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/smp_lock.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/highmem.h>
  38 #include <linux/spinlock.h>
  39 #include <linux/key.h>
  40 #include <linux/personality.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/utsname.h>
  43 #include <linux/pid_namespace.h>
  44 #include <linux/module.h>
  45 #include <linux/namei.h>
  46 #include <linux/proc_fs.h>
  47 #include <linux/mount.h>
  48 #include <linux/security.h>
  49 #include <linux/ima.h>
  50 #include <linux/syscalls.h>
  51 #include <linux/tsacct_kern.h>
  52 #include <linux/cn_proc.h>
  53 #include <linux/audit.h>
  54 #include <linux/tracehook.h>
  55 #include <linux/kmod.h>
  56 #include <linux/fsnotify.h>
  57 #include <linux/fs_struct.h>
  58 #include <linux/pipe_fs_i.h>
  59
  60 #include <asm/uaccess.h>
  61 #include <asm/mmu_context.h>
  62 #include <asm/tlb.h>
  63 #include "internal.h"
  64
  65 int core_uses_pid;
  66 char core_pattern[CORENAME_MAX_SIZE] = "core";
  67 unsigned int core_pipe_limit;
  68 int suid_dumpable = 0;
  69
  70 /* The maximal length of core_pattern is also specified in sysctl.c */
  71
  72 static LIST_HEAD(formats);
  73 static DEFINE_RWLOCK(binfmt_lock);
  74
  75 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  76 {
  77         if (!fmt)
  78                 return -EINVAL;
  79         write_lock(&binfmt_lock);
  80         insert ? list_add(&fmt->lh, &formats) :
  81                  list_add_tail(&fmt->lh, &formats);
  82         write_unlock(&binfmt_lock);
  83         return 0;
  84 }
  85
  86 EXPORT_SYMBOL(__register_binfmt);
  87
  88 void unregister_binfmt(struct linux_binfmt * fmt)
  89 {
  90         write_lock(&binfmt_lock);
  91         list_del(&fmt->lh);
  92         write_unlock(&binfmt_lock);
  93 }
  94
  95 EXPORT_SYMBOL(unregister_binfmt);
  96
  97 static inline void put_binfmt(struct linux_binfmt * fmt)
  98 {
  99         module_put(fmt->module);
 100 }
 101
 102 /*
 103  * Note that a shared library must be both readable and executable due to
 104  * security reasons.
 105  *
 106  * Also note that we take the address to load from from the file itself.
 107  */
 108 SYSCALL_DEFINE1(uselib, const char __user *, library)
 109 {
 110         struct file *file;
 111         char *tmp = getname(library);
 112         int error = PTR_ERR(tmp);
 113
 114         if (IS_ERR(tmp))
 115                 goto out;
 116
 117         file = do_filp_open(AT_FDCWD, tmp,
 118                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 119                                 MAY_READ | MAY_EXEC | MAY_OPEN);
 120         putname(tmp);
 121         error = PTR_ERR(file);
 122         if (IS_ERR(file))
 123                 goto out;
 124
 125         error = -EINVAL;
 126         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 127                 goto exit;
 128
 129         error = -EACCES;
 130         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 131                 goto exit;
 132
 133         fsnotify_open(file->f_path.dentry);
 134
 135         error = -ENOEXEC;
 136         if(file->f_op) {
 137                 struct linux_binfmt * fmt;
 138
 139                 read_lock(&binfmt_lock);
 140                 list_for_each_entry(fmt, &formats, lh) {
 141                         if (!fmt->load_shlib)
 142                                 continue;
 143                         if (!try_module_get(fmt->module))
 144                                 continue;
 145                         read_unlock(&binfmt_lock);
 146                         error = fmt->load_shlib(file);
 147                         read_lock(&binfmt_lock);
 148                         put_binfmt(fmt);
 149                         if (error != -ENOEXEC)
 150                                 break;
 151                 }
 152                 read_unlock(&binfmt_lock);
 153         }
 154 exit:
 155         fput(file);
 156 out:
 157         return error;
 158 }
 159
 160 #ifdef CONFIG_MMU
 161
 162 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 163                 int write)
 164 {
 165         struct page *page;
 166         int ret;
 167
 168 #ifdef CONFIG_STACK_GROWSUP
 169         if (write) {
 170                 ret = expand_stack_downwards(bprm->vma, pos);
 171                 if (ret < 0)
 172                         return NULL;
 173         }
 174 #endif
 175         ret = get_user_pages(current, bprm->mm, pos,
 176                         1, write, 1, &page, NULL);
 177         if (ret <= 0)
 178                 return NULL;
 179
 180         if (write) {
 181                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 182                 struct rlimit *rlim;
 183
 184                 /*
 185                  * We've historically supported up to 32 pages (ARG_MAX)
 186                  * of argument strings even with small stacks
 187                  */
 188                 if (size <= ARG_MAX)
 189                         return page;
 190
 191                 /*
 192                  * Limit to 1/4-th the stack size for the argv+env strings.
 193                  * This ensures that:
 194                  *  - the remaining binfmt code will not run out of stack space,
 195                  *  - the program will have a reasonable amount of stack left
 196                  *    to work from.
 197                  */
 198                 rlim = current->signal->rlim;
 199                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 200                         put_page(page);
 201                         return NULL;
 202                 }
 203         }
 204
 205         return page;
 206 }
 207
 208 static void put_arg_page(struct page *page)
 209 {
 210         put_page(page);
 211 }
 212
 213 static void free_arg_page(struct linux_binprm *bprm, int i)
 214 {
 215 }
 216
 217 static void free_arg_pages(struct linux_binprm *bprm)
 218 {
 219 }
 220
 221 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 222                 struct page *page)
 223 {
 224         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 225 }
 226
 227 static int __bprm_mm_init(struct linux_binprm *bprm)
 228 {
 229         int err;
 230         struct vm_area_struct *vma = NULL;
 231         struct mm_struct *mm = bprm->mm;
 232
 233         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 234         if (!vma)
 235                 return -ENOMEM;
 236
 237         down_write(&mm->mmap_sem);
 238         vma->vm_mm = mm;
 239
 240         /*
 241          * Place the stack at the largest stack address the architecture
 242          * supports. Later, we'll move this to an appropriate place. We don't
 243          * use STACK_TOP because that can depend on attributes which aren't
 244          * configured yet.
