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/syscalls.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53 #include <linux/tracehook.h>
  54 #include <linux/kmod.h>
  55 #include <linux/fsnotify.h>
  56 #include <linux/fs_struct.h>
  57 #include <linux/pipe_fs_i.h>
  58
  59 #include <asm/uaccess.h>
  60 #include <asm/mmu_context.h>
  61 #include <asm/tlb.h>
  62 #include "internal.h"
  63
  64 int core_uses_pid;
  65 char core_pattern[CORENAME_MAX_SIZE] = "core";
  66 unsigned int core_pipe_limit;
  67 int suid_dumpable = 0;
  68
  69 /* The maximal length of core_pattern is also specified in sysctl.c */
  70
  71 static LIST_HEAD(formats);
  72 static DEFINE_RWLOCK(binfmt_lock);
  73
  74 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  75 {
  76         if (!fmt)
  77                 return -EINVAL;
  78         write_lock(&binfmt_lock);
  79         insert ? list_add(&fmt->lh, &formats) :
  80                  list_add_tail(&fmt->lh, &formats);
  81         write_unlock(&binfmt_lock);
  82         return 0;
  83 }
  84
  85 EXPORT_SYMBOL(__register_binfmt);
  86
  87 void unregister_binfmt(struct linux_binfmt * fmt)
  88 {
  89         write_lock(&binfmt_lock);
  90         list_del(&fmt->lh);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(unregister_binfmt);
  95
  96 static inline void put_binfmt(struct linux_binfmt * fmt)
  97 {
  98         module_put(fmt->module);
  99 }
 100
 101 /*
 102  * Note that a shared library must be both readable and executable due to
 103  * security reasons.
 104  *
 105  * Also note that we take the address to load from from the file itself.
 106  */
 107 SYSCALL_DEFINE1(uselib, const char __user *, library)
 108 {
 109         struct file *file;
 110         char *tmp = getname(library);
 111         int error = PTR_ERR(tmp);
 112
 113         if (IS_ERR(tmp))
 114                 goto out;
 115
 116         file = do_filp_open(AT_FDCWD, tmp,
 117                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 118                                 MAY_READ | MAY_EXEC | MAY_OPEN);
 119         putname(tmp);
 120         error = PTR_ERR(file);
 121         if (IS_ERR(file))
 122                 goto out;
 123
 124         error = -EINVAL;
 125         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 126                 goto exit;
 127
 128         error = -EACCES;
 129         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 130                 goto exit;
 131
 132         fsnotify_open(file->f_path.dentry);
 133
 134         error = -ENOEXEC;
 135         if(file->f_op) {
 136                 struct linux_binfmt * fmt;
 137
 138                 read_lock(&binfmt_lock);
 139                 list_for_each_entry(fmt, &formats, lh) {
 140                         if (!fmt->load_shlib)
 141                                 continue;
 142                         if (!try_module_get(fmt->module))
 143                                 continue;
 144                         read_unlock(&binfmt_lock);
 145                         error = fmt->load_shlib(file);
 146                         read_lock(&binfmt_lock);
 147                         put_binfmt(fmt);
 148                         if (error != -ENOEXEC)
 149                                 break;
 150                 }
 151                 read_unlock(&binfmt_lock);
 152         }
 153 exit:
 154         fput(file);
 155 out:
 156         return error;
 157 }
 158
 159 #ifdef CONFIG_MMU
 160
 161 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 162                 int write)
 163 {
 164         struct page *page;
 165         int ret;
 166
 167 #ifdef CONFIG_STACK_GROWSUP
 168         if (write) {
 169                 ret = expand_stack_downwards(bprm->vma, pos);
 170                 if (ret < 0)
 171                         return NULL;
 172         }
 173 #endif
 174         ret = get_user_pages(current, bprm->mm, pos,
 175                         1, write, 1, &page, NULL);
 176         if (ret <= 0)
 177                 return NULL;
 178
 179         if (write) {
 180                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 181                 struct rlimit *rlim;
 182
 183                 /*
 184                  * We've historically supported up to 32 pages (ARG_MAX)
 185                  * of argument strings even with small stacks
 186                  */
 187                 if (size <= ARG_MAX)
 188                         return page;
 189
 190                 /*
 191                  * Limit to 1/4-th the stack size for the argv+env strings.
 192                  * This ensures that:
 193                  *  - the remaining binfmt code will not run out of stack space,
 194                  *  - the program will have a reasonable amount of stack left
 195                  *    to work from.
 196                  */
 197                 rlim = current->signal->rlim;
 198                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 199                         put_page(page);
 200                         return NULL;
 201                 }
 202         }
 203
 204         return page;
 205 }
 206
 207 static void put_arg_page(struct page *page)
 208 {
 209         put_page(page);
 210 }
 211
 212 static void free_arg_page(struct linux_binprm *bprm, int i)
 213 {
 214 }
 215
 216 static void free_arg_pages(struct linux_binprm *bprm)
 217 {
 218 }
 219
 220 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 221                 struct page *page)
 222 {
 223         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 224 }
 225
 226 static int __bprm_mm_init(struct linux_binprm *bprm)
 227 {
 228         int err;
 229         struct vm_area_struct *vma = NULL;
 230         struct mm_struct *mm = bprm->mm;
 231
 232         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 233         if (!vma)
 234                 return -ENOMEM;
 235
 236         down_write(&mm->mmap_sem);
 237         vma->vm_mm = mm;
 238
 239         /*
 240          * Place the stack at the largest stack address the architecture
 241          * supports. Later, we'll move this to an appropriate place. We don't
 242          * use STACK_TOP because that can depend on attributes which aren't
 243          * configured yet.
 244          */
 245         BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 246         vma->vm_end = STACK_TOP_MAX;
 247         vma->vm_start = vma->vm_end - PAGE_SIZE;
 248         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 249         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 250         INIT_LIST_HEAD(&vma->anon_vma_chain);
 251         err = insert_vm_struct(mm, vma);
 252         if (err)
 253                 goto err;
 254
 255         mm->stack_vm = mm->total_vm = 1;
 256         up_write(&mm->mmap_sem);
 257         bprm->p = vma->vm_end - sizeof(void *);
 258         return 0;
 259 err:
 260         up_write(&mm->mmap_sem);
 261         bprm->vma = NULL;
 262         kmem_cache_free(vm_area_cachep, vma);
 263         return err;
 264 }
 265
 266 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 267 {
 268         return len <= MAX_ARG_STRLEN;
 269 }
 270
 271 #else
 272
 273 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 274                 int write)
 275 {
 276         struct page *page;
 277
 278         page = bprm->page[pos / PAGE_SIZE];
 279         if (!page && write) {
 280                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 281                 if (!page)
 282                         return NULL;
 283                 bprm->page[pos / PAGE_SIZE] = page;
 284         }
 285
 286         return page;
 287 }
 288
 289 static void put_arg_page(struct page *page)
 290 {
 291 }
 292
 293 static void free_arg_page(struct linux_binprm *bprm, int i)
 294 {
 295         if (bprm->page[i]) {
 296                 __free_page(bprm->page[i]);
 297                 bprm->page[i] = NULL;
 298         }
 299 }
 300
 301 static void free_arg_pages(struct linux_binprm *bprm)
 302 {
 303         int i;
 304
 305         for (i = 0; i < MAX_ARG_PAGES; i++)
 306                 free_arg_page(bprm, i);
 307 }
 308
 309 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 310                 struct page *page)
 311 {
 312 }
 313
 314 static int __bprm_mm_init(struct linux_binprm *bprm)
 315 {
 316         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 317         return 0;
 318 }
 319
 320 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 321 {
 322         return len <= bprm->p;
 323 }
 324
 325 #endif /* CONFIG_MMU */
 326
 327 /*
 328  * Create a new mm_struct and populate it with a temporary stack
 329  * vm_area_struct.  We don't have enough context at this point to set the stack
 330  * flags, permissions, and offset, so we use temporary values.  We'll update
 331  * them later in setup_arg_pages().
