fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/fdtable.h>
  28 #include <linux/mm.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/swap.h>
  32 #include <linux/string.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/perf_event.h>
  36 #include <linux/highmem.h>
  37 #include <linux/spinlock.h>
  38 #include <linux/key.h>
  39 #include <linux/personality.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/utsname.h>
  42 #include <linux/pid_namespace.h>
  43 #include <linux/module.h>
  44 #include <linux/namei.h>
  45 #include <linux/proc_fs.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/tsacct_kern.h>
  50 #include <linux/cn_proc.h>
  51 #include <linux/audit.h>
  52 #include <linux/tracehook.h>
  53 #include <linux/kmod.h>
  54 #include <linux/fsnotify.h>
  55 #include <linux/fs_struct.h>
  56 #include <linux/pipe_fs_i.h>
  57
  58 #include <asm/uaccess.h>
  59 #include <asm/mmu_context.h>
  60 #include <asm/tlb.h>
  61 #include "internal.h"
  62
  63 int core_uses_pid;
  64 char core_pattern[CORENAME_MAX_SIZE] = "core";
  65 unsigned int core_pipe_limit;
  66 int suid_dumpable = 0;
  67
  68 /* The maximal length of core_pattern is also specified in sysctl.c */
  69
  70 static LIST_HEAD(formats);
  71 static DEFINE_RWLOCK(binfmt_lock);
  72
  73 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  74 {
  75         if (!fmt)
  76                 return -EINVAL;
  77         write_lock(&binfmt_lock);
  78         insert ? list_add(&fmt->lh, &formats) :
  79                  list_add_tail(&fmt->lh, &formats);
  80         write_unlock(&binfmt_lock);
  81         return 0;
  82 }
  83
  84 EXPORT_SYMBOL(__register_binfmt);
  85
  86 void unregister_binfmt(struct linux_binfmt * fmt)
  87 {
  88         write_lock(&binfmt_lock);
  89         list_del(&fmt->lh);
  90         write_unlock(&binfmt_lock);
  91 }
  92
  93 EXPORT_SYMBOL(unregister_binfmt);
  94
  95 static inline void put_binfmt(struct linux_binfmt * fmt)
  96 {
  97         module_put(fmt->module);
  98 }
  99
 100 /*
 101  * Note that a shared library must be both readable and executable due to
 102  * security reasons.
 103  *
 104  * Also note that we take the address to load from from the file itself.
 105  */
 106 SYSCALL_DEFINE1(uselib, const char __user *, library)
 107 {
 108         struct file *file;
 109         char *tmp = getname(library);
 110         int error = PTR_ERR(tmp);
 111
 112         if (IS_ERR(tmp))
 113                 goto out;
 114
 115         file = do_filp_open(AT_FDCWD, tmp,
 116                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 117                                 MAY_READ | MAY_EXEC | MAY_OPEN);
 118         putname(tmp);
 119         error = PTR_ERR(file);
 120         if (IS_ERR(file))
 121                 goto out;
 122
 123         error = -EINVAL;
 124         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 125                 goto exit;
 126
 127         error = -EACCES;
 128         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 129                 goto exit;
 130
 131         fsnotify_open(file);
 132
 133         error = -ENOEXEC;
 134         if(file->f_op) {
 135                 struct linux_binfmt * fmt;
 136
 137                 read_lock(&binfmt_lock);
 138                 list_for_each_entry(fmt, &formats, lh) {
 139                         if (!fmt->load_shlib)
 140                                 continue;
 141                         if (!try_module_get(fmt->module))
 142                                 continue;
 143                         read_unlock(&binfmt_lock);
 144                         error = fmt->load_shlib(file);
 145                         read_lock(&binfmt_lock);
 146                         put_binfmt(fmt);
 147                         if (error != -ENOEXEC)
 148                                 break;
 149                 }
 150                 read_unlock(&binfmt_lock);
 151         }
 152 exit:
 153         fput(file);
 154 out:
 155         return error;
 156 }
 157
 158 #ifdef CONFIG_MMU
 159
 160 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 161                 int write)
 162 {
 163         struct page *page;
 164         int ret;
 165
 166 #ifdef CONFIG_STACK_GROWSUP
 167         if (write) {
 168                 ret = expand_stack_downwards(bprm->vma, pos);
 169                 if (ret < 0)
 170                         return NULL;
 171         }
 172 #endif
 173         ret = get_user_pages(current, bprm->mm, pos,
 174                         1, write, 1, &page, NULL);
 175         if (ret <= 0)
 176                 return NULL;
 177
 178         if (write) {
 179                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 180                 struct rlimit *rlim;
 181
 182                 /*
 183                  * We've historically supported up to 32 pages (ARG_MAX)
 184                  * of argument strings even with small stacks
 185                  */
 186                 if (size <= ARG_MAX)
 187                         return page;
 188
 189                 /*
 190                  * Limit to 1/4-th the stack size for the argv+env strings.
 191                  * This ensures that:
 192                  *  - the remaining binfmt code will not run out of stack space,
 193                  *  - the program will have a reasonable amount of stack left
 194                  *    to work from.
 195                  */
 196                 rlim = current->signal->rlim;
 197                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 198                         put_page(page);
 199                         return NULL;
 200                 }
 201         }
 202
 203         return page;
 204 }
 205
 206 static void put_arg_page(struct page *page)
 207 {
 208         put_page(page);
 209 }
 210
 211 static void free_arg_page(struct linux_binprm *bprm, int i)
 212 {
 213 }
 214
 215 static void free_arg_pages(struct linux_binprm *bprm)
 216 {
 217 }
 218
 219 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 220                 struct page *page)
 221 {
 222         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 223 }
 224
 225 static int __bprm_mm_init(struct linux_binprm *bprm)
 226 {
 227         int err;
 228         struct vm_area_struct *vma = NULL;
 229         struct mm_struct *mm = bprm->mm;
 230
 231         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 232         if (!vma)
 233                 return -ENOMEM;
 234
 235         down_write(&mm->mmap_sem);
 236         vma->vm_mm = mm;
 237
 238         /*
 239          * Place the stack at the largest stack address the architecture
 240          * supports. Later, we'll move this to an appropriate place. We don't
 241          * use STACK_TOP because that can depend on attributes which aren't
 242          * configured yet.
 243          */
 244         BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 245         vma->vm_end = STACK_TOP_MAX;
 246         vma->vm_start = vma->vm_end - PAGE_SIZE;
 247         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 248         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 249         INIT_LIST_HEAD(&vma->anon_vma_chain);
 250         err = insert_vm_struct(mm, vma);
 251         if (err)
 252                 goto err;
 253
 254         mm->stack_vm = mm->total_vm = 1;
 255         up_write(&mm->mmap_sem);
 256         bprm->p = vma->vm_end - sizeof(void *);
 257         return 0;
 258 err:
 259         up_write(&mm->mmap_sem);
 260         bprm->vma = NULL;
 261         kmem_cache_free(vm_area_cachep, vma);
 262         return err;
 263 }
 264
 265 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 266 {
 267         return len <= MAX_ARG_STRLEN;
 268 }
 269
 270 #else
 271
 272 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 273                 int write)
 274 {
 275         struct page *page;
 276
 277         page = bprm->page[pos / PAGE_SIZE];
 278         if (!page && write) {
 279                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 280                 if (!page)
 281                         return NULL;
 282                 bprm->page[pos / PAGE_SIZE] = page;
 283         }
 284
 285         return page;
 286 }
 287
 288 static void put_arg_page(struct page *page)
 289 {
 290 }
 291
 292 static void free_arg_page(struct linux_binprm *bprm, int i)
 293 {
 294         if (bprm->page[i]) {
 295                 __free_page(bprm->page[i]);
 296                 bprm->page[i] = NULL;
 297         }
 298 }
 299
 300 static void free_arg_pages(struct linux_binprm *bprm)
 301 {
 302         int i;
 303
 304         for (i = 0; i < MAX_ARG_PAGES; i++)
 305                 free_arg_page(bprm, i);
 306 }
 307
 308 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 309                 struct page *page)
 310 {
 311 }
 312
 313 static int __bprm_mm_init(struct linux_binprm *bprm)
 314 {
 315         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 316         return 0;
 317 }
 318
 319 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 320 {
 321         return len <= bprm->p;
 322 }
 323
 324 #endif /* CONFIG_MMU */
 325
 326 /*
 327  * Create a new mm_struct and populate it with a temporary stack
 328  * vm_area_struct.  We don't have enough context at this point to set the stack
 329  * flags, permissions, and offset, so we use temporary values.  We'll update
 330  * them later in setup_arg_pages().
