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/vmacache.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/sched/mm.h>
  36 #include <linux/sched/coredump.h>
  37 #include <linux/sched/signal.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/perf_event.h>
  40 #include <linux/highmem.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/key.h>
  43 #include <linux/personality.h>
  44 #include <linux/binfmts.h>
  45 #include <linux/utsname.h>
  46 #include <linux/pid_namespace.h>
  47 #include <linux/module.h>
  48 #include <linux/namei.h>
  49 #include <linux/mount.h>
  50 #include <linux/security.h>
  51 #include <linux/syscalls.h>
  52 #include <linux/tsacct_kern.h>
  53 #include <linux/cn_proc.h>
  54 #include <linux/audit.h>
  55 #include <linux/tracehook.h>
  56 #include <linux/kmod.h>
  57 #include <linux/fsnotify.h>
  58 #include <linux/fs_struct.h>
  59 #include <linux/pipe_fs_i.h>
  60 #include <linux/oom.h>
  61 #include <linux/compat.h>
  62 #include <linux/vmalloc.h>
  63
  64 #include <linux/uaccess.h>
  65 #include <asm/mmu_context.h>
  66 #include <asm/tlb.h>
  67
  68 #include <trace/events/task.h>
  69 #include "internal.h"
  70
  71 #include <trace/events/sched.h>
  72
  73 int suid_dumpable = 0;
  74
  75 static LIST_HEAD(formats);
  76 static DEFINE_RWLOCK(binfmt_lock);
  77
  78 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  79 {
  80         BUG_ON(!fmt);
  81         if (WARN_ON(!fmt->load_binary))
  82                 return;
  83         write_lock(&binfmt_lock);
  84         insert ? list_add(&fmt->lh, &formats) :
  85                  list_add_tail(&fmt->lh, &formats);
  86         write_unlock(&binfmt_lock);
  87 }
  88
  89 EXPORT_SYMBOL(__register_binfmt);
  90
  91 void unregister_binfmt(struct linux_binfmt * fmt)
  92 {
  93         write_lock(&binfmt_lock);
  94         list_del(&fmt->lh);
  95         write_unlock(&binfmt_lock);
  96 }
  97
  98 EXPORT_SYMBOL(unregister_binfmt);
  99
 100 static inline void put_binfmt(struct linux_binfmt * fmt)
 101 {
 102         module_put(fmt->module);
 103 }
 104
 105 bool path_noexec(const struct path *path)
 106 {
 107         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 108                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 109 }
 110
 111 #ifdef CONFIG_USELIB
 112 /*
 113  * Note that a shared library must be both readable and executable due to
 114  * security reasons.
 115  *
 116  * Also note that we take the address to load from from the file itself.
 117  */
 118 SYSCALL_DEFINE1(uselib, const char __user *, library)
 119 {
 120         struct linux_binfmt *fmt;
 121         struct file *file;
 122         struct filename *tmp = getname(library);
 123         int error = PTR_ERR(tmp);
 124         static const struct open_flags uselib_flags = {
 125                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 126                 .acc_mode = MAY_READ | MAY_EXEC,
 127                 .intent = LOOKUP_OPEN,
 128                 .lookup_flags = LOOKUP_FOLLOW,
 129         };
 130
 131         if (IS_ERR(tmp))
 132                 goto out;
 133
 134         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 135         putname(tmp);
 136         error = PTR_ERR(file);
 137         if (IS_ERR(file))
 138                 goto out;
 139
 140         error = -EINVAL;
 141         if (!S_ISREG(file_inode(file)->i_mode))
 142                 goto exit;
 143
 144         error = -EACCES;
 145         if (path_noexec(&file->f_path))
 146                 goto exit;
 147
 148         fsnotify_open(file);
 149
 150         error = -ENOEXEC;
 151
 152         read_lock(&binfmt_lock);
 153         list_for_each_entry(fmt, &formats, lh) {
 154                 if (!fmt->load_shlib)
 155                         continue;
 156                 if (!try_module_get(fmt->module))
 157                         continue;
 158                 read_unlock(&binfmt_lock);
 159                 error = fmt->load_shlib(file);
 160                 read_lock(&binfmt_lock);
 161                 put_binfmt(fmt);
 162                 if (error != -ENOEXEC)
 163                         break;
 164         }
 165         read_unlock(&binfmt_lock);
 166 exit:
 167         fput(file);
 168 out:
 169         return error;
 170 }
 171 #endif /* #ifdef CONFIG_USELIB */
 172
 173 #ifdef CONFIG_MMU
 174 /*
 175  * The nascent bprm->mm is not visible until exec_mmap() but it can
 176  * use a lot of memory, account these pages in current->mm temporary
 177  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 178  * change the counter back via acct_arg_size(0).