 332  */
 333 int bprm_mm_init(struct linux_binprm *bprm)
 334 {
 335         int err;
 336         struct mm_struct *mm = NULL;
 337
 338         bprm->mm = mm = mm_alloc();
 339         err = -ENOMEM;
 340         if (!mm)
 341                 goto err;
 342
 343         err = init_new_context(current, mm);
 344         if (err)
 345                 goto err;
 346
 347         err = __bprm_mm_init(bprm);
 348         if (err)
 349                 goto err;
 350
 351         return 0;
 352
 353 err:
 354         if (mm) {
 355                 bprm->mm = NULL;
 356                 mmdrop(mm);
 357         }
 358
 359         return err;
 360 }
 361
 362 /*
 363  * count() counts the number of strings in array ARGV.
 364  */
 365 static int count(char __user * __user * argv, int max)
 366 {
 367         int i = 0;
 368
 369         if (argv != NULL) {
 370                 for (;;) {
 371                         char __user * p;
 372
 373                         if (get_user(p, argv))
 374                                 return -EFAULT;
 375                         if (!p)
 376                                 break;
 377                         argv++;
 378                         if (i++ >= max)
 379                                 return -E2BIG;
 380                         cond_resched();
 381                 }
 382         }
 383         return i;
 384 }
 385
 386 /*
 387  * 'copy_strings()' copies argument/environment strings from the old
 388  * processes's memory to the new process's stack.  The call to get_user_pages()
 389  * ensures the destination page is created and not swapped out.
 390  */
 391 static int copy_strings(int argc, char __user * __user * argv,
 392                         struct linux_binprm *bprm)
 393 {
 394         struct page *kmapped_page = NULL;
 395         char *kaddr = NULL;
 396         unsigned long kpos = 0;
 397         int ret;
 398
 399         while (argc-- > 0) {
 400                 char __user *str;
 401                 int len;
 402                 unsigned long pos;
 403
 404                 if (get_user(str, argv+argc) ||
 405                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 406                         ret = -EFAULT;
 407                         goto out;
 408                 }
 409
 410                 if (!valid_arg_len(bprm, len)) {
 411                         ret = -E2BIG;
 412                         goto out;
 413                 }
 414
 415                 /* We're going to work our way backwords. */
 416                 pos = bprm->p;
 417                 str += len;
 418                 bprm->p -= len;
 419
 420                 while (len > 0) {
 421                         int offset, bytes_to_copy;
 422
 423                         offset = pos % PAGE_SIZE;
 424                         if (offset == 0)
 425                                 offset = PAGE_SIZE;
 426
 427                         bytes_to_copy = offset;
 428                         if (bytes_to_copy > len)
 429                                 bytes_to_copy = len;
 430
 431                         offset -= bytes_to_copy;
 432                         pos -= bytes_to_copy;
 433                         str -= bytes_to_copy;
 434                         len -= bytes_to_copy;
 435
 436                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 437                                 struct page *page;
 438
 439                                 page = get_arg_page(bprm, pos, 1);
 440                                 if (!page) {
 441                                         ret = -E2BIG;
 442                                         goto out;
 443                                 }
 444
 445                                 if (kmapped_page) {
 446                                         flush_kernel_dcache_page(kmapped_page);
 447                                         kunmap(kmapped_page);
 448                                         put_arg_page(kmapped_page);
 449                                 }
 450                                 kmapped_page = page;
 451                                 kaddr = kmap(kmapped_page);
 452                                 kpos = pos & PAGE_MASK;
 453                                 flush_arg_page(bprm, kpos, kmapped_page);
 454                         }
 455                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 456                                 ret = -EFAULT;
 457                                 goto out;
 458                         }
 459                 }
 460         }
 461         ret = 0;
 462 out:
 463         if (kmapped_page) {
 464                 flush_kernel_dcache_page(kmapped_page);
 465                 kunmap(kmapped_page);
 466                 put_arg_page(kmapped_page);
 467         }
 468         return ret;
 469 }
 470
 471 /*
 472  * Like copy_strings, but get argv and its values from kernel memory.
 473  */
 474 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 475 {
 476         int r;
 477         mm_segment_t oldfs = get_fs();
 478         set_fs(KERNEL_DS);
 479         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 480         set_fs(oldfs);
 481         return r;
 482 }
 483 EXPORT_SYMBOL(copy_strings_kernel);
 484
 485 #ifdef CONFIG_MMU
 486
 487 /*
 488  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 489  * the binfmt code determines where the new stack should reside, we shift it to
 490  * its final location.  The process proceeds as follows:
 491  *
 492  * 1) Use shift to calculate the new vma endpoints.
 493  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 494  *    arguments passed to subsequent functions are consistent.
 495  * 3) Move vma's page tables to the new range.
 496  * 4) Free up any cleared pgd range.
 497  * 5) Shrink the vma to cover only the new range.
 498  */
 499 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 500 {
 501         struct mm_struct *mm = vma->vm_mm;
 502         unsigned long old_start = vma->vm_start;
 503         unsigned long old_end = vma->vm_end;
 504         unsigned long length = old_end - old_start;
 505         unsigned long new_start = old_start - shift;
 506         unsigned long new_end = old_end - shift;
 507         struct mmu_gather *tlb;
 508
 509         BUG_ON(new_start > new_end);
 510
 511         /*
 512          * ensure there are no vmas between where we want to go
 513          * and where we are
 514          */
 515         if (vma != find_vma(mm, new_start))
 516                 return -EFAULT;
 517
 518         /*
 519          * cover the whole range: [new_start, old_end)
 520          */
 521         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 522                 return -ENOMEM;
 523
 524         /*
 525          * move the page tables downwards, on failure we rely on
 526          * process cleanup to remove whatever mess we made.
 527          */
 528         if (length != move_page_tables(vma, old_start,
 529                                        vma, new_start, length))
 530                 return -ENOMEM;
 531
 532         lru_add_drain();
 533         tlb = tlb_gather_mmu(mm, 0);
 534         if (new_end > old_start) {
 535                 /*
 536                  * when the old and new regions overlap clear from new_end.
 537                  */
 538                 free_pgd_range(tlb, new_end, old_end, new_end,
 539                         vma->vm_next ? vma->vm_next->vm_start : 0);
 540         } else {
 541                 /*
 542                  * otherwise, clean from old_start; this is done to not touch
 543                  * the address space in [new_end, old_start) some architectures
 544                  * have constraints on va-space that make this illegal (IA64) -
 545                  * for the others its just a little faster.
 546                  */
 547                 free_pgd_range(tlb, old_start, old_end, new_end,
 548                         vma->vm_next ? vma->vm_next->vm_start : 0);
 549         }
 550         tlb_finish_mmu(tlb, new_end, old_end);
 551
 552         /*
 553          * Shrink the vma to just the new range.  Always succeeds.