 245          */
 246         vma->vm_end = STACK_TOP_MAX;
 247         vma->vm_start = vma->vm_end - PAGE_SIZE;
 248         vma->vm_flags = VM_STACK_FLAGS;
 249         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 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(char __user * __user * argv, int max)
 365 {
 366         int i = 0;
 367
 368         if (argv != NULL) {
 369                 for (;;) {
 370                         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, char __user * __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                 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                         offset = pos % PAGE_SIZE;
 423                         if (offset == 0)
 424                                 offset = PAGE_SIZE;
 425
 426                         bytes_to_copy = offset;
 427                         if (bytes_to_copy > len)
 428                                 bytes_to_copy = len;
 429
 430                         offset -= bytes_to_copy;
 431                         pos -= bytes_to_copy;
 432                         str -= bytes_to_copy;
 433                         len -= bytes_to_copy;
 434
 435                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 436                                 struct page *page;
 437
 438                                 page = get_arg_page(bprm, pos, 1);
 439                                 if (!page) {
 440                                         ret = -E2BIG;
 441                                         goto out;
 442                                 }
 443
 444                                 if (kmapped_page) {
 445                                         flush_kernel_dcache_page(kmapped_page);
 446                                         kunmap(kmapped_page);
 447                                         put_arg_page(kmapped_page);
 448                                 }
 449                                 kmapped_page = page;
 450                                 kaddr = kmap(kmapped_page);
 451                                 kpos = pos & PAGE_MASK;
 452                                 flush_arg_page(bprm, kpos, kmapped_page);
 453                         }
 454                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 455                                 ret = -EFAULT;
 456                                 goto out;
 457                         }
 458                 }
 459         }
 460         ret = 0;
 461 out:
 462         if (kmapped_page) {
 463                 flush_kernel_dcache_page(kmapped_page);
 464                 kunmap(kmapped_page);
 465                 put_arg_page(kmapped_page);
 466         }
 467         return ret;
 468 }
 469
 470 /*
 471  * Like copy_strings, but get argv and its values from kernel memory.
 472  */
 473 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 474 {
 475         int r;
 476         mm_segment_t oldfs = get_fs();
 477         set_fs(KERNEL_DS);
 478         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 479         set_fs(oldfs);
 480         return r;
 481 }
 482 EXPORT_SYMBOL(copy_strings_kernel);
 483
 484 #ifdef CONFIG_MMU
 485
 486 /*
 487  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 488  * the binfmt code determines where the new stack should reside, we shift it to
 489  * its final location.  The process proceeds as follows:
 490  *
 491  * 1) Use shift to calculate the new vma endpoints.
 492  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 493  *    arguments passed to subsequent functions are consistent.
 494  * 3) Move vma's page tables to the new range.
 495  * 4) Free up any cleared pgd range.
 496  * 5) Shrink the vma to cover only the new range.
 497  */
 498 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 499 {
 500         struct mm_struct *mm = vma->vm_mm;
 501         unsigned long old_start = vma->vm_start;
 502         unsigned long old_end = vma->vm_end;
 503         unsigned long length = old_end - old_start;
 504         unsigned long new_start = old_start - shift;
 505         unsigned long new_end = old_end - shift;
 506         struct mmu_gather *tlb;
 507
 508         BUG_ON(new_start > new_end);
 509
 510         /*
 511          * ensure there are no vmas between where we want to go
 512          * and where we are
 513          */
 514         if (vma != find_vma(mm, new_start))
 515                 return -EFAULT;
 516
 517         /*
 518          * cover the whole range: [new_start, old_end)
 519          */
 520         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 521
 522         /*
 523          * move the page tables downwards, on failure we rely on
 524          * process cleanup to remove whatever mess we made.
 525          */
 526         if (length != move_page_tables(vma, old_start,
 527                                        vma, new_start, length))
 528                 return -ENOMEM;
 529
 530         lru_add_drain();
 531         tlb = tlb_gather_mmu(mm, 0);
 532         if (new_end > old_start) {
 533                 /*
 534                  * when the old and new regions overlap clear from new_end.
 535                  */
 536                 free_pgd_range(tlb, new_end, old_end, new_end,
 537                         vma->vm_next ? vma->vm_next->vm_start : 0);
 538         } else {
 539                 /*
 540                  * otherwise, clean from old_start; this is done to not touch
 541                  * the address space in [new_end, old_start) some architectures
 542                  * have constraints on va-space that make this illegal (IA64) -
 543                  * for the others its just a little faster.
 544                  */
 545                 free_pgd_range(tlb, old_start, old_end, new_end,
 546                         vma->vm_next ? vma->vm_next->vm_start : 0);
 547         }
 548         tlb_finish_mmu(tlb, new_end, old_end);
 549
 550         /*
 551          * shrink the vma to just the new range.
 552          */
 553         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 554
 555         return 0;
 556 }
 557
 558 #define EXTRA_STACK_VM_PAGES    20      /* random */
 559
 560 /*
 561  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 562  * the stack is optionally relocated, and some extra space is added.
 563  */
 564 int setup_arg_pages(struct linux_binprm *bprm,
 565                     unsigned long stack_top,
 566                     int executable_stack)
 567 {
 568         unsigned long ret;
 569         unsigned long stack_shift;
 570         struct mm_struct *mm = current->mm;
 571         struct vm_area_struct *vma = bprm->vma;
 572         struct vm_area_struct *prev = NULL;
 573         unsigned long vm_flags;
 574         unsigned long stack_base;
 575         unsigned long stack_size;
 576         unsigned long stack_expand;
 577         unsigned long rlim_stack;
 578
 579 #ifdef CONFIG_STACK_GROWSUP
 580         /* Limit stack size to 1GB */
 581         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 582         if (stack_base > (1 << 30))
 583                 stack_base = 1 << 30;
 584
 585         /* Make sure we didn't let the argument array grow too large. */
 586         if (vma->vm_end - vma->vm_start > stack_base)
 587                 return -ENOMEM;
 588
 589         stack_base = PAGE_ALIGN(stack_top - stack_base);
 590
 591         stack_shift = vma->vm_start - stack_base;
 592         mm->arg_start = bprm->p - stack_shift;
 593         bprm->p = vma->vm_end - stack_shift;
 594 #else
 595         stack_top = arch_align_stack(stack_top);
 596         stack_top = PAGE_ALIGN(stack_top);
 597
 598         if (unlikely(stack_top < mmap_min_addr) ||
 599             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 600                 return -ENOMEM;
 601
 602         stack_shift = vma->vm_end - stack_top;
 603
 604         bprm->p -= stack_shift;
 605         mm->arg_start = bprm->p;
 606 #endif
 607
 608         if (bprm->loader)
 609                 bprm->loader -= stack_shift;
 610         bprm->exec -= stack_shift;
 611
 612         down_write(&mm->mmap_sem);
 613         vm_flags = VM_STACK_FLAGS;
 614
 615         /*
 616          * Adjust stack execute permissions; explicitly enable for
 617          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 618          * (arch default) otherwise.
 619          */
 620         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 621                 vm_flags |= VM_EXEC;
 622         else if (executable_stack == EXSTACK_DISABLE_X)
 623                 vm_flags &= ~VM_EXEC;
 624         vm_flags |= mm->def_flags;
 625
 626         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 627                         vm_flags);
 628         if (ret)
 629                 goto out_unlock;
 630         BUG_ON(prev != vma);
 631
 632         /* Move stack pages down in memory. */
 633         if (stack_shift) {
 634                 ret = shift_arg_pages(vma, stack_shift);
 635                 if (ret)
 636                         goto out_unlock;
 637         }
 638
 639         stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 640         stack_size = vma->vm_end - vma->vm_start;
 641         /*
 642          * Align this down to a page boundary as expand_stack
 643          * will align it up.
