size = 0;
 464                         if (!*interp_map_addr)
 465                                 *interp_map_addr = map_addr;
 466                         error = map_addr;
 467                         if (BAD_ADDR(map_addr))
 468                                 goto out_close;
 469
 470                         if (!load_addr_set &&
 471                             interp_elf_ex->e_type == ET_DYN) {
 472                                 load_addr = map_addr - ELF_PAGESTART(vaddr);
 473                                 load_addr_set = 1;
 474                         }
 475
 476                         /*
 477                          * Check to see if the section's size will overflow the
 478                          * allowed task size. Note that p_filesz must always be
 479                          * <= p_memsize so it's only necessary to check p_memsz.
 480                          */
 481                         k = load_addr + eppnt->p_vaddr;
 482                         if (BAD_ADDR(k) ||
 483                             eppnt->p_filesz > eppnt->p_memsz ||
 484                             eppnt->p_memsz > TASK_SIZE ||
 485                             TASK_SIZE - eppnt->p_memsz < k) {
 486                                 error = -ENOMEM;
 487                                 goto out_close;
 488                         }
 489
 490                         /*
 491                          * Find the end of the file mapping for this phdr, and
 492                          * keep track of the largest address we see for this.
 493                          */
 494                         k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
 495                         if (k > elf_bss)
 496                                 elf_bss = k;
 497
 498                         /*
 499                          * Do the same thing for the memory mapping - between
 500                          * elf_bss and last_bss is the bss section.
 501                          */
 502                         k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
 503                         if (k > last_bss)
 504                                 last_bss = k;
 505                 }
 506         }
 507
 508         if (last_bss > elf_bss) {
 509                 /*
 510                  * Now fill out the bss section.  First pad the last page up
 511                  * to the page boundary, and then perform a mmap to make sure
 512                  * that there are zero-mapped pages up to and including the
 513                  * last bss page.
 514                  */
 515                 if (padzero(elf_bss)) {
 516                         error = -EFAULT;
 517                         goto out_close;
 518                 }
 519
 520                 /* What we have mapped so far */
 521                 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
 522
 523                 /* Map the last of the bss segment */
 524                 error = vm_brk(elf_bss, last_bss - elf_bss);
 525                 if (BAD_ADDR(error))
 526                         goto out_close;
 527         }
 528
 529         error = load_addr;
 530
 531 out_close:
 532         kfree(elf_phdata);
 533 out:
 534         return error;
 535 }
 536
 537 /*
 538  * These are the functions used to load ELF style executables and shared
 539  * libraries.  There is no binary dependent code anywhere else.
 540  */
 541
 542 #define INTERPRETER_NONE 0
 543 #define INTERPRETER_ELF 2
 544
 545 #ifndef STACK_RND_MASK
 546 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))     /* 8MB of VA */
 547 #endif
 548
 549 static unsigned long randomize_stack_top(unsigned long stack_top)
 550 {
 551         unsigned int random_variable = 0;
 552
 553         if ((current->flags & PF_RANDOMIZE) &&
 554                 !(current->personality & ADDR_NO_RANDOMIZE)) {
 555                 random_variable = get_random_int() & STACK_RND_MASK;
 556                 random_variable <<= PAGE_SHIFT;
 557         }
 558 #ifdef CONFIG_STACK_GROWSUP
 559         return PAGE_ALIGN(stack_top) + random_variable;
 560 #else
 561         return PAGE_ALIGN(stack_top) - random_variable;
 562 #endif
 563 }
 564
 565 static int load_elf_binary(struct linux_binprm *bprm)
 566 {
 567         struct file *interpreter = NULL; /* to shut gcc up */
 568         unsigned long load_addr = 0, load_bias = 0;
 569         int load_addr_set = 0;
 570         char * elf_interpreter = NULL;
 571         unsigned long error;
 572         struct elf_phdr *elf_ppnt, *elf_phdata;
 573         unsigned long elf_bss, elf_brk;
 574         int retval, i;
 575         unsigned int size;
 576         unsigned long elf_entry;
 577         unsigned long interp_load_addr = 0;
 578         unsigned long start_code, end_code, start_data, end_data;
 579         unsigned long reloc_func_desc __maybe_unused = 0;
 580         int executable_stack = EXSTACK_DEFAULT;
 581         unsigned long def_flags = 0;
 582         struct pt_regs *regs = current_pt_regs();
 583         struct {
 584                 struct elfhdr elf_ex;
 585                 struct elfhdr interp_elf_ex;
 586         } *loc;
 587
 588         loc = kmalloc(sizeof(*loc), GFP_KERNEL);
 589         if (!loc) {
 590                 retval = -ENOMEM;
 591                 goto out_ret;
 592         }
 593
 594         /* Get the exec-header */
 595         loc->elf_ex = *((struct elfhdr *)bprm->buf);
 596
 597         retval = -ENOEXEC;
 598         /* First of all, some simple consistency checks */
 599         if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 600                 goto out;
 601
 602         if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
 603                 goto out;
 604         if (!elf_check_arch(&loc->elf_ex))
 605                 goto out;
 606         if (!bprm->file->f_op || !bprm->file->f_op->mmap)
 607                 goto out;
 608
 609         /* Now read in all of the header information */
 610         if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
 611                 goto out;
 612         if (loc->elf_ex.e_phnum < 1 ||
 613                 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
 614                 goto out;
 615         size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
 616         retval = -ENOMEM;
 617         elf_phdata = kmalloc(size, GFP_KERNEL);
 618         if (!elf_phdata)
 619                 goto out;
 620
 621         retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
 622                              (char *)elf_phdata, size);
 623         if (retval != size) {
 624                 if (retval >= 0)
 625                         retval = -EIO;
 626                 goto out_free_ph;
 627         }
 628
 629         elf_ppnt = elf_phdata;
 630         elf_bss = 0;
 631         elf_brk = 0;
 632
 633         start_code = ~0UL;
 634         end_code = 0;
 635         start_data = 0;
 636         end_data = 0;
 637
 638         for (i = 0; i < loc->elf_ex.e_phnum; i++) {
 639                 if (elf_ppnt->p_type == PT_INTERP) {
 640                         /* This is the program interpreter used for
 641                          * shared libraries - for now assume that this
 642                          * is an a.out format binary
 643                          */
 644                         retval = -ENOEXEC;
 645                         if (elf_ppnt->p_filesz > PATH_MAX ||
 646                             elf_ppnt->p_filesz < 2)
 647                                 goto out_free_ph;
 648
 649                         retval = -ENOMEM;
 650                         elf_interpreter = kmalloc(elf_ppnt->p_filesz,
 651                                                   GFP_KERNEL);
 652                         if (!elf_interpreter)
 653                                 goto out_free_ph;
 654
 655                         retval = kernel_read(bprm->file, elf_ppnt->p_offset,
 656                                              elf_interpreter,
 657                                              elf_ppnt->p_filesz);
 658                         if (retval != elf_ppnt->p_filesz) {
 659                                 if (retval >= 0)
 660                                         retval = -EIO;
 661                                 goto out_free_interp;
 662                         }
 663                         /* make sure path is NULL terminated */
 664                         retval = -ENOEXEC;
 665                         if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
 666                                 goto out_free_interp;
 667
 668                         interpreter = open_exec(elf_interpreter);
 669                         retval = PTR_ERR(interpreter);
 670                         if (IS_ERR(interpreter))
 671                                 goto out_free_interp;
 672
 673                         /*
 674                          * If the binary is not readable then enforce
 675                          * mm->dumpable = 0 regardless of the interpreter's
 676                          * permissions.