 554          */
 555         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 556
 557         return 0;
 558 }
 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 = rlimit_max(RLIMIT_STACK);
 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         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 626
 627         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 628                         vm_flags);
 629         if (ret)
 630                 goto out_unlock;
 631         BUG_ON(prev != vma);
 632
 633         /* Move stack pages down in memory. */
 634         if (stack_shift) {
 635                 ret = shift_arg_pages(vma, stack_shift);
 636                 if (ret)
 637                         goto out_unlock;
 638         }
 639
 640         /* mprotect_fixup is overkill to remove the temporary stack flags */
 641         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 642
 643         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 644         stack_size = vma->vm_end - vma->vm_start;
 645         /*
 646          * Align this down to a page boundary as expand_stack
 647          * will align it up.
 648          */
 649         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 650 #ifdef CONFIG_STACK_GROWSUP
 651         if (stack_size + stack_expand > rlim_stack)
 652                 stack_base = vma->vm_start + rlim_stack;
 653         else
 654                 stack_base = vma->vm_end + stack_expand;
 655 #else
 656         if (stack_size + stack_expand > rlim_stack)
 657                 stack_base = vma->vm_end - rlim_stack;
 658         else
 659                 stack_base = vma->vm_start - stack_expand;
 660 #endif
 661         ret = expand_stack(vma, stack_base);
 662         if (ret)
 663                 ret = -EFAULT;
 664
 665 out_unlock:
 666         up_write(&mm->mmap_sem);
 667         return ret;
 668 }
 669 EXPORT_SYMBOL(setup_arg_pages);
 670
 671 #endif /* CONFIG_MMU */
 672
 673 struct file *open_exec(const char *name)
 674 {
 675         struct file *file;
 676         int err;
 677
 678         file = do_filp_open(AT_FDCWD, name,
 679                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 680                                 MAY_EXEC | MAY_OPEN);
 681         if (IS_ERR(file))
 682                 goto out;
 683
 684         err = -EACCES;
 685         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 686                 goto exit;
 687
 688         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 689                 goto exit;
 690
 691         fsnotify_open(file->f_path.dentry);
 692
 693         err = deny_write_access(file);
 694         if (err)
 695                 goto exit;
 696
 697 out:
 698         return file;
 699
 700 exit:
 701         fput(file);
 702         return ERR_PTR(err);
 703 }
 704 EXPORT_SYMBOL(open_exec);
 705
 706 int kernel_read(struct file *file, loff_t offset,
 707                 char *addr, unsigned long count)
 708 {
 709         mm_segment_t old_fs;
 710         loff_t pos = offset;
 711         int result;
 712
 713         old_fs = get_fs();
 714         set_fs(get_ds());
 715         /* The cast to a user pointer is valid due to the set_fs() */
 716         result = vfs_read(file, (void __user *)addr, count, &pos);
 717         set_fs(old_fs);
 718         return result;
 719 }
 720
 721 EXPORT_SYMBOL(kernel_read);
 722
 723 static int exec_mmap(struct mm_struct *mm)
 724 {
 725         struct task_struct *tsk;
 726         struct mm_struct * old_mm, *active_mm;
 727
 728         /* Notify parent that we're no longer interested in the old VM */
 729         tsk = current;
 730         old_mm = current->mm;
 731         sync_mm_rss(tsk, old_mm);
 732         mm_release(tsk, old_mm);
 733
 734         if (old_mm) {
 735                 /*
 736                  * Make sure that if there is a core dump in progress
 737                  * for the old mm, we get out and die instead of going
 738                  * through with the exec.  We must hold mmap_sem around
 739                  * checking core_state and changing tsk->mm.
 740                  */
 741                 down_read(&old_mm->mmap_sem);
 742                 if (unlikely(old_mm->core_state)) {
 743                         up_read(&old_mm->mmap_sem);
 744                         return -EINTR;
 745                 }
 746         }
 747         task_lock(tsk);
 748         active_mm = tsk->active_mm;
 749         tsk->mm = mm;
 750         tsk->active_mm = mm;
 751         activate_mm(active_mm, mm);
 752         task_unlock(tsk);
 753         arch_pick_mmap_layout(mm);
 754         if (old_mm) {
 755                 up_read(&old_mm->mmap_sem);
 756                 BUG_ON(active_mm != old_mm);
 757                 mm_update_next_owner(old_mm);
 758                 mmput(old_mm);
 759                 return 0;
 760         }
 761         mmdrop(active_mm);
 762         return 0;
 763 }
 764
 765 /*
 766  * This function makes sure the current process has its own signal table,
 767  * so that flush_signal_handlers can later reset the handlers without
 768  * disturbing other processes.  (Other processes might share the signal
 769  * table via the CLONE_SIGHAND option to clone().)
 770  */
 771 static int de_thread(struct task_struct *tsk)
 772 {
 773         struct signal_struct *sig = tsk->signal;
 774         struct sighand_struct *oldsighand = tsk->sighand;
 775         spinlock_t *lock = &oldsighand->siglock;
 776
 777         if (thread_group_empty(tsk))
 778                 goto no_thread_group;
 779
 780         /*
 781          * Kill all other threads in the thread group.
 782          */
 783         spin_lock_irq(lock);
 784         if (signal_group_exit(sig)) {
 785                 /*
 786                  * Another group action in progress, just
 787                  * return so that the signal is processed.
 788                  */
 789                 spin_unlock_irq(lock);
 790                 return -EAGAIN;
 791         }
 792
 793         sig->group_exit_task = tsk;
 794         sig->notify_count = zap_other_threads(tsk);
 795         if (!thread_group_leader(tsk))
 796                 sig->notify_count--;
 797
 798         while (sig->notify_count) {
 799                 __set_current_state(TASK_UNINTERRUPTIBLE);
 800                 spin_unlock_irq(lock);
 801                 schedule();
 802                 spin_lock_irq(lock);
 803         }
 804         spin_unlock_irq(lock);
 805
 806         /*
 807          * At this point all other threads have exited, all we have to
 808          * do is to wait for the thread group leader to become inactive,
 809          * and to assume its PID:
 810          */
 811         if (!thread_group_leader(tsk)) {
 812                 struct task_struct *leader = tsk->group_leader;
 813
 814                 sig->notify_count = -1; /* for exit_notify() */
 815                 for (;;) {
 816                         write_lock_irq(&tasklist_lock);
 817                         if (likely(leader->exit_state))
 818                                 break;
 819                         __set_current_state(TASK_UNINTERRUPTIBLE);
 820                         write_unlock_irq(&tasklist_lock);
 821                         schedule();
 822                 }
 823
 824                 /*
 825                  * The only record we have of the real-time age of a
 826                  * process, regardless of execs it's done, is start_time.
 827                  * All the past CPU time is accumulated in signal_struct
 828                  * from sister threads now dead.  But in this non-leader
 829                  * exec, nothing survives from the original leader thread,
 830                  * whose birth marks the true age of this process now.
 831                  * When we take on its identity by switching to its PID, we
 832                  * also take its birthdate (always earlier than our own).
 833                  */
 834                 tsk->start_time = leader->start_time;
 835
 836                 BUG_ON(!same_thread_group(leader, tsk));
 837                 BUG_ON(has_group_leader_pid(tsk));
 838                 /*
 839                  * An exec() starts a new thread group with the
 840                  * TGID of the previous thread group. Rehash the
 841                  * two threads with a switched PID, and release
 842                  * the former thread group leader:
 843                  */
 844
 845                 /* Become a process group leader with the old leader's pid.
 846                  * The old leader becomes a thread of the this thread group.
 847                  * Note: The old leader also uses this pid until release_task
 848                  *       is called.  Odd but simple and correct.