 179  */
 180 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 181 {
 182         struct mm_struct *mm = current->mm;
 183         long diff = (long)(pages - bprm->vma_pages);
 184
 185         if (!mm || !diff)
 186                 return;
 187
 188         bprm->vma_pages = pages;
 189         add_mm_counter(mm, MM_ANONPAGES, diff);
 190 }
 191
 192 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 193                 int write)
 194 {
 195         struct page *page;
 196         int ret;
 197         unsigned int gup_flags = FOLL_FORCE;
 198
 199 #ifdef CONFIG_STACK_GROWSUP
 200         if (write) {
 201                 ret = expand_downwards(bprm->vma, pos);
 202                 if (ret < 0)
 203                         return NULL;
 204         }
 205 #endif
 206
 207         if (write)
 208                 gup_flags |= FOLL_WRITE;
 209
 210         /*
 211          * We are doing an exec().  'current' is the process
 212          * doing the exec and bprm->mm is the new process's mm.
 213          */
 214         ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
 215                         &page, NULL, NULL);
 216         if (ret <= 0)
 217                 return NULL;
 218
 219         if (write) {
 220                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 221                 struct rlimit *rlim;
 222
 223                 acct_arg_size(bprm, size / PAGE_SIZE);
 224
 225                 /*
 226                  * We've historically supported up to 32 pages (ARG_MAX)
 227                  * of argument strings even with small stacks
 228                  */
 229                 if (size <= ARG_MAX)
 230                         return page;
 231
 232                 /*
 233                  * Limit to 1/4-th the stack size for the argv+env strings.
 234                  * This ensures that:
 235                  *  - the remaining binfmt code will not run out of stack space,
 236                  *  - the program will have a reasonable amount of stack left
 237                  *    to work from.
 238                  */
 239                 rlim = current->signal->rlim;
 240                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 241                         put_page(page);
 242                         return NULL;
 243                 }
 244         }
 245
 246         return page;
 247 }
 248
 249 static void put_arg_page(struct page *page)
 250 {
 251         put_page(page);
 252 }
 253
 254 static void free_arg_pages(struct linux_binprm *bprm)
 255 {
 256 }
 257
 258 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 259                 struct page *page)
 260 {
 261         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 262 }
 263
 264 static int __bprm_mm_init(struct linux_binprm *bprm)
 265 {
 266         int err;
 267         struct vm_area_struct *vma = NULL;
 268         struct mm_struct *mm = bprm->mm;
 269
 270         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 271         if (!vma)
 272                 return -ENOMEM;
 273
 274         if (down_write_killable(&mm->mmap_sem)) {
 275                 err = -EINTR;
 276                 goto err_free;
 277         }
 278         vma->vm_mm = mm;
 279
 280         /*
 281          * Place the stack at the largest stack address the architecture
 282          * supports. Later, we'll move this to an appropriate place. We don't
 283          * use STACK_TOP because that can depend on attributes which aren't
 284          * configured yet.
 285          */
 286         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 287         vma->vm_end = STACK_TOP_MAX;
 288         vma->vm_start = vma->vm_end - PAGE_SIZE;
 289         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 290         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 291         INIT_LIST_HEAD(&vma->anon_vma_chain);
 292
 293         err = insert_vm_struct(mm, vma);
 294         if (err)
 295                 goto err;
 296
 297         mm->stack_vm = mm->total_vm = 1;
 298         arch_bprm_mm_init(mm, vma);
 299         up_write(&mm->mmap_sem);
 300         bprm->p = vma->vm_end - sizeof(void *);
 301         return 0;
 302 err:
 303         up_write(&mm->mmap_sem);
 304 err_free:
 305         bprm->vma = NULL;
 306         kmem_cache_free(vm_area_cachep, vma);
 307         return err;
 308 }
 309
 310 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 311 {
 312         return len <= MAX_ARG_STRLEN;
 313 }
 314
 315 #else
 316
 317 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 318 {
 319 }
 320
 321 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 322                 int write)
 323 {
 324         struct page *page;
 325
 326         page = bprm->page[pos / PAGE_SIZE];
 327         if (!page && write) {
 328                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 329                 if (!page)
 330                         return NULL;
 331                 bprm->page[pos / PAGE_SIZE] = page;
 332         }
 333
 334         return page;
 335 }
 336
 337 static void put_arg_page(struct page *page)
 338 {
 339 }
 340
 341 static void free_arg_page(struct linux_binprm *bprm, int i)
 342 {
 343         if (bprm->page[i]) {
 344                 __free_page(bprm->page[i]);
 345                 bprm->page[i] = NULL;
 346         }
 347 }
 348
 349 static void free_arg_pages(struct linux_binprm *bprm)
 350 {
 351         int i;
 352
 353         for (i = 0; i < MAX_ARG_PAGES; i++)
 354                 free_arg_page(bprm, i);
 355 }
 356
 357 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 358                 struct page *page)
 359 {
 360 }
 361
 362 static int __bprm_mm_init(struct linux_binprm *bprm)
 363 {
 364         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 365         return 0;
 366 }
 367
 368 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 369 {
 370         return len <= bprm->p;
 371 }
 372
 373 #endif /* CONFIG_MMU */
 374
 375 /*
 376  * Create a new mm_struct and populate it with a temporary stack
 377  * vm_area_struct.  We don't have enough context at this point to set the stack
 378  * flags, permissions, and offset, so we use temporary values.  We'll update
 379  * them later in setup_arg_pages().