 644          */
 645         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 646 #ifdef CONFIG_STACK_GROWSUP
 647         if (stack_size + stack_expand > rlim_stack)
 648                 stack_base = vma->vm_start + rlim_stack;
 649         else
 650                 stack_base = vma->vm_end + stack_expand;
 651 #else
 652         if (stack_size + stack_expand > rlim_stack)
 653                 stack_base = vma->vm_end - rlim_stack;
 654         else
 655                 stack_base = vma->vm_start - stack_expand;
 656 #endif
 657         ret = expand_stack(vma, stack_base);
 658         if (ret)
 659                 ret = -EFAULT;
 660
 661 out_unlock:
 662         up_write(&mm->mmap_sem);
 663         return ret;
 664 }
 665 EXPORT_SYMBOL(setup_arg_pages);
 666
 667 #endif /* CONFIG_MMU */
 668
 669 struct file *open_exec(const char *name)
 670 {
 671         struct file *file;
 672         int err;
 673
 674         file = do_filp_open(AT_FDCWD, name,
 675                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 676                                 MAY_EXEC | MAY_OPEN);
 677         if (IS_ERR(file))
 678                 goto out;
 679
 680         err = -EACCES;
 681         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 682                 goto exit;
 683
 684         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 685                 goto exit;
 686
 687         fsnotify_open(file->f_path.dentry);
 688
 689         err = deny_write_access(file);
 690         if (err)
 691                 goto exit;
 692
 693 out:
 694         return file;
 695
 696 exit:
 697         fput(file);
 698         return ERR_PTR(err);
 699 }
 700 EXPORT_SYMBOL(open_exec);
 701
 702 int kernel_read(struct file *file, loff_t offset,
 703                 char *addr, unsigned long count)
 704 {
 705         mm_segment_t old_fs;
 706         loff_t pos = offset;
 707         int result;
 708
 709         old_fs = get_fs();
 710         set_fs(get_ds());
 711         /* The cast to a user pointer is valid due to the set_fs() */
 712         result = vfs_read(file, (void __user *)addr, count, &pos);
 713         set_fs(old_fs);
 714         return result;
 715 }
 716
 717 EXPORT_SYMBOL(kernel_read);
 718
 719 static int exec_mmap(struct mm_struct *mm)
 720 {
 721         struct task_struct *tsk;
 722         struct mm_struct * old_mm, *active_mm;
 723
 724         /* Notify parent that we're no longer interested in the old VM */
 725         tsk = current;
 726         old_mm = current->mm;
 727         mm_release(tsk, old_mm);
 728
 729         if (old_mm) {
 730                 /*
 731                  * Make sure that if there is a core dump in progress
 732                  * for the old mm, we get out and die instead of going
 733                  * through with the exec.  We must hold mmap_sem around
 734                  * checking core_state and changing tsk->mm.
 735                  */
 736                 down_read(&old_mm->mmap_sem);
 737                 if (unlikely(old_mm->core_state)) {
 738                         up_read(&old_mm->mmap_sem);
 739                         return -EINTR;
 740                 }
 741         }
 742         task_lock(tsk);
 743         active_mm = tsk->active_mm;
 744         tsk->mm = mm;
 745         tsk->active_mm = mm;
 746         activate_mm(active_mm, mm);
 747         task_unlock(tsk);
 748         arch_pick_mmap_layout(mm);
 749         if (old_mm) {
 750                 up_read(&old_mm->mmap_sem);
 751                 BUG_ON(active_mm != old_mm);
 752                 mm_update_next_owner(old_mm);
 753                 mmput(old_mm);
 754                 return 0;
 755         }
 756         mmdrop(active_mm);
 757         return 0;
 758 }
 759
 760 /*
 761  * This function makes sure the current process has its own signal table,
 762  * so that flush_signal_handlers can later reset the handlers without
 763  * disturbing other processes.  (Other processes might share the signal
 764  * table via the CLONE_SIGHAND option to clone().)
 765  */
 766 static int de_thread(struct task_struct *tsk)
 767 {
 768         struct signal_struct *sig = tsk->signal;
 769         struct sighand_struct *oldsighand = tsk->sighand;
 770         spinlock_t *lock = &oldsighand->siglock;
 771         int count;
 772
 773         if (thread_group_empty(tsk))
 774                 goto no_thread_group;
 775
 776         /*
 777          * Kill all other threads in the thread group.
 778          */
 779         spin_lock_irq(lock);
 780         if (signal_group_exit(sig)) {
 781                 /*
 782                  * Another group action in progress, just
 783                  * return so that the signal is processed.
 784                  */
 785                 spin_unlock_irq(lock);
 786                 return -EAGAIN;
 787         }
 788         sig->group_exit_task = tsk;
 789         zap_other_threads(tsk);
 790
 791         /* Account for the thread group leader hanging around: */
 792         count = thread_group_leader(tsk) ? 1 : 2;
 793         sig->notify_count = count;
 794         while (atomic_read(&sig->count) > count) {
 795                 __set_current_state(TASK_UNINTERRUPTIBLE);
 796                 spin_unlock_irq(lock);
 797                 schedule();
 798                 spin_lock_irq(lock);
 799         }
 800         spin_unlock_irq(lock);
 801
 802         /*
 803          * At this point all other threads have exited, all we have to
 804          * do is to wait for the thread group leader to become inactive,
 805          * and to assume its PID:
 806          */
 807         if (!thread_group_leader(tsk)) {
 808                 struct task_struct *leader = tsk->group_leader;
 809
 810                 sig->notify_count = -1; /* for exit_notify() */
 811                 for (;;) {
 812                         write_lock_irq(&tasklist_lock);
 813                         if (likely(leader->exit_state))
 814                                 break;
 815                         __set_current_state(TASK_UNINTERRUPTIBLE);
 816                         write_unlock_irq(&tasklist_lock);
 817                         schedule();
 818                 }
 819
 820                 /*
 821                  * The only record we have of the real-time age of a
 822                  * process, regardless of execs it's done, is start_time.
 823                  * All the past CPU time is accumulated in signal_struct
 824                  * from sister threads now dead.  But in this non-leader
 825                  * exec, nothing survives from the original leader thread,
 826                  * whose birth marks the true age of this process now.
 827                  * When we take on its identity by switching to its PID, we
 828                  * also take its birthdate (always earlier than our own).
 829                  */
 830                 tsk->start_time = leader->start_time;
 831
 832                 BUG_ON(!same_thread_group(leader, tsk));
 833                 BUG_ON(has_group_leader_pid(tsk));
 834                 /*
 835                  * An exec() starts a new thread group with the
 836                  * TGID of the previous thread group. Rehash the
 837                  * two threads with a switched PID, and release
 838                  * the former thread group leader:
 839                  */
 840
 841                 /* Become a process group leader with the old leader's pid.
 842                  * The old leader becomes a thread of the this thread group.
 843                  * Note: The old leader also uses this pid until release_task
 844                  *       is called.  Odd but simple and correct.