 849                  */
 850                 detach_pid(tsk, PIDTYPE_PID);
 851                 tsk->pid = leader->pid;
 852                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 853                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 854                 transfer_pid(leader, tsk, PIDTYPE_SID);
 855
 856                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 857                 list_replace_init(&leader->sibling, &tsk->sibling);
 858
 859                 tsk->group_leader = tsk;
 860                 leader->group_leader = tsk;
 861
 862                 tsk->exit_signal = SIGCHLD;
 863
 864                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 865                 leader->exit_state = EXIT_DEAD;
 866                 write_unlock_irq(&tasklist_lock);
 867
 868                 release_task(leader);
 869         }
 870
 871         sig->group_exit_task = NULL;
 872         sig->notify_count = 0;
 873
 874 no_thread_group:
 875         if (current->mm)
 876                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 877
 878         exit_itimers(sig);
 879         flush_itimer_signals();
 880
 881         if (atomic_read(&oldsighand->count) != 1) {
 882                 struct sighand_struct *newsighand;
 883                 /*
 884                  * This ->sighand is shared with the CLONE_SIGHAND
 885                  * but not CLONE_THREAD task, switch to the new one.
 886                  */
 887                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 888                 if (!newsighand)
 889                         return -ENOMEM;
 890
 891                 atomic_set(&newsighand->count, 1);
 892                 memcpy(newsighand->action, oldsighand->action,
 893                        sizeof(newsighand->action));
 894
 895                 write_lock_irq(&tasklist_lock);
 896                 spin_lock(&oldsighand->siglock);
 897                 rcu_assign_pointer(tsk->sighand, newsighand);
 898                 spin_unlock(&oldsighand->siglock);
 899                 write_unlock_irq(&tasklist_lock);
 900
 901                 __cleanup_sighand(oldsighand);
 902         }
 903
 904         BUG_ON(!thread_group_leader(tsk));
 905         return 0;
 906 }
 907
 908 /*
 909  * These functions flushes out all traces of the currently running executable
 910  * so that a new one can be started
 911  */
 912 static void flush_old_files(struct files_struct * files)
 913 {
 914         long j = -1;
 915         struct fdtable *fdt;
 916
 917         spin_lock(&files->file_lock);
 918         for (;;) {
 919                 unsigned long set, i;
 920
 921                 j++;
 922                 i = j * __NFDBITS;
 923                 fdt = files_fdtable(files);
 924                 if (i >= fdt->max_fds)
 925                         break;
 926                 set = fdt->close_on_exec->fds_bits[j];
 927                 if (!set)
 928                         continue;
 929                 fdt->close_on_exec->fds_bits[j] = 0;
 930                 spin_unlock(&files->file_lock);
 931                 for ( ; set ; i++,set >>= 1) {
 932                         if (set & 1) {
 933                                 sys_close(i);
 934                         }
 935                 }
 936                 spin_lock(&files->file_lock);
 937
 938         }
 939         spin_unlock(&files->file_lock);
 940 }
 941
 942 char *get_task_comm(char *buf, struct task_struct *tsk)
 943 {
 944         /* buf must be at least sizeof(tsk->comm) in size */
 945         task_lock(tsk);
 946         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 947         task_unlock(tsk);
 948         return buf;
 949 }
 950
 951 void set_task_comm(struct task_struct *tsk, char *buf)
 952 {
 953         task_lock(tsk);
 954
 955         /*
 956          * Threads may access current->comm without holding
 957          * the task lock, so write the string carefully.
 958          * Readers without a lock may see incomplete new
 959          * names but are safe from non-terminating string reads.
 960          */
 961         memset(tsk->comm, 0, TASK_COMM_LEN);
 962         wmb();
 963         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 964         task_unlock(tsk);
 965         perf_event_comm(tsk);
 966 }
 967
 968 int flush_old_exec(struct linux_binprm * bprm)
 969 {
 970         int retval;
 971
 972         /*
 973          * Make sure we have a private signal table and that
 974          * we are unassociated from the previous thread group.
 975          */
 976         retval = de_thread(current);
 977         if (retval)
 978                 goto out;
 979
 980         set_mm_exe_file(bprm->mm, bprm->file);
 981
 982         /*
 983          * Release all of the old mmap stuff
 984          */
 985         retval = exec_mmap(bprm->mm);
 986         if (retval)
 987                 goto out;
 988
 989         bprm->mm = NULL;                /* We're using it now */
 990
 991         current->flags &= ~PF_RANDOMIZE;
 992         flush_thread();
 993         current->personality &= ~bprm->per_clear;
 994
 995         return 0;
 996
 997 out:
 998         return retval;
 999 }
1000 EXPORT_SYMBOL(flush_old_exec);
1001
1002 void setup_new_exec(struct linux_binprm * bprm)
1003 {
1004         int i, ch;
1005         char * name;
1006         char tcomm[sizeof(current->comm)];
1007
1008         arch_pick_mmap_layout(current->mm);
1009
1010         /* This is the point of no return */
1011         current->sas_ss_sp = current->sas_ss_size = 0;
1012
1013         if (current_euid() == current_uid() && current_egid() == current_gid())
1014                 set_dumpable(current->mm, 1);
1015         else
1016                 set_dumpable(current->mm, suid_dumpable);
1017
1018         name = bprm->filename;
1019
1020         /* Copies the binary name from after last slash */
1021         for (i=0; (ch = *(name++)) != '\0';) {
1022                 if (ch == '/')
1023                         i = 0; /* overwrite what we wrote */
1024                 else
1025                         if (i < (sizeof(tcomm) - 1))
1026                                 tcomm[i++] = ch;
1027         }
1028         tcomm[i] = '\0';
1029         set_task_comm(current, tcomm);
1030
1031         /* Set the new mm task size. We have to do that late because it may
1032          * depend on TIF_32BIT which is only updated in flush_thread() on
1033          * some architectures like powerpc
1034          */
1035         current->mm->task_size = TASK_SIZE;
1036
1037         /* install the new credentials */
1038         if (bprm->cred->uid != current_euid() ||
1039             bprm->cred->gid != current_egid()) {
1040                 current->pdeath_signal = 0;
1041         } else if (file_permission(bprm->file, MAY_READ) ||
1042                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1043                 set_dumpable(current->mm, suid_dumpable);
1044         }
1045
1046         /*
1047          * Flush performance counters when crossing a
1048          * security domain:
1049          */
1050         if (!get_dumpable(current->mm))
1051                 perf_event_exit_task(current);
1052
1053         /* An exec changes our domain. We are no longer part of the thread
1054            group */
1055
1056         current->self_exec_id++;
1057
1058         flush_signal_handlers(current, 0);
1059         flush_old_files(current->files);
1060 }
1061 EXPORT_SYMBOL(setup_new_exec);
1062
1063 /*
1064  * Prepare credentials and lock ->cred_guard_mutex.
1065  * install_exec_creds() commits the new creds and drops the lock.
1066  * Or, if exec fails before, free_bprm() should release ->cred and
1067  * and unlock.
1068  */
1069 int prepare_bprm_creds(struct linux_binprm *bprm)
1070 {
1071         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1072                 return -ERESTARTNOINTR;
1073
1074         bprm->cred = prepare_exec_creds();
1075         if (likely(bprm->cred))
1076                 return 0;
1077
1078         mutex_unlock(&current->cred_guard_mutex);
1079         return -ENOMEM;
1080 }
1081
1082 void free_bprm(struct linux_binprm *bprm)
1083 {
1084         free_arg_pages(bprm);
1085         if (bprm->cred) {
1086                 mutex_unlock(&current->cred_guard_mutex);
1087                 abort_creds(bprm->cred);
1088         }
1089         kfree(bprm);
1090 }
1091
1092 /*
1093  * install the new credentials for this executable
1094  */
1095 void install_exec_creds(struct linux_binprm *bprm)
1096 {
1097         security_bprm_committing_creds(bprm);
1098
1099         commit_creds(bprm->cred);
1100         bprm->cred = NULL;
1101         /*
1102          * cred_guard_mutex must be held at least to this point to prevent
1103          * ptrace_attach() from altering our determination of the task's
1104          * credentials; any time after this it may be unlocked.