 380  */
 381 static int bprm_mm_init(struct linux_binprm *bprm)
 382 {
 383         int err;
 384         struct mm_struct *mm = NULL;
 385
 386         bprm->mm = mm = mm_alloc();
 387         err = -ENOMEM;
 388         if (!mm)
 389                 goto err;
 390
 391         err = __bprm_mm_init(bprm);
 392         if (err)
 393                 goto err;
 394
 395         return 0;
 396
 397 err:
 398         if (mm) {
 399                 bprm->mm = NULL;
 400                 mmdrop(mm);
 401         }
 402
 403         return err;
 404 }
 405
 406 struct user_arg_ptr {
 407 #ifdef CONFIG_COMPAT
 408         bool is_compat;
 409 #endif
 410         union {
 411                 const char __user *const __user *native;
 412 #ifdef CONFIG_COMPAT
 413                 const compat_uptr_t __user *compat;
 414 #endif
 415         } ptr;
 416 };
 417
 418 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 419 {
 420         const char __user *native;
 421
 422 #ifdef CONFIG_COMPAT
 423         if (unlikely(argv.is_compat)) {
 424                 compat_uptr_t compat;
 425
 426                 if (get_user(compat, argv.ptr.compat + nr))
 427                         return ERR_PTR(-EFAULT);
 428
 429                 return compat_ptr(compat);
 430         }
 431 #endif
 432
 433         if (get_user(native, argv.ptr.native + nr))
 434                 return ERR_PTR(-EFAULT);
 435
 436         return native;
 437 }
 438
 439 /*
 440  * count() counts the number of strings in array ARGV.
 441  */
 442 static int count(struct user_arg_ptr argv, int max)
 443 {
 444         int i = 0;
 445
 446         if (argv.ptr.native != NULL) {
 447                 for (;;) {
 448                         const char __user *p = get_user_arg_ptr(argv, i);
 449
 450                         if (!p)
 451                                 break;
 452
 453                         if (IS_ERR(p))
 454                                 return -EFAULT;
 455
 456                         if (i >= max)
 457                                 return -E2BIG;
 458                         ++i;
 459
 460                         if (fatal_signal_pending(current))
 461                                 return -ERESTARTNOHAND;
 462                         cond_resched();
 463                 }
 464         }
 465         return i;
 466 }
 467
 468 /*
 469  * 'copy_strings()' copies argument/environment strings from the old
 470  * processes's memory to the new process's stack.  The call to get_user_pages()
 471  * ensures the destination page is created and not swapped out.
 472  */
 473 static int copy_strings(int argc, struct user_arg_ptr argv,
 474                         struct linux_binprm *bprm)
 475 {
 476         struct page *kmapped_page = NULL;
 477         char *kaddr = NULL;
 478         unsigned long kpos = 0;
 479         int ret;
 480
 481         while (argc-- > 0) {
 482                 const char __user *str;
 483                 int len;
 484                 unsigned long pos;
 485
 486                 ret = -EFAULT;
 487                 str = get_user_arg_ptr(argv, argc);
 488                 if (IS_ERR(str))
 489                         goto out;
 490
 491                 len = strnlen_user(str, MAX_ARG_STRLEN);
 492                 if (!len)
 493                         goto out;
 494
 495                 ret = -E2BIG;
 496                 if (!valid_arg_len(bprm, len))
 497                         goto out;
 498
 499                 /* We're going to work our way backwords. */
 500                 pos = bprm->p;
 501                 str += len;
 502                 bprm->p -= len;
 503
 504                 while (len > 0) {
 505                         int offset, bytes_to_copy;
 506
 507                         if (fatal_signal_pending(current)) {
 508                                 ret = -ERESTARTNOHAND;
 509                                 goto out;
 510                         }
 511                         cond_resched();
 512
 513                         offset = pos % PAGE_SIZE;
 514                         if (offset == 0)
 515                                 offset = PAGE_SIZE;
 516
 517                         bytes_to_copy = offset;
 518                         if (bytes_to_copy > len)
 519                                 bytes_to_copy = len;
 520
 521                         offset -= bytes_to_copy;
 522                         pos -= bytes_to_copy;
 523                         str -= bytes_to_copy;
 524                         len -= bytes_to_copy;
 525
 526                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 527                                 struct page *page;
 528
 529                                 page = get_arg_page(bprm, pos, 1);
 530                                 if (!page) {
 531                                         ret = -E2BIG;
 532                                         goto out;
 533                                 }
 534
 535                                 if (kmapped_page) {
 536                                         flush_kernel_dcache_page(kmapped_page);
 537                                         kunmap(kmapped_page);
 538                                         put_arg_page(kmapped_page);
 539                                 }
 540                                 kmapped_page = page;
 541                                 kaddr = kmap(kmapped_page);
 542                                 kpos = pos & PAGE_MASK;
 543                                 flush_arg_page(bprm, kpos, kmapped_page);
 544                         }
 545                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 546                                 ret = -EFAULT;
 547                                 goto out;
 548                         }
 549                 }
 550         }
 551         ret = 0;
 552 out:
 553         if (kmapped_page) {
 554                 flush_kernel_dcache_page(kmapped_page);
 555                 kunmap(kmapped_page);
 556                 put_arg_page(kmapped_page);
 557         }
 558         return ret;
 559 }
 560
 561 /*
 562  * Like copy_strings, but get argv and its values from kernel memory.