 331  */
 332 int bprm_mm_init(struct linux_binprm *bprm)
 333 {
 334         int err;
 335         struct mm_struct *mm = NULL;
 336
 337         bprm->mm = mm = mm_alloc();
 338         err = -ENOMEM;
 339         if (!mm)
 340                 goto err;
 341
 342         err = init_new_context(current, mm);
 343         if (err)
 344                 goto err;
 345
 346         err = __bprm_mm_init(bprm);
 347         if (err)
 348                 goto err;
 349
 350         return 0;
 351
 352 err:
 353         if (mm) {
 354                 bprm->mm = NULL;
 355                 mmdrop(mm);
 356         }
 357
 358         return err;
 359 }
 360
 361 /*
 362  * count() counts the number of strings in array ARGV.
 363  */
 364 static int count(const char __user * const __user * argv, int max)
 365 {
 366         int i = 0;
 367
 368         if (argv != NULL) {
 369                 for (;;) {
 370                         const char __user * p;
 371
 372                         if (get_user(p, argv))
 373                                 return -EFAULT;
 374                         if (!p)
 375                                 break;
 376                         argv++;
 377                         if (i++ >= max)
 378                                 return -E2BIG;
 379                         cond_resched();
 380                 }
 381         }
 382         return i;
 383 }
 384
 385 /*
 386  * 'copy_strings()' copies argument/environment strings from the old
 387  * processes's memory to the new process's stack.  The call to get_user_pages()
 388  * ensures the destination page is created and not swapped out.
 389  */
 390 static int copy_strings(int argc, const char __user *const __user *argv,
 391                         struct linux_binprm *bprm)
 392 {
 393         struct page *kmapped_page = NULL;
 394         char *kaddr = NULL;
 395         unsigned long kpos = 0;
 396         int ret;
 397
 398         while (argc-- > 0) {
 399                 const char __user *str;
 400                 int len;
 401                 unsigned long pos;
 402
 403                 if (get_user(str, argv+argc) ||
 404                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 405                         ret = -EFAULT;
 406                         goto out;
 407                 }
 408
 409                 if (!valid_arg_len(bprm, len)) {
 410                         ret = -E2BIG;
 411                         goto out;
 412                 }
 413
 414                 /* We're going to work our way backwords. */
 415                 pos = bprm->p;
 416                 str += len;
 417                 bprm->p -= len;
 418
 419                 while (len > 0) {
 420                         int offset, bytes_to_copy;
 421
 422                         cond_resched();
 423
 424                         offset = pos % PAGE_SIZE;
 425                         if (offset == 0)
 426                                 offset = PAGE_SIZE;
 427
 428                         bytes_to_copy = offset;
 429                         if (bytes_to_copy > len)
 430                                 bytes_to_copy = len;
 431
 432                         offset -= bytes_to_copy;
 433                         pos -= bytes_to_copy;
 434                         str -= bytes_to_copy;
 435                         len -= bytes_to_copy;
 436
 437                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 438                                 struct page *page;
 439
 440                                 page = get_arg_page(bprm, pos, 1);
 441                                 if (!page) {
 442                                         ret = -E2BIG;
 443                                         goto out;
 444                                 }
 445
 446                                 if (kmapped_page) {
 447                                         flush_kernel_dcache_page(kmapped_page);
 448                                         kunmap(kmapped_page);
 449                                         put_arg_page(kmapped_page);
 450                                 }
 451                                 kmapped_page = page;
 452                                 kaddr = kmap(kmapped_page);
 453                                 kpos = pos & PAGE_MASK;
 454                                 flush_arg_page(bprm, kpos, kmapped_page);
 455                         }
 456                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 457                                 ret = -EFAULT;
 458                                 goto out;
 459                         }
 460                 }
 461         }
 462         ret = 0;
 463 out:
 464         if (kmapped_page) {
 465                 flush_kernel_dcache_page(kmapped_page);
 466                 kunmap(kmapped_page);
 467                 put_arg_page(kmapped_page);
 468         }
 469         return ret;
 470 }
 471
 472 /*
 473  * Like copy_strings, but get argv and its values from kernel memory.
 474  */
 475 int copy_strings_kernel(int argc, const char *const *argv,
 476                         struct linux_binprm *bprm)
 477 {
 478         int r;
 479         mm_segment_t oldfs = get_fs();
 480         set_fs(KERNEL_DS);
 481         r = copy_strings(argc, (const char __user *const  __user *)argv, bprm);
 482         set_fs(oldfs);
 483         return r;
 484 }
 485 EXPORT_SYMBOL(copy_strings_kernel);
 486
 487 #ifdef CONFIG_MMU
 488
 489 /*
 490  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 491  * the binfmt code determines where the new stack should reside, we shift it to
 492  * its final location.  The process proceeds as follows:
 493  *
 494  * 1) Use shift to calculate the new vma endpoints.
 495  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 496  *    arguments passed to subsequent functions are consistent.
 497  * 3) Move vma's page tables to the new range.
 498  * 4) Free up any cleared pgd range.
 499  * 5) Shrink the vma to cover only the new range.
 500  */
 501 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 502 {
 503         struct mm_struct *mm = vma->vm_mm;
 504         unsigned long old_start = vma->vm_start;
 505         unsigned long old_end = vma->vm_end;
 506         unsigned long length = old_end - old_start;
 507         unsigned long new_start = old_start - shift;
 508         unsigned long new_end = old_end - shift;
 509         struct mmu_gather *tlb;
 510
 511         BUG_ON(new_start > new_end);
 512
 513         /*
 514          * ensure there are no vmas between where we want to go
 515          * and where we are
 516          */
 517         if (vma != find_vma(mm, new_start))
 518                 return -EFAULT;
 519
 520         /*
 521          * cover the whole range: [new_start, old_end)
 522          */
 523         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 524                 return -ENOMEM;
 525
 526         /*
 527          * move the page tables downwards, on failure we rely on
 528          * process cleanup to remove whatever mess we made.
 529          */
 530         if (length != move_page_tables(vma, old_start,
 531                                        vma, new_start, length))
 532                 return -ENOMEM;
 533
 534         lru_add_drain();
 535         tlb = tlb_gather_mmu(mm, 0);
 536         if (new_end > old_start) {
 537                 /*
 538                  * when the old and new regions overlap clear from new_end.
 539                  */
 540                 free_pgd_range(tlb, new_end, old_end, new_end,
 541                         vma->vm_next ? vma->vm_next->vm_start : 0);
 542         } else {
 543                 /*
 544                  * otherwise, clean from old_start; this is done to not touch
 545                  * the address space in [new_end, old_start) some architectures
 546                  * have constraints on va-space that make this illegal (IA64) -
 547                  * for the others its just a little faster.
 548                  */
 549                 free_pgd_range(tlb, old_start, old_end, new_end,
 550                         vma->vm_next ? vma->vm_next->vm_start : 0);
 551         }
 552         tlb_finish_mmu(tlb, new_end, old_end);
 553
 554         /*
 555          * Shrink the vma to just the new range.  Always succeeds.
 556          */
 557         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 558
 559         return 0;
 560 }
 561
 562 /*
 563  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 564  * the stack is optionally relocated, and some extra space is added.