1105          */
1106         security_bprm_committed_creds(bprm);
1107         mutex_unlock(&current->cred_guard_mutex);
1108 }
1109 EXPORT_SYMBOL(install_exec_creds);
1110
1111 /*
1112  * determine how safe it is to execute the proposed program
1113  * - the caller must hold current->cred_guard_mutex to protect against
1114  *   PTRACE_ATTACH
1115  */
1116 int check_unsafe_exec(struct linux_binprm *bprm)
1117 {
1118         struct task_struct *p = current, *t;
1119         unsigned n_fs;
1120         int res = 0;
1121
1122         bprm->unsafe = tracehook_unsafe_exec(p);
1123
1124         n_fs = 1;
1125         write_lock(&p->fs->lock);
1126         rcu_read_lock();
1127         for (t = next_thread(p); t != p; t = next_thread(t)) {
1128                 if (t->fs == p->fs)
1129                         n_fs++;
1130         }
1131         rcu_read_unlock();
1132
1133         if (p->fs->users > n_fs) {
1134                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1135         } else {
1136                 res = -EAGAIN;
1137                 if (!p->fs->in_exec) {
1138                         p->fs->in_exec = 1;
1139                         res = 1;
1140                 }
1141         }
1142         write_unlock(&p->fs->lock);
1143
1144         return res;
1145 }
1146
1147 /*
1148  * Fill the binprm structure from the inode.
1149  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1150  *
1151  * This may be called multiple times for binary chains (scripts for example).
1152  */
1153 int prepare_binprm(struct linux_binprm *bprm)
1154 {
1155         umode_t mode;
1156         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1157         int retval;
1158
1159         mode = inode->i_mode;
1160         if (bprm->file->f_op == NULL)
1161                 return -EACCES;
1162
1163         /* clear any previous set[ug]id data from a previous binary */
1164         bprm->cred->euid = current_euid();
1165         bprm->cred->egid = current_egid();
1166
1167         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1168                 /* Set-uid? */
1169                 if (mode & S_ISUID) {
1170                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1171                         bprm->cred->euid = inode->i_uid;
1172                 }
1173
1174                 /* Set-gid? */
1175                 /*
1176                  * If setgid is set but no group execute bit then this
1177                  * is a candidate for mandatory locking, not a setgid
1178                  * executable.
1179                  */
1180                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1181                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1182                         bprm->cred->egid = inode->i_gid;
1183                 }
1184         }
1185
1186         /* fill in binprm security blob */
1187         retval = security_bprm_set_creds(bprm);
1188         if (retval)
1189                 return retval;
1190         bprm->cred_prepared = 1;
1191
1192         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1193         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1194 }
1195
1196 EXPORT_SYMBOL(prepare_binprm);
1197
1198 /*
1199  * Arguments are '\0' separated strings found at the location bprm->p
1200  * points to; chop off the first by relocating brpm->p to right after
1201  * the first '\0' encountered.
1202  */
1203 int remove_arg_zero(struct linux_binprm *bprm)
1204 {
1205         int ret = 0;
1206         unsigned long offset;
1207         char *kaddr;
1208         struct page *page;
1209
1210         if (!bprm->argc)
1211                 return 0;
1212
1213         do {
1214                 offset = bprm->p & ~PAGE_MASK;
1215                 page = get_arg_page(bprm, bprm->p, 0);
1216                 if (!page) {
1217                         ret = -EFAULT;
1218                         goto out;
1219                 }
1220                 kaddr = kmap_atomic(page, KM_USER0);
1221
1222                 for (; offset < PAGE_SIZE && kaddr[offset];
1223                                 offset++, bprm->p++)
1224                         ;
1225
1226                 kunmap_atomic(kaddr, KM_USER0);
1227                 put_arg_page(page);
1228
1229                 if (offset == PAGE_SIZE)
1230                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1231         } while (offset == PAGE_SIZE);
1232
1233         bprm->p++;
1234         bprm->argc--;
1235         ret = 0;
1236
1237 out:
1238         return ret;
1239 }
1240 EXPORT_SYMBOL(remove_arg_zero);
1241
1242 /*
1243  * cycle the list of binary formats handler, until one recognizes the image
1244  */
1245 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1246 {
1247         unsigned int depth = bprm->recursion_depth;
1248         int try,retval;
1249         struct linux_binfmt *fmt;
1250
1251         retval = security_bprm_check(bprm);
1252         if (retval)
1253                 return retval;
1254
1255         /* kernel module loader fixup */
1256         /* so we don't try to load run modprobe in kernel space. */
1257         set_fs(USER_DS);
1258
1259         retval = audit_bprm(bprm);
1260         if (retval)
1261                 return retval;
1262
1263         retval = -ENOENT;
1264         for (try=0; try<2; try++) {
1265                 read_lock(&binfmt_lock);
1266                 list_for_each_entry(fmt, &formats, lh) {
1267                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1268                         if (!fn)
1269                                 continue;
1270                         if (!try_module_get(fmt->module))
1271                                 continue;
1272                         read_unlock(&binfmt_lock);
1273                         retval = fn(bprm, regs);
1274                         /*
1275                          * Restore the depth counter to its starting value
1276                          * in this call, so we don't have to rely on every
1277                          * load_binary function to restore it on return.
1278                          */
1279                         bprm->recursion_depth = depth;
1280                         if (retval >= 0) {
1281                                 if (depth == 0)
1282                                         tracehook_report_exec(fmt, bprm, regs);
1283                                 put_binfmt(fmt);
1284                                 allow_write_access(bprm->file);
1285                                 if (bprm->file)
1286                                         fput(bprm->file);
1287                                 bprm->file = NULL;
1288                                 current->did_exec = 1;
1289                                 proc_exec_connector(current);
1290                                 return retval;
1291                         }
1292                         read_lock(&binfmt_lock);
1293                         put_binfmt(fmt);
1294                         if (retval != -ENOEXEC || bprm->mm == NULL)
1295                                 break;
1296                         if (!bprm->file) {
1297                                 read_unlock(&binfmt_lock);
1298                                 return retval;
1299                         }
1300                 }
1301                 read_unlock(&binfmt_lock);
1302                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1303                         break;
1304 #ifdef CONFIG_MODULES
1305                 } else {
1306 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1307                         if (printable(bprm->buf[0]) &&
1308                             printable(bprm->buf[1]) &&
1309                             printable(bprm->buf[2]) &&
1310                             printable(bprm->buf[3]))
1311                                 break; /* -ENOEXEC */
1312                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1313 #endif
1314                 }
1315         }
1316         return retval;
1317 }
1318
1319 EXPORT_SYMBOL(search_binary_handler);
1320
1321 /*
1322  * sys_execve() executes a new program.