 677                          */
 678                         would_dump(bprm, interpreter);
 679
 680                         retval = kernel_read(interpreter, 0, bprm->buf,
 681                                              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_elf_ex = *((struct elfhdr *)bprm->buf);
 690                         break;
 691                 }
 692                 elf_ppnt++;
 693         }
 694
 695         elf_ppnt = elf_phdata;
 696         for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
 697                 if (elf_ppnt->p_type == PT_GNU_STACK) {
 698                         if (elf_ppnt->p_flags & PF_X)
 699                                 executable_stack = EXSTACK_ENABLE_X;
 700                         else
 701                                 executable_stack = EXSTACK_DISABLE_X;
 702                         break;
 703                 }
 704
 705         /* Some simple consistency checks for the interpreter */
 706         if (elf_interpreter) {
 707                 retval = -ELIBBAD;
 708                 /* Not an ELF interpreter */
 709                 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
 710                         goto out_free_dentry;
 711                 /* Verify the interpreter has a valid arch */
 712                 if (!elf_check_arch(&loc->interp_elf_ex))
 713                         goto out_free_dentry;
 714         }
 715
 716         /* Flush all traces of the currently running executable */
 717         retval = flush_old_exec(bprm);
 718         if (retval)
 719                 goto out_free_dentry;
 720
 721         /* OK, This is the point of no return */
 722         current->mm->def_flags = def_flags;
 723
 724         /* Do this immediately, since STACK_TOP as used in setup_arg_pages
 725            may depend on the personality.  */
 726         SET_PERSONALITY(loc->elf_ex);
 727         if (elf_read_implies_exec(loc->elf_ex, executable_stack))
 728                 current->personality |= READ_IMPLIES_EXEC;
 729
 730         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 731                 current->flags |= PF_RANDOMIZE;
 732
 733         setup_new_exec(bprm);
 734
 735         /* Do this so that we can load the interpreter, if need be.  We will
 736            change some of these later */
 737         current->mm->free_area_cache = current->mm->mmap_base;
 738         current->mm->cached_hole_size = 0;
 739         retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
 740                                  executable_stack);
 741         if (retval < 0) {
 742                 send_sig(SIGKILL, current, 0);
 743                 goto out_free_dentry;
 744         }
 745
 746         current->mm->start_stack = bprm->p;
 747
 748         /* Now we do a little grungy work by mmapping the ELF image into
 749            the correct location in memory. */
 750         for(i = 0, elf_ppnt = elf_phdata;
 751             i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
 752                 int elf_prot = 0, elf_flags;
 753                 unsigned long k, vaddr;
 754
 755                 if (elf_ppnt->p_type != PT_LOAD)
 756                         continue;
 757
 758                 if (unlikely (elf_brk > elf_bss)) {
 759                         unsigned long nbyte;
 760
 761                         /* There was a PT_LOAD segment with p_memsz > p_filesz
 762                            before this one. Map anonymous pages, if needed,
 763                            and clear the area.  */
 764                         retval = set_brk(elf_bss + load_bias,
 765                                          elf_brk + load_bias);
 766                         if (retval) {
 767                                 send_sig(SIGKILL, current, 0);
 768                                 goto out_free_dentry;
 769                         }
 770                         nbyte = ELF_PAGEOFFSET(elf_bss);
 771                         if (nbyte) {
 772                                 nbyte = ELF_MIN_ALIGN - nbyte;
 773                                 if (nbyte > elf_brk - elf_bss)
 774                                         nbyte = elf_brk - elf_bss;
 775                                 if (clear_user((void __user *)elf_bss +
 776                                                         load_bias, nbyte)) {
 777                                         /*
 778                                          * This bss-zeroing can fail if the ELF
 779                                          * file specifies odd protections. So
 780                                          * we don't check the return value
 781                                          */
 782                                 }
 783                         }
 784                 }
 785
 786                 if (elf_ppnt->p_flags & PF_R)
 787                         elf_prot |= PROT_READ;
 788                 if (elf_ppnt->p_flags & PF_W)
 789                         elf_prot |= PROT_WRITE;
 790                 if (elf_ppnt->p_flags & PF_X)
 791                         elf_prot |= PROT_EXEC;
 792
 793                 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
 794
 795                 vaddr = elf_ppnt->p_vaddr;
 796                 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
 797                         elf_flags |= MAP_FIXED;
 798                 } else if (loc->elf_ex.e_type == ET_DYN) {
 799                         /* Try and get dynamic programs out of the way of the
 800                          * default mmap base, as well as whatever program they
 801                          * might try to exec.  This is because the brk will
 802                          * follow the loader, and is not movable.  */
 803 #ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
 804                         /* Memory randomization might have been switched off
 805                          * in runtime via sysctl.
 806                          * If that is the case, retain the original non-zero
 807                          * load_bias value in order to establish proper
 808                          * non-randomized mappings.
 809                          */
 810                         if (current->flags & PF_RANDOMIZE)
 811                                 load_bias = 0;
 812                         else
 813                                 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 814 #else
 815                         load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
 816 #endif
 817                 }
 818
 819                 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
 820                                 elf_prot, elf_flags, 0);
 821                 if (BAD_ADDR(error)) {
 822                         send_sig(SIGKILL, current, 0);
 823                         retval = IS_ERR((void *)error) ?
 824                                 PTR_ERR((void*)error) : -EINVAL;
 825                         goto out_free_dentry;
 826                 }
 827
 828                 if (!load_addr_set) {
 829                         load_addr_set = 1;
 830                         load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
 831                         if (loc->elf_ex.e_type == ET_DYN) {
 832                                 load_bias += error -
 833                                              ELF_PAGESTART(load_bias + vaddr);
 834                                 load_addr += load_bias;
 835                                 reloc_func_desc = load_bias;
 836                         }
 837                 }
 838                 k = elf_ppnt->p_vaddr;
 839                 if (k < start_code)
 840                         start_code = k;
 841                 if (start_data < k)
 842                         start_data = k;
 843
 844                 /*
 845                  * Check to see if the section's size will overflow the
 846                  * allowed task size. Note that p_filesz must always be
 847                  * <= p_memsz so it is only necessary to check p_memsz.
 848                  */
 849                 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
 850                     elf_ppnt->p_memsz > TASK_SIZE ||
 851                     TASK_SIZE - elf_ppnt->p_memsz < k) {
 852                         /* set_brk can never work. Avoid overflows. */
 853                         send_sig(SIGKILL, current, 0);
 854                         retval = -EINVAL;
 855                         goto out_free_dentry;
 856                 }
 857
 858                 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
 859
 860                 if (k > elf_bss)
 861                         elf_bss = k;
 862                 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
 863                         end_code = k;
 864                 if (end_data < k)
 865                         end_data = k;
 866                 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
 867                 if (k > elf_brk)
 868                         elf_brk = k;
 869         }
 870
 871         loc->elf_ex.e_entry += load_bias;
 872         elf_bss += load_bias;
 873         elf_brk += load_bias;
 874         start_code += load_bias;
 875         end_code += load_bias;
 876         start_data += load_bias;
 877         end_data += load_bias;
 878
 879         /* Calling set_brk effectively mmaps the pages that we need
 880          * for the bss and break sections.  We must do this before
 881          * mapping in the interpreter, to make sure it doesn't wind
 882          * up getting placed where the bss needs to go.