 845                  */
 846                 detach_pid(tsk, PIDTYPE_PID);
 847                 tsk->pid = leader->pid;
 848                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 849                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 850                 transfer_pid(leader, tsk, PIDTYPE_SID);
 851                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 852
 853                 tsk->group_leader = tsk;
 854                 leader->group_leader = tsk;
 855
 856                 tsk->exit_signal = SIGCHLD;
 857
 858                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 859                 leader->exit_state = EXIT_DEAD;
 860                 write_unlock_irq(&tasklist_lock);
 861
 862                 release_task(leader);
 863         }
 864
 865         sig->group_exit_task = NULL;
 866         sig->notify_count = 0;
 867
 868 no_thread_group:
 869         if (current->mm)
 870                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 871
 872         exit_itimers(sig);
 873         flush_itimer_signals();
 874
 875         if (atomic_read(&oldsighand->count) != 1) {
 876                 struct sighand_struct *newsighand;
 877                 /*
 878                  * This ->sighand is shared with the CLONE_SIGHAND
 879                  * but not CLONE_THREAD task, switch to the new one.
 880                  */
 881                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 882                 if (!newsighand)
 883                         return -ENOMEM;
 884
 885                 atomic_set(&newsighand->count, 1);
 886                 memcpy(newsighand->action, oldsighand->action,
 887                        sizeof(newsighand->action));
 888
 889                 write_lock_irq(&tasklist_lock);
 890                 spin_lock(&oldsighand->siglock);
 891                 rcu_assign_pointer(tsk->sighand, newsighand);
 892                 spin_unlock(&oldsighand->siglock);
 893                 write_unlock_irq(&tasklist_lock);
 894
 895                 __cleanup_sighand(oldsighand);
 896         }
 897
 898         BUG_ON(!thread_group_leader(tsk));
 899         return 0;
 900 }
 901
 902 /*
 903  * These functions flushes out all traces of the currently running executable
 904  * so that a new one can be started
 905  */
 906 static void flush_old_files(struct files_struct * files)
 907 {
 908         long j = -1;
 909         struct fdtable *fdt;
 910
 911         spin_lock(&files->file_lock);
 912         for (;;) {
 913                 unsigned long set, i;
 914
 915                 j++;
 916                 i = j * __NFDBITS;
 917                 fdt = files_fdtable(files);
 918                 if (i >= fdt->max_fds)
 919                         break;
 920                 set = fdt->close_on_exec->fds_bits[j];
 921                 if (!set)
 922                         continue;
 923                 fdt->close_on_exec->fds_bits[j] = 0;
 924                 spin_unlock(&files->file_lock);
 925                 for ( ; set ; i++,set >>= 1) {
 926                         if (set & 1) {
 927                                 sys_close(i);
 928                         }
 929                 }
 930                 spin_lock(&files->file_lock);
 931
 932         }
 933         spin_unlock(&files->file_lock);
 934 }
 935
 936 char *get_task_comm(char *buf, struct task_struct *tsk)
 937 {
 938         /* buf must be at least sizeof(tsk->comm) in size */
 939         task_lock(tsk);
 940         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 941         task_unlock(tsk);
 942         return buf;
 943 }
 944
 945 void set_task_comm(struct task_struct *tsk, char *buf)
 946 {
 947         task_lock(tsk);
 948         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 949         task_unlock(tsk);
 950         perf_event_comm(tsk);
 951 }
 952
 953 int flush_old_exec(struct linux_binprm * bprm)
 954 {
 955         int retval;
 956
 957         /*
 958          * Make sure we have a private signal table and that
 959          * we are unassociated from the previous thread group.
 960          */
 961         retval = de_thread(current);
 962         if (retval)
 963                 goto out;
 964
 965         set_mm_exe_file(bprm->mm, bprm->file);
 966
 967         /*
 968          * Release all of the old mmap stuff
 969          */
 970         retval = exec_mmap(bprm->mm);
 971         if (retval)
 972                 goto out;
 973
 974         bprm->mm = NULL;                /* We're using it now */
 975
 976         current->flags &= ~PF_RANDOMIZE;
 977         flush_thread();
 978         current->personality &= ~bprm->per_clear;
 979
 980         return 0;
 981
 982 out:
 983         return retval;
 984 }
 985 EXPORT_SYMBOL(flush_old_exec);
 986
 987 void setup_new_exec(struct linux_binprm * bprm)
 988 {
 989         int i, ch;
 990         char * name;
 991         char tcomm[sizeof(current->comm)];
 992
 993         arch_pick_mmap_layout(current->mm);
 994
 995         /* This is the point of no return */
 996         current->sas_ss_sp = current->sas_ss_size = 0;
 997
 998         if (current_euid() == current_uid() && current_egid() == current_gid())
 999                 set_dumpable(current->mm, 1);
1000         else
1001                 set_dumpable(current->mm, suid_dumpable);
1002
1003         name = bprm->filename;
1004
1005         /* Copies the binary name from after last slash */
1006         for (i=0; (ch = *(name++)) != '\0';) {
1007                 if (ch == '/')
1008                         i = 0; /* overwrite what we wrote */
1009                 else
1010                         if (i < (sizeof(tcomm) - 1))
1011                                 tcomm[i++] = ch;
1012         }
1013         tcomm[i] = '\0';
1014         set_task_comm(current, tcomm);
1015
1016         /* Set the new mm task size. We have to do that late because it may
1017          * depend on TIF_32BIT which is only updated in flush_thread() on
1018          * some architectures like powerpc
1019          */
1020         current->mm->task_size = TASK_SIZE;
1021
1022         /* install the new credentials */
1023         if (bprm->cred->uid != current_euid() ||
1024             bprm->cred->gid != current_egid()) {
1025                 current->pdeath_signal = 0;
1026         } else if (file_permission(bprm->file, MAY_READ) ||
1027                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1028                 set_dumpable(current->mm, suid_dumpable);
1029         }
1030
1031         /*
1032          * Flush performance counters when crossing a
1033          * security domain:
1034          */
1035         if (!get_dumpable(current->mm))
1036                 perf_event_exit_task(current);
1037
1038         /* An exec changes our domain. We are no longer part of the thread
1039            group */
1040
1041         current->self_exec_id++;
1042
1043         flush_signal_handlers(current, 0);
1044         flush_old_files(current->files);
1045 }
1046 EXPORT_SYMBOL(setup_new_exec);
1047
1048 /*
1049  * Prepare credentials and lock ->cred_guard_mutex.
1050  * install_exec_creds() commits the new creds and drops the lock.
1051  * Or, if exec fails before, free_bprm() should release ->cred and
1052  * and unlock.