 565  */
 566 int setup_arg_pages(struct linux_binprm *bprm,
 567                     unsigned long stack_top,
 568                     int executable_stack)
 569 {
 570         unsigned long ret;
 571         unsigned long stack_shift;
 572         struct mm_struct *mm = current->mm;
 573         struct vm_area_struct *vma = bprm->vma;
 574         struct vm_area_struct *prev = NULL;
 575         unsigned long vm_flags;
 576         unsigned long stack_base;
 577         unsigned long stack_size;
 578         unsigned long stack_expand;
 579         unsigned long rlim_stack;
 580
 581 #ifdef CONFIG_STACK_GROWSUP
 582         /* Limit stack size to 1GB */
 583         stack_base = rlimit_max(RLIMIT_STACK);
 584         if (stack_base > (1 << 30))
 585                 stack_base = 1 << 30;
 586
 587         /* Make sure we didn't let the argument array grow too large. */
 588         if (vma->vm_end - vma->vm_start > stack_base)
 589                 return -ENOMEM;
 590
 591         stack_base = PAGE_ALIGN(stack_top - stack_base);
 592
 593         stack_shift = vma->vm_start - stack_base;
 594         mm->arg_start = bprm->p - stack_shift;
 595         bprm->p = vma->vm_end - stack_shift;
 596 #else
 597         stack_top = arch_align_stack(stack_top);
 598         stack_top = PAGE_ALIGN(stack_top);
 599
 600         if (unlikely(stack_top < mmap_min_addr) ||
 601             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 602                 return -ENOMEM;
 603
 604         stack_shift = vma->vm_end - stack_top;
 605
 606         bprm->p -= stack_shift;
 607         mm->arg_start = bprm->p;
 608 #endif
 609
 610         if (bprm->loader)
 611                 bprm->loader -= stack_shift;
 612         bprm->exec -= stack_shift;
 613
 614         down_write(&mm->mmap_sem);
 615         vm_flags = VM_STACK_FLAGS;
 616
 617         /*
 618          * Adjust stack execute permissions; explicitly enable for
 619          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 620          * (arch default) otherwise.
 621          */
 622         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 623                 vm_flags |= VM_EXEC;
 624         else if (executable_stack == EXSTACK_DISABLE_X)
 625                 vm_flags &= ~VM_EXEC;
 626         vm_flags |= mm->def_flags;
 627         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 628
 629         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 630                         vm_flags);
 631         if (ret)
 632                 goto out_unlock;
 633         BUG_ON(prev != vma);
 634
 635         /* Move stack pages down in memory. */
 636         if (stack_shift) {
 637                 ret = shift_arg_pages(vma, stack_shift);
 638                 if (ret)
 639                         goto out_unlock;
 640         }
 641
 642         /* mprotect_fixup is overkill to remove the temporary stack flags */
 643         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 644
 645         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 646         stack_size = vma->vm_end - vma->vm_start;
 647         /*
 648          * Align this down to a page boundary as expand_stack
 649          * will align it up.
 650          */
 651         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 652 #ifdef CONFIG_STACK_GROWSUP
 653         if (stack_size + stack_expand > rlim_stack)
 654                 stack_base = vma->vm_start + rlim_stack;
 655         else
 656                 stack_base = vma->vm_end + stack_expand;
 657 #else
 658         if (stack_size + stack_expand > rlim_stack)
 659                 stack_base = vma->vm_end - rlim_stack;
 660         else
 661                 stack_base = vma->vm_start - stack_expand;
 662 #endif
 663         current->mm->start_stack = bprm->p;
 664         ret = expand_stack(vma, stack_base);
 665         if (ret)
 666                 ret = -EFAULT;
 667
 668 out_unlock:
 669         up_write(&mm->mmap_sem);
 670         return ret;
 671 }
 672 EXPORT_SYMBOL(setup_arg_pages);
 673
 674 #endif /* CONFIG_MMU */
 675
 676 struct file *open_exec(const char *name)
 677 {
 678         struct file *file;
 679         int err;
 680
 681         file = do_filp_open(AT_FDCWD, name,
 682                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 683                                 MAY_EXEC | MAY_OPEN);
 684         if (IS_ERR(file))
 685                 goto out;
 686
 687         err = -EACCES;
 688         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 689                 goto exit;
 690
 691         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 692                 goto exit;
 693
 694         fsnotify_open(file);
 695
 696         err = deny_write_access(file);
 697         if (err)
 698                 goto exit;
 699
 700 out:
 701         return file;
 702
 703 exit:
 704         fput(file);
 705         return ERR_PTR(err);
 706 }
 707 EXPORT_SYMBOL(open_exec);
 708
 709 int kernel_read(struct file *file, loff_t offset,
 710                 char *addr, unsigned long count)
 711 {
 712         mm_segment_t old_fs;
 713         loff_t pos = offset;
 714         int result;
 715
 716         old_fs = get_fs();
 717         set_fs(get_ds());
 718         /* The cast to a user pointer is valid due to the set_fs() */
 719         result = vfs_read(file, (void __user *)addr, count, &pos);
 720         set_fs(old_fs);
 721         return result;
 722 }
 723
 724 EXPORT_SYMBOL(kernel_read);
 725
 726 static int exec_mmap(struct mm_struct *mm)
 727 {
 728         struct task_struct *tsk;
 729         struct mm_struct * old_mm, *active_mm;
 730
 731         /* Notify parent that we're no longer interested in the old VM */
 732         tsk = current;
 733         old_mm = current->mm;
 734         sync_mm_rss(tsk, old_mm);
 735         mm_release(tsk, old_mm);
 736
 737         if (old_mm) {
 738                 /*
 739                  * Make sure that if there is a core dump in progress
 740                  * for the old mm, we get out and die instead of going
 741                  * through with the exec.  We must hold mmap_sem around
 742                  * checking core_state and changing tsk->mm.
 743                  */
 744                 down_read(&old_mm->mmap_sem);
 745                 if (unlikely(old_mm->core_state)) {
 746                         up_read(&old_mm->mmap_sem);
 747                         return -EINTR;
 748                 }
 749         }
 750         task_lock(tsk);
 751         active_mm = tsk->active_mm;
 752         tsk->mm = mm;
 753         tsk->active_mm = mm;
 754         activate_mm(active_mm, mm);
 755         task_unlock(tsk);
 756         arch_pick_mmap_layout(mm);
 757         if (old_mm) {
 758                 up_read(&old_mm->mmap_sem);
 759                 BUG_ON(active_mm != old_mm);
 760                 mm_update_next_owner(old_mm);
 761                 mmput(old_mm);
 762                 return 0;
 763         }
 764         mmdrop(active_mm);
 765         return 0;
 766 }
 767
 768 /*
 769  * This function makes sure the current process has its own signal table,
 770  * so that flush_signal_handlers can later reset the handlers without
 771  * disturbing other processes.  (Other processes might share the signal
 772  * table via the CLONE_SIGHAND option to clone().)
 773  */
 774 static int de_thread(struct task_struct *tsk)
 775 {
 776         struct signal_struct *sig = tsk->signal;
 777         struct sighand_struct *oldsighand = tsk->sighand;
 778         spinlock_t *lock = &oldsighand->siglock;
 779
 780         if (thread_group_empty(tsk))
 781                 goto no_thread_group;
 782
 783         /*
 784          * Kill all other threads in the thread group.
 785          */
 786         spin_lock_irq(lock);
 787         if (signal_group_exit(sig)) {
 788                 /*
 789                  * Another group action in progress, just
 790                  * return so that the signal is processed.