 883          */
 884         retval = set_brk(elf_bss, elf_brk);
 885         if (retval) {
 886                 send_sig(SIGKILL, current, 0);
 887                 goto out_free_dentry;
 888         }
 889         if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
 890                 send_sig(SIGSEGV, current, 0);
 891                 retval = -EFAULT; /* Nobody gets to see this, but.. */
 892                 goto out_free_dentry;
 893         }
 894
 895         if (elf_interpreter) {
 896                 unsigned long interp_map_addr = 0;
 897
 898                 elf_entry = load_elf_interp(&loc->interp_elf_ex,
 899                                             interpreter,
 900                                             &interp_map_addr,
 901                                             load_bias);
 902                 if (!IS_ERR((void *)elf_entry)) {
 903                         /*
 904                          * load_elf_interp() returns relocation
 905                          * adjustment
 906                          */
 907                         interp_load_addr = elf_entry;
 908                         elf_entry += loc->interp_elf_ex.e_entry;
 909                 }
 910                 if (BAD_ADDR(elf_entry)) {
 911                         force_sig(SIGSEGV, current);
 912                         retval = IS_ERR((void *)elf_entry) ?
 913                                         (int)elf_entry : -EINVAL;
 914                         goto out_free_dentry;
 915                 }
 916                 reloc_func_desc = interp_load_addr;
 917
 918                 allow_write_access(interpreter);
 919                 fput(interpreter);
 920                 kfree(elf_interpreter);
 921         } else {
 922                 elf_entry = loc->elf_ex.e_entry;
 923                 if (BAD_ADDR(elf_entry)) {
 924                         force_sig(SIGSEGV, current);
 925                         retval = -EINVAL;
 926                         goto out_free_dentry;
 927                 }
 928         }
 929
 930         kfree(elf_phdata);
 931
 932         set_binfmt(&elf_format);
 933
 934 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
 935         retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
 936         if (retval < 0) {
 937                 send_sig(SIGKILL, current, 0);
 938                 goto out;
 939         }
 940 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
 941
 942         install_exec_creds(bprm);
 943         retval = create_elf_tables(bprm, &loc->elf_ex,
 944                           load_addr, interp_load_addr);
 945         if (retval < 0) {
 946                 send_sig(SIGKILL, current, 0);
 947                 goto out;
 948         }
 949         /* N.B. passed_fileno might not be initialized? */
 950         current->mm->end_code = end_code;
 951         current->mm->start_code = start_code;
 952         current->mm->start_data = start_data;
 953         current->mm->end_data = end_data;
 954         current->mm->start_stack = bprm->p;
 955
 956 #ifdef arch_randomize_brk
 957         if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
 958                 current->mm->brk = current->mm->start_brk =
 959                         arch_randomize_brk(current->mm);
 960 #ifdef CONFIG_COMPAT_BRK
 961                 current->brk_randomized = 1;
 962 #endif
 963         }
 964 #endif
 965
 966         if (current->personality & MMAP_PAGE_ZERO) {
 967                 /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
 968                    and some applications "depend" upon this behavior.
 969                    Since we do not have the power to recompile these, we
 970                    emulate the SVr4 behavior. Sigh. */
 971                 error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
 972                                 MAP_FIXED | MAP_PRIVATE, 0);
 973         }
 974
 975 #ifdef ELF_PLAT_INIT
 976         /*
 977          * The ABI may specify that certain registers be set up in special
 978          * ways (on i386 %edx is the address of a DT_FINI function, for
 979          * example.  In addition, it may also specify (eg, PowerPC64 ELF)
 980          * that the e_entry field is the address of the function descriptor
 981          * for the startup routine, rather than the address of the startup
 982          * routine itself.  This macro performs whatever initialization to
 983          * the regs structure is required as well as any relocations to the
 984          * function descriptor entries when executing dynamically links apps.
 985          */
 986         ELF_PLAT_INIT(regs, reloc_func_desc);
 987 #endif
 988
 989         start_thread(regs, elf_entry, bprm->p);
 990         retval = 0;
 991 out:
 992         kfree(loc);
 993 out_ret:
 994         return retval;
 995
 996         /* error cleanup */
 997 out_free_dentry:
 998         allow_write_access(interpreter);
 999         if (interpreter)
1000                 fput(interpreter);
1001 out_free_interp:
1002         kfree(elf_interpreter);
1003 out_free_ph:
1004         kfree(elf_phdata);
1005         goto out;
1006 }
1007
1008 /* This is really simpleminded and specialized - we are loading an
1009    a.out library that is given an ELF header. */
1010 static int load_elf_library(struct file *file)
1011 {
1012         struct elf_phdr *elf_phdata;
1013         struct elf_phdr *eppnt;
1014         unsigned long elf_bss, bss, len;
1015         int retval, error, i, j;
1016         struct elfhdr elf_ex;
1017
1018         error = -ENOEXEC;
1019         retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1020         if (retval != sizeof(elf_ex))
1021                 goto out;
1022
1023         if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1024                 goto out;
1025
1026         /* First of all, some simple consistency checks */
1027         if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1028             !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1029                 goto out;
1030
1031         /* Now read in all of the header information */
1032
1033         j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1034         /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1035
1036         error = -ENOMEM;
1037         elf_phdata = kmalloc(j, GFP_KERNEL);
1038         if (!elf_phdata)
1039                 goto out;
1040
1041         eppnt = elf_phdata;
1042         error = -ENOEXEC;
1043         retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1044         if (retval != j)
1045                 goto out_free_ph;
1046
1047         for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1048                 if ((eppnt + i)->p_type == PT_LOAD)
1049                         j++;
1050         if (j != 1)
1051                 goto out_free_ph;
1052
1053         while (eppnt->p_type != PT_LOAD)
1054                 eppnt++;
1055
1056         /* Now use mmap to map the library into memory. */
1057         error = vm_mmap(file,
1058                         ELF_PAGESTART(eppnt->p_vaddr),
1059                         (eppnt->p_filesz +
1060                          ELF_PAGEOFFSET(eppnt->p_vaddr)),
1061                         PROT_READ | PROT_WRITE | PROT_EXEC,
1062                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1063                         (eppnt->p_offset -
1064                          ELF_PAGEOFFSET(eppnt->p_vaddr)));
1065         if (error != ELF_PAGESTART(eppnt->p_vaddr))
1066                 goto out_free_ph;
1067
1068         elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1069         if (padzero(elf_bss)) {
1070                 error = -EFAULT;
1071                 goto out_free_ph;
1072         }
1073
1074         len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1075                             ELF_MIN_ALIGN - 1);
1076         bss = eppnt->p_memsz + eppnt->p_vaddr;
1077         if (bss > len)
1078                 vm_brk(len, bss - len);
1079         error = 0;
1080
1081 out_free_ph:
1082         kfree(elf_phdata);
1083 out:
1084         return error;
1085 }
1086
1087 #ifdef CONFIG_ELF_CORE
1088 /*
1089  * ELF core dumper
1090  *
1091  * Modelled on fs/exec.c:aout_core_dump()
1092  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1093  */
1094
1095 /*
1096  * The purpose of always_dump_vma() is to make sure that special kernel mappings
1097  * that are useful for post-mortem analysis are included in every core dump.
