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
 380                         if (fatal_signal_pending(current))
 381                                 return -ERESTARTNOHAND;
 382                         cond_resched();
 383                 }
 384         }
 385         return i;
 386 }
 387
 388 /*
 389  * 'copy_strings()' copies argument/environment strings from the old
 390  * processes's memory to the new process's stack.  The call to get_user_pages()
 391  * ensures the destination page is created and not swapped out.
 392  */
 393 static int copy_strings(int argc, char __user * __user * argv,
 394                         struct linux_binprm *bprm)
 395 {
 396         struct page *kmapped_page = NULL;
 397         char *kaddr = NULL;
 398         unsigned long kpos = 0;
 399         int ret;
 400
 401         while (argc-- > 0) {
 402                 char __user *str;
 403                 int len;
 404                 unsigned long pos;
 405
 406                 if (get_user(str, argv+argc) ||
 407                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 408                         ret = -EFAULT;
 409                         goto out;
 410                 }
 411
 412                 if (!valid_arg_len(bprm, len)) {
 413                         ret = -E2BIG;
 414                         goto out;
 415                 }
 416
 417                 /* We're going to work our way backwords. */
 418                 pos = bprm->p;
 419                 str += len;
 420                 bprm->p -= len;
 421
 422                 while (len > 0) {
 423                         int offset, bytes_to_copy;
 424
 425                         if (fatal_signal_pending(current)) {
 426                                 ret = -ERESTARTNOHAND;
 427                                 goto out;
 428                         }
 429                         cond_resched();
 430
 431                         offset = pos % PAGE_SIZE;
 432                         if (offset == 0)
 433                                 offset = PAGE_SIZE;
 434
 435                         bytes_to_copy = offset;
 436                         if (bytes_to_copy > len)
 437                                 bytes_to_copy = len;
 438
 439                         offset -= bytes_to_copy;
 440                         pos -= bytes_to_copy;
 441                         str -= bytes_to_copy;
 442                         len -= bytes_to_copy;
 443
 444                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 445                                 struct page *page;
 446
 447                                 page = get_arg_page(bprm, pos, 1);
 448                                 if (!page) {
 449                                         ret = -E2BIG;
 450                                         goto out;
 451                                 }
 452
 453                                 if (kmapped_page) {
 454                                         flush_kernel_dcache_page(kmapped_page);
 455                                         kunmap(kmapped_page);
 456                                         put_arg_page(kmapped_page);
 457                                 }
 458                                 kmapped_page = page;
 459                                 kaddr = kmap(kmapped_page);
 460                                 kpos = pos & PAGE_MASK;
 461                                 flush_arg_page(bprm, kpos, kmapped_page);
 462                         }
 463                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 464                                 ret = -EFAULT;
 465                                 goto out;
 466                         }
 467                 }
 468         }
 469         ret = 0;
 470 out:
 471         if (kmapped_page) {
 472                 flush_kernel_dcache_page(kmapped_page);
 473                 kunmap(kmapped_page);
 474                 put_arg_page(kmapped_page);
 475         }
 476         return ret;
 477 }
 478
 479 /*
 480  * Like copy_strings, but get argv and its values from kernel memory.
 481  */
 482 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 483 {
 484         int r;
 485         mm_segment_t oldfs = get_fs();
 486         set_fs(KERNEL_DS);
 487         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 488         set_fs(oldfs);
 489         return r;
 490 }
 491 EXPORT_SYMBOL(copy_strings_kernel);
 492
 493 #ifdef CONFIG_MMU
 494
 495 /*
 496  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 497  * the binfmt code determines where the new stack should reside, we shift it to
 498  * its final location.  The process proceeds as follows:
 499  *
 500  * 1) Use shift to calculate the new vma endpoints.
 501  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 502  *    arguments passed to subsequent functions are consistent.
 503  * 3) Move vma's page tables to the new range.
 504  * 4) Free up any cleared pgd range.
 505  * 5) Shrink the vma to cover only the new range.
 506  */
 507 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 508 {
 509         struct mm_struct *mm = vma->vm_mm;
 510         unsigned long old_start = vma->vm_start;
 511         unsigned long old_end = vma->vm_end;
 512         unsigned long length = old_end - old_start;
 513         unsigned long new_start = old_start - shift;
 514         unsigned long new_end = old_end - shift;
 515         struct mmu_gather *tlb;
 516
 517         BUG_ON(new_start > new_end);
 518
 519         /*
 520          * ensure there are no vmas between where we want to go
 521          * and where we are
 522          */
 523         if (vma != find_vma(mm, new_start))
 524                 return -EFAULT;
 525
 526         /*
 527          * cover the whole range: [new_start, old_end)
 528          */
 529         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 530
 531         /*
 532          * move the page tables downwards, on failure we rely on
 533          * process cleanup to remove whatever mess we made.
 534          */
 535         if (length != move_page_tables(vma, old_start,
 536                                        vma, new_start, length))
 537                 return -ENOMEM;
 538
 539         lru_add_drain();
 540         tlb = tlb_gather_mmu(mm, 0);
 541         if (new_end > old_start) {
 542                 /*
 543                  * when the old and new regions overlap clear from new_end.
 544                  */
 545                 free_pgd_range(tlb, new_end, old_end, new_end,
 546                         vma->vm_next ? vma->vm_next->vm_start : 0);
 547         } else {
 548                 /*
 549                  * otherwise, clean from old_start; this is done to not touch
 550                  * the address space in [new_end, old_start) some architectures
 551                  * have constraints on va-space that make this illegal (IA64) -
 552                  * for the others its just a little faster.
 553                  */
 554                 free_pgd_range(tlb, old_start, old_end, new_end,
 555                         vma->vm_next ? vma->vm_next->vm_start : 0);
 556         }
 557         tlb_finish_mmu(tlb, new_end, old_end);
 558
 559         /*
 560          * shrink the vma to just the new range.
 561          */
 562         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 563
 564         return 0;
 565 }
 566
 567 #define EXTRA_STACK_VM_PAGES    20      /* random */
 568
 569 /*
 570  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 571  * the stack is optionally relocated, and some extra space is added.