1053  */
1054 int prepare_bprm_creds(struct linux_binprm *bprm)
1055 {
1056         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1057                 return -ERESTARTNOINTR;
1058
1059         bprm->cred = prepare_exec_creds();
1060         if (likely(bprm->cred))
1061                 return 0;
1062
1063         mutex_unlock(&current->cred_guard_mutex);
1064         return -ENOMEM;
1065 }
1066
1067 void free_bprm(struct linux_binprm *bprm)
1068 {
1069         free_arg_pages(bprm);
1070         if (bprm->cred) {
1071                 mutex_unlock(&current->cred_guard_mutex);
1072                 abort_creds(bprm->cred);
1073         }
1074         kfree(bprm);
1075 }
1076
1077 /*
1078  * install the new credentials for this executable
1079  */
1080 void install_exec_creds(struct linux_binprm *bprm)
1081 {
1082         security_bprm_committing_creds(bprm);
1083
1084         commit_creds(bprm->cred);
1085         bprm->cred = NULL;
1086         /*
1087          * cred_guard_mutex must be held at least to this point to prevent
1088          * ptrace_attach() from altering our determination of the task's
1089          * credentials; any time after this it may be unlocked.
1090          */
1091         security_bprm_committed_creds(bprm);
1092         mutex_unlock(&current->cred_guard_mutex);
1093 }
1094 EXPORT_SYMBOL(install_exec_creds);
1095
1096 /*
1097  * determine how safe it is to execute the proposed program
1098  * - the caller must hold current->cred_guard_mutex to protect against
1099  *   PTRACE_ATTACH
1100  */
1101 int check_unsafe_exec(struct linux_binprm *bprm)
1102 {
1103         struct task_struct *p = current, *t;
1104         unsigned n_fs;
1105         int res = 0;
1106
1107         bprm->unsafe = tracehook_unsafe_exec(p);
1108
1109         n_fs = 1;
1110         write_lock(&p->fs->lock);
1111         rcu_read_lock();
1112         for (t = next_thread(p); t != p; t = next_thread(t)) {
1113                 if (t->fs == p->fs)
1114                         n_fs++;
1115         }
1116         rcu_read_unlock();
1117
1118         if (p->fs->users > n_fs) {
1119                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1120         } else {
1121                 res = -EAGAIN;
1122                 if (!p->fs->in_exec) {
1123                         p->fs->in_exec = 1;
1124                         res = 1;
1125                 }
1126         }
1127         write_unlock(&p->fs->lock);
1128
1129         return res;
1130 }
1131
1132 /*
1133  * Fill the binprm structure from the inode.
1134  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1135  *
1136  * This may be called multiple times for binary chains (scripts for example).
1137  */
1138 int prepare_binprm(struct linux_binprm *bprm)
1139 {
1140         umode_t mode;
1141         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1142         int retval;
1143
1144         mode = inode->i_mode;
1145         if (bprm->file->f_op == NULL)
1146                 return -EACCES;
1147
1148         /* clear any previous set[ug]id data from a previous binary */
1149         bprm->cred->euid = current_euid();
1150         bprm->cred->egid = current_egid();
1151
1152         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1153                 /* Set-uid? */
1154                 if (mode & S_ISUID) {
1155                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1156                         bprm->cred->euid = inode->i_uid;
1157                 }
1158
1159                 /* Set-gid? */
1160                 /*
1161                  * If setgid is set but no group execute bit then this
1162                  * is a candidate for mandatory locking, not a setgid
1163                  * executable.
1164                  */
1165                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1166                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1167                         bprm->cred->egid = inode->i_gid;
1168                 }
1169         }
1170
1171         /* fill in binprm security blob */
1172         retval = security_bprm_set_creds(bprm);
1173         if (retval)
1174                 return retval;
1175         bprm->cred_prepared = 1;
1176
1177         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1178         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1179 }
1180
1181 EXPORT_SYMBOL(prepare_binprm);
1182
1183 /*
1184  * Arguments are '\0' separated strings found at the location bprm->p
1185  * points to; chop off the first by relocating brpm->p to right after
1186  * the first '\0' encountered.
1187  */
1188 int remove_arg_zero(struct linux_binprm *bprm)
1189 {
1190         int ret = 0;
1191         unsigned long offset;
1192         char *kaddr;
1193         struct page *page;
1194
1195         if (!bprm->argc)
1196                 return 0;
1197
1198         do {
1199                 offset = bprm->p & ~PAGE_MASK;
1200                 page = get_arg_page(bprm, bprm->p, 0);
1201                 if (!page) {
1202                         ret = -EFAULT;
1203                         goto out;
1204                 }
1205                 kaddr = kmap_atomic(page, KM_USER0);
1206
1207                 for (; offset < PAGE_SIZE && kaddr[offset];
1208                                 offset++, bprm->p++)
1209                         ;
1210
1211                 kunmap_atomic(kaddr, KM_USER0);
1212                 put_arg_page(page);
1213
1214                 if (offset == PAGE_SIZE)
1215                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1216         } while (offset == PAGE_SIZE);
1217
1218         bprm->p++;
1219         bprm->argc--;
1220         ret = 0;
1221
1222 out:
1223         return ret;
1224 }
1225 EXPORT_SYMBOL(remove_arg_zero);
1226
1227 /*
1228  * cycle the list of binary formats handler, until one recognizes the image
1229  */
1230 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1231 {
1232         unsigned int depth = bprm->recursion_depth;
1233         int try,retval;
1234         struct linux_binfmt *fmt;
1235
1236         retval = security_bprm_check(bprm);
1237         if (retval)
1238                 return retval;
1239         retval = ima_bprm_check(bprm);
1240         if (retval)
1241                 return retval;
1242
1243         /* kernel module loader fixup */
1244         /* so we don't try to load run modprobe in kernel space. */
1245         set_fs(USER_DS);
1246
1247         retval = audit_bprm(bprm);
1248         if (retval)
1249                 return retval;
1250
1251         retval = -ENOENT;
1252         for (try=0; try<2; try++) {
1253                 read_lock(&binfmt_lock);
1254                 list_for_each_entry(fmt, &formats, lh) {
1255                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1256                         if (!fn)
1257                                 continue;
1258                         if (!try_module_get(fmt->module))
1259                                 continue;
1260                         read_unlock(&binfmt_lock);
1261                         retval = fn(bprm, regs);
1262                         /*
1263                          * Restore the depth counter to its starting value
1264                          * in this call, so we don't have to rely on every
1265                          * load_binary function to restore it on return.
1266                          */
1267                         bprm->recursion_depth = depth;
1268                         if (retval >= 0) {
1269                                 if (depth == 0)
1270                                         tracehook_report_exec(fmt, bprm, regs);
1271                                 put_binfmt(fmt);
1272                                 allow_write_access(bprm->file);
1273                                 if (bprm->file)
1274                                         fput(bprm->file);
1275                                 bprm->file = NULL;
1276                                 current->did_exec = 1;
1277                                 proc_exec_connector(current);
1278                                 return retval;
1279                         }
1280                         read_lock(&binfmt_lock);
1281                         put_binfmt(fmt);
1282                         if (retval != -ENOEXEC || bprm->mm == NULL)
1283                                 break;
1284                         if (!bprm->file) {
1285                                 read_unlock(&binfmt_lock);
1286                                 return retval;
1287                         }
1288                 }
1289                 read_unlock(&binfmt_lock);
1290                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1291                         break;
1292 #ifdef CONFIG_MODULES
1293                 } else {
1294 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1295                         if (printable(bprm->buf[0]) &&
1296                             printable(bprm->buf[1]) &&
1297                             printable(bprm->buf[2]) &&
1298                             printable(bprm->buf[3]))
1299                                 break; /* -ENOEXEC */
1300                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1301 #endif
1302                 }
1303         }
1304         return retval;
1305 }
1306
1307 EXPORT_SYMBOL(search_binary_handler);
1308
1309 /*
1310  * sys_execve() executes a new program.