1098  * In that way we ensure that the core dump is fully interpretable later
1099  * without matching up the same kernel and hardware config to see what PC values
1100  * meant. These special mappings include - vDSO, vsyscall, and other
1101  * architecture specific mappings
1102  */
1103 static bool always_dump_vma(struct vm_area_struct *vma)
1104 {
1105         /* Any vsyscall mappings? */
1106         if (vma == get_gate_vma(vma->vm_mm))
1107                 return true;
1108         /*
1109          * arch_vma_name() returns non-NULL for special architecture mappings,
1110          * such as vDSO sections.
1111          */
1112         if (arch_vma_name(vma))
1113                 return true;
1114
1115         return false;
1116 }
1117
1118 /*
1119  * Decide what to dump of a segment, part, all or none.
1120  */
1121 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1122                                    unsigned long mm_flags)
1123 {
1124 #define FILTER(type)    (mm_flags & (1UL << MMF_DUMP_##type))
1125
1126         /* always dump the vdso and vsyscall sections */
1127         if (always_dump_vma(vma))
1128                 goto whole;
1129
1130         if (vma->vm_flags & VM_DONTDUMP)
1131                 return 0;
1132
1133         /* Hugetlb memory check */
1134         if (vma->vm_flags & VM_HUGETLB) {
1135                 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1136                         goto whole;
1137                 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1138                         goto whole;
1139         }
1140
1141         /* Do not dump I/O mapped devices or special mappings */
1142         if (vma->vm_flags & VM_IO)
1143                 return 0;
1144
1145         /* By default, dump shared memory if mapped from an anonymous file. */
1146         if (vma->vm_flags & VM_SHARED) {
1147                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1148                     FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1149                         goto whole;
1150                 return 0;
1151         }
1152
1153         /* Dump segments that have been written to.  */
1154         if (vma->anon_vma && FILTER(ANON_PRIVATE))
1155                 goto whole;
1156         if (vma->vm_file == NULL)
1157                 return 0;
1158
1159         if (FILTER(MAPPED_PRIVATE))
1160                 goto whole;
1161
1162         /*
1163          * If this looks like the beginning of a DSO or executable mapping,
1164          * check for an ELF header.  If we find one, dump the first page to
1165          * aid in determining what was mapped here.
1166          */
1167         if (FILTER(ELF_HEADERS) &&
1168             vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
1169                 u32 __user *header = (u32 __user *) vma->vm_start;
1170                 u32 word;
1171                 mm_segment_t fs = get_fs();
1172                 /*
1173                  * Doing it this way gets the constant folded by GCC.
1174                  */
1175                 union {
1176                         u32 cmp;
1177                         char elfmag[SELFMAG];
1178                 } magic;
1179                 BUILD_BUG_ON(SELFMAG != sizeof word);
1180                 magic.elfmag[EI_MAG0] = ELFMAG0;
1181                 magic.elfmag[EI_MAG1] = ELFMAG1;
1182                 magic.elfmag[EI_MAG2] = ELFMAG2;
1183                 magic.elfmag[EI_MAG3] = ELFMAG3;
1184                 /*
1185                  * Switch to the user "segment" for get_user(),
1186                  * then put back what elf_core_dump() had in place.
1187                  */
1188                 set_fs(USER_DS);
1189                 if (unlikely(get_user(word, header)))
1190                         word = 0;
1191                 set_fs(fs);
1192                 if (word == magic.cmp)
1193                         return PAGE_SIZE;
1194         }
1195
1196 #undef  FILTER
1197
1198         return 0;
1199
1200 whole:
1201         return vma->vm_end - vma->vm_start;
1202 }
1203
1204 /* An ELF note in memory */
1205 struct memelfnote
1206 {
1207         const char *name;
1208         int type;
1209         unsigned int datasz;
1210         void *data;
1211 };
1212
1213 static int notesize(struct memelfnote *en)
1214 {
1215         int sz;
1216
1217         sz = sizeof(struct elf_note);
1218         sz += roundup(strlen(en->name) + 1, 4);
1219         sz += roundup(en->datasz, 4);
1220
1221         return sz;
1222 }
1223
1224 #define DUMP_WRITE(addr, nr, foffset)   \
1225         do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1226
1227 static int alignfile(struct file *file, loff_t *foffset)
1228 {
1229         static const char buf[4] = { 0, };
1230         DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1231         return 1;
1232 }
1233
1234 static int writenote(struct memelfnote *men, struct file *file,
1235                         loff_t *foffset)
1236 {
1237         struct elf_note en;
1238         en.n_namesz = strlen(men->name) + 1;
1239         en.n_descsz = men->datasz;
1240         en.n_type = men->type;
1241
1242         DUMP_WRITE(&en, sizeof(en), foffset);
1243         DUMP_WRITE(men->name, en.n_namesz, foffset);
1244         if (!alignfile(file, foffset))
1245                 return 0;
1246         DUMP_WRITE(men->data, men->datasz, foffset);
1247         if (!alignfile(file, foffset))
1248                 return 0;
1249
1250         return 1;
1251 }
1252 #undef DUMP_WRITE
1253
1254 static void fill_elf_header(struct elfhdr *elf, int segs,
1255                             u16 machine, u32 flags, u8 osabi)
1256 {
1257         memset(elf, 0, sizeof(*elf));
1258
1259         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1260         elf->e_ident[EI_CLASS] = ELF_CLASS;
1261         elf->e_ident[EI_DATA] = ELF_DATA;
1262         elf->e_ident[EI_VERSION] = EV_CURRENT;
1263         elf->e_ident[EI_OSABI] = ELF_OSABI;
1264
1265         elf->e_type = ET_CORE;
1266         elf->e_machine = machine;
1267         elf->e_version = EV_CURRENT;
1268         elf->e_phoff = sizeof(struct elfhdr);
1269         elf->e_flags = flags;
1270         elf->e_ehsize = sizeof(struct elfhdr);
1271         elf->e_phentsize = sizeof(struct elf_phdr);
1272         elf->e_phnum = segs;
1273
1274         return;
1275 }
1276
1277 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1278 {
1279         phdr->p_type = PT_NOTE;
1280         phdr->p_offset = offset;
1281         phdr->p_vaddr = 0;
1282         phdr->p_paddr = 0;
1283         phdr->p_filesz = sz;
1284         phdr->p_memsz = 0;
1285         phdr->p_flags = 0;
1286         phdr->p_align = 0;
1287         return;
1288 }
1289
1290 static void fill_note(struct memelfnote *note, const char *name, int type,
1291                 unsigned int sz, void *data)
1292 {
1293         note->name = name;
1294         note->type = type;
1295         note->datasz = sz;
1296         note->data = data;
1297         return;
1298 }
1299
1300 /*
1301  * fill up all the fields in prstatus from the given task struct, except
1302  * registers which need to be filled up separately.
1303  */
1304 static void fill_prstatus(struct elf_prstatus *prstatus,
1305                 struct task_struct *p, long signr)
1306 {
1307         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1308         prstatus->pr_sigpend = p->pending.signal.sig[0];
1309         prstatus->pr_sighold = p->blocked.sig[0];
1310         rcu_read_lock();
1311         prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1312         rcu_read_unlock();
1313         prstatus->pr_pid = task_pid_vnr(p);
1314         prstatus->pr_pgrp = task_pgrp_vnr(p);
1315         prstatus->pr_sid = task_session_vnr(p);
1316         if (thread_group_leader(p)) {
1317                 struct task_cputime cputime;
1318
1319                 /*
1320                  * This is the record for the group leader.  It shows the
1321                  * group-wide total, not its individual thread total.