 572  */
 573 int setup_arg_pages(struct linux_binprm *bprm,
 574                     unsigned long stack_top,
 575                     int executable_stack)
 576 {
 577         unsigned long ret;
 578         unsigned long stack_shift;
 579         struct mm_struct *mm = current->mm;
 580         struct vm_area_struct *vma = bprm->vma;
 581         struct vm_area_struct *prev = NULL;
 582         unsigned long vm_flags;
 583         unsigned long stack_base;
 584         unsigned long stack_size;
 585         unsigned long stack_expand;
 586         unsigned long rlim_stack;
 587
 588 #ifdef CONFIG_STACK_GROWSUP
 589         /* Limit stack size to 1GB */
 590         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 591         if (stack_base > (1 << 30))
 592                 stack_base = 1 << 30;
 593
 594         /* Make sure we didn't let the argument array grow too large. */
 595         if (vma->vm_end - vma->vm_start > stack_base)
 596                 return -ENOMEM;
 597
 598         stack_base = PAGE_ALIGN(stack_top - stack_base);
 599
 600         stack_shift = vma->vm_start - stack_base;
 601         mm->arg_start = bprm->p - stack_shift;
 602         bprm->p = vma->vm_end - stack_shift;
 603 #else
 604         stack_top = arch_align_stack(stack_top);
 605         stack_top = PAGE_ALIGN(stack_top);
 606
 607         if (unlikely(stack_top < mmap_min_addr) ||
 608             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 609                 return -ENOMEM;
 610
 611         stack_shift = vma->vm_end - stack_top;
 612
 613         bprm->p -= stack_shift;
 614         mm->arg_start = bprm->p;
 615 #endif
 616
 617         if (bprm->loader)
 618                 bprm->loader -= stack_shift;
 619         bprm->exec -= stack_shift;
 620
 621         down_write(&mm->mmap_sem);
 622         vm_flags = VM_STACK_FLAGS;
 623
 624         /*
 625          * Adjust stack execute permissions; explicitly enable for
 626          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 627          * (arch default) otherwise.
 628          */
 629         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 630                 vm_flags |= VM_EXEC;
 631         else if (executable_stack == EXSTACK_DISABLE_X)
 632                 vm_flags &= ~VM_EXEC;
 633         vm_flags |= mm->def_flags;
 634
 635         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 636                         vm_flags);
 637         if (ret)
 638                 goto out_unlock;
 639         BUG_ON(prev != vma);
 640
 641         /* Move stack pages down in memory. */
 642         if (stack_shift) {
 643                 ret = shift_arg_pages(vma, stack_shift);
 644                 if (ret)
 645                         goto out_unlock;
 646         }
 647
 648         stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 649         stack_size = vma->vm_end - vma->vm_start;
 650         /*
 651          * Align this down to a page boundary as expand_stack
 652          * will align it up.
 653          */
 654         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 655 #ifdef CONFIG_STACK_GROWSUP
 656         if (stack_size + stack_expand > rlim_stack)
 657                 stack_base = vma->vm_start + rlim_stack;
 658         else
 659                 stack_base = vma->vm_end + stack_expand;
 660 #else
 661         if (stack_size + stack_expand > rlim_stack)
 662                 stack_base = vma->vm_end - rlim_stack;
 663         else
 664                 stack_base = vma->vm_start - stack_expand;
 665 #endif
 666         ret = expand_stack(vma, stack_base);
 667         if (ret)
 668                 ret = -EFAULT;
 669
 670 out_unlock:
 671         up_write(&mm->mmap_sem);
 672         return ret;
 673 }
 674 EXPORT_SYMBOL(setup_arg_pages);
 675
 676 #endif /* CONFIG_MMU */
 677
 678 struct file *open_exec(const char *name)
 679 {
 680         struct file *file;
 681         int err;
 682
 683         file = do_filp_open(AT_FDCWD, name,
 684                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 685                                 MAY_EXEC | MAY_OPEN);
 686         if (IS_ERR(file))
 687                 goto out;
 688
 689         err = -EACCES;
 690         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 691                 goto exit;
 692
 693         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 694                 goto exit;
 695
 696         fsnotify_open(file->f_path.dentry);
 697
 698         err = deny_write_access(file);
 699         if (err)
 700                 goto exit;
 701
 702 out:
 703         return file;
 704
 705 exit:
 706         fput(file);
 707         return ERR_PTR(err);
 708 }
 709 EXPORT_SYMBOL(open_exec);
 710
 711 int kernel_read(struct file *file, loff_t offset,
 712                 char *addr, unsigned long count)
 713 {
 714         mm_segment_t old_fs;
 715         loff_t pos = offset;
 716         int result;
 717
 718         old_fs = get_fs();
 719         set_fs(get_ds());
 720         /* The cast to a user pointer is valid due to the set_fs() */
 721         result = vfs_read(file, (void __user *)addr, count, &pos);
 722         set_fs(old_fs);
 723         return result;
 724 }
 725
 726 EXPORT_SYMBOL(kernel_read);
 727
 728 static int exec_mmap(struct mm_struct *mm)
 729 {
 730         struct task_struct *tsk;
 731         struct mm_struct * old_mm, *active_mm;
 732
 733         /* Notify parent that we're no longer interested in the old VM */
 734         tsk = current;
 735         old_mm = current->mm;
 736         mm_release(tsk, old_mm);
 737
 738         if (old_mm) {
 739                 /*
 740                  * Make sure that if there is a core dump in progress
 741                  * for the old mm, we get out and die instead of going
 742                  * through with the exec.  We must hold mmap_sem around
 743                  * checking core_state and changing tsk->mm.