1323  */
1324 int do_execve(char * filename,
1325         char __user *__user *argv,
1326         char __user *__user *envp,
1327         struct pt_regs * regs)
1328 {
1329         struct linux_binprm *bprm;
1330         struct file *file;
1331         struct files_struct *displaced;
1332         bool clear_in_exec;
1333         int retval;
1334
1335         retval = unshare_files(&displaced);
1336         if (retval)
1337                 goto out_ret;
1338
1339         retval = -ENOMEM;
1340         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1341         if (!bprm)
1342                 goto out_files;
1343
1344         retval = prepare_bprm_creds(bprm);
1345         if (retval)
1346                 goto out_free;
1347
1348         retval = check_unsafe_exec(bprm);
1349         if (retval < 0)
1350                 goto out_free;
1351         clear_in_exec = retval;
1352         current->in_execve = 1;
1353
1354         file = open_exec(filename);
1355         retval = PTR_ERR(file);
1356         if (IS_ERR(file))
1357                 goto out_unmark;
1358
1359         sched_exec();
1360
1361         bprm->file = file;
1362         bprm->filename = filename;
1363         bprm->interp = filename;
1364
1365         retval = bprm_mm_init(bprm);
1366         if (retval)
1367                 goto out_file;
1368
1369         bprm->argc = count(argv, MAX_ARG_STRINGS);
1370         if ((retval = bprm->argc) < 0)
1371                 goto out;
1372
1373         bprm->envc = count(envp, MAX_ARG_STRINGS);
1374         if ((retval = bprm->envc) < 0)
1375                 goto out;
1376
1377         retval = prepare_binprm(bprm);
1378         if (retval < 0)
1379                 goto out;
1380
1381         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1382         if (retval < 0)
1383                 goto out;
1384
1385         bprm->exec = bprm->p;
1386         retval = copy_strings(bprm->envc, envp, bprm);
1387         if (retval < 0)
1388                 goto out;
1389
1390         retval = copy_strings(bprm->argc, argv, bprm);
1391         if (retval < 0)
1392                 goto out;
1393
1394         current->flags &= ~PF_KTHREAD;
1395         retval = search_binary_handler(bprm,regs);
1396         if (retval < 0)
1397                 goto out;
1398
1399         /* execve succeeded */
1400         current->fs->in_exec = 0;
1401         current->in_execve = 0;
1402         acct_update_integrals(current);
1403         free_bprm(bprm);
1404         if (displaced)
1405                 put_files_struct(displaced);
1406         return retval;
1407
1408 out:
1409         if (bprm->mm)
1410                 mmput (bprm->mm);
1411
1412 out_file:
1413         if (bprm->file) {
1414                 allow_write_access(bprm->file);
1415                 fput(bprm->file);
1416         }
1417
1418 out_unmark:
1419         if (clear_in_exec)
1420                 current->fs->in_exec = 0;
1421         current->in_execve = 0;
1422
1423 out_free:
1424         free_bprm(bprm);
1425
1426 out_files:
1427         if (displaced)
1428                 reset_files_struct(displaced);
1429 out_ret:
1430         return retval;
1431 }
1432
1433 void set_binfmt(struct linux_binfmt *new)
1434 {
1435         struct mm_struct *mm = current->mm;
1436
1437         if (mm->binfmt)
1438                 module_put(mm->binfmt->module);
1439
1440         mm->binfmt = new;
1441         if (new)
1442                 __module_get(new->module);
1443 }
1444
1445 EXPORT_SYMBOL(set_binfmt);
1446
1447 /* format_corename will inspect the pattern parameter, and output a
1448  * name into corename, which must have space for at least
1449  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1450  */
1451 static int format_corename(char *corename, long signr)
1452 {
1453         const struct cred *cred = current_cred();
1454         const char *pat_ptr = core_pattern;
1455         int ispipe = (*pat_ptr == '|');
1456         char *out_ptr = corename;
1457         char *const out_end = corename + CORENAME_MAX_SIZE;
1458         int rc;
1459         int pid_in_pattern = 0;
1460
1461         /* Repeat as long as we have more pattern to process and more output
1462            space */
1463         while (*pat_ptr) {
1464                 if (*pat_ptr != '%') {
1465                         if (out_ptr == out_end)
1466                                 goto out;
1467                         *out_ptr++ = *pat_ptr++;
1468                 } else {
1469                         switch (*++pat_ptr) {
1470                         case 0:
1471                                 goto out;
1472                         /* Double percent, output one percent */
1473                         case '%':
1474                                 if (out_ptr == out_end)
1475                                         goto out;
1476                                 *out_ptr++ = '%';
1477                                 break;
1478                         /* pid */
1479                         case 'p':
1480                                 pid_in_pattern = 1;
1481                                 rc = snprintf(out_ptr, out_end - out_ptr,
1482                                               "%d", task_tgid_vnr(current));
1483                                 if (rc > out_end - out_ptr)
1484                                         goto out;
1485                                 out_ptr += rc;
1486                                 break;
1487                         /* uid */
1488                         case 'u':
1489                                 rc = snprintf(out_ptr, out_end - out_ptr,
1490                                               "%d", cred->uid);
1491                                 if (rc > out_end - out_ptr)
1492                                         goto out;
1493                                 out_ptr += rc;
1494                                 break;
1495                         /* gid */
1496                         case 'g':
1497                                 rc = snprintf(out_ptr, out_end - out_ptr,
1498                                               "%d", cred->gid);
1499                                 if (rc > out_end - out_ptr)
1500                                         goto out;
1501                                 out_ptr += rc;
1502                                 break;
1503                         /* signal that caused the coredump */
1504                         case 's':
1505                                 rc = snprintf(out_ptr, out_end - out_ptr,
1506                                               "%ld", signr);
1507                                 if (rc > out_end - out_ptr)
1508                                         goto out;
1509                                 out_ptr += rc;
1510                                 break;
1511                         /* UNIX time of coredump */
1512                         case 't': {
1513                                 struct timeval tv;
1514                                 do_gettimeofday(&tv);
1515                                 rc = snprintf(out_ptr, out_end - out_ptr,
1516                                               "%lu", tv.tv_sec);
1517                                 if (rc > out_end - out_ptr)
1518                                         goto out;
1519                                 out_ptr += rc;
1520                                 break;
1521                         }
1522                         /* hostname */
1523                         case 'h':
1524                                 down_read(&uts_sem);
1525                                 rc = snprintf(out_ptr, out_end - out_ptr,
1526                                               "%s", utsname()->nodename);
1527                                 up_read(&uts_sem);
1528                                 if (rc > out_end - out_ptr)
1529                                         goto out;
1530                                 out_ptr += rc;
1531                                 break;
1532                         /* executable */
1533                         case 'e':
1534                                 rc = snprintf(out_ptr, out_end - out_ptr,
1535                                               "%s", current->comm);
1536                                 if (rc > out_end - out_ptr)
1537                                         goto out;
1538                                 out_ptr += rc;
1539                                 break;
1540                         /* core limit size */
1541                         case 'c':
1542                                 rc = snprintf(out_ptr, out_end - out_ptr,
1543                                               "%lu", rlimit(RLIMIT_CORE));
1544                                 if (rc > out_end - out_ptr)
1545                                         goto out;
1546                                 out_ptr += rc;
1547                                 break;
1548                         default:
1549                                 break;
1550                         }
1551                         ++pat_ptr;
1552                 }
1553         }
1554         /* Backward compatibility with core_uses_pid:
1555          *
1556          * If core_pattern does not include a %p (as is the default)
1557          * and core_uses_pid is set, then .%pid will be appended to
1558          * the filename. Do not do this for piped commands. */
1559         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1560                 rc = snprintf(out_ptr, out_end - out_ptr,
1561                               ".%d", task_tgid_vnr(current));
1562                 if (rc > out_end - out_ptr)
1563                         goto out;
1564                 out_ptr += rc;
1565         }
1566 out:
1567         *out_ptr = 0;
1568         return ispipe;
1569 }
1570
1571 static int zap_process(struct task_struct *start, int exit_code)
1572 {
1573         struct task_struct *t;
1574         int nr = 0;
1575
1576         start->signal->flags = SIGNAL_GROUP_EXIT;
1577         start->signal->group_exit_code = exit_code;
1578         start->signal->group_stop_count = 0;
1579
1580         t = start;
1581         do {
1582                 if (t != current && t->mm) {
1583                         sigaddset(&t->pending.signal, SIGKILL);
1584                         signal_wake_up(t, 1);
1585                         nr++;
1586                 }
1587         } while_each_thread(start, t);
1588
1589         return nr;
1590 }
1591
1592 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1593                                 struct core_state *core_state, int exit_code)
1594 {
1595         struct task_struct *g, *p;
1596         unsigned long flags;
1597         int nr = -EAGAIN;
1598
1599         spin_lock_irq(&tsk->sighand->siglock);
1600         if (!signal_group_exit(tsk->signal)) {
1601                 mm->core_state = core_state;
1602                 nr = zap_process(tsk, exit_code);
1603         }
1604         spin_unlock_irq(&tsk->sighand->siglock);
1605         if (unlikely(nr < 0))
1606                 return nr;
1607
1608         if (atomic_read(&mm->mm_users) == nr + 1)
1609                 goto done;
1610         /*
1611          * We should find and kill all tasks which use this mm, and we should
1612          * count them correctly into ->nr_threads. We don't take tasklist
1613          * lock, but this is safe wrt:
1614          *
1615          * fork:
1616          *      None of sub-threads can fork after zap_process(leader). All
1617          *      processes which were created before this point should be
1618          *      visible to zap_threads() because copy_process() adds the new
1619          *      process to the tail of init_task.tasks list, and lock/unlock
1620          *      of ->siglock provides a memory barrier.