 563  */
 564 int copy_strings_kernel(int argc, const char *const *__argv,
 565                         struct linux_binprm *bprm)
 566 {
 567         int r;
 568         mm_segment_t oldfs = get_fs();
 569         struct user_arg_ptr argv = {
 570                 .ptr.native = (const char __user *const  __user *)__argv,
 571         };
 572
 573         set_fs(KERNEL_DS);
 574         r = copy_strings(argc, argv, bprm);
 575         set_fs(oldfs);
 576
 577         return r;
 578 }
 579 EXPORT_SYMBOL(copy_strings_kernel);
 580
 581 #ifdef CONFIG_MMU
 582
 583 /*
 584  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 585  * the binfmt code determines where the new stack should reside, we shift it to
 586  * its final location.  The process proceeds as follows:
 587  *
 588  * 1) Use shift to calculate the new vma endpoints.
 589  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 590  *    arguments passed to subsequent functions are consistent.
 591  * 3) Move vma's page tables to the new range.
 592  * 4) Free up any cleared pgd range.
 593  * 5) Shrink the vma to cover only the new range.
 594  */
 595 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 596 {
 597         struct mm_struct *mm = vma->vm_mm;
 598         unsigned long old_start = vma->vm_start;
 599         unsigned long old_end = vma->vm_end;
 600         unsigned long length = old_end - old_start;
 601         unsigned long new_start = old_start - shift;
 602         unsigned long new_end = old_end - shift;
 603         struct mmu_gather tlb;
 604
 605         BUG_ON(new_start > new_end);
 606
 607         /*
 608          * ensure there are no vmas between where we want to go
 609          * and where we are
 610          */
 611         if (vma != find_vma(mm, new_start))
 612                 return -EFAULT;
 613
 614         /*
 615          * cover the whole range: [new_start, old_end)
 616          */
 617         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 618                 return -ENOMEM;
 619
 620         /*
 621          * move the page tables downwards, on failure we rely on
 622          * process cleanup to remove whatever mess we made.
 623          */
 624         if (length != move_page_tables(vma, old_start,
 625                                        vma, new_start, length, false))
 626                 return -ENOMEM;
 627
 628         lru_add_drain();
 629         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 630         if (new_end > old_start) {
 631                 /*
 632                  * when the old and new regions overlap clear from new_end.
 633                  */
 634                 free_pgd_range(&tlb, new_end, old_end, new_end,
 635                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 636         } else {
 637                 /*
 638                  * otherwise, clean from old_start; this is done to not touch
 639                  * the address space in [new_end, old_start) some architectures
 640                  * have constraints on va-space that make this illegal (IA64) -
 641                  * for the others its just a little faster.
 642                  */
 643                 free_pgd_range(&tlb, old_start, old_end, new_end,
 644                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 645         }
 646         tlb_finish_mmu(&tlb, old_start, old_end);
 647
 648         /*
 649          * Shrink the vma to just the new range.  Always succeeds.
 650          */
 651         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 652
 653         return 0;
 654 }
 655
 656 /*
 657  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 658  * the stack is optionally relocated, and some extra space is added.