 744                  */
 745                 down_read(&old_mm->mmap_sem);
 746                 if (unlikely(old_mm->core_state)) {
 747                         up_read(&old_mm->mmap_sem);
 748                         return -EINTR;
 749                 }
 750         }
 751         task_lock(tsk);
 752         active_mm = tsk->active_mm;
 753         tsk->mm = mm;
 754         tsk->active_mm = mm;
 755         activate_mm(active_mm, mm);
 756         task_unlock(tsk);
 757         arch_pick_mmap_layout(mm);
 758         if (old_mm) {
 759                 up_read(&old_mm->mmap_sem);
 760                 BUG_ON(active_mm != old_mm);
 761                 mm_update_next_owner(old_mm);
 762                 mmput(old_mm);
 763                 return 0;
 764         }
 765         mmdrop(active_mm);
 766         return 0;
 767 }
 768
 769 /*
 770  * This function makes sure the current process has its own signal table,
 771  * so that flush_signal_handlers can later reset the handlers without
 772  * disturbing other processes.  (Other processes might share the signal
 773  * table via the CLONE_SIGHAND option to clone().)
 774  */
 775 static int de_thread(struct task_struct *tsk)
 776 {
 777         struct signal_struct *sig = tsk->signal;
 778         struct sighand_struct *oldsighand = tsk->sighand;
 779         spinlock_t *lock = &oldsighand->siglock;
 780         int count;
 781
 782         if (thread_group_empty(tsk))
 783                 goto no_thread_group;
 784
 785         /*
 786          * Kill all other threads in the thread group.
 787          */
 788         spin_lock_irq(lock);
 789         if (signal_group_exit(sig)) {
 790                 /*
 791                  * Another group action in progress, just
 792                  * return so that the signal is processed.
 793                  */
 794                 spin_unlock_irq(lock);
 795                 return -EAGAIN;
 796         }
 797         sig->group_exit_task = tsk;
 798         zap_other_threads(tsk);
 799
 800         /* Account for the thread group leader hanging around: */
 801         count = thread_group_leader(tsk) ? 1 : 2;
 802         sig->notify_count = count;
 803         while (atomic_read(&sig->count) > count) {
 804                 __set_current_state(TASK_UNINTERRUPTIBLE);
 805                 spin_unlock_irq(lock);
 806                 schedule();
 807                 spin_lock_irq(lock);
 808         }
 809         spin_unlock_irq(lock);
 810
 811         /*
 812          * At this point all other threads have exited, all we have to
 813          * do is to wait for the thread group leader to become inactive,
 814          * and to assume its PID:
 815          */
 816         if (!thread_group_leader(tsk)) {
 817                 struct task_struct *leader = tsk->group_leader;
 818
 819                 sig->notify_count = -1; /* for exit_notify() */
 820                 for (;;) {
 821                         write_lock_irq(&tasklist_lock);
 822                         if (likely(leader->exit_state))
 823                                 break;
 824                         __set_current_state(TASK_UNINTERRUPTIBLE);
 825                         write_unlock_irq(&tasklist_lock);
 826                         schedule();
 827                 }
 828
 829                 /*
 830                  * The only record we have of the real-time age of a
 831                  * process, regardless of execs it's done, is start_time.
 832                  * All the past CPU time is accumulated in signal_struct
 833                  * from sister threads now dead.  But in this non-leader
 834                  * exec, nothing survives from the original leader thread,
 835                  * whose birth marks the true age of this process now.
 836                  * When we take on its identity by switching to its PID, we
 837                  * also take its birthdate (always earlier than our own).
 838                  */
 839                 tsk->start_time = leader->start_time;
 840
 841                 BUG_ON(!same_thread_group(leader, tsk));
 842                 BUG_ON(has_group_leader_pid(tsk));
 843                 /*
 844                  * An exec() starts a new thread group with the
 845                  * TGID of the previous thread group. Rehash the
 846                  * two threads with a switched PID, and release
 847                  * the former thread group leader:
 848                  */
 849
 850                 /* Become a process group leader with the old leader's pid.
 851                  * The old leader becomes a thread of the this thread group.
 852                  * Note: The old leader also uses this pid until release_task
 853                  *       is called.  Odd but simple and correct.
 854                  */
 855                 detach_pid(tsk, PIDTYPE_PID);
 856                 tsk->pid = leader->pid;
 857                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 858                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 859                 transfer_pid(leader, tsk, PIDTYPE_SID);
 860                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 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         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 958         task_unlock(tsk);
 959         perf_event_comm(tsk);
 960 }
 961
 962 int flush_old_exec(struct linux_binprm * bprm)
 963 {
 964         int retval;
 965
 966         /*
 967          * Make sure we have a private signal table and that
 968          * we are unassociated from the previous thread group.
 969          */
 970         retval = de_thread(current);
 971         if (retval)
 972                 goto out;
 973
 974         set_mm_exe_file(bprm->mm, bprm->file);
 975
 976         /*
 977          * Release all of the old mmap stuff
 978          */
 979         retval = exec_mmap(bprm->mm);
 980         if (retval)
 981                 goto out;
 982
 983         bprm->mm = NULL;                /* We're using it now */
 984
 985         current->flags &= ~PF_RANDOMIZE;
 986         flush_thread();
 987         current->personality &= ~bprm->per_clear;
 988
 989         return 0;
 990
 991 out:
 992         return retval;
 993 }
 994 EXPORT_SYMBOL(flush_old_exec);
 995
 996 void setup_new_exec(struct linux_binprm * bprm)
 997 {
 998         int i, ch;
 999         char * name;
1000         char tcomm[sizeof(current->comm)];
1001
1002         arch_pick_mmap_layout(current->mm);
1003
1004         /* This is the point of no return */
1005         current->sas_ss_sp = current->sas_ss_size = 0;
1006
1007         if (current_euid() == current_uid() && current_egid() == current_gid())
1008                 set_dumpable(current->mm, 1);
1009         else
1010                 set_dumpable(current->mm, suid_dumpable);
1011
1012         name = bprm->filename;
1013
1014         /* Copies the binary name from after last slash */
1015         for (i=0; (ch = *(name++)) != '\0';) {
1016                 if (ch == '/')
1017                         i = 0; /* overwrite what we wrote */
1018                 else
1019                         if (i < (sizeof(tcomm) - 1))
1020                                 tcomm[i++] = ch;
1021         }
1022         tcomm[i] = '\0';
1023         set_task_comm(current, tcomm);
1024
1025         /* Set the new mm task size. We have to do that late because it may
1026          * depend on TIF_32BIT which is only updated in flush_thread() on
1027          * some architectures like powerpc
1028          */
1029         current->mm->task_size = TASK_SIZE;
1030
1031         /* install the new credentials */
1032         if (bprm->cred->uid != current_euid() ||
1033             bprm->cred->gid != current_egid()) {
1034                 current->pdeath_signal = 0;
1035         } else if (file_permission(bprm->file, MAY_READ) ||
1036                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1037                 set_dumpable(current->mm, suid_dumpable);
1038         }
1039
1040         /*
1041          * Flush performance counters when crossing a
1042          * security domain:
1043          */
1044         if (!get_dumpable(current->mm))
1045                 perf_event_exit_task(current);
1046
1047         /* An exec changes our domain. We are no longer part of the thread
1048            group */
1049
1050         current->self_exec_id++;
1051
1052         flush_signal_handlers(current, 0);
1053         flush_old_files(current->files);
1054 }
1055 EXPORT_SYMBOL(setup_new_exec);
1056
1057 /*
1058  * Prepare credentials and lock ->cred_guard_mutex.