 791                  */
 792                 spin_unlock_irq(lock);
 793                 return -EAGAIN;
 794         }
 795
 796         sig->group_exit_task = tsk;
 797         sig->notify_count = zap_other_threads(tsk);
 798         if (!thread_group_leader(tsk))
 799                 sig->notify_count--;
 800
 801         while (sig->notify_count) {
 802                 __set_current_state(TASK_UNINTERRUPTIBLE);
 803                 spin_unlock_irq(lock);
 804                 schedule();
 805                 spin_lock_irq(lock);
 806         }
 807         spin_unlock_irq(lock);
 808
 809         /*
 810          * At this point all other threads have exited, all we have to
 811          * do is to wait for the thread group leader to become inactive,
 812          * and to assume its PID:
 813          */
 814         if (!thread_group_leader(tsk)) {
 815                 struct task_struct *leader = tsk->group_leader;
 816
 817                 sig->notify_count = -1; /* for exit_notify() */
 818                 for (;;) {
 819                         write_lock_irq(&tasklist_lock);
 820                         if (likely(leader->exit_state))
 821                                 break;
 822                         __set_current_state(TASK_UNINTERRUPTIBLE);
 823                         write_unlock_irq(&tasklist_lock);
 824                         schedule();
 825                 }
 826
 827                 /*
 828                  * The only record we have of the real-time age of a
 829                  * process, regardless of execs it's done, is start_time.
 830                  * All the past CPU time is accumulated in signal_struct
 831                  * from sister threads now dead.  But in this non-leader
 832                  * exec, nothing survives from the original leader thread,
 833                  * whose birth marks the true age of this process now.
 834                  * When we take on its identity by switching to its PID, we
 835                  * also take its birthdate (always earlier than our own).
 836                  */
 837                 tsk->start_time = leader->start_time;
 838
 839                 BUG_ON(!same_thread_group(leader, tsk));
 840                 BUG_ON(has_group_leader_pid(tsk));
 841                 /*
 842                  * An exec() starts a new thread group with the
 843                  * TGID of the previous thread group. Rehash the
 844                  * two threads with a switched PID, and release
 845                  * the former thread group leader:
 846                  */
 847
 848                 /* Become a process group leader with the old leader's pid.
 849                  * The old leader becomes a thread of the this thread group.
 850                  * Note: The old leader also uses this pid until release_task
 851                  *       is called.  Odd but simple and correct.
 852                  */
 853                 detach_pid(tsk, PIDTYPE_PID);
 854                 tsk->pid = leader->pid;
 855                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 856                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 857                 transfer_pid(leader, tsk, PIDTYPE_SID);
 858
 859                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 860                 list_replace_init(&leader->sibling, &tsk->sibling);
 861
 862                 tsk->group_leader = tsk;
 863                 leader->group_leader = tsk;
 864
 865                 tsk->exit_signal = SIGCHLD;
 866
 867                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 868                 leader->exit_state = EXIT_DEAD;
 869                 write_unlock_irq(&tasklist_lock);
 870
 871                 release_task(leader);
 872         }
 873
 874         sig->group_exit_task = NULL;
 875         sig->notify_count = 0;
 876
 877 no_thread_group:
 878         if (current->mm)
 879                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 880
 881         exit_itimers(sig);
 882         flush_itimer_signals();
 883
 884         if (atomic_read(&oldsighand->count) != 1) {
 885                 struct sighand_struct *newsighand;
 886                 /*
 887                  * This ->sighand is shared with the CLONE_SIGHAND
 888                  * but not CLONE_THREAD task, switch to the new one.
 889                  */
 890                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 891                 if (!newsighand)
 892                         return -ENOMEM;
 893
 894                 atomic_set(&newsighand->count, 1);
 895                 memcpy(newsighand->action, oldsighand->action,
 896                        sizeof(newsighand->action));
 897
 898                 write_lock_irq(&tasklist_lock);
 899                 spin_lock(&oldsighand->siglock);
 900                 rcu_assign_pointer(tsk->sighand, newsighand);
 901                 spin_unlock(&oldsighand->siglock);
 902                 write_unlock_irq(&tasklist_lock);
 903
 904                 __cleanup_sighand(oldsighand);
 905         }
 906
 907         BUG_ON(!thread_group_leader(tsk));
 908         return 0;
 909 }
 910
 911 /*
 912  * These functions flushes out all traces of the currently running executable
 913  * so that a new one can be started
 914  */
 915 static void flush_old_files(struct files_struct * files)
 916 {
 917         long j = -1;
 918         struct fdtable *fdt;
 919
 920         spin_lock(&files->file_lock);
 921         for (;;) {
 922                 unsigned long set, i;
 923
 924                 j++;
 925                 i = j * __NFDBITS;
 926                 fdt = files_fdtable(files);
 927                 if (i >= fdt->max_fds)
 928                         break;
 929                 set = fdt->close_on_exec->fds_bits[j];
 930                 if (!set)
 931                         continue;
 932                 fdt->close_on_exec->fds_bits[j] = 0;
 933                 spin_unlock(&files->file_lock);
 934                 for ( ; set ; i++,set >>= 1) {
 935                         if (set & 1) {
 936                                 sys_close(i);
 937                         }
 938                 }
 939                 spin_lock(&files->file_lock);
 940
 941         }
 942         spin_unlock(&files->file_lock);
 943 }
 944
 945 char *get_task_comm(char *buf, struct task_struct *tsk)
 946 {
 947         /* buf must be at least sizeof(tsk->comm) in size */
 948         task_lock(tsk);
 949         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 950         task_unlock(tsk);
 951         return buf;
 952 }
 953
 954 void set_task_comm(struct task_struct *tsk, char *buf)
 955 {
 956         task_lock(tsk);
 957
 958         /*
 959          * Threads may access current->comm without holding
 960          * the task lock, so write the string carefully.
 961          * Readers without a lock may see incomplete new
 962          * names but are safe from non-terminating string reads.
 963          */
 964         memset(tsk->comm, 0, TASK_COMM_LEN);
 965         wmb();
 966         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 967         task_unlock(tsk);
 968         perf_event_comm(tsk);
 969 }
 970
 971 int flush_old_exec(struct linux_binprm * bprm)
 972 {
 973         int retval;
 974
 975         /*
 976          * Make sure we have a private signal table and that
 977          * we are unassociated from the previous thread group.
 978          */
 979         retval = de_thread(current);
 980         if (retval)
 981                 goto out;
 982
 983         set_mm_exe_file(bprm->mm, bprm->file);
 984
 985         /*
 986          * Release all of the old mmap stuff
 987          */
 988         retval = exec_mmap(bprm->mm);
 989         if (retval)
 990                 goto out;
 991
 992         bprm->mm = NULL;                /* We're using it now */
 993
 994         current->flags &= ~PF_RANDOMIZE;
 995         flush_thread();
 996         current->personality &= ~bprm->per_clear;
 997
 998         return 0;
 999
1000 out:
1001         return retval;
1002 }
1003 EXPORT_SYMBOL(flush_old_exec);
1004
1005 void setup_new_exec(struct linux_binprm * bprm)
1006 {
1007         int i, ch;
1008         const char *name;
1009         char tcomm[sizeof(current->comm)];
1010
1011         arch_pick_mmap_layout(current->mm);
1012
1013         /* This is the point of no return */
1014         current->sas_ss_sp = current->sas_ss_size = 0;
1015
1016         if (current_euid() == current_uid() && current_egid() == current_gid())
1017                 set_dumpable(current->mm, 1);
1018         else
1019                 set_dumpable(current->mm, suid_dumpable);
1020
1021         name = bprm->filename;
1022
1023         /* Copies the binary name from after last slash */
1024         for (i=0; (ch = *(name++)) != '\0';) {
1025                 if (ch == '/')
1026                         i = 0; /* overwrite what we wrote */
1027                 else
1028                         if (i < (sizeof(tcomm) - 1))
1029                                 tcomm[i++] = ch;
1030         }
1031         tcomm[i] = '\0';
1032         set_task_comm(current, tcomm);
1033
1034         /* Set the new mm task size. We have to do that late because it may
1035          * depend on TIF_32BIT which is only updated in flush_thread() on
1036          * some architectures like powerpc
1037          */
1038         current->mm->task_size = TASK_SIZE;
1039
1040         /* install the new credentials */
1041         if (bprm->cred->uid != current_euid() ||
1042             bprm->cred->gid != current_egid()) {
1043                 current->pdeath_signal = 0;
1044         } else if (file_permission(bprm->file, MAY_READ) ||
1045                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1046                 set_dumpable(current->mm, suid_dumpable);
1047         }
1048
1049         /*
1050          * Flush performance counters when crossing a
1051          * security domain:
1052          */
1053         if (!get_dumpable(current->mm))
1054                 perf_event_exit_task(current);
1055
1056         /* An exec changes our domain. We are no longer part of the thread
1057            group */
1058
1059         current->self_exec_id++;
1060
1061         flush_signal_handlers(current, 0);
1062         flush_old_files(current->files);
1063 }
1064 EXPORT_SYMBOL(setup_new_exec);
1065
1066 /*
1067  * Prepare credentials and lock ->cred_guard_mutex.