 659  */
 660 int setup_arg_pages(struct linux_binprm *bprm,
 661                     unsigned long stack_top,
 662                     int executable_stack)
 663 {
 664         unsigned long ret;
 665         unsigned long stack_shift;
 666         struct mm_struct *mm = current->mm;
 667         struct vm_area_struct *vma = bprm->vma;
 668         struct vm_area_struct *prev = NULL;
 669         unsigned long vm_flags;
 670         unsigned long stack_base;
 671         unsigned long stack_size;
 672         unsigned long stack_expand;
 673         unsigned long rlim_stack;
 674
 675 #ifdef CONFIG_STACK_GROWSUP
 676         /* Limit stack size */
 677         stack_base = rlimit_max(RLIMIT_STACK);
 678         if (stack_base > STACK_SIZE_MAX)
 679                 stack_base = STACK_SIZE_MAX;
 680
 681         /* Add space for stack randomization. */
 682         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 683
 684         /* Make sure we didn't let the argument array grow too large. */
 685         if (vma->vm_end - vma->vm_start > stack_base)
 686                 return -ENOMEM;
 687
 688         stack_base = PAGE_ALIGN(stack_top - stack_base);
 689
 690         stack_shift = vma->vm_start - stack_base;
 691         mm->arg_start = bprm->p - stack_shift;
 692         bprm->p = vma->vm_end - stack_shift;
 693 #else
 694         stack_top = arch_align_stack(stack_top);
 695         stack_top = PAGE_ALIGN(stack_top);
 696
 697         if (unlikely(stack_top < mmap_min_addr) ||
 698             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 699                 return -ENOMEM;
 700
 701         stack_shift = vma->vm_end - stack_top;
 702
 703         bprm->p -= stack_shift;
 704         mm->arg_start = bprm->p;
 705 #endif
 706
 707         if (bprm->loader)
 708                 bprm->loader -= stack_shift;
 709         bprm->exec -= stack_shift;
 710
 711         if (down_write_killable(&mm->mmap_sem))
 712                 return -EINTR;
 713
 714         vm_flags = VM_STACK_FLAGS;
 715
 716         /*
 717          * Adjust stack execute permissions; explicitly enable for
 718          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 719          * (arch default) otherwise.
 720          */
 721         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 722                 vm_flags |= VM_EXEC;
 723         else if (executable_stack == EXSTACK_DISABLE_X)
 724                 vm_flags &= ~VM_EXEC;
 725         vm_flags |= mm->def_flags;
 726         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 727
 728         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 729                         vm_flags);
 730         if (ret)
 731                 goto out_unlock;
 732         BUG_ON(prev != vma);
 733
 734         /* Move stack pages down in memory. */
 735         if (stack_shift) {
 736                 ret = shift_arg_pages(vma, stack_shift);
 737                 if (ret)
 738                         goto out_unlock;
 739         }
 740
 741         /* mprotect_fixup is overkill to remove the temporary stack flags */
 742         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 743
 744         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 745         stack_size = vma->vm_end - vma->vm_start;
 746         /*
 747          * Align this down to a page boundary as expand_stack
 748          * will align it up.
 749          */
 750         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 751 #ifdef CONFIG_STACK_GROWSUP
 752         if (stack_size + stack_expand > rlim_stack)
 753                 stack_base = vma->vm_start + rlim_stack;
 754         else
 755                 stack_base = vma->vm_end + stack_expand;
 756 #else
 757         if (stack_size + stack_expand > rlim_stack)
 758                 stack_base = vma->vm_end - rlim_stack;
 759         else
 760                 stack_base = vma->vm_start - stack_expand;
 761 #endif
 762         current->mm->start_stack = bprm->p;
 763         ret = expand_stack(vma, stack_base);
 764         if (ret)
 765                 ret = -EFAULT;
 766
 767 out_unlock:
 768         up_write(&mm->mmap_sem);
 769         return ret;
 770 }
 771 EXPORT_SYMBOL(setup_arg_pages);
 772
 773 #else
 774
 775 /*
 776  * Transfer the program arguments and environment from the holding pages
 777  * onto the stack. The provided stack pointer is adjusted accordingly.
 778  */
 779 int transfer_args_to_stack(struct linux_binprm *bprm,
 780                            unsigned long *sp_location)
 781 {
 782         unsigned long index, stop, sp;
 783         int ret = 0;
 784
 785         stop = bprm->p >> PAGE_SHIFT;
 786         sp = *sp_location;
 787
 788         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 789                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 790                 char *src = kmap(bprm->page[index]) + offset;
 791                 sp -= PAGE_SIZE - offset;
 792                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 793                         ret = -EFAULT;
 794                 kunmap(bprm->page[index]);
 795                 if (ret)
 796                         goto out;
 797         }
 798
 799         *sp_location = sp;
 800
 801 out:
 802         return ret;
 803 }
 804 EXPORT_SYMBOL(transfer_args_to_stack);
 805
 806 #endif /* CONFIG_MMU */
 807
 808 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 809 {
 810         struct file *file;
 811         int err;
 812         struct open_flags open_exec_flags = {
 813                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 814                 .acc_mode = MAY_EXEC,
 815                 .intent = LOOKUP_OPEN,
 816                 .lookup_flags = LOOKUP_FOLLOW,
 817         };
 818
 819         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 820                 return ERR_PTR(-EINVAL);
 821         if (flags & AT_SYMLINK_NOFOLLOW)