1311  */
1312 int do_execve(char * filename,
1313         char __user *__user *argv,
1314         char __user *__user *envp,
1315         struct pt_regs * regs)
1316 {
1317         struct linux_binprm *bprm;
1318         struct file *file;
1319         struct files_struct *displaced;
1320         bool clear_in_exec;
1321         int retval;
1322
1323         retval = unshare_files(&displaced);
1324         if (retval)
1325                 goto out_ret;
1326
1327         retval = -ENOMEM;
1328         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1329         if (!bprm)
1330                 goto out_files;
1331
1332         retval = prepare_bprm_creds(bprm);
1333         if (retval)
1334                 goto out_free;
1335
1336         retval = check_unsafe_exec(bprm);
1337         if (retval < 0)
1338                 goto out_free;
1339         clear_in_exec = retval;
1340         current->in_execve = 1;
1341
1342         file = open_exec(filename);
1343         retval = PTR_ERR(file);
1344         if (IS_ERR(file))
1345                 goto out_unmark;
1346
1347         sched_exec();
1348
1349         bprm->file = file;
1350         bprm->filename = filename;
1351         bprm->interp = filename;
1352
1353         retval = bprm_mm_init(bprm);
1354         if (retval)
1355                 goto out_file;
1356
1357         bprm->argc = count(argv, MAX_ARG_STRINGS);
1358         if ((retval = bprm->argc) < 0)
1359                 goto out;
1360
1361         bprm->envc = count(envp, MAX_ARG_STRINGS);
1362         if ((retval = bprm->envc) < 0)
1363                 goto out;
1364
1365         retval = prepare_binprm(bprm);
1366         if (retval < 0)
1367                 goto out;
1368
1369         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1370         if (retval < 0)
1371                 goto out;
1372
1373         bprm->exec = bprm->p;
1374         retval = copy_strings(bprm->envc, envp, bprm);
1375         if (retval < 0)
1376                 goto out;
1377
1378         retval = copy_strings(bprm->argc, argv, bprm);
1379         if (retval < 0)
1380                 goto out;
1381
1382         current->flags &= ~PF_KTHREAD;
1383         retval = search_binary_handler(bprm,regs);
1384         if (retval < 0)
1385                 goto out;
1386
1387         /* execve succeeded */
1388         current->fs->in_exec = 0;
1389         current->in_execve = 0;
1390         acct_update_integrals(current);
1391         free_bprm(bprm);
1392         if (displaced)
1393                 put_files_struct(displaced);
1394         return retval;
1395
1396 out:
1397         if (bprm->mm)
1398                 mmput (bprm->mm);
1399
1400 out_file:
1401         if (bprm->file) {
1402                 allow_write_access(bprm->file);
1403                 fput(bprm->file);
1404         }
1405
1406 out_unmark:
1407         if (clear_in_exec)
1408                 current->fs->in_exec = 0;
1409         current->in_execve = 0;
1410
1411 out_free:
1412         free_bprm(bprm);
1413
1414 out_files:
1415         if (displaced)
1416                 reset_files_struct(displaced);
1417 out_ret:
1418         return retval;
1419 }
1420
1421 void set_binfmt(struct linux_binfmt *new)
1422 {
1423         struct mm_struct *mm = current->mm;
1424
1425         if (mm->binfmt)
1426                 module_put(mm->binfmt->module);
1427
1428         mm->binfmt = new;
1429         if (new)
1430                 __module_get(new->module);
1431 }
1432
1433 EXPORT_SYMBOL(set_binfmt);
1434
1435 /* format_corename will inspect the pattern parameter, and output a
1436  * name into corename, which must have space for at least
1437  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1438  */
1439 static int format_corename(char *corename, long signr)
1440 {
1441         const struct cred *cred = current_cred();
1442         const char *pat_ptr = core_pattern;
1443         int ispipe = (*pat_ptr == '|');
1444         char *out_ptr = corename;
1445         char *const out_end = corename + CORENAME_MAX_SIZE;
1446         int rc;
1447         int pid_in_pattern = 0;
1448
1449         /* Repeat as long as we have more pattern to process and more output
1450            space */
1451         while (*pat_ptr) {
1452                 if (*pat_ptr != '%') {
1453                         if (out_ptr == out_end)
1454                                 goto out;
1455                         *out_ptr++ = *pat_ptr++;
1456                 } else {
1457                         switch (*++pat_ptr) {
1458                         case 0:
1459                                 goto out;
1460                         /* Double percent, output one percent */
1461                         case '%':
1462                                 if (out_ptr == out_end)
1463                                         goto out;
1464                                 *out_ptr++ = '%';
1465                                 break;
1466                         /* pid */
1467                         case 'p':
1468                                 pid_in_pattern = 1;
1469                                 rc = snprintf(out_ptr, out_end - out_ptr,
1470                                               "%d", task_tgid_vnr(current));
1471                                 if (rc > out_end - out_ptr)
1472                                         goto out;
1473                                 out_ptr += rc;
1474                                 break;
1475                         /* uid */
1476                         case 'u':
1477                                 rc = snprintf(out_ptr, out_end - out_ptr,
1478                                               "%d", cred->uid);
1479                                 if (rc > out_end - out_ptr)
1480                                         goto out;
1481                                 out_ptr += rc;
1482                                 break;
1483                         /* gid */
1484                         case 'g':
1485                                 rc = snprintf(out_ptr, out_end - out_ptr,
1486                                               "%d", cred->gid);
1487                                 if (rc > out_end - out_ptr)
1488                                         goto out;
1489                                 out_ptr += rc;
1490                                 break;
1491                         /* signal that caused the coredump */
1492                         case 's':
1493                                 rc = snprintf(out_ptr, out_end - out_ptr,
1494                                               "%ld", signr);
1495                                 if (rc > out_end - out_ptr)
1496                                         goto out;
1497                                 out_ptr += rc;
1498                                 break;
1499                         /* UNIX time of coredump */
1500                         case 't': {
1501                                 struct timeval tv;
1502                                 do_gettimeofday(&tv);
1503                                 rc = snprintf(out_ptr, out_end - out_ptr,
1504                                               "%lu", tv.tv_sec);
1505                                 if (rc > out_end - out_ptr)
1506                                         goto out;
1507                                 out_ptr += rc;
1508                                 break;
1509                         }
1510                         /* hostname */
1511                         case 'h':
1512                                 down_read(&uts_sem);
1513                                 rc = snprintf(out_ptr, out_end - out_ptr,
1514                                               "%s", utsname()->nodename);
1515                                 up_read(&uts_sem);
1516                                 if (rc > out_end - out_ptr)
1517                                         goto out;
1518                                 out_ptr += rc;
1519                                 break;
1520                         /* executable */
1521                         case 'e':
1522                                 rc = snprintf(out_ptr, out_end - out_ptr,