1068  * install_exec_creds() commits the new creds and drops the lock.
1069  * Or, if exec fails before, free_bprm() should release ->cred and
1070  * and unlock.
1071  */
1072 int prepare_bprm_creds(struct linux_binprm *bprm)
1073 {
1074         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1075                 return -ERESTARTNOINTR;
1076
1077         bprm->cred = prepare_exec_creds();
1078         if (likely(bprm->cred))
1079                 return 0;
1080
1081         mutex_unlock(&current->cred_guard_mutex);
1082         return -ENOMEM;
1083 }
1084
1085 void free_bprm(struct linux_binprm *bprm)
1086 {
1087         free_arg_pages(bprm);
1088         if (bprm->cred) {
1089                 mutex_unlock(&current->cred_guard_mutex);
1090                 abort_creds(bprm->cred);
1091         }
1092         kfree(bprm);
1093 }
1094
1095 /*
1096  * install the new credentials for this executable
1097  */
1098 void install_exec_creds(struct linux_binprm *bprm)
1099 {
1100         security_bprm_committing_creds(bprm);
1101
1102         commit_creds(bprm->cred);
1103         bprm->cred = NULL;
1104         /*
1105          * cred_guard_mutex must be held at least to this point to prevent
1106          * ptrace_attach() from altering our determination of the task's
1107          * credentials; any time after this it may be unlocked.
1108          */
1109         security_bprm_committed_creds(bprm);
1110         mutex_unlock(&current->cred_guard_mutex);
1111 }
1112 EXPORT_SYMBOL(install_exec_creds);
1113
1114 /*
1115  * determine how safe it is to execute the proposed program
1116  * - the caller must hold current->cred_guard_mutex to protect against
1117  *   PTRACE_ATTACH
1118  */
1119 int check_unsafe_exec(struct linux_binprm *bprm)
1120 {
1121         struct task_struct *p = current, *t;
1122         unsigned n_fs;
1123         int res = 0;
1124
1125         bprm->unsafe = tracehook_unsafe_exec(p);
1126
1127         n_fs = 1;
1128         spin_lock(&p->fs->lock);
1129         rcu_read_lock();
1130         for (t = next_thread(p); t != p; t = next_thread(t)) {
1131                 if (t->fs == p->fs)
1132                         n_fs++;
1133         }
1134         rcu_read_unlock();
1135
1136         if (p->fs->users > n_fs) {
1137                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1138         } else {
1139                 res = -EAGAIN;
1140                 if (!p->fs->in_exec) {
1141                         p->fs->in_exec = 1;
1142                         res = 1;
1143                 }
1144         }
1145         spin_unlock(&p->fs->lock);
1146
1147         return res;
1148 }
1149
1150 /*
1151  * Fill the binprm structure from the inode.
1152  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1153  *
1154  * This may be called multiple times for binary chains (scripts for example).
1155  */
1156 int prepare_binprm(struct linux_binprm *bprm)
1157 {
1158         umode_t mode;
1159         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1160         int retval;
1161
1162         mode = inode->i_mode;
1163         if (bprm->file->f_op == NULL)
1164                 return -EACCES;
1165
1166         /* clear any previous set[ug]id data from a previous binary */
1167         bprm->cred->euid = current_euid();
1168         bprm->cred->egid = current_egid();
1169
1170         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1171                 /* Set-uid? */
1172                 if (mode & S_ISUID) {
1173                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1174                         bprm->cred->euid = inode->i_uid;
1175                 }
1176
1177                 /* Set-gid? */
1178                 /*
1179                  * If setgid is set but no group execute bit then this
1180                  * is a candidate for mandatory locking, not a setgid
1181                  * executable.
1182                  */
1183                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1184                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1185                         bprm->cred->egid = inode->i_gid;
1186                 }
1187         }
1188
1189         /* fill in binprm security blob */
1190         retval = security_bprm_set_creds(bprm);
1191         if (retval)
1192                 return retval;
1193         bprm->cred_prepared = 1;
1194
1195         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1196         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1197 }
1198
1199 EXPORT_SYMBOL(prepare_binprm);
1200
1201 /*
1202  * Arguments are '\0' separated strings found at the location bprm->p
1203  * points to; chop off the first by relocating brpm->p to right after
1204  * the first '\0' encountered.
1205  */
1206 int remove_arg_zero(struct linux_binprm *bprm)
1207 {
1208         int ret = 0;
1209         unsigned long offset;
1210         char *kaddr;
1211         struct page *page;
1212
1213         if (!bprm->argc)
1214                 return 0;
1215
1216         do {
1217                 offset = bprm->p & ~PAGE_MASK;
1218                 page = get_arg_page(bprm, bprm->p, 0);
1219                 if (!page) {
1220                         ret = -EFAULT;
1221                         goto out;
1222                 }
1223                 kaddr = kmap_atomic(page, KM_USER0);
1224
1225                 for (; offset < PAGE_SIZE && kaddr[offset];
1226                                 offset++, bprm->p++)
1227                         ;
1228
1229                 kunmap_atomic(kaddr, KM_USER0);
1230                 put_arg_page(page);
1231
1232                 if (offset == PAGE_SIZE)
1233                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1234         } while (offset == PAGE_SIZE);
1235
1236         bprm->p++;
1237         bprm->argc--;
1238         ret = 0;
1239
1240 out:
1241         return ret;
1242 }
1243 EXPORT_SYMBOL(remove_arg_zero);
1244
1245 /*
1246  * cycle the list of binary formats handler, until one recognizes the image
1247  */
1248 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1249 {
1250         unsigned int depth = bprm->recursion_depth;
1251         int try,retval;
1252         struct linux_binfmt *fmt;
1253
1254         retval = security_bprm_check(bprm);
1255         if (retval)
1256                 return retval;
1257
1258         /* kernel module loader fixup */
1259         /* so we don't try to load run modprobe in kernel space. */
1260         set_fs(USER_DS);
1261
1262         retval = audit_bprm(bprm);
1263         if (retval)
1264                 return retval;
1265
1266         retval = -ENOENT;
1267         for (try=0; try<2; try++) {
1268                 read_lock(&binfmt_lock);
1269                 list_for_each_entry(fmt, &formats, lh) {
1270                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1271                         if (!fn)
1272                                 continue;
1273                         if (!try_module_get(fmt->module))
1274                                 continue;
1275                         read_unlock(&binfmt_lock);
1276                         retval = fn(bprm, regs);
1277                         /*
1278                          * Restore the depth counter to its starting value
1279                          * in this call, so we don't have to rely on every
1280                          * load_binary function to restore it on return.