1059  * install_exec_creds() commits the new creds and drops the lock.
1060  * Or, if exec fails before, free_bprm() should release ->cred and
1061  * and unlock.
1062  */
1063 int prepare_bprm_creds(struct linux_binprm *bprm)
1064 {
1065         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1066                 return -ERESTARTNOINTR;
1067
1068         bprm->cred = prepare_exec_creds();
1069         if (likely(bprm->cred))
1070                 return 0;
1071
1072         mutex_unlock(&current->cred_guard_mutex);
1073         return -ENOMEM;
1074 }
1075
1076 void free_bprm(struct linux_binprm *bprm)
1077 {
1078         free_arg_pages(bprm);
1079         if (bprm->cred) {
1080                 mutex_unlock(&current->cred_guard_mutex);
1081                 abort_creds(bprm->cred);
1082         }
1083         kfree(bprm);
1084 }
1085
1086 /*
1087  * install the new credentials for this executable
1088  */
1089 void install_exec_creds(struct linux_binprm *bprm)
1090 {
1091         security_bprm_committing_creds(bprm);
1092
1093         commit_creds(bprm->cred);
1094         bprm->cred = NULL;
1095         /*
1096          * cred_guard_mutex must be held at least to this point to prevent
1097          * ptrace_attach() from altering our determination of the task's
1098          * credentials; any time after this it may be unlocked.
1099          */
1100         security_bprm_committed_creds(bprm);
1101         mutex_unlock(&current->cred_guard_mutex);
1102 }
1103 EXPORT_SYMBOL(install_exec_creds);
1104
1105 /*
1106  * determine how safe it is to execute the proposed program
1107  * - the caller must hold current->cred_guard_mutex to protect against
1108  *   PTRACE_ATTACH
1109  */
1110 int check_unsafe_exec(struct linux_binprm *bprm)
1111 {
1112         struct task_struct *p = current, *t;
1113         unsigned n_fs;
1114         int res = 0;
1115
1116         bprm->unsafe = tracehook_unsafe_exec(p);
1117
1118         n_fs = 1;
1119         write_lock(&p->fs->lock);
1120         rcu_read_lock();
1121         for (t = next_thread(p); t != p; t = next_thread(t)) {
1122                 if (t->fs == p->fs)
1123                         n_fs++;
1124         }
1125         rcu_read_unlock();
1126
1127         if (p->fs->users > n_fs) {
1128                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1129         } else {
1130                 res = -EAGAIN;
1131                 if (!p->fs->in_exec) {
1132                         p->fs->in_exec = 1;
1133                         res = 1;
1134                 }
1135         }
1136         write_unlock(&p->fs->lock);
1137
1138         return res;
1139 }
1140
1141 /*
1142  * Fill the binprm structure from the inode.
1143  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1144  *
1145  * This may be called multiple times for binary chains (scripts for example).
1146  */
1147 int prepare_binprm(struct linux_binprm *bprm)
1148 {
1149         umode_t mode;
1150         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1151         int retval;
1152
1153         mode = inode->i_mode;
1154         if (bprm->file->f_op == NULL)
1155                 return -EACCES;
1156
1157         /* clear any previous set[ug]id data from a previous binary */
1158         bprm->cred->euid = current_euid();
1159         bprm->cred->egid = current_egid();
1160
1161         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1162                 /* Set-uid? */
1163                 if (mode & S_ISUID) {
1164                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1165                         bprm->cred->euid = inode->i_uid;
1166                 }
1167
1168                 /* Set-gid? */
1169                 /*
1170                  * If setgid is set but no group execute bit then this
1171                  * is a candidate for mandatory locking, not a setgid
1172                  * executable.
1173                  */
1174                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1175                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1176                         bprm->cred->egid = inode->i_gid;
1177                 }
1178         }
1179
1180         /* fill in binprm security blob */
1181         retval = security_bprm_set_creds(bprm);
1182         if (retval)
1183                 return retval;
1184         bprm->cred_prepared = 1;
1185
1186         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1187         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1188 }
1189
1190 EXPORT_SYMBOL(prepare_binprm);
1191
1192 /*
1193  * Arguments are '\0' separated strings found at the location bprm->p
1194  * points to; chop off the first by relocating brpm->p to right after
1195  * the first '\0' encountered.