1621          *
1622          * do_exit:
1623          *      The caller holds mm->mmap_sem. This means that the task which
1624          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1625          *      its ->mm.
1626          *
1627          * de_thread:
1628          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1629          *      we must see either old or new leader, this does not matter.
1630          *      However, it can change p->sighand, so lock_task_sighand(p)
1631          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1632          *      it can't fail.
1633          *
1634          *      Note also that "g" can be the old leader with ->mm == NULL
1635          *      and already unhashed and thus removed from ->thread_group.
1636          *      This is OK, __unhash_process()->list_del_rcu() does not
1637          *      clear the ->next pointer, we will find the new leader via
1638          *      next_thread().
1639          */
1640         rcu_read_lock();
1641         for_each_process(g) {
1642                 if (g == tsk->group_leader)
1643                         continue;
1644                 if (g->flags & PF_KTHREAD)
1645                         continue;
1646                 p = g;
1647                 do {
1648                         if (p->mm) {
1649                                 if (unlikely(p->mm == mm)) {
1650                                         lock_task_sighand(p, &flags);
1651                                         nr += zap_process(p, exit_code);
1652                                         unlock_task_sighand(p, &flags);
1653                                 }
1654                                 break;
1655                         }
1656                 } while_each_thread(g, p);
1657         }
1658         rcu_read_unlock();
1659 done:
1660         atomic_set(&core_state->nr_threads, nr);
1661         return nr;
1662 }
1663
1664 static int coredump_wait(int exit_code, struct core_state *core_state)
1665 {
1666         struct task_struct *tsk = current;
1667         struct mm_struct *mm = tsk->mm;
1668         struct completion *vfork_done;
1669         int core_waiters = -EBUSY;
1670
1671         init_completion(&core_state->startup);
1672         core_state->dumper.task = tsk;
1673         core_state->dumper.next = NULL;
1674
1675         down_write(&mm->mmap_sem);
1676         if (!mm->core_state)
1677                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1678         up_write(&mm->mmap_sem);
1679
1680         if (unlikely(core_waiters < 0))
1681                 goto fail;
1682
1683         /*
1684          * Make sure nobody is waiting for us to release the VM,
1685          * otherwise we can deadlock when we wait on each other
1686          */
1687         vfork_done = tsk->vfork_done;
1688         if (vfork_done) {
1689                 tsk->vfork_done = NULL;
1690                 complete(vfork_done);
1691         }
1692
1693         if (core_waiters)
1694                 wait_for_completion(&core_state->startup);
1695 fail:
1696         return core_waiters;
1697 }
1698
1699 static void coredump_finish(struct mm_struct *mm)
1700 {
1701         struct core_thread *curr, *next;
1702         struct task_struct *task;
1703
1704         next = mm->core_state->dumper.next;
1705         while ((curr = next) != NULL) {
1706                 next = curr->next;
1707                 task = curr->task;
1708                 /*
1709                  * see exit_mm(), curr->task must not see
1710                  * ->task == NULL before we read ->next.
1711                  */
1712                 smp_mb();
1713                 curr->task = NULL;
1714                 wake_up_process(task);
1715         }
1716
1717         mm->core_state = NULL;
1718 }
1719
1720 /*
1721  * set_dumpable converts traditional three-value dumpable to two flags and
1722  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1723  * these bits are not changed atomically.  So get_dumpable can observe the
1724  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1725  * return either old dumpable or new one by paying attention to the order of
1726  * modifying the bits.
1727  *
1728  * dumpable |   mm->flags (binary)
1729  * old  new | initial interim  final
1730  * ---------+-----------------------
1731  *  0    1  |   00      01      01
1732  *  0    2  |   00      10(*)   11
1733  *  1    0  |   01      00      00
1734  *  1    2  |   01      11      11
1735  *  2    0  |   11      10(*)   00
1736  *  2    1  |   11      11      01
1737  *
1738  * (*) get_dumpable regards interim value of 10 as 11.
1739  */
1740 void set_dumpable(struct mm_struct *mm, int value)
1741 {
1742         switch (value) {
1743         case 0:
1744                 clear_bit(MMF_DUMPABLE, &mm->flags);
1745                 smp_wmb();
1746                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1747                 break;
1748         case 1:
1749                 set_bit(MMF_DUMPABLE, &mm->flags);
1750                 smp_wmb();
1751                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1752                 break;
1753         case 2:
1754                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1755                 smp_wmb();
1756                 set_bit(MMF_DUMPABLE, &mm->flags);
1757                 break;
1758         }
1759 }
1760
1761 static int __get_dumpable(unsigned long mm_flags)
1762 {
1763         int ret;
1764
1765         ret = mm_flags & MMF_DUMPABLE_MASK;
1766         return (ret >= 2) ? 2 : ret;
1767 }
1768
1769 int get_dumpable(struct mm_struct *mm)
1770 {
1771         return __get_dumpable(mm->flags);
1772 }
1773
1774 static void wait_for_dump_helpers(struct file *file)
1775 {
1776         struct pipe_inode_info *pipe;
1777
1778         pipe = file->f_path.dentry->d_inode->i_pipe;
1779
1780         pipe_lock(pipe);
1781         pipe->readers++;
1782         pipe->writers--;
1783
1784         while ((pipe->readers > 1) && (!signal_pending(current))) {
1785                 wake_up_interruptible_sync(&pipe->wait);
1786                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1787                 pipe_wait(pipe);
1788         }
1789
1790         pipe->readers--;
1791         pipe->writers++;
1792         pipe_unlock(pipe);
1793
1794 }
1795
1796
1797 /*
1798  * uhm_pipe_setup
1799  * helper function to customize the process used
1800  * to collect the core in userspace.  Specifically
1801  * it sets up a pipe and installs it as fd 0 (stdin)
1802  * for the process.  Returns 0 on success, or
1803  * PTR_ERR on failure.