1322                  */
1323                 thread_group_cputime(p, &cputime);
1324                 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1325                 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
1326         } else {
1327                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1328                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1329         }
1330         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1331         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1332 }
1333
1334 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1335                        struct mm_struct *mm)
1336 {
1337         const struct cred *cred;
1338         unsigned int i, len;
1339
1340         /* first copy the parameters from user space */
1341         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1342
1343         len = mm->arg_end - mm->arg_start;
1344         if (len >= ELF_PRARGSZ)
1345                 len = ELF_PRARGSZ-1;
1346         if (copy_from_user(&psinfo->pr_psargs,
1347                            (const char __user *)mm->arg_start, len))
1348                 return -EFAULT;
1349         for(i = 0; i < len; i++)
1350                 if (psinfo->pr_psargs[i] == 0)
1351                         psinfo->pr_psargs[i] = ' ';
1352         psinfo->pr_psargs[len] = 0;
1353
1354         rcu_read_lock();
1355         psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1356         rcu_read_unlock();
1357         psinfo->pr_pid = task_pid_vnr(p);
1358         psinfo->pr_pgrp = task_pgrp_vnr(p);
1359         psinfo->pr_sid = task_session_vnr(p);
1360
1361         i = p->state ? ffz(~p->state) + 1 : 0;
1362         psinfo->pr_state = i;
1363         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1364         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1365         psinfo->pr_nice = task_nice(p);
1366         psinfo->pr_flag = p->flags;
1367         rcu_read_lock();
1368         cred = __task_cred(p);
1369         SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
1370         SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
1371         rcu_read_unlock();
1372         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1373
1374         return 0;
1375 }
1376
1377 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1378 {
1379         elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1380         int i = 0;
1381         do
1382                 i += 2;
1383         while (auxv[i - 2] != AT_NULL);
1384         fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1385 }
1386
1387 static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
1388                 siginfo_t *siginfo)
1389 {
1390         mm_segment_t old_fs = get_fs();
1391         set_fs(KERNEL_DS);
1392         copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
1393         set_fs(old_fs);
1394         fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
1395 }
1396
1397 #define MAX_FILE_NOTE_SIZE (4*1024*1024)
1398 /*
1399  * Format of NT_FILE note:
1400  *
1401  * long count     -- how many files are mapped
1402  * long page_size -- units for file_ofs
1403  * array of [COUNT] elements of
1404  *   long start
1405  *   long end
1406  *   long file_ofs
1407  * followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
1408  */
1409 static void fill_files_note(struct memelfnote *note)
1410 {
1411         struct vm_area_struct *vma;
1412         unsigned count, size, names_ofs, remaining, n;
1413         user_long_t *data;
1414         user_long_t *start_end_ofs;
1415         char *name_base, *name_curpos;
1416
1417         /* *Estimated* file count and total data size needed */
1418         count = current->mm->map_count;
1419         size = count * 64;
1420
1421         names_ofs = (2 + 3 * count) * sizeof(data[0]);
1422  alloc:
1423         if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
1424                 goto err;
1425         size = round_up(size, PAGE_SIZE);
1426         data = vmalloc(size);
1427         if (!data)
1428                 goto err;
1429
1430         start_end_ofs = data + 2;
1431         name_base = name_curpos = ((char *)data) + names_ofs;
1432         remaining = size - names_ofs;
1433         count = 0;
1434         for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
1435                 struct file *file;
1436                 const char *filename;
1437
1438                 file = vma->vm_file;
1439                 if (!file)
1440                         continue;
1441                 filename = d_path(&file->f_path, name_curpos, remaining);
1442                 if (IS_ERR(filename)) {
1443                         if (PTR_ERR(filename) == -ENAMETOOLONG) {
1444                                 vfree(data);
1445                                 size = size * 5 / 4;
1446                                 goto alloc;
1447                         }
1448                         continue;
1449                 }
1450
1451                 /* d_path() fills at the end, move name down */
1452                 /* n = strlen(filename) + 1: */
1453                 n = (name_curpos + remaining) - filename;
1454                 remaining = filename - name_curpos;
1455                 memmove(name_curpos, filename, n);
1456                 name_curpos += n;
1457
1458                 *start_end_ofs++ = vma->vm_start;
1459                 *start_end_ofs++ = vma->vm_end;
1460                 *start_end_ofs++ = vma->vm_pgoff;
1461                 count++;
1462         }
1463
1464         /* Now we know exact count of files, can store it */
1465         data[0] = count;
1466         data[1] = PAGE_SIZE;
1467         /*
1468          * Count usually is less than current->mm->map_count,
1469          * we need to move filenames down.
1470          */
1471         n = current->mm->map_count - count;
1472         if (n != 0) {
1473                 unsigned shift_bytes = n * 3 * sizeof(data[0]);
1474                 memmove(name_base - shift_bytes, name_base,
1475                         name_curpos - name_base);
1476                 name_curpos -= shift_bytes;
1477         }
1478
1479         size = name_curpos - (char *)data;
1480         fill_note(note, "CORE", NT_FILE, size, data);
1481  err: ;
1482 }
1483
1484 #ifdef CORE_DUMP_USE_REGSET
1485 #include <linux/regset.h>
1486
1487 struct elf_thread_core_info {
1488         struct elf_thread_core_info *next;
1489         struct task_struct *task;
1490         struct elf_prstatus prstatus;
1491         struct memelfnote notes[0];
1492 };
1493
1494 struct elf_note_info {
1495         struct elf_thread_core_info *thread;
1496         struct memelfnote psinfo;
1497         struct memelfnote signote;
1498         struct memelfnote auxv;
1499         struct memelfnote files;
1500         user_siginfo_t csigdata;
1501         size_t size;
1502         int thread_notes;
1503 };
1504
1505 /*
1506  * When a regset has a writeback hook, we call it on each thread before
1507  * dumping user memory.  On register window machines, this makes sure the
1508  * user memory backing the register data is up to date before we read it.
1509  */
1510 static void do_thread_regset_writeback(struct task_struct *task,
1511                                        const struct user_regset *regset)
1512 {
1513         if (regset->writeback)
1514                 regset->writeback(task, regset, 1);
1515 }
1516
1517 #ifndef PR_REG_SIZE
1518 #define PR_REG_SIZE(S) sizeof(S)
1519 #endif
1520
1521 #ifndef PRSTATUS_SIZE
1522 #define PRSTATUS_SIZE(S) sizeof(S)
1523 #endif
1524
1525 #ifndef PR_REG_PTR
1526 #define PR_REG_PTR(S) (&((S)->pr_reg))
1527 #endif
1528
1529 #ifndef SET_PR_FPVALID
1530 #define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
1531 #endif
1532
1533 static int fill_thread_core_info(struct elf_thread_core_info *t,
1534                                  const struct user_regset_view *view,
1535                                  long signr, size_t *total)
1536 {
1537         unsigned int i;
1538
1539         /*
1540          * NT_PRSTATUS is the one special case, because the regset data
1541          * goes into the pr_reg field inside the note contents, rather
1542          * than being the whole note contents.  We fill the reset in here.
1543          * We assume that regset 0 is NT_PRSTATUS.