1196  */
1197 int remove_arg_zero(struct linux_binprm *bprm)
1198 {
1199         int ret = 0;
1200         unsigned long offset;
1201         char *kaddr;
1202         struct page *page;
1203
1204         if (!bprm->argc)
1205                 return 0;
1206
1207         do {
1208                 offset = bprm->p & ~PAGE_MASK;
1209                 page = get_arg_page(bprm, bprm->p, 0);
1210                 if (!page) {
1211                         ret = -EFAULT;
1212                         goto out;
1213                 }
1214                 kaddr = kmap_atomic(page, KM_USER0);
1215
1216                 for (; offset < PAGE_SIZE && kaddr[offset];
1217                                 offset++, bprm->p++)
1218                         ;
1219
1220                 kunmap_atomic(kaddr, KM_USER0);
1221                 put_arg_page(page);
1222
1223                 if (offset == PAGE_SIZE)
1224                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1225         } while (offset == PAGE_SIZE);
1226
1227         bprm->p++;
1228         bprm->argc--;
1229         ret = 0;
1230
1231 out:
1232         return ret;
1233 }
1234 EXPORT_SYMBOL(remove_arg_zero);
1235
1236 /*
1237  * cycle the list of binary formats handler, until one recognizes the image
1238  */
1239 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1240 {
1241         unsigned int depth = bprm->recursion_depth;
1242         int try,retval;
1243         struct linux_binfmt *fmt;
1244
1245         retval = security_bprm_check(bprm);
1246         if (retval)
1247                 return retval;
1248         retval = ima_bprm_check(bprm);
1249         if (retval)
1250                 return retval;
1251
1252         /* kernel module loader fixup */
1253         /* so we don't try to load run modprobe in kernel space. */
1254         set_fs(USER_DS);
1255
1256         retval = audit_bprm(bprm);
1257         if (retval)
1258                 return retval;
1259
1260         retval = -ENOENT;
1261         for (try=0; try<2; try++) {
1262                 read_lock(&binfmt_lock);
1263                 list_for_each_entry(fmt, &formats, lh) {
1264                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1265                         if (!fn)
1266                                 continue;
1267                         if (!try_module_get(fmt->module))
1268                                 continue;
1269                         read_unlock(&binfmt_lock);
1270                         retval = fn(bprm, regs);
1271                         /*
1272                          * Restore the depth counter to its starting value
1273                          * in this call, so we don't have to rely on every
1274                          * load_binary function to restore it on return.
1275                          */
1276                         bprm->recursion_depth = depth;
1277                         if (retval >= 0) {
1278                                 if (depth == 0)
1279                                         tracehook_report_exec(fmt, bprm, regs);
1280                                 put_binfmt(fmt);
1281                                 allow_write_access(bprm->file);
1282                                 if (bprm->file)
1283                                         fput(bprm->file);
1284                                 bprm->file = NULL;
1285                                 current->did_exec = 1;
1286                                 proc_exec_connector(current);
1287                                 return retval;
1288                         }
1289                         read_lock(&binfmt_lock);
1290                         put_binfmt(fmt);
1291                         if (retval != -ENOEXEC || bprm->mm == NULL)
1292                                 break;
1293                         if (!bprm->file) {
1294                                 read_unlock(&binfmt_lock);
1295                                 return retval;
1296                         }
1297                 }
1298                 read_unlock(&binfmt_lock);
1299                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1300                         break;
1301 #ifdef CONFIG_MODULES
1302                 } else {
1303 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1304                         if (printable(bprm->buf[0]) &&
1305                             printable(bprm->buf[1]) &&
1306                             printable(bprm->buf[2]) &&
1307                             printable(bprm->buf[3]))
1308                                 break; /* -ENOEXEC */
1309                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1310 #endif
1311                 }
1312         }
1313         return retval;
1314 }
1315
1316 EXPORT_SYMBOL(search_binary_handler);
1317
1318 /*
1319  * sys_execve() executes a new program.
1320  */
1321 int do_execve(char * filename,
1322         char __user *__user *argv,
1323         char __user *__user *envp,
1324         struct pt_regs * regs)
1325 {
1326         struct linux_binprm *bprm;
1327         struct file *file;
1328         struct files_struct *displaced;
1329         bool clear_in_exec;
1330         int retval;
1331
1332         retval = unshare_files(&displaced);
1333         if (retval)
1334                 goto out_ret;
1335
1336         retval = -ENOMEM;
1337         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1338         if (!bprm)
1339                 goto out_files;
1340
1341         retval = prepare_bprm_creds(bprm);
1342         if (retval)
1343                 goto out_free;
1344
1345         retval = check_unsafe_exec(bprm);
1346         if (retval < 0)
1347                 goto out_free;
1348         clear_in_exec = retval;
1349         current->in_execve = 1;
1350
1351         file = open_exec(filename);
1352         retval = PTR_ERR(file);
1353         if (IS_ERR(file))
1354                 goto out_unmark;
1355
1356         sched_exec();
1357
1358         bprm->file = file;
1359         bprm->filename = filename;
1360         bprm->interp = filename;
1361
1362         retval = bprm_mm_init(bprm);
1363         if (retval)
1364                 goto out_file;
1365
1366         bprm->argc = count(argv, MAX_ARG_STRINGS);
1367         if ((retval = bprm->argc) < 0)
1368                 goto out;
1369
1370         bprm->envc = count(envp, MAX_ARG_STRINGS);
1371         if ((retval = bprm->envc) < 0)
1372                 goto out;
1373
1374         retval = prepare_binprm(bprm);
1375         if (retval < 0)
1376                 goto out;
1377
1378         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1379         if (retval < 0)
1380                 goto out;
1381
1382         bprm->exec = bprm->p;
1383         retval = copy_strings(bprm->envc, envp, bprm);
1384         if (retval < 0)
1385                 goto out;
1386
1387         retval = copy_strings(bprm->argc, argv, bprm);
1388         if (retval < 0)
1389                 goto out;
1390
1391         current->flags &= ~PF_KTHREAD;
1392         retval = search_binary_handler(bprm,regs);
1393         if (retval < 0)
1394                 goto out;
1395
1396         /* execve succeeded */
1397         current->fs->in_exec = 0;
1398         current->in_execve = 0;
1399         acct_update_integrals(current);
1400         free_bprm(bprm);
1401         if (displaced)
1402                 put_files_struct(displaced);
1403         return retval;
1404
1405 out:
1406         if (bprm->mm)
1407                 mmput (bprm->mm);
1408
1409 out_file:
1410         if (bprm->file) {
1411                 allow_write_access(bprm->file);
1412                 fput(bprm->file);
1413         }
1414
1415 out_unmark:
1416         if (clear_in_exec)
1417                 current->fs->in_exec = 0;
1418         current->in_execve = 0;
1419
1420 out_free:
1421         free_bprm(bprm);
1422
1423 out_files:
1424         if (displaced)
1425                 reset_files_struct(displaced);
1426 out_ret:
1427         return retval;
1428 }
1429
1430 void set_binfmt(struct linux_binfmt *new)
1431 {
1432         struct mm_struct *mm = current->mm;
1433
1434         if (mm->binfmt)
1435                 module_put(mm->binfmt->module);
1436
1437         mm->binfmt = new;
1438         if (new)
1439                 __module_get(new->module);