1523                                               "%s", current->comm);
1524                                 if (rc > out_end - out_ptr)
1525                                         goto out;
1526                                 out_ptr += rc;
1527                                 break;
1528                         /* core limit size */
1529                         case 'c':
1530                                 rc = snprintf(out_ptr, out_end - out_ptr,
1531                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1532                                 if (rc > out_end - out_ptr)
1533                                         goto out;
1534                                 out_ptr += rc;
1535                                 break;
1536                         default:
1537                                 break;
1538                         }
1539                         ++pat_ptr;
1540                 }
1541         }
1542         /* Backward compatibility with core_uses_pid:
1543          *
1544          * If core_pattern does not include a %p (as is the default)
1545          * and core_uses_pid is set, then .%pid will be appended to
1546          * the filename. Do not do this for piped commands. */
1547         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1548                 rc = snprintf(out_ptr, out_end - out_ptr,
1549                               ".%d", task_tgid_vnr(current));
1550                 if (rc > out_end - out_ptr)
1551                         goto out;
1552                 out_ptr += rc;
1553         }
1554 out:
1555         *out_ptr = 0;
1556         return ispipe;
1557 }
1558
1559 static int zap_process(struct task_struct *start)
1560 {
1561         struct task_struct *t;
1562         int nr = 0;
1563
1564         start->signal->flags = SIGNAL_GROUP_EXIT;
1565         start->signal->group_stop_count = 0;
1566
1567         t = start;
1568         do {
1569                 if (t != current && t->mm) {
1570                         sigaddset(&t->pending.signal, SIGKILL);
1571                         signal_wake_up(t, 1);
1572                         nr++;
1573                 }
1574         } while_each_thread(start, t);
1575
1576         return nr;
1577 }
1578
1579 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1580                                 struct core_state *core_state, int exit_code)
1581 {
1582         struct task_struct *g, *p;
1583         unsigned long flags;
1584         int nr = -EAGAIN;
1585
1586         spin_lock_irq(&tsk->sighand->siglock);
1587         if (!signal_group_exit(tsk->signal)) {
1588                 mm->core_state = core_state;
1589                 tsk->signal->group_exit_code = exit_code;
1590                 nr = zap_process(tsk);
1591         }
1592         spin_unlock_irq(&tsk->sighand->siglock);
1593         if (unlikely(nr < 0))
1594                 return nr;
1595
1596         if (atomic_read(&mm->mm_users) == nr + 1)
1597                 goto done;
1598         /*
1599          * We should find and kill all tasks which use this mm, and we should
1600          * count them correctly into ->nr_threads. We don't take tasklist
1601          * lock, but this is safe wrt:
1602          *
1603          * fork:
1604          *      None of sub-threads can fork after zap_process(leader). All
1605          *      processes which were created before this point should be
1606          *      visible to zap_threads() because copy_process() adds the new
1607          *      process to the tail of init_task.tasks list, and lock/unlock
1608          *      of ->siglock provides a memory barrier.
1609          *
1610          * do_exit:
1611          *      The caller holds mm->mmap_sem. This means that the task which
1612          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1613          *      its ->mm.
1614          *
1615          * de_thread:
1616          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1617          *      we must see either old or new leader, this does not matter.
1618          *      However, it can change p->sighand, so lock_task_sighand(p)
1619          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1620          *      it can't fail.
1621          *
1622          *      Note also that "g" can be the old leader with ->mm == NULL
1623          *      and already unhashed and thus removed from ->thread_group.
1624          *      This is OK, __unhash_process()->list_del_rcu() does not
1625          *      clear the ->next pointer, we will find the new leader via
1626          *      next_thread().
1627          */
1628         rcu_read_lock();
1629         for_each_process(g) {
1630                 if (g == tsk->group_leader)
1631                         continue;
1632                 if (g->flags & PF_KTHREAD)
1633                         continue;
1634                 p = g;
1635                 do {
1636                         if (p->mm) {
1637                                 if (unlikely(p->mm == mm)) {
1638                                         lock_task_sighand(p, &flags);
1639                                         nr += zap_process(p);
1640                                         unlock_task_sighand(p, &flags);
1641                                 }
1642                                 break;
1643                         }
1644                 } while_each_thread(g, p);
1645         }
1646         rcu_read_unlock();
1647 done:
1648         atomic_set(&core_state->nr_threads, nr);
1649         return nr;
1650 }
1651
1652 static int coredump_wait(int exit_code, struct core_state *core_state)
1653 {
1654         struct task_struct *tsk = current;
1655         struct mm_struct *mm = tsk->mm;
1656         struct completion *vfork_done;
1657         int core_waiters;
1658
1659         init_completion(&core_state->startup);
1660         core_state->dumper.task = tsk;
1661         core_state->dumper.next = NULL;
1662         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1663         up_write(&mm->mmap_sem);
1664
1665         if (unlikely(core_waiters < 0))
1666                 goto fail;
1667
1668         /*
1669          * Make sure nobody is waiting for us to release the VM,
1670          * otherwise we can deadlock when we wait on each other
1671          */
1672         vfork_done = tsk->vfork_done;
1673         if (vfork_done) {
1674                 tsk->vfork_done = NULL;
1675                 complete(vfork_done);
1676         }
1677
1678         if (core_waiters)
1679                 wait_for_completion(&core_state->startup);
1680 fail:
1681         return core_waiters;
1682 }
1683
1684 static void coredump_finish(struct mm_struct *mm)
1685 {
1686         struct core_thread *curr, *next;
1687         struct task_struct *task;
1688
1689         next = mm->core_state->dumper.next;
1690         while ((curr = next) != NULL) {
1691                 next = curr->next;
1692                 task = curr->task;
1693                 /*
1694                  * see exit_mm(), curr->task must not see
1695                  * ->task == NULL before we read ->next.
1696                  */
1697                 smp_mb();
1698                 curr->task = NULL;
1699                 wake_up_process(task);
1700         }
1701
1702         mm->core_state = NULL;
1703 }
1704
1705 /*
1706  * set_dumpable converts traditional three-value dumpable to two flags and
1707  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1708  * these bits are not changed atomically.  So get_dumpable can observe the
1709  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1710  * return either old dumpable or new one by paying attention to the order of
1711  * modifying the bits.
1712  *
1713  * dumpable |   mm->flags (binary)
1714  * old  new | initial interim  final
1715  * ---------+-----------------------
1716  *  0    1  |   00      01      01
1717  *  0    2  |   00      10(*)   11
1718  *  1    0  |   01      00      00
1719  *  1    2  |   01      11      11
1720  *  2    0  |   11      10(*)   00
1721  *  2    1  |   11      11      01
1722  *
1723  * (*) get_dumpable regards interim value of 10 as 11.