1804  * Note that it also sets the core limit to 1.  This
1805  * is a special value that we use to trap recursive
1806  * core dumps
1807  */
1808 static int umh_pipe_setup(struct subprocess_info *info)
1809 {
1810         struct file *rp, *wp;
1811         struct fdtable *fdt;
1812         struct coredump_params *cp = (struct coredump_params *)info->data;
1813         struct files_struct *cf = current->files;
1814
1815         wp = create_write_pipe(0);
1816         if (IS_ERR(wp))
1817                 return PTR_ERR(wp);
1818
1819         rp = create_read_pipe(wp, 0);
1820         if (IS_ERR(rp)) {
1821                 free_write_pipe(wp);
1822                 return PTR_ERR(rp);
1823         }
1824
1825         cp->file = wp;
1826
1827         sys_close(0);
1828         fd_install(0, rp);
1829         spin_lock(&cf->file_lock);
1830         fdt = files_fdtable(cf);
1831         FD_SET(0, fdt->open_fds);
1832         FD_CLR(0, fdt->close_on_exec);
1833         spin_unlock(&cf->file_lock);
1834
1835         /* and disallow core files too */
1836         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1837
1838         return 0;
1839 }
1840
1841 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1842 {
1843         struct core_state core_state;
1844         char corename[CORENAME_MAX_SIZE + 1];
1845         struct mm_struct *mm = current->mm;
1846         struct linux_binfmt * binfmt;
1847         const struct cred *old_cred;
1848         struct cred *cred;
1849         int retval = 0;
1850         int flag = 0;
1851         int ispipe;
1852         static atomic_t core_dump_count = ATOMIC_INIT(0);
1853         struct coredump_params cprm = {
1854                 .signr = signr,
1855                 .regs = regs,
1856                 .limit = rlimit(RLIMIT_CORE),
1857                 /*
1858                  * We must use the same mm->flags while dumping core to avoid
1859                  * inconsistency of bit flags, since this flag is not protected
1860                  * by any locks.
1861                  */
1862                 .mm_flags = mm->flags,
1863         };
1864
1865         audit_core_dumps(signr);
1866
1867         binfmt = mm->binfmt;
1868         if (!binfmt || !binfmt->core_dump)
1869                 goto fail;
1870         if (!__get_dumpable(cprm.mm_flags))
1871                 goto fail;
1872
1873         cred = prepare_creds();
1874         if (!cred)
1875                 goto fail;
1876         /*
1877          *      We cannot trust fsuid as being the "true" uid of the
1878          *      process nor do we know its entire history. We only know it
1879          *      was tainted so we dump it as root in mode 2.
1880          */
1881         if (__get_dumpable(cprm.mm_flags) == 2) {
1882                 /* Setuid core dump mode */
1883                 flag = O_EXCL;          /* Stop rewrite attacks */
1884                 cred->fsuid = 0;        /* Dump root private */
1885         }
1886
1887         retval = coredump_wait(exit_code, &core_state);
1888         if (retval < 0)
1889                 goto fail_creds;
1890
1891         old_cred = override_creds(cred);
1892
1893         /*
1894          * Clear any false indication of pending signals that might
1895          * be seen by the filesystem code called to write the core file.
1896          */
1897         clear_thread_flag(TIF_SIGPENDING);
1898
1899         /*
1900          * lock_kernel() because format_corename() is controlled by sysctl, which
1901          * uses lock_kernel()
1902          */
1903         lock_kernel();
1904         ispipe = format_corename(corename, signr);
1905         unlock_kernel();
1906
1907         if (ispipe) {
1908                 int dump_count;
1909                 char **helper_argv;
1910
1911                 if (cprm.limit == 1) {
1912                         /*
1913                          * Normally core limits are irrelevant to pipes, since
1914                          * we're not writing to the file system, but we use
1915                          * cprm.limit of 1 here as a speacial value. Any
1916                          * non-1 limit gets set to RLIM_INFINITY below, but
1917                          * a limit of 0 skips the dump.  This is a consistent
1918                          * way to catch recursive crashes.  We can still crash
1919                          * if the core_pattern binary sets RLIM_CORE =  !1
1920                          * but it runs as root, and can do lots of stupid things
1921                          * Note that we use task_tgid_vnr here to grab the pid
1922                          * of the process group leader.  That way we get the
1923                          * right pid if a thread in a multi-threaded
1924                          * core_pattern process dies.
1925                          */
1926                         printk(KERN_WARNING
1927                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
1928                                 task_tgid_vnr(current), current->comm);
1929                         printk(KERN_WARNING "Aborting core\n");
1930                         goto fail_unlock;
1931                 }
1932                 cprm.limit = RLIM_INFINITY;
1933
1934                 dump_count = atomic_inc_return(&core_dump_count);
1935                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1936                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1937                                task_tgid_vnr(current), current->comm);
1938                         printk(KERN_WARNING "Skipping core dump\n");
1939                         goto fail_dropcount;
1940                 }
1941
1942                 helper_argv = argv_split(GFP_KERNEL, corename+1, NULL);
1943                 if (!helper_argv) {
1944                         printk(KERN_WARNING "%s failed to allocate memory\n",
1945                                __func__);
1946                         goto fail_dropcount;
1947                 }
1948
1949                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
1950                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
1951                                         NULL, &cprm);
1952                 argv_free(helper_argv);
1953                 if (retval) {
1954                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1955                                corename);
1956                         goto close_fail;
1957                 }
1958         } else {
1959                 struct inode *inode;
1960
1961                 if (cprm.limit < binfmt->min_coredump)
1962                         goto fail_unlock;
1963
1964                 cprm.file = filp_open(corename,
1965                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1966                                  0600);
1967                 if (IS_ERR(cprm.file))
1968                         goto fail_unlock;
1969
1970                 inode = cprm.file->f_path.dentry->d_inode;
1971                 if (inode->i_nlink > 1)
1972                         goto close_fail;
1973                 if (d_unhashed(cprm.file->f_path.dentry))
1974                         goto close_fail;
1975                 /*
1976                  * AK: actually i see no reason to not allow this for named
1977                  * pipes etc, but keep the previous behaviour for now.
1978                  */
1979                 if (!S_ISREG(inode->i_mode))
1980                         goto close_fail;
1981                 /*
1982                  * Dont allow local users get cute and trick others to coredump
1983                  * into their pre-created files.
1984                  */
1985                 if (inode->i_uid != current_fsuid())
1986                         goto close_fail;
1987                 if (!cprm.file->f_op || !cprm.file->f_op->write)
1988                         goto close_fail;
1989                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
1990                         goto close_fail;
1991         }
1992
1993         retval = binfmt->core_dump(&cprm);
1994         if (retval)
1995                 current->signal->group_exit_code |= 0x80;
1996
1997         if (ispipe && core_pipe_limit)
1998                 wait_for_dump_helpers(cprm.file);
1999 close_fail:
2000         if (cprm.file)
2001                 filp_close(cprm.file, NULL);
2002 fail_dropcount:
2003         if (ispipe)
2004                 atomic_dec(&core_dump_count);
2005 fail_unlock:
2006         coredump_finish(mm);
2007         revert_creds(old_cred);
2008 fail_creds:
2009         put_cred(cred);
2010 fail:
2011         return;
2012 }