1440 }
1441
1442 EXPORT_SYMBOL(set_binfmt);
1443
1444 /* format_corename will inspect the pattern parameter, and output a
1445  * name into corename, which must have space for at least
1446  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1447  */
1448 static int format_corename(char *corename, long signr)
1449 {
1450         const struct cred *cred = current_cred();
1451         const char *pat_ptr = core_pattern;
1452         int ispipe = (*pat_ptr == '|');
1453         char *out_ptr = corename;
1454         char *const out_end = corename + CORENAME_MAX_SIZE;
1455         int rc;
1456         int pid_in_pattern = 0;
1457
1458         /* Repeat as long as we have more pattern to process and more output
1459            space */
1460         while (*pat_ptr) {
1461                 if (*pat_ptr != '%') {
1462                         if (out_ptr == out_end)
1463                                 goto out;
1464                         *out_ptr++ = *pat_ptr++;
1465                 } else {
1466                         switch (*++pat_ptr) {
1467                         case 0:
1468                                 goto out;
1469                         /* Double percent, output one percent */
1470                         case '%':
1471                                 if (out_ptr == out_end)
1472                                         goto out;
1473                                 *out_ptr++ = '%';
1474                                 break;
1475                         /* pid */
1476                         case 'p':
1477                                 pid_in_pattern = 1;
1478                                 rc = snprintf(out_ptr, out_end - out_ptr,
1479                                               "%d", task_tgid_vnr(current));
1480                                 if (rc > out_end - out_ptr)
1481                                         goto out;
1482                                 out_ptr += rc;
1483                                 break;
1484                         /* uid */
1485                         case 'u':
1486                                 rc = snprintf(out_ptr, out_end - out_ptr,
1487                                               "%d", cred->uid);
1488                                 if (rc > out_end - out_ptr)
1489                                         goto out;
1490                                 out_ptr += rc;
1491                                 break;
1492                         /* gid */
1493                         case 'g':
1494                                 rc = snprintf(out_ptr, out_end - out_ptr,
1495                                               "%d", cred->gid);
1496                                 if (rc > out_end - out_ptr)
1497                                         goto out;
1498                                 out_ptr += rc;
1499                                 break;
1500                         /* signal that caused the coredump */
1501                         case 's':
1502                                 rc = snprintf(out_ptr, out_end - out_ptr,
1503                                               "%ld", signr);
1504                                 if (rc > out_end - out_ptr)
1505                                         goto out;
1506                                 out_ptr += rc;
1507                                 break;
1508                         /* UNIX time of coredump */
1509                         case 't': {
1510                                 struct timeval tv;
1511                                 do_gettimeofday(&tv);
1512                                 rc = snprintf(out_ptr, out_end - out_ptr,
1513                                               "%lu", tv.tv_sec);
1514                                 if (rc > out_end - out_ptr)
1515                                         goto out;
1516                                 out_ptr += rc;
1517                                 break;
1518                         }
1519                         /* hostname */
1520                         case 'h':
1521                                 down_read(&uts_sem);
1522                                 rc = snprintf(out_ptr, out_end - out_ptr,
1523                                               "%s", utsname()->nodename);
1524                                 up_read(&uts_sem);
1525                                 if (rc > out_end - out_ptr)
1526                                         goto out;
1527                                 out_ptr += rc;
1528                                 break;
1529                         /* executable */
1530                         case 'e':
1531                                 rc = snprintf(out_ptr, out_end - out_ptr,
1532                                               "%s", current->comm);
1533                                 if (rc > out_end - out_ptr)
1534                                         goto out;
1535                                 out_ptr += rc;
1536                                 break;
1537                         /* core limit size */
1538                         case 'c':
1539                                 rc = snprintf(out_ptr, out_end - out_ptr,
1540                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1541                                 if (rc > out_end - out_ptr)
1542                                         goto out;
1543                                 out_ptr += rc;
1544                                 break;
1545                         default:
1546                                 break;
1547                         }
1548                         ++pat_ptr;
1549                 }
1550         }
1551         /* Backward compatibility with core_uses_pid:
1552          *
1553          * If core_pattern does not include a %p (as is the default)
1554          * and core_uses_pid is set, then .%pid will be appended to
1555          * the filename. Do not do this for piped commands. */
1556         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1557                 rc = snprintf(out_ptr, out_end - out_ptr,
1558                               ".%d", task_tgid_vnr(current));
1559                 if (rc > out_end - out_ptr)
1560                         goto out;
1561                 out_ptr += rc;
1562         }
1563 out:
1564         *out_ptr = 0;
1565         return ispipe;
1566 }
1567
1568 static int zap_process(struct task_struct *start)
1569 {
1570         struct task_struct *t;
1571         int nr = 0;
1572
1573         start->signal->flags = SIGNAL_GROUP_EXIT;
1574         start->signal->group_stop_count = 0;
1575
1576         t = start;
1577         do {
1578                 if (t != current && t->mm) {
1579                         sigaddset(&t->pending.signal, SIGKILL);
1580                         signal_wake_up(t, 1);
1581                         nr++;
1582                 }
1583         } while_each_thread(start, t);
1584
1585         return nr;
1586 }
1587
1588 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1589                                 struct core_state *core_state, int exit_code)
1590 {
1591         struct task_struct *g, *p;
1592         unsigned long flags;
1593         int nr = -EAGAIN;
1594
1595         spin_lock_irq(&tsk->sighand->siglock);
1596         if (!signal_group_exit(tsk->signal)) {
1597                 mm->core_state = core_state;
1598                 tsk->signal->group_exit_code = exit_code;
1599                 nr = zap_process(tsk);
1600         }
1601         spin_unlock_irq(&tsk->sighand->siglock);
1602         if (unlikely(nr < 0))
1603                 return nr;
1604
1605         if (atomic_read(&mm->mm_users) == nr + 1)
1606                 goto done;
1607         /*
1608          * We should find and kill all tasks which use this mm, and we should
1609          * count them correctly into ->nr_threads. We don't take tasklist
1610          * lock, but this is safe wrt:
1611          *
1612          * fork:
1613          *      None of sub-threads can fork after zap_process(leader). All
1614          *      processes which were created before this point should be
1615          *      visible to zap_threads() because copy_process() adds the new
1616          *      process to the tail of init_task.tasks list, and lock/unlock
1617          *      of ->siglock provides a memory barrier.