 822                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 823         if (flags & AT_EMPTY_PATH)
 824                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 825
 826         file = do_filp_open(fd, name, &open_exec_flags);
 827         if (IS_ERR(file))
 828                 goto out;
 829
 830         err = -EACCES;
 831         if (!S_ISREG(file_inode(file)->i_mode))
 832                 goto exit;
 833
 834         if (path_noexec(&file->f_path))
 835                 goto exit;
 836
 837         err = deny_write_access(file);
 838         if (err)
 839                 goto exit;
 840
 841         if (name->name[0] != '\0')
 842                 fsnotify_open(file);
 843
 844 out:
 845         return file;
 846
 847 exit:
 848         fput(file);
 849         return ERR_PTR(err);
 850 }
 851
 852 struct file *open_exec(const char *name)
 853 {
 854         struct filename *filename = getname_kernel(name);
 855         struct file *f = ERR_CAST(filename);
 856
 857         if (!IS_ERR(filename)) {
 858                 f = do_open_execat(AT_FDCWD, filename, 0);
 859                 putname(filename);
 860         }
 861         return f;
 862 }
 863 EXPORT_SYMBOL(open_exec);
 864
 865 int kernel_read(struct file *file, loff_t offset,
 866                 char *addr, unsigned long count)
 867 {
 868         mm_segment_t old_fs;
 869         loff_t pos = offset;
 870         int result;
 871
 872         old_fs = get_fs();
 873         set_fs(get_ds());
 874         /* The cast to a user pointer is valid due to the set_fs() */
 875         result = vfs_read(file, (void __user *)addr, count, &pos);
 876         set_fs(old_fs);
 877         return result;
 878 }
 879
 880 EXPORT_SYMBOL(kernel_read);
 881
 882 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 883                      loff_t max_size, enum kernel_read_file_id id)
 884 {
 885         loff_t i_size, pos;
 886         ssize_t bytes = 0;
 887         int ret;
 888
 889         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 890                 return -EINVAL;
 891
 892         ret = security_kernel_read_file(file, id);
 893         if (ret)
 894                 return ret;
 895
 896         ret = deny_write_access(file);
 897         if (ret)
 898                 return ret;
 899
 900         i_size = i_size_read(file_inode(file));
 901         if (max_size > 0 && i_size > max_size) {
 902                 ret = -EFBIG;
 903                 goto out;
 904         }
 905         if (i_size <= 0) {
 906                 ret = -EINVAL;
 907                 goto out;
 908         }
 909
 910         if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 911                 *buf = vmalloc(i_size);
 912         if (!*buf) {
 913                 ret = -ENOMEM;
 914                 goto out;
 915         }
 916
 917         pos = 0;
 918         while (pos < i_size) {
 919                 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
 920                                     i_size - pos);
 921                 if (bytes < 0) {
 922                         ret = bytes;
 923                         goto out;
 924                 }
 925
 926                 if (bytes == 0)
 927                         break;
 928                 pos += bytes;
 929         }
 930
 931         if (pos != i_size) {
 932                 ret = -EIO;
 933                 goto out_free;
 934         }
 935
 936         ret = security_kernel_post_read_file(file, *buf, i_size, id);
 937         if (!ret)
 938                 *size = pos;
 939
 940 out_free:
 941         if (ret < 0) {
 942                 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 943                         vfree(*buf);
 944                         *buf = NULL;
 945                 }
 946         }
 947
 948 out:
 949         allow_write_access(file);
 950         return ret;
 951 }
 952 EXPORT_SYMBOL_GPL(kernel_read_file);
 953
 954 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
 955                                loff_t max_size, enum kernel_read_file_id id)
 956 {
 957         struct file *file;
 958         int ret;
 959
 960         if (!path || !*path)
 961                 return -EINVAL;
 962
 963         file = filp_open(path, O_RDONLY, 0);
 964         if (IS_ERR(file))
 965                 return PTR_ERR(file);
 966
 967         ret = kernel_read_file(file, buf, size, max_size, id);
 968         fput(file);
 969         return ret;
 970 }
 971 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 972
 973 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 974                              enum kernel_read_file_id id)
 975 {
 976         struct fd f = fdget(fd);
 977         int ret = -EBADF;
 978
 979         if (!f.file)
 980                 goto out;
 981
 982         ret = kernel_read_file(f.file, buf, size, max_size, id);
 983 out:
 984         fdput(f);
 985         return ret;
 986 }
 987 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
 988
 989 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 990 {
 991         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 992         if (res > 0)
 993                 flush_icache_range(addr, addr + len);
 994         return res;
 995 }
 996 EXPORT_SYMBOL(read_code);
 997
 998 static int exec_mmap(struct mm_struct *mm)
 999 {
1000         struct task_struct *tsk;
1001         struct mm_struct *old_mm, *active_mm;
1002
1003         /* Notify parent that we're no longer interested in the old VM */
1004         tsk = current;
1005         old_mm = current->mm;
1006         mm_release(tsk, old_mm);
1007
1008         if (old_mm) {
1009                 sync_mm_rss(old_mm);
1010                 /*
1011                  * Make sure that if there is a core dump in progress
1012                  * for the old mm, we get out and die instead of going
1013                  * through with the exec.  We must hold mmap_sem around
1014                  * checking core_state and changing tsk->mm.