1724  */
1725 void set_dumpable(struct mm_struct *mm, int value)
1726 {
1727         switch (value) {
1728         case 0:
1729                 clear_bit(MMF_DUMPABLE, &mm->flags);
1730                 smp_wmb();
1731                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1732                 break;
1733         case 1:
1734                 set_bit(MMF_DUMPABLE, &mm->flags);
1735                 smp_wmb();
1736                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1737                 break;
1738         case 2:
1739                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1740                 smp_wmb();
1741                 set_bit(MMF_DUMPABLE, &mm->flags);
1742                 break;
1743         }
1744 }
1745
1746 int get_dumpable(struct mm_struct *mm)
1747 {
1748         int ret;
1749
1750         ret = mm->flags & 0x3;
1751         return (ret >= 2) ? 2 : ret;
1752 }
1753
1754 static void wait_for_dump_helpers(struct file *file)
1755 {
1756         struct pipe_inode_info *pipe;
1757
1758         pipe = file->f_path.dentry->d_inode->i_pipe;
1759
1760         pipe_lock(pipe);
1761         pipe->readers++;
1762         pipe->writers--;
1763
1764         while ((pipe->readers > 1) && (!signal_pending(current))) {
1765                 wake_up_interruptible_sync(&pipe->wait);
1766                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1767                 pipe_wait(pipe);
1768         }
1769
1770         pipe->readers--;
1771         pipe->writers++;
1772         pipe_unlock(pipe);
1773
1774 }
1775
1776
1777 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1778 {
1779         struct core_state core_state;
1780         char corename[CORENAME_MAX_SIZE + 1];
1781         struct mm_struct *mm = current->mm;
1782         struct linux_binfmt * binfmt;
1783         struct inode * inode;
1784         struct file * file;
1785         const struct cred *old_cred;
1786         struct cred *cred;
1787         int retval = 0;
1788         int flag = 0;
1789         int ispipe = 0;
1790         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1791         char **helper_argv = NULL;
1792         int helper_argc = 0;
1793         int dump_count = 0;
1794         static atomic_t core_dump_count = ATOMIC_INIT(0);
1795
1796         audit_core_dumps(signr);
1797
1798         binfmt = mm->binfmt;
1799         if (!binfmt || !binfmt->core_dump)
1800                 goto fail;
1801
1802         cred = prepare_creds();
1803         if (!cred) {
1804                 retval = -ENOMEM;
1805                 goto fail;
1806         }
1807
1808         down_write(&mm->mmap_sem);
1809         /*
1810          * If another thread got here first, or we are not dumpable, bail out.
1811          */
1812         if (mm->core_state || !get_dumpable(mm)) {
1813                 up_write(&mm->mmap_sem);
1814                 put_cred(cred);
1815                 goto fail;
1816         }
1817
1818         /*
1819          *      We cannot trust fsuid as being the "true" uid of the
1820          *      process nor do we know its entire history. We only know it
1821          *      was tainted so we dump it as root in mode 2.
1822          */
1823         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1824                 flag = O_EXCL;          /* Stop rewrite attacks */
1825                 cred->fsuid = 0;        /* Dump root private */
1826         }
1827
1828         retval = coredump_wait(exit_code, &core_state);
1829         if (retval < 0) {
1830                 put_cred(cred);
1831                 goto fail;
1832         }
1833
1834         old_cred = override_creds(cred);
1835
1836         /*
1837          * Clear any false indication of pending signals that might
1838          * be seen by the filesystem code called to write the core file.
1839          */
1840         clear_thread_flag(TIF_SIGPENDING);
1841
1842         /*
1843          * lock_kernel() because format_corename() is controlled by sysctl, which
1844          * uses lock_kernel()
1845          */
1846         lock_kernel();
1847         ispipe = format_corename(corename, signr);
1848         unlock_kernel();
1849
1850         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1851                 goto fail_unlock;
1852
1853         if (ispipe) {
1854                 if (core_limit == 0) {
1855                         /*
1856                          * Normally core limits are irrelevant to pipes, since
1857                          * we're not writing to the file system, but we use
1858                          * core_limit of 0 here as a speacial value. Any
1859                          * non-zero limit gets set to RLIM_INFINITY below, but
1860                          * a limit of 0 skips the dump.  This is a consistent
1861                          * way to catch recursive crashes.  We can still crash
1862                          * if the core_pattern binary sets RLIM_CORE =  !0
1863                          * but it runs as root, and can do lots of stupid things
1864                          * Note that we use task_tgid_vnr here to grab the pid
1865                          * of the process group leader.  That way we get the
1866                          * right pid if a thread in a multi-threaded
1867                          * core_pattern process dies.
1868                          */
1869                         printk(KERN_WARNING
1870                                 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1871                                 task_tgid_vnr(current), current->comm);
1872                         printk(KERN_WARNING "Aborting core\n");
1873                         goto fail_unlock;
1874                 }
1875
1876                 dump_count = atomic_inc_return(&core_dump_count);
1877                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1878                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1879                                task_tgid_vnr(current), current->comm);
1880                         printk(KERN_WARNING "Skipping core dump\n");
1881                         goto fail_dropcount;
1882                 }
1883
1884                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1885                 if (!helper_argv) {
1886                         printk(KERN_WARNING "%s failed to allocate memory\n",
1887                                __func__);
1888                         goto fail_dropcount;
1889                 }
1890
1891                 core_limit = RLIM_INFINITY;
1892
1893                 /* SIGPIPE can happen, but it's just never processed */
1894                 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1895                                 &file)) {
1896                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1897                                corename);
1898                         goto fail_dropcount;
1899                 }
1900         } else
1901                 file = filp_open(corename,
1902                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1903                                  0600);
1904         if (IS_ERR(file))
1905                 goto fail_dropcount;
1906         inode = file->f_path.dentry->d_inode;
1907         if (inode->i_nlink > 1)
1908                 goto close_fail;        /* multiple links - don't dump */
1909         if (!ispipe && d_unhashed(file->f_path.dentry))
1910                 goto close_fail;
1911
1912         /* AK: actually i see no reason to not allow this for named pipes etc.,
1913            but keep the previous behaviour for now. */
1914         if (!ispipe && !S_ISREG(inode->i_mode))
1915                 goto close_fail;
1916         /*
1917          * Dont allow local users get cute and trick others to coredump
1918          * into their pre-created files:
1919          * Note, this is not relevant for pipes
1920          */
1921         if (!ispipe && (inode->i_uid != current_fsuid()))
1922                 goto close_fail;
1923         if (!file->f_op)
1924                 goto close_fail;
1925         if (!file->f_op->write)
1926                 goto close_fail;
1927         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1928                 goto close_fail;
1929
1930         retval = binfmt->core_dump(signr, regs, file, core_limit);
1931
1932         if (retval)
1933                 current->signal->group_exit_code |= 0x80;
1934 close_fail:
1935         if (ispipe && core_pipe_limit)
1936                 wait_for_dump_helpers(file);
1937         filp_close(file, NULL);
1938 fail_dropcount:
1939         if (dump_count)
1940                 atomic_dec(&core_dump_count);
1941 fail_unlock:
1942         if (helper_argv)
1943                 argv_free(helper_argv);
1944
1945         revert_creds(old_cred);
1946         put_cred(cred);
1947         coredump_finish(mm);
1948 fail:
1949         return;
1950 }