1618          *
1619          * do_exit:
1620          *      The caller holds mm->mmap_sem. This means that the task which
1621          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1622          *      its ->mm.
1623          *
1624          * de_thread:
1625          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1626          *      we must see either old or new leader, this does not matter.
1627          *      However, it can change p->sighand, so lock_task_sighand(p)
1628          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1629          *      it can't fail.
1630          *
1631          *      Note also that "g" can be the old leader with ->mm == NULL
1632          *      and already unhashed and thus removed from ->thread_group.
1633          *      This is OK, __unhash_process()->list_del_rcu() does not
1634          *      clear the ->next pointer, we will find the new leader via
1635          *      next_thread().
1636          */
1637         rcu_read_lock();
1638         for_each_process(g) {
1639                 if (g == tsk->group_leader)
1640                         continue;
1641                 if (g->flags & PF_KTHREAD)
1642                         continue;
1643                 p = g;
1644                 do {
1645                         if (p->mm) {
1646                                 if (unlikely(p->mm == mm)) {
1647                                         lock_task_sighand(p, &flags);
1648                                         nr += zap_process(p);
1649                                         unlock_task_sighand(p, &flags);
1650                                 }
1651                                 break;
1652                         }
1653                 } while_each_thread(g, p);
1654         }
1655         rcu_read_unlock();
1656 done:
1657         atomic_set(&core_state->nr_threads, nr);
1658         return nr;
1659 }
1660
1661 static int coredump_wait(int exit_code, struct core_state *core_state)
1662 {
1663         struct task_struct *tsk = current;
1664         struct mm_struct *mm = tsk->mm;
1665         struct completion *vfork_done;
1666         int core_waiters;
1667
1668         init_completion(&core_state->startup);
1669         core_state->dumper.task = tsk;
1670         core_state->dumper.next = NULL;
1671         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1672         up_write(&mm->mmap_sem);
1673
1674         if (unlikely(core_waiters < 0))
1675                 goto fail;
1676
1677         /*
1678          * Make sure nobody is waiting for us to release the VM,
1679          * otherwise we can deadlock when we wait on each other
1680          */
1681         vfork_done = tsk->vfork_done;
1682         if (vfork_done) {
1683                 tsk->vfork_done = NULL;
1684                 complete(vfork_done);
1685         }
1686
1687         if (core_waiters)
1688                 wait_for_completion(&core_state->startup);
1689 fail:
1690         return core_waiters;
1691 }
1692
1693 static void coredump_finish(struct mm_struct *mm)
1694 {
1695         struct core_thread *curr, *next;
1696         struct task_struct *task;
1697
1698         next = mm->core_state->dumper.next;
1699         while ((curr = next) != NULL) {
1700                 next = curr->next;
1701                 task = curr->task;
1702                 /*
1703                  * see exit_mm(), curr->task must not see
1704                  * ->task == NULL before we read ->next.
1705                  */
1706                 smp_mb();
1707                 curr->task = NULL;
1708                 wake_up_process(task);
1709         }
1710
1711         mm->core_state = NULL;
1712 }
1713
1714 /*
1715  * set_dumpable converts traditional three-value dumpable to two flags and
1716  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1717  * these bits are not changed atomically.  So get_dumpable can observe the
1718  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1719  * return either old dumpable or new one by paying attention to the order of
1720  * modifying the bits.
1721  *
1722  * dumpable |   mm->flags (binary)
1723  * old  new | initial interim  final
1724  * ---------+-----------------------
1725  *  0    1  |   00      01      01
1726  *  0    2  |   00      10(*)   11
1727  *  1    0  |   01      00      00
1728  *  1    2  |   01      11      11
1729  *  2    0  |   11      10(*)   00
1730  *  2    1  |   11      11      01
1731  *
1732  * (*) get_dumpable regards interim value of 10 as 11.
1733  */
1734 void set_dumpable(struct mm_struct *mm, int value)
1735 {
1736         switch (value) {
1737         case 0:
1738                 clear_bit(MMF_DUMPABLE, &mm->flags);
1739                 smp_wmb();
1740                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1741                 break;
1742         case 1:
1743                 set_bit(MMF_DUMPABLE, &mm->flags);
1744                 smp_wmb();
1745                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1746                 break;
1747         case 2:
1748                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1749                 smp_wmb();
1750                 set_bit(MMF_DUMPABLE, &mm->flags);
1751                 break;
1752         }
1753 }
1754
1755 int get_dumpable(struct mm_struct *mm)
1756 {
1757         int ret;
1758
1759         ret = mm->flags & 0x3;
1760         return (ret >= 2) ? 2 : ret;
1761 }
1762
1763 static void wait_for_dump_helpers(struct file *file)
1764 {
1765         struct pipe_inode_info *pipe;
1766
1767         pipe = file->f_path.dentry->d_inode->i_pipe;
1768
1769         pipe_lock(pipe);
1770         pipe->readers++;
1771         pipe->writers--;
1772
1773         while ((pipe->readers > 1) && (!signal_pending(current))) {
1774                 wake_up_interruptible_sync(&pipe->wait);
1775                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1776                 pipe_wait(pipe);
1777         }
1778
1779         pipe->readers--;
1780         pipe->writers++;
1781         pipe_unlock(pipe);
1782
1783 }
1784
1785
1786 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1787 {
1788         struct core_state core_state;
1789         char corename[CORENAME_MAX_SIZE + 1];
1790         struct mm_struct *mm = current->mm;
1791         struct linux_binfmt * binfmt;
1792         struct inode * inode;
1793         struct file * file;
1794         const struct cred *old_cred;
1795         struct cred *cred;
1796         int retval = 0;
1797         int flag = 0;
1798         int ispipe = 0;
1799         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1800         char **helper_argv = NULL;
1801         int helper_argc = 0;
1802         int dump_count = 0;
1803         static atomic_t core_dump_count = ATOMIC_INIT(0);
1804
1805         audit_core_dumps(signr);
1806
1807         binfmt = mm->binfmt;
1808         if (!binfmt || !binfmt->core_dump)
1809                 goto fail;
1810
1811         cred = prepare_creds();
1812         if (!cred) {
1813                 retval = -ENOMEM;
1814                 goto fail;
1815         }
1816
1817         down_write(&mm->mmap_sem);
1818         /*
1819          * If another thread got here first, or we are not dumpable, bail out.