1015                  */
1016                 down_read(&old_mm->mmap_sem);
1017                 if (unlikely(old_mm->core_state)) {
1018                         up_read(&old_mm->mmap_sem);
1019                         return -EINTR;
1020                 }
1021         }
1022         task_lock(tsk);
1023         active_mm = tsk->active_mm;
1024         tsk->mm = mm;
1025         tsk->active_mm = mm;
1026         activate_mm(active_mm, mm);
1027         tsk->mm->vmacache_seqnum = 0;
1028         vmacache_flush(tsk);
1029         task_unlock(tsk);
1030         if (old_mm) {
1031                 up_read(&old_mm->mmap_sem);
1032                 BUG_ON(active_mm != old_mm);
1033                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1034                 mm_update_next_owner(old_mm);
1035                 mmput(old_mm);
1036                 return 0;
1037         }
1038         mmdrop(active_mm);
1039         return 0;
1040 }
1041
1042 /*
1043  * This function makes sure the current process has its own signal table,
1044  * so that flush_signal_handlers can later reset the handlers without
1045  * disturbing other processes.  (Other processes might share the signal
1046  * table via the CLONE_SIGHAND option to clone().)
1047  */
1048 static int de_thread(struct task_struct *tsk)
1049 {
1050         struct signal_struct *sig = tsk->signal;
1051         struct sighand_struct *oldsighand = tsk->sighand;
1052         spinlock_t *lock = &oldsighand->siglock;
1053
1054         if (thread_group_empty(tsk))
1055                 goto no_thread_group;
1056
1057         /*
1058          * Kill all other threads in the thread group.
1059          */
1060         spin_lock_irq(lock);
1061         if (signal_group_exit(sig)) {
1062                 /*
1063                  * Another group action in progress, just
1064                  * return so that the signal is processed.
1065                  */
1066                 spin_unlock_irq(lock);
1067                 return -EAGAIN;
1068         }
1069
1070         sig->group_exit_task = tsk;
1071         sig->notify_count = zap_other_threads(tsk);
1072         if (!thread_group_leader(tsk))
1073                 sig->notify_count--;
1074
1075         while (sig->notify_count) {
1076                 __set_current_state(TASK_KILLABLE);
1077                 spin_unlock_irq(lock);
1078                 schedule();
1079                 if (unlikely(__fatal_signal_pending(tsk)))
1080                         goto killed;
1081                 spin_lock_irq(lock);
1082         }
1083         spin_unlock_irq(lock);
1084
1085         /*
1086          * At this point all other threads have exited, all we have to
1087          * do is to wait for the thread group leader to become inactive,
1088          * and to assume its PID:
1089          */
1090         if (!thread_group_leader(tsk)) {
1091                 struct task_struct *leader = tsk->group_leader;
1092
1093                 for (;;) {
1094                         cgroup_threadgroup_change_begin(tsk);
1095                         write_lock_irq(&tasklist_lock);
1096                         /*
1097                          * Do this under tasklist_lock to ensure that
1098                          * exit_notify() can't miss ->group_exit_task
1099                          */
1100                         sig->notify_count = -1;
1101                         if (likely(leader->exit_state))
1102                                 break;
1103                         __set_current_state(TASK_KILLABLE);
1104                         write_unlock_irq(&tasklist_lock);
1105                         cgroup_threadgroup_change_end(tsk);
1106                         schedule();
1107                         if (unlikely(__fatal_signal_pending(tsk)))
1108                                 goto killed;
1109                 }
1110
1111                 /*
1112                  * The only record we have of the real-time age of a
1113                  * process, regardless of execs it's done, is start_time.
1114                  * All the past CPU time is accumulated in signal_struct
1115                  * from sister threads now dead.  But in this non-leader
1116                  * exec, nothing survives from the original leader thread,
1117                  * whose birth marks the true age of this process now.
1118                  * When we take on its identity by switching to its PID, we
1119                  * also take its birthdate (always earlier than our own).
1120                  */
1121                 tsk->start_time = leader->start_time;
1122                 tsk->real_start_time = leader->real_start_time;
1123
1124                 BUG_ON(!same_thread_group(leader, tsk));
1125                 BUG_ON(has_group_leader_pid(tsk));
1126                 /*
1127                  * An exec() starts a new thread group with the
1128                  * TGID of the previous thread group. Rehash the
1129                  * two threads with a switched PID, and release
1130                  * the former thread group leader:
1131                  */
1132
1133                 /* Become a process group leader with the old leader's pid.
1134                  * The old leader becomes a thread of the this thread group.
1135                  * Note: The old leader also uses this pid until release_task
1136                  *       is called.  Odd but simple and correct.