1281                          */
1282                         bprm->recursion_depth = depth;
1283                         if (retval >= 0) {
1284                                 if (depth == 0)
1285                                         tracehook_report_exec(fmt, bprm, regs);
1286                                 put_binfmt(fmt);
1287                                 allow_write_access(bprm->file);
1288                                 if (bprm->file)
1289                                         fput(bprm->file);
1290                                 bprm->file = NULL;
1291                                 current->did_exec = 1;
1292                                 proc_exec_connector(current);
1293                                 return retval;
1294                         }
1295                         read_lock(&binfmt_lock);
1296                         put_binfmt(fmt);
1297                         if (retval != -ENOEXEC || bprm->mm == NULL)
1298                                 break;
1299                         if (!bprm->file) {
1300                                 read_unlock(&binfmt_lock);
1301                                 return retval;
1302                         }
1303                 }
1304                 read_unlock(&binfmt_lock);
1305                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1306                         break;
1307 #ifdef CONFIG_MODULES
1308                 } else {
1309 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1310                         if (printable(bprm->buf[0]) &&
1311                             printable(bprm->buf[1]) &&
1312                             printable(bprm->buf[2]) &&
1313                             printable(bprm->buf[3]))
1314                                 break; /* -ENOEXEC */
1315                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1316 #endif
1317                 }
1318         }
1319         return retval;
1320 }
1321
1322 EXPORT_SYMBOL(search_binary_handler);
1323
1324 /*
1325  * sys_execve() executes a new program.
1326  */
1327 int do_execve(const char * filename,
1328         const char __user *const __user *argv,
1329         const char __user *const __user *envp,
1330         struct pt_regs * regs)
1331 {
1332         struct linux_binprm *bprm;
1333         struct file *file;
1334         struct files_struct *displaced;
1335         bool clear_in_exec;
1336         int retval;
1337
1338         retval = unshare_files(&displaced);
1339         if (retval)
1340                 goto out_ret;
1341
1342         retval = -ENOMEM;
1343         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1344         if (!bprm)
1345                 goto out_files;
1346
1347         retval = prepare_bprm_creds(bprm);
1348         if (retval)
1349                 goto out_free;
1350
1351         retval = check_unsafe_exec(bprm);
1352         if (retval < 0)
1353                 goto out_free;
1354         clear_in_exec = retval;
1355         current->in_execve = 1;
1356
1357         file = open_exec(filename);
1358         retval = PTR_ERR(file);
1359         if (IS_ERR(file))
1360                 goto out_unmark;
1361
1362         sched_exec();
1363
1364         bprm->file = file;
1365         bprm->filename = filename;
1366         bprm->interp = filename;
1367
1368         retval = bprm_mm_init(bprm);
1369         if (retval)
1370                 goto out_file;
1371
1372         bprm->argc = count(argv, MAX_ARG_STRINGS);
1373         if ((retval = bprm->argc) < 0)
1374                 goto out;
1375
1376         bprm->envc = count(envp, MAX_ARG_STRINGS);
1377         if ((retval = bprm->envc) < 0)
1378                 goto out;
1379
1380         retval = prepare_binprm(bprm);
1381         if (retval < 0)
1382                 goto out;
1383
1384         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1385         if (retval < 0)
1386                 goto out;
1387
1388         bprm->exec = bprm->p;
1389         retval = copy_strings(bprm->envc, envp, bprm);
1390         if (retval < 0)
1391                 goto out;
1392
1393         retval = copy_strings(bprm->argc, argv, bprm);
1394         if (retval < 0)
1395                 goto out;
1396
1397         current->flags &= ~PF_KTHREAD;
1398         retval = search_binary_handler(bprm,regs);
1399         if (retval < 0)
1400                 goto out;
1401
1402         /* execve succeeded */
1403         current->fs->in_exec = 0;
1404         current->in_execve = 0;
1405         acct_update_integrals(current);
1406         free_bprm(bprm);
1407         if (displaced)
1408                 put_files_struct(displaced);
1409         return retval;
1410
1411 out:
1412         if (bprm->mm)
1413                 mmput (bprm->mm);
1414
1415 out_file:
1416         if (bprm->file) {
1417                 allow_write_access(bprm->file);
1418                 fput(bprm->file);
1419         }
1420
1421 out_unmark:
1422         if (clear_in_exec)
1423                 current->fs->in_exec = 0;
1424         current->in_execve = 0;
1425
1426 out_free:
1427         free_bprm(bprm);
1428
1429 out_files:
1430         if (displaced)
1431                 reset_files_struct(displaced);
1432 out_ret:
1433         return retval;
1434 }
1435
1436 void set_binfmt(struct linux_binfmt *new)
1437 {
1438         struct mm_struct *mm = current->mm;
1439
1440         if (mm->binfmt)
1441                 module_put(mm->binfmt->module);
1442
1443         mm->binfmt = new;
1444         if (new)
1445                 __module_get(new->module);
1446 }
1447
1448 EXPORT_SYMBOL(set_binfmt);
1449
1450 /* format_corename will inspect the pattern parameter, and output a
1451  * name into corename, which must have space for at least
1452  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1453  */
1454 static int format_corename(char *corename, long signr)
1455 {
1456         const struct cred *cred = current_cred();
1457         const char *pat_ptr = core_pattern;
1458         int ispipe = (*pat_ptr == '|');
1459         char *out_ptr = corename;
1460         char *const out_end = corename + CORENAME_MAX_SIZE;
1461         int rc;
1462         int pid_in_pattern = 0;
1463
1464         /* Repeat as long as we have more pattern to process and more output
1465            space */
1466         while (*pat_ptr) {
1467                 if (*pat_ptr != '%') {
1468                         if (out_ptr == out_end)
1469                                 goto out;
1470                         *out_ptr++ = *pat_ptr++;
1471                 } else {
1472                         switch (*++pat_ptr) {
1473                         case 0:
1474                                 goto out;
1475                         /* Double percent, output one percent */
1476                         case '%':
1477                                 if (out_ptr == out_end)
1478                                         goto out;
1479                                 *out_ptr++ = '%';
1480                                 break;
1481                         /* pid */
1482                         case 'p':
1483                                 pid_in_pattern = 1;
1484                                 rc = snprintf(out_ptr, out_end - out_ptr,
1485                                               "%d", task_tgid_vnr(current));
1486                                 if (rc > out_end - out_ptr)
1487                                         goto out;
1488                                 out_ptr += rc;
1489                                 break;
1490                         /* uid */
1491                         case 'u':
1492                                 rc = snprintf(out_ptr, out_end - out_ptr,
1493                                               "%d", cred->uid);
1494                                 if (rc > out_end - out_ptr)
1495                                         goto out;
1496                                 out_ptr += rc;
1497                                 break;
1498                         /* gid */
1499                         case 'g':
1500                                 rc = snprintf(out_ptr, out_end - out_ptr,
1501                                               "%d", cred->gid);
1502                                 if (rc > out_end - out_ptr)
1503                                         goto out;
1504                                 out_ptr += rc;
1505                                 break;
1506                         /* signal that caused the coredump */
1507                         case 's':
1508                                 rc = snprintf(out_ptr, out_end - out_ptr,
1509                                               "%ld", signr);
1510                                 if (rc > out_end - out_ptr)
1511                                         goto out;
1512                                 out_ptr += rc;
1513                                 break;
1514                         /* UNIX time of coredump */
1515                         case 't': {
1516                                 struct timeval tv;
1517                                 do_gettimeofday(&tv);
1518                                 rc = snprintf(out_ptr, out_end - out_ptr,
1519                                               "%lu", tv.tv_sec);
1520                                 if (rc > out_end - out_ptr)
1521                                         goto out;
1522                                 out_ptr += rc;
1523                                 break;
1524                         }
1525                         /* hostname */
1526                         case 'h':
1527                                 down_read(&uts_sem);
1528                                 rc = snprintf(out_ptr, out_end - out_ptr,
1529                                               "%s", utsname()->nodename);
1530                                 up_read(&uts_sem);
1531                                 if (rc > out_end - out_ptr)
1532                                         goto out;
1533                                 out_ptr += rc;
1534                                 break;
1535                         /* executable */
1536                         case 'e':
1537                                 rc = snprintf(out_ptr, out_end - out_ptr,