1544          */
1545         fill_prstatus(&t->prstatus, t->task, signr);
1546         (void) view->regsets[0].get(t->task, &view->regsets[0],
1547                                     0, PR_REG_SIZE(t->prstatus.pr_reg),
1548                                     PR_REG_PTR(&t->prstatus), NULL);
1549
1550         fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1551                   PRSTATUS_SIZE(t->prstatus), &t->prstatus);
1552         *total += notesize(&t->notes[0]);
1553
1554         do_thread_regset_writeback(t->task, &view->regsets[0]);
1555
1556         /*
1557          * Each other regset might generate a note too.  For each regset
1558          * that has no core_note_type or is inactive, we leave t->notes[i]
1559          * all zero and we'll know to skip writing it later.
1560          */
1561         for (i = 1; i < view->n; ++i) {
1562                 const struct user_regset *regset = &view->regsets[i];
1563                 do_thread_regset_writeback(t->task, regset);
1564                 if (regset->core_note_type && regset->get &&
1565                     (!regset->active || regset->active(t->task, regset))) {
1566                         int ret;
1567                         size_t size = regset->n * regset->size;
1568                         void *data = kmalloc(size, GFP_KERNEL);
1569                         if (unlikely(!data))
1570                                 return 0;
1571                         ret = regset->get(t->task, regset,
1572                                           0, size, data, NULL);
1573                         if (unlikely(ret))
1574                                 kfree(data);
1575                         else {
1576                                 if (regset->core_note_type != NT_PRFPREG)
1577                                         fill_note(&t->notes[i], "LINUX",
1578                                                   regset->core_note_type,
1579                                                   size, data);
1580                                 else {
1581                                         SET_PR_FPVALID(&t->prstatus, 1);
1582                                         fill_note(&t->notes[i], "CORE",
1583                                                   NT_PRFPREG, size, data);
1584                                 }
1585                                 *total += notesize(&t->notes[i]);
1586                         }
1587                 }
1588         }
1589
1590         return 1;
1591 }
1592
1593 static int fill_note_info(struct elfhdr *elf, int phdrs,
1594                           struct elf_note_info *info,
1595                           siginfo_t *siginfo, struct pt_regs *regs)
1596 {
1597         struct task_struct *dump_task = current;
1598         const struct user_regset_view *view = task_user_regset_view(dump_task);
1599         struct elf_thread_core_info *t;
1600         struct elf_prpsinfo *psinfo;
1601         struct core_thread *ct;
1602         unsigned int i;
1603
1604         info->size = 0;
1605         info->thread = NULL;
1606
1607         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1608         if (psinfo == NULL) {
1609                 info->psinfo.data = NULL; /* So we don't free this wrongly */
1610                 return 0;
1611         }
1612
1613         fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1614
1615         /*
1616          * Figure out how many notes we're going to need for each thread.
1617          */
1618         info->thread_notes = 0;
1619         for (i = 0; i < view->n; ++i)
1620                 if (view->regsets[i].core_note_type != 0)
1621                         ++info->thread_notes;
1622
1623         /*
1624          * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1625          * since it is our one special case.
1626          */
1627         if (unlikely(info->thread_notes == 0) ||
1628             unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1629                 WARN_ON(1);
1630                 return 0;
1631         }
1632
1633         /*
1634          * Initialize the ELF file header.
1635          */
1636         fill_elf_header(elf, phdrs,
1637                         view->e_machine, view->e_flags, view->ei_osabi);
1638
1639         /*
1640          * Allocate a structure for each thread.
1641          */
1642         for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1643                 t = kzalloc(offsetof(struct elf_thread_core_info,
1644                                      notes[info->thread_notes]),
1645                             GFP_KERNEL);
1646                 if (unlikely(!t))
1647                         return 0;
1648
1649                 t->task = ct->task;
1650                 if (ct->task == dump_task || !info->thread) {
1651                         t->next = info->thread;
1652                         info->thread = t;
1653                 } else {
1654                         /*
1655                          * Make sure to keep the original task at
1656                          * the head of the list.
1657                          */
1658                         t->next = info->thread->next;
1659                         info->thread->next = t;
1660                 }
1661         }
1662
1663         /*
1664          * Now fill in each thread's information.
1665          */
1666         for (t = info->thread; t != NULL; t = t->next)
1667                 if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
1668                         return 0;
1669
1670         /*
1671          * Fill in the two process-wide notes.
1672          */
1673         fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1674         info->size += notesize(&info->psinfo);
1675
1676         fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
1677         info->size += notesize(&info->signote);
1678
1679         fill_auxv_note(&info->auxv, current->mm);
1680         info->size += notesize(&info->auxv);
1681
1682         fill_files_note(&info->files);
1683         info->size += notesize(&info->files);
1684
1685         return 1;
1686 }
1687
1688 static size_t get_note_info_size(struct elf_note_info *info)
1689 {
1690         return info->size;
1691 }
1692
1693 /*
1694  * Write all the notes for each thread.  When writing the first thread, the
1695  * process-wide notes are interleaved after the first thread-specific note.
1696  */
1697 static int write_note_info(struct elf_note_info *info,
1698                            struct file *file, loff_t *foffset)
1699 {
1700         bool first = 1;
1701         struct elf_thread_core_info *t = info->thread;
1702
1703         do {
1704                 int i;
1705
1706                 if (!writenote(&t->notes[0], file, foffset))
1707                         return 0;
1708
1709                 if (first && !writenote(&info->psinfo, file, foffset))
1710                         return 0;
1711                 if (first && !writenote(&info->signote, file, foffset))
1712                         return 0;
1713                 if (first && !writenote(&info->auxv, file, foffset))
1714                         return 0;
1715                 if (first && !writenote(&info->files, file, foffset))
1716                         return 0;
1717
1718                 for (i = 1; i < info->thread_notes; ++i)
1719                         if (t->notes[i].data &&
1720                             !writenote(&t->notes[i], file, foffset))
1721                                 return 0;
1722
1723                 first = 0;
1724                 t = t->next;
1725         } while (t);
1726
1727         return 1;
1728 }
1729
1730 static void free_note_info(struct elf_note_info *info)
1731 {
1732         struct elf_thread_core_info *threads = info->thread;
1733         while (threads) {
1734                 unsigned int i;
1735                 struct elf_thread_core_info *t = threads;
1736                 threads = t->next;
1737                 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1738                 for (i = 1; i < info->thread_notes; ++i)
1739                         kfree(t->notes[i].data);
1740                 kfree(t);
1741         }
1742         kfree(info->psinfo.data);
1743         vfree(info->files.data);
1744 }
1745
1746 #else
1747
1748 /* Here is the structure in which status of each thread is captured. */
1749 struct elf_thread_status
1750 {
1751         struct list_head list;
1752         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1753         elf_fpregset_t fpu;             /* NT_PRFPREG */
1754         struct task_struct *thread;
1755 #ifdef ELF_CORE_COPY_XFPREGS
1756         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1757 #endif
1758         struct memelfnote notes[3];
1759         int num_notes;
1760 };
1761
1762 /*
1763  * In order to add the specific thread information for the elf file format,
1764  * we need to keep a linked list of every threads pr_status and then create
1765  * a single section for them in the final core file.