1820          */
1821         if (mm->core_state || !get_dumpable(mm)) {
1822                 up_write(&mm->mmap_sem);
1823                 put_cred(cred);
1824                 goto fail;
1825         }
1826
1827         /*
1828          *      We cannot trust fsuid as being the "true" uid of the
1829          *      process nor do we know its entire history. We only know it
1830          *      was tainted so we dump it as root in mode 2.
1831          */
1832         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1833                 flag = O_EXCL;          /* Stop rewrite attacks */
1834                 cred->fsuid = 0;        /* Dump root private */
1835         }
1836
1837         retval = coredump_wait(exit_code, &core_state);
1838         if (retval < 0) {
1839                 put_cred(cred);
1840                 goto fail;
1841         }
1842
1843         old_cred = override_creds(cred);
1844
1845         /*
1846          * Clear any false indication of pending signals that might
1847          * be seen by the filesystem code called to write the core file.
1848          */
1849         clear_thread_flag(TIF_SIGPENDING);
1850
1851         /*
1852          * lock_kernel() because format_corename() is controlled by sysctl, which
1853          * uses lock_kernel()
1854          */
1855         lock_kernel();
1856         ispipe = format_corename(corename, signr);
1857         unlock_kernel();
1858
1859         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1860                 goto fail_unlock;
1861
1862         if (ispipe) {
1863                 if (core_limit == 0) {
1864                         /*
1865                          * Normally core limits are irrelevant to pipes, since
1866                          * we're not writing to the file system, but we use
1867                          * core_limit of 0 here as a speacial value. Any
1868                          * non-zero limit gets set to RLIM_INFINITY below, but
1869                          * a limit of 0 skips the dump.  This is a consistent
1870                          * way to catch recursive crashes.  We can still crash
1871                          * if the core_pattern binary sets RLIM_CORE =  !0
1872                          * but it runs as root, and can do lots of stupid things
1873                          * Note that we use task_tgid_vnr here to grab the pid
1874                          * of the process group leader.  That way we get the
1875                          * right pid if a thread in a multi-threaded
1876                          * core_pattern process dies.
1877                          */
1878                         printk(KERN_WARNING
1879                                 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1880                                 task_tgid_vnr(current), current->comm);
1881                         printk(KERN_WARNING "Aborting core\n");
1882                         goto fail_unlock;
1883                 }
1884
1885                 dump_count = atomic_inc_return(&core_dump_count);
1886                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1887                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1888                                task_tgid_vnr(current), current->comm);
1889                         printk(KERN_WARNING "Skipping core dump\n");
1890                         goto fail_dropcount;
1891                 }
1892
1893                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1894                 if (!helper_argv) {
1895                         printk(KERN_WARNING "%s failed to allocate memory\n",
1896                                __func__);
1897                         goto fail_dropcount;
1898                 }
1899
1900                 core_limit = RLIM_INFINITY;
1901
1902                 /* SIGPIPE can happen, but it's just never processed */
1903                 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1904                                 &file)) {
1905                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1906                                corename);
1907                         goto fail_dropcount;
1908                 }
1909         } else
1910                 file = filp_open(corename,
1911                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1912                                  0600);
1913         if (IS_ERR(file))
1914                 goto fail_dropcount;
1915         inode = file->f_path.dentry->d_inode;
1916         if (inode->i_nlink > 1)
1917                 goto close_fail;        /* multiple links - don't dump */
1918         if (!ispipe && d_unhashed(file->f_path.dentry))
1919                 goto close_fail;
1920
1921         /* AK: actually i see no reason to not allow this for named pipes etc.,
1922            but keep the previous behaviour for now. */
1923         if (!ispipe && !S_ISREG(inode->i_mode))
1924                 goto close_fail;
1925         /*
1926          * Dont allow local users get cute and trick others to coredump
1927          * into their pre-created files:
1928          * Note, this is not relevant for pipes
1929          */
1930         if (!ispipe && (inode->i_uid != current_fsuid()))
1931                 goto close_fail;
1932         if (!file->f_op)
1933                 goto close_fail;
1934         if (!file->f_op->write)
1935                 goto close_fail;
1936         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1937                 goto close_fail;
1938
1939         retval = binfmt->core_dump(signr, regs, file, core_limit);
1940
1941         if (retval)
1942                 current->signal->group_exit_code |= 0x80;
1943 close_fail:
1944         if (ispipe && core_pipe_limit)
1945                 wait_for_dump_helpers(file);
1946         filp_close(file, NULL);
1947 fail_dropcount:
1948         if (dump_count)
1949                 atomic_dec(&core_dump_count);
1950 fail_unlock:
1951         if (helper_argv)
1952                 argv_free(helper_argv);
1953
1954         revert_creds(old_cred);
1955         put_cred(cred);
1956         coredump_finish(mm);
1957 fail:
1958         return;
1959 }