1538                                               "%s", current->comm);
1539                                 if (rc > out_end - out_ptr)
1540                                         goto out;
1541                                 out_ptr += rc;
1542                                 break;
1543                         /* core limit size */
1544                         case 'c':
1545                                 rc = snprintf(out_ptr, out_end - out_ptr,
1546                                               "%lu", rlimit(RLIMIT_CORE));
1547                                 if (rc > out_end - out_ptr)
1548                                         goto out;
1549                                 out_ptr += rc;
1550                                 break;
1551                         default:
1552                                 break;
1553                         }
1554                         ++pat_ptr;
1555                 }
1556         }
1557         /* Backward compatibility with core_uses_pid:
1558          *
1559          * If core_pattern does not include a %p (as is the default)
1560          * and core_uses_pid is set, then .%pid will be appended to
1561          * the filename. Do not do this for piped commands. */
1562         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1563                 rc = snprintf(out_ptr, out_end - out_ptr,
1564                               ".%d", task_tgid_vnr(current));
1565                 if (rc > out_end - out_ptr)
1566                         goto out;
1567                 out_ptr += rc;
1568         }
1569 out:
1570         *out_ptr = 0;
1571         return ispipe;
1572 }
1573
1574 static int zap_process(struct task_struct *start, int exit_code)
1575 {
1576         struct task_struct *t;
1577         int nr = 0;
1578
1579         start->signal->flags = SIGNAL_GROUP_EXIT;
1580         start->signal->group_exit_code = exit_code;
1581         start->signal->group_stop_count = 0;
1582
1583         t = start;
1584         do {
1585                 if (t != current && t->mm) {
1586                         sigaddset(&t->pending.signal, SIGKILL);
1587                         signal_wake_up(t, 1);
1588                         nr++;
1589                 }
1590         } while_each_thread(start, t);
1591
1592         return nr;
1593 }
1594
1595 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1596                                 struct core_state *core_state, int exit_code)
1597 {
1598         struct task_struct *g, *p;
1599         unsigned long flags;
1600         int nr = -EAGAIN;
1601
1602         spin_lock_irq(&tsk->sighand->siglock);
1603         if (!signal_group_exit(tsk->signal)) {
1604                 mm->core_state = core_state;
1605                 nr = zap_process(tsk, exit_code);
1606         }
1607         spin_unlock_irq(&tsk->sighand->siglock);
1608         if (unlikely(nr < 0))
1609                 return nr;
1610
1611         if (atomic_read(&mm->mm_users) == nr + 1)
1612                 goto done;
1613         /*
1614          * We should find and kill all tasks which use this mm, and we should
1615          * count them correctly into ->nr_threads. We don't take tasklist
1616          * lock, but this is safe wrt:
1617          *
1618          * fork:
1619          *      None of sub-threads can fork after zap_process(leader). All
1620          *      processes which were created before this point should be
1621          *      visible to zap_threads() because copy_process() adds the new
1622          *      process to the tail of init_task.tasks list, and lock/unlock
1623          *      of ->siglock provides a memory barrier.
1624          *
1625          * do_exit:
1626          *      The caller holds mm->mmap_sem. This means that the task which
1627          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1628          *      its ->mm.
1629          *
1630          * de_thread:
1631          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1632          *      we must see either old or new leader, this does not matter.
1633          *      However, it can change p->sighand, so lock_task_sighand(p)
1634          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1635          *      it can't fail.
1636          *
1637          *      Note also that "g" can be the old leader with ->mm == NULL
1638          *      and already unhashed and thus removed from ->thread_group.
1639          *      This is OK, __unhash_process()->list_del_rcu() does not
1640          *      clear the ->next pointer, we will find the new leader via
1641          *      next_thread().
1642          */
1643         rcu_read_lock();
1644         for_each_process(g) {
1645                 if (g == tsk->group_leader)
1646                         continue;
1647                 if (g->flags & PF_KTHREAD)
1648                         continue;
1649                 p = g;
1650                 do {
1651                         if (p->mm) {
1652                                 if (unlikely(p->mm == mm)) {
1653                                         lock_task_sighand(p, &flags);
1654                                         nr += zap_process(p, exit_code);
1655                                         unlock_task_sighand(p, &flags);
1656                                 }
1657                                 break;
1658                         }
1659                 } while_each_thread(g, p);
1660         }
1661         rcu_read_unlock();
1662 done:
1663         atomic_set(&core_state->nr_threads, nr);
1664         return nr;
1665 }
1666
1667 static int coredump_wait(int exit_code, struct core_state *core_state)
1668 {
1669         struct task_struct *tsk = current;
1670         struct mm_struct *mm = tsk->mm;
1671         struct completion *vfork_done;
1672         int core_waiters = -EBUSY;
1673
1674         init_completion(&core_state->startup);
1675         core_state->dumper.task = tsk;
1676         core_state->dumper.next = NULL;
1677
1678         down_write(&mm->mmap_sem);
1679         if (!mm->core_state)
1680                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1681         up_write(&mm->mmap_sem);
1682
1683         if (unlikely(core_waiters < 0))
1684                 goto fail;
1685
1686         /*
1687          * Make sure nobody is waiting for us to release the VM,
1688          * otherwise we can deadlock when we wait on each other
1689          */
1690         vfork_done = tsk->vfork_done;
1691         if (vfork_done) {
1692                 tsk->vfork_done = NULL;
1693                 complete(vfork_done);
1694         }
1695
1696         if (core_waiters)
1697                 wait_for_completion(&core_state->startup);
1698 fail:
1699         return core_waiters;
1700 }
1701
1702 static void coredump_finish(struct mm_struct *mm)
1703 {
1704         struct core_thread *curr, *next;
1705         struct task_struct *task;
1706
1707         next = mm->core_state->dumper.next;
1708         while ((curr = next) != NULL) {
1709                 next = curr->next;
1710                 task = curr->task;
1711                 /*
1712                  * see exit_mm(), curr->task must not see
1713                  * ->task == NULL before we read ->next.
1714                  */
1715                 smp_mb();
1716                 curr->task = NULL;
1717                 wake_up_process(task);
1718         }
1719
1720         mm->core_state = NULL;
1721 }
1722
1723 /*
1724  * set_dumpable converts traditional three-value dumpable to two flags and
1725  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1726  * these bits are not changed atomically.  So get_dumpable can observe the
1727  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1728  * return either old dumpable or new one by paying attention to the order of
1729  * modifying the bits.
1730  *
1731  * dumpable |   mm->flags (binary)
1732  * old  new | initial interim  final
1733  * ---------+-----------------------
1734  *  0    1  |   00      01      01
1735  *  0    2  |   00      10(*)   11
1736  *  1    0  |   01      00      00
1737  *  1    2  |   01      11      11
1738  *  2    0  |   11      10(*)   00
1739  *  2    1  |   11      11      01
1740  *
1741  * (*) get_dumpable regards interim value of 10 as 11.
1742  */
1743 void set_dumpable(struct mm_struct *mm, int value)
1744 {
1745         switch (value) {
1746         case 0:
1747                 clear_bit(MMF_DUMPABLE, &mm->flags);
1748                 smp_wmb();
1749                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1750                 break;
1751         case 1:
1752                 set_bit(MMF_DUMPABLE, &mm->flags);
1753                 smp_wmb();
1754                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1755                 break;
1756         case 2:
1757                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1758                 smp_wmb();
1759                 set_bit(MMF_DUMPABLE, &mm->flags);
1760                 break;
1761         }
1762 }
1763
1764 static int __get_dumpable(unsigned long mm_flags)
1765 {
1766         int ret;
1767
1768         ret = mm_flags & MMF_DUMPABLE_MASK;
1769         return (ret >= 2) ? 2 : ret;
1770 }
1771
1772 int get_dumpable(struct mm_struct *mm)
1773 {
1774         return __get_dumpable(mm->flags);
1775 }
1776
1777 static void wait_for_dump_helpers(struct file *file)
1778 {
1779         struct pipe_inode_info *pipe;
1780
1781         pipe = file->f_path.dentry->d_inode->i_pipe;
1782
1783         pipe_lock(pipe);
1784         pipe->readers++;
1785         pipe->writers--;
1786
1787         while ((pipe->readers > 1) && (!signal_pending(current))) {
1788                 wake_up_interruptible_sync(&pipe->wait);
1789                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1790                 pipe_wait(pipe);
1791         }
1792
1793         pipe->readers--;
1794         pipe->writers++;
1795         pipe_unlock(pipe);
1796
1797 }
1798
1799
1800 /*
1801  * uhm_pipe_setup
1802  * helper function to customize the process used
1803  * to collect the core in userspace.  Specifically
1804  * it sets up a pipe and installs it as fd 0 (stdin)
1805  * for the process.  Returns 0 on success, or
1806  * PTR_ERR on failure.