1766  */
1767 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1768 {
1769         int sz = 0;
1770         struct task_struct *p = t->thread;
1771         t->num_notes = 0;
1772
1773         fill_prstatus(&t->prstatus, p, signr);
1774         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1775
1776         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1777                   &(t->prstatus));
1778         t->num_notes++;
1779         sz += notesize(&t->notes[0]);
1780
1781         if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1782                                                                 &t->fpu))) {
1783                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1784                           &(t->fpu));
1785                 t->num_notes++;
1786                 sz += notesize(&t->notes[1]);
1787         }
1788
1789 #ifdef ELF_CORE_COPY_XFPREGS
1790         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1791                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1792                           sizeof(t->xfpu), &t->xfpu);
1793                 t->num_notes++;
1794                 sz += notesize(&t->notes[2]);
1795         }
1796 #endif
1797         return sz;
1798 }
1799
1800 struct elf_note_info {
1801         struct memelfnote *notes;
1802         struct elf_prstatus *prstatus;  /* NT_PRSTATUS */
1803         struct elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
1804         struct list_head thread_list;
1805         elf_fpregset_t *fpu;
1806 #ifdef ELF_CORE_COPY_XFPREGS
1807         elf_fpxregset_t *xfpu;
1808 #endif
1809         user_siginfo_t csigdata;
1810         int thread_status_size;
1811         int numnote;
1812 };
1813
1814 static int elf_note_info_init(struct elf_note_info *info)
1815 {
1816         memset(info, 0, sizeof(*info));
1817         INIT_LIST_HEAD(&info->thread_list);
1818
1819         /* Allocate space for ELF notes */
1820         info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
1821         if (!info->notes)
1822                 return 0;
1823         info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1824         if (!info->psinfo)
1825                 return 0;
1826         info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1827         if (!info->prstatus)
1828                 return 0;
1829         info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1830         if (!info->fpu)
1831                 return 0;
1832 #ifdef ELF_CORE_COPY_XFPREGS
1833         info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1834         if (!info->xfpu)
1835                 return 0;
1836 #endif
1837         return 1;
1838 }
1839
1840 static int fill_note_info(struct elfhdr *elf, int phdrs,
1841                           struct elf_note_info *info,
1842                           siginfo_t *siginfo, struct pt_regs *regs)
1843 {
1844         struct list_head *t;
1845
1846         if (!elf_note_info_init(info))
1847                 return 0;
1848
1849         if (siginfo->si_signo) {
1850                 struct core_thread *ct;
1851                 struct elf_thread_status *ets;
1852
1853                 for (ct = current->mm->core_state->dumper.next;
1854                                                 ct; ct = ct->next) {
1855                         ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1856                         if (!ets)
1857                                 return 0;
1858
1859                         ets->thread = ct->task;
1860                         list_add(&ets->list, &info->thread_list);
1861                 }
1862
1863                 list_for_each(t, &info->thread_list) {
1864                         int sz;
1865
1866                         ets = list_entry(t, struct elf_thread_status, list);
1867                         sz = elf_dump_thread_status(siginfo->si_signo, ets);
1868                         info->thread_status_size += sz;
1869                 }
1870         }
1871         /* now collect the dump for the current */
1872         memset(info->prstatus, 0, sizeof(*info->prstatus));
1873         fill_prstatus(info->prstatus, current, siginfo->si_signo);
1874         elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1875
1876         /* Set up header */
1877         fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1878
1879         /*
1880          * Set up the notes in similar form to SVR4 core dumps made
1881          * with info from their /proc.
1882          */
1883
1884         fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1885                   sizeof(*info->prstatus), info->prstatus);
1886         fill_psinfo(info->psinfo, current->group_leader, current->mm);
1887         fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1888                   sizeof(*info->psinfo), info->psinfo);
1889
1890         fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
1891         fill_auxv_note(info->notes + 3, current->mm);
1892         fill_files_note(info->notes + 4);
1893
1894         info->numnote = 5;
1895
1896         /* Try to dump the FPU. */
1897         info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1898                                                                info->fpu);
1899         if (info->prstatus->pr_fpvalid)
1900                 fill_note(info->notes + info->numnote++,
1901                           "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1902 #ifdef ELF_CORE_COPY_XFPREGS
1903         if (elf_core_copy_task_xfpregs(current, info->xfpu))
1904                 fill_note(info->notes + info->numnote++,
1905                           "LINUX", ELF_CORE_XFPREG_TYPE,
1906                           sizeof(*info->xfpu), info->xfpu);
1907 #endif
1908
1909         return 1;
1910 }
1911
1912 static size_t get_note_info_size(struct elf_note_info *info)
1913 {
1914         int sz = 0;
1915         int i;
1916
1917         for (i = 0; i < info->numnote; i++)
1918                 sz += notesize(info->notes + i);
1919
1920         sz += info->thread_status_size;
1921
1922         return sz;
1923 }
1924
1925 static int write_note_info(struct elf_note_info *info,
1926                            struct file *file, loff_t *foffset)
1927 {
1928         int i;
1929         struct list_head *t;
1930
1931         for (i = 0; i < info->numnote; i++)
1932                 if (!writenote(info->notes + i, file, foffset))
1933                         return 0;
1934
1935         /* write out the thread status notes section */
1936         list_for_each(t, &info->thread_list) {
1937                 struct elf_thread_status *tmp =
1938                                 list_entry(t, struct elf_thread_status, list);
1939
1940                 for (i = 0; i < tmp->num_notes; i++)
1941                         if (!writenote(&tmp->notes[i], file, foffset))
1942                                 return 0;
1943         }
1944
1945         return 1;
1946 }
1947
1948 static void free_note_info(struct elf_note_info *info)
1949 {
1950         while (!list_empty(&info->thread_list)) {
1951                 struct list_head *tmp = info->thread_list.next;
1952                 list_del(tmp);
1953                 kfree(list_entry(tmp, struct elf_thread_status, list));
1954         }
1955
1956         /* Free data allocated by fill_files_note(): */
1957         vfree(info->notes[4].data);
1958
1959         kfree(info->prstatus);
1960         kfree(info->psinfo);
1961         kfree(info->notes);
1962         kfree(info->fpu);
1963 #ifdef ELF_CORE_COPY_XFPREGS
1964         kfree(info->xfpu);
1965 #endif
1966 }
1967
1968 #endif
1969
1970 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1971                                         struct vm_area_struct *gate_vma)
1972 {
1973         struct vm_area_struct *ret = tsk->mm->mmap;
1974
1975         if (ret)
1976                 return ret;
1977         return gate_vma;
1978 }
1979 /*
1980  * Helper function for iterating across a vma list.  It ensures that the caller
1981  * will visit `gate_vma' prior to terminating the search.
1982  */
1983 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1984                                         struct vm_area_struct *gate_vma)
1985 {
1986         struct vm_area_struct *ret;
1987
1988         ret = this_vma->vm_next;
1989         if (ret)
1990                 return ret;
1991         if (this_vma == gate_vma)
1992                 return NULL;
1993         return gate_vma;
1994 }
1995
1996 static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1997                              elf_addr_t e_shoff, int segs)
1998 {
1999         elf->e_shoff = e_shoff;
2000         elf->e_shentsize = sizeof(*shdr4extnum);
2001         elf->e_shnum = 1;
2002         elf->e_shstrndx = SHN_UNDEF;
2003
2004         memset(shdr4extnum, 0, sizeof(*shdr4extnum));
2005
2006         shdr4extnum->sh_type = SHT_NULL;
2007         shdr4extnum->sh_size = elf->e_shnum;
2008         shdr4extnum->sh_link = elf->e_shstrndx;
2009         shdr4extnum->sh_info = segs;
2010 }
2011
2012 static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
2013                                      unsigned long mm_flags)
2014 {
2015         struct vm_area_struct *vma;
2016         size_t size = 0;
2017
2018         for (vma = first_vma(current, gate_vma); vma != NULL;
2019              vma = next_vma(vma, gate_vma))
2020                 size += vma_dump_size(vma, mm_flags);
2021         return size;
2022 }
2023
2024 /*
2025  * Actual dumper
2026  *
2027  * This is a two-pass process; first we find the offsets of the bits,
2028  * and then they are actually written out.  If we run out of core limit
2029  * we just truncate.