1807  * Note that it also sets the core limit to 1.  This
1808  * is a special value that we use to trap recursive
1809  * core dumps
1810  */
1811 static int umh_pipe_setup(struct subprocess_info *info)
1812 {
1813         struct file *rp, *wp;
1814         struct fdtable *fdt;
1815         struct coredump_params *cp = (struct coredump_params *)info->data;
1816         struct files_struct *cf = current->files;
1817
1818         wp = create_write_pipe(0);
1819         if (IS_ERR(wp))
1820                 return PTR_ERR(wp);
1821
1822         rp = create_read_pipe(wp, 0);
1823         if (IS_ERR(rp)) {
1824                 free_write_pipe(wp);
1825                 return PTR_ERR(rp);
1826         }
1827
1828         cp->file = wp;
1829
1830         sys_close(0);
1831         fd_install(0, rp);
1832         spin_lock(&cf->file_lock);
1833         fdt = files_fdtable(cf);
1834         FD_SET(0, fdt->open_fds);
1835         FD_CLR(0, fdt->close_on_exec);
1836         spin_unlock(&cf->file_lock);
1837
1838         /* and disallow core files too */
1839         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1840
1841         return 0;
1842 }
1843
1844 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1845 {
1846         struct core_state core_state;
1847         char corename[CORENAME_MAX_SIZE + 1];
1848         struct mm_struct *mm = current->mm;
1849         struct linux_binfmt * binfmt;
1850         const struct cred *old_cred;
1851         struct cred *cred;
1852         int retval = 0;
1853         int flag = 0;
1854         int ispipe;
1855         static atomic_t core_dump_count = ATOMIC_INIT(0);
1856         struct coredump_params cprm = {
1857                 .signr = signr,
1858                 .regs = regs,
1859                 .limit = rlimit(RLIMIT_CORE),
1860                 /*
1861                  * We must use the same mm->flags while dumping core to avoid
1862                  * inconsistency of bit flags, since this flag is not protected
1863                  * by any locks.
1864                  */
1865                 .mm_flags = mm->flags,
1866         };
1867
1868         audit_core_dumps(signr);
1869
1870         binfmt = mm->binfmt;
1871         if (!binfmt || !binfmt->core_dump)
1872                 goto fail;
1873         if (!__get_dumpable(cprm.mm_flags))
1874                 goto fail;
1875
1876         cred = prepare_creds();
1877         if (!cred)
1878                 goto fail;
1879         /*
1880          *      We cannot trust fsuid as being the "true" uid of the
1881          *      process nor do we know its entire history. We only know it
1882          *      was tainted so we dump it as root in mode 2.
1883          */
1884         if (__get_dumpable(cprm.mm_flags) == 2) {
1885                 /* Setuid core dump mode */
1886                 flag = O_EXCL;          /* Stop rewrite attacks */
1887                 cred->fsuid = 0;        /* Dump root private */
1888         }
1889
1890         retval = coredump_wait(exit_code, &core_state);
1891         if (retval < 0)
1892                 goto fail_creds;
1893
1894         old_cred = override_creds(cred);
1895
1896         /*
1897          * Clear any false indication of pending signals that might
1898          * be seen by the filesystem code called to write the core file.
1899          */
1900         clear_thread_flag(TIF_SIGPENDING);
1901
1902         ispipe = format_corename(corename, signr);
1903
1904         if (ispipe) {
1905                 int dump_count;
1906                 char **helper_argv;
1907
1908                 if (cprm.limit == 1) {
1909                         /*
1910                          * Normally core limits are irrelevant to pipes, since
1911                          * we're not writing to the file system, but we use
1912                          * cprm.limit of 1 here as a speacial value. Any
1913                          * non-1 limit gets set to RLIM_INFINITY below, but
1914                          * a limit of 0 skips the dump.  This is a consistent
1915                          * way to catch recursive crashes.  We can still crash
1916                          * if the core_pattern binary sets RLIM_CORE =  !1
1917                          * but it runs as root, and can do lots of stupid things
1918                          * Note that we use task_tgid_vnr here to grab the pid
1919                          * of the process group leader.  That way we get the
1920                          * right pid if a thread in a multi-threaded
1921                          * core_pattern process dies.
1922                          */
1923                         printk(KERN_WARNING
1924                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
1925                                 task_tgid_vnr(current), current->comm);
1926                         printk(KERN_WARNING "Aborting core\n");
1927                         goto fail_unlock;
1928                 }
1929                 cprm.limit = RLIM_INFINITY;
1930
1931                 dump_count = atomic_inc_return(&core_dump_count);
1932                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1933                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1934                                task_tgid_vnr(current), current->comm);
1935                         printk(KERN_WARNING "Skipping core dump\n");
1936                         goto fail_dropcount;
1937                 }
1938
1939                 helper_argv = argv_split(GFP_KERNEL, corename+1, NULL);
1940                 if (!helper_argv) {
1941                         printk(KERN_WARNING "%s failed to allocate memory\n",
1942                                __func__);
1943                         goto fail_dropcount;
1944                 }
1945
1946                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
1947                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
1948                                         NULL, &cprm);
1949                 argv_free(helper_argv);
1950                 if (retval) {
1951                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1952                                corename);
1953                         goto close_fail;
1954                 }
1955         } else {
1956                 struct inode *inode;
1957
1958                 if (cprm.limit < binfmt->min_coredump)
1959                         goto fail_unlock;
1960
1961                 cprm.file = filp_open(corename,
1962                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1963                                  0600);
1964                 if (IS_ERR(cprm.file))
1965                         goto fail_unlock;
1966
1967                 inode = cprm.file->f_path.dentry->d_inode;
1968                 if (inode->i_nlink > 1)
1969                         goto close_fail;
1970                 if (d_unhashed(cprm.file->f_path.dentry))
1971                         goto close_fail;
1972                 /*
1973                  * AK: actually i see no reason to not allow this for named
1974                  * pipes etc, but keep the previous behaviour for now.
1975                  */
1976                 if (!S_ISREG(inode->i_mode))
1977                         goto close_fail;
1978                 /*
1979                  * Dont allow local users get cute and trick others to coredump
1980                  * into their pre-created files.
1981                  */
1982                 if (inode->i_uid != current_fsuid())
1983                         goto close_fail;
1984                 if (!cprm.file->f_op || !cprm.file->f_op->write)
1985                         goto close_fail;
1986                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
1987                         goto close_fail;
1988         }
1989
1990         retval = binfmt->core_dump(&cprm);
1991         if (retval)
1992                 current->signal->group_exit_code |= 0x80;
1993
1994         if (ispipe && core_pipe_limit)
1995                 wait_for_dump_helpers(cprm.file);
1996 close_fail:
1997         if (cprm.file)
1998                 filp_close(cprm.file, NULL);
1999 fail_dropcount:
2000         if (ispipe)
2001                 atomic_dec(&core_dump_count);
2002 fail_unlock:
2003         coredump_finish(mm);
2004         revert_creds(old_cred);
2005 fail_creds:
2006         put_cred(cred);
2007 fail:
2008         return;
2009 }