2030  */
2031 static int elf_core_dump(struct coredump_params *cprm)
2032 {
2033         int has_dumped = 0;
2034         mm_segment_t fs;
2035         int segs;
2036         size_t size = 0;
2037         struct vm_area_struct *vma, *gate_vma;
2038         struct elfhdr *elf = NULL;
2039         loff_t offset = 0, dataoff, foffset;
2040         struct elf_note_info info;
2041         struct elf_phdr *phdr4note = NULL;
2042         struct elf_shdr *shdr4extnum = NULL;
2043         Elf_Half e_phnum;
2044         elf_addr_t e_shoff;
2045
2046         /*
2047          * We no longer stop all VM operations.
2048          *
2049          * This is because those proceses that could possibly change map_count
2050          * or the mmap / vma pages are now blocked in do_exit on current
2051          * finishing this core dump.
2052          *
2053          * Only ptrace can touch these memory addresses, but it doesn't change
2054          * the map_count or the pages allocated. So no possibility of crashing
2055          * exists while dumping the mm->vm_next areas to the core file.
2056          */
2057
2058         /* alloc memory for large data structures: too large to be on stack */
2059         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
2060         if (!elf)
2061                 goto out;
2062         /*
2063          * The number of segs are recored into ELF header as 16bit value.
2064          * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
2065          */
2066         segs = current->mm->map_count;
2067         segs += elf_core_extra_phdrs();
2068
2069         gate_vma = get_gate_vma(current->mm);
2070         if (gate_vma != NULL)
2071                 segs++;
2072
2073         /* for notes section */
2074         segs++;
2075
2076         /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
2077          * this, kernel supports extended numbering. Have a look at
2078          * include/linux/elf.h for further information. */
2079         e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
2080
2081         /*
2082          * Collect all the non-memory information about the process for the
2083          * notes.  This also sets up the file header.
2084          */
2085         if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
2086                 goto cleanup;
2087
2088         has_dumped = 1;
2089         current->flags |= PF_DUMPCORE;
2090
2091         fs = get_fs();
2092         set_fs(KERNEL_DS);
2093
2094         offset += sizeof(*elf);                         /* Elf header */
2095         offset += segs * sizeof(struct elf_phdr);       /* Program headers */
2096         foffset = offset;
2097
2098         /* Write notes phdr entry */
2099         {
2100                 size_t sz = get_note_info_size(&info);
2101
2102                 sz += elf_coredump_extra_notes_size();
2103
2104                 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
2105                 if (!phdr4note)
2106                         goto end_coredump;
2107
2108                 fill_elf_note_phdr(phdr4note, sz, offset);
2109                 offset += sz;
2110         }
2111
2112         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
2113
2114         offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
2115         offset += elf_core_extra_data_size();
2116         e_shoff = offset;
2117
2118         if (e_phnum == PN_XNUM) {
2119                 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
2120                 if (!shdr4extnum)
2121                         goto end_coredump;
2122                 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
2123         }
2124
2125         offset = dataoff;
2126
2127         size += sizeof(*elf);
2128         if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
2129                 goto end_coredump;
2130
2131         size += sizeof(*phdr4note);
2132         if (size > cprm->limit
2133             || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
2134                 goto end_coredump;
2135
2136         /* Write program headers for segments dump */
2137         for (vma = first_vma(current, gate_vma); vma != NULL;
2138                         vma = next_vma(vma, gate_vma)) {
2139                 struct elf_phdr phdr;
2140
2141                 phdr.p_type = PT_LOAD;
2142                 phdr.p_offset = offset;
2143                 phdr.p_vaddr = vma->vm_start;
2144                 phdr.p_paddr = 0;
2145                 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
2146                 phdr.p_memsz = vma->vm_end - vma->vm_start;
2147                 offset += phdr.p_filesz;
2148                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
2149                 if (vma->vm_flags & VM_WRITE)
2150                         phdr.p_flags |= PF_W;
2151                 if (vma->vm_flags & VM_EXEC)
2152                         phdr.p_flags |= PF_X;
2153                 phdr.p_align = ELF_EXEC_PAGESIZE;
2154
2155                 size += sizeof(phdr);
2156                 if (size > cprm->limit
2157                     || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2158                         goto end_coredump;
2159         }
2160
2161         if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2162                 goto end_coredump;
2163
2164         /* write out the notes section */
2165         if (!write_note_info(&info, cprm->file, &foffset))
2166                 goto end_coredump;
2167
2168         if (elf_coredump_extra_notes_write(cprm->file, &foffset))
2169                 goto end_coredump;
2170
2171         /* Align to page */
2172         if (!dump_seek(cprm->file, dataoff - foffset))
2173                 goto end_coredump;
2174
2175         for (vma = first_vma(current, gate_vma); vma != NULL;
2176                         vma = next_vma(vma, gate_vma)) {
2177                 unsigned long addr;
2178                 unsigned long end;
2179
2180                 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
2181
2182                 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2183                         struct page *page;
2184                         int stop;
2185
2186                         page = get_dump_page(addr);
2187                         if (page) {
2188                                 void *kaddr = kmap(page);
2189                                 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2190                                         !dump_write(cprm->file, kaddr,
2191                                                     PAGE_SIZE);
2192                                 kunmap(page);
2193                                 page_cache_release(page);
2194                         } else
2195                                 stop = !dump_seek(cprm->file, PAGE_SIZE);
2196                         if (stop)
2197                                 goto end_coredump;
2198                 }
2199         }
2200
2201         if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2202                 goto end_coredump;
2203
2204         if (e_phnum == PN_XNUM) {
2205                 size += sizeof(*shdr4extnum);
2206                 if (size > cprm->limit
2207                     || !dump_write(cprm->file, shdr4extnum,
2208                                    sizeof(*shdr4extnum)))
2209                         goto end_coredump;
2210         }
2211
2212 end_coredump:
2213         set_fs(fs);
2214
2215 cleanup:
2216         free_note_info(&info);
2217         kfree(shdr4extnum);
2218         kfree(phdr4note);
2219         kfree(elf);
2220 out:
2221         return has_dumped;
2222 }
2223
2224 #endif          /* CONFIG_ELF_CORE */
2225
2226 static int __init init_elf_binfmt(void)
2227 {
2228         register_binfmt(&elf_format);
2229         return 0;
2230 }
2231
2232 static void __exit exit_elf_binfmt(void)
2233 {
2234         /* Remove the COFF and ELF loaders. */
2235         unregister_binfmt(&elf_format);
2236 }
2237
2238 core_initcall(init_elf_binfmt);
2239 module_exit(exit_elf_binfmt);
2240 MODULE_LICENSE("GPL");