1137                  */
1138                 tsk->pid = leader->pid;
1139                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1140                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1141                 transfer_pid(leader, tsk, PIDTYPE_SID);
1142
1143                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1144                 list_replace_init(&leader->sibling, &tsk->sibling);
1145
1146                 tsk->group_leader = tsk;
1147                 leader->group_leader = tsk;
1148
1149                 tsk->exit_signal = SIGCHLD;
1150                 leader->exit_signal = -1;
1151
1152                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1153                 leader->exit_state = EXIT_DEAD;
1154
1155                 /*
1156                  * We are going to release_task()->ptrace_unlink() silently,
1157                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1158                  * the tracer wont't block again waiting for this thread.
1159                  */
1160                 if (unlikely(leader->ptrace))
1161                         __wake_up_parent(leader, leader->parent);
1162                 write_unlock_irq(&tasklist_lock);
1163                 cgroup_threadgroup_change_end(tsk);
1164
1165                 release_task(leader);
1166         }
1167
1168         sig->group_exit_task = NULL;
1169         sig->notify_count = 0;
1170
1171 no_thread_group:
1172         /* we have changed execution domain */
1173         tsk->exit_signal = SIGCHLD;
1174
1175 #ifdef CONFIG_POSIX_TIMERS
1176         exit_itimers(sig);
1177         flush_itimer_signals();
1178 #endif
1179
1180         if (atomic_read(&oldsighand->count) != 1) {
1181                 struct sighand_struct *newsighand;
1182                 /*
1183                  * This ->sighand is shared with the CLONE_SIGHAND
1184                  * but not CLONE_THREAD task, switch to the new one.
1185                  */
1186                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1187                 if (!newsighand)
1188                         return -ENOMEM;
1189
1190                 atomic_set(&newsighand->count, 1);
1191                 memcpy(newsighand->action, oldsighand->action,
1192                        sizeof(newsighand->action));
1193
1194                 write_lock_irq(&tasklist_lock);
1195                 spin_lock(&oldsighand->siglock);
1196                 rcu_assign_pointer(tsk->sighand, newsighand);
1197                 spin_unlock(&oldsighand->siglock);
1198                 write_unlock_irq(&tasklist_lock);
1199
1200                 __cleanup_sighand(oldsighand);
1201         }
1202
1203         BUG_ON(!thread_group_leader(tsk));
1204         return 0;
1205
1206 killed:
1207         /* protects against exit_notify() and __exit_signal() */
1208         read_lock(&tasklist_lock);
1209         sig->group_exit_task = NULL;
1210         sig->notify_count = 0;
1211         read_unlock(&tasklist_lock);
1212         return -EAGAIN;
1213 }
1214
1215 char *get_task_comm(char *buf, struct task_struct *tsk)
1216 {
1217         /* buf must be at least sizeof(tsk->comm) in size */
1218         task_lock(tsk);
1219         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1220         task_unlock(tsk);
1221         return buf;
1222 }
1223 EXPORT_SYMBOL_GPL(get_task_comm);
1224
1225 /*
1226  * These functions flushes out all traces of the currently running executable
1227  * so that a new one can be started
1228  */
1229
1230 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1231 {
1232         task_lock(tsk);
1233         trace_task_rename(tsk, buf);
1234         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1235         task_unlock(tsk);
1236         perf_event_comm(tsk, exec);
1237 }
1238
1239 int flush_old_exec(struct linux_binprm * bprm)
1240 {
1241         int retval;
1242
1243         /*
1244          * Make sure we have a private signal table and that
1245          * we are unassociated from the previous thread group.
1246          */
1247         retval = de_thread(current);
1248         if (retval)
1249                 goto out;
1250
1251         /*
1252          * Must be called _before_ exec_mmap() as bprm->mm is
1253          * not visibile until then. This also enables the update
1254          * to be lockless.
1255          */
1256         set_mm_exe_file(bprm->mm, bprm->file);
1257
1258         /*
1259          * Release all of the old mmap stuff
1260          */
1261         acct_arg_size(bprm, 0);
1262         retval = exec_mmap(bprm->mm);
1263         if (retval)
1264                 goto out;
1265
1266         bprm->mm = NULL;                /* We're using it now */
1267
1268         set_fs(USER_DS);
1269         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1270                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1271         flush_thread();
1272         current->personality &= ~bprm->per_clear;
1273
1274         /*
1275          * We have to apply CLOEXEC before we change whether the process is
1276          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1277          * trying to access the should-be-closed file descriptors of a process
1278          * undergoing exec(2).
1279          */
1280         do_close_on_exec(current->files);
1281         return 0;
1282
1283 out:
1284         return retval;
1285 }
1286 EXPORT_SYMBOL(flush_old_exec);
1287
1288 void would_dump(struct linux_binprm *bprm, struct file *file)
1289 {
1290         struct inode *inode = file_inode(file);
1291         if (inode_permission(inode, MAY_READ) < 0) {
1292                 struct user_namespace *old, *user_ns;
1293                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1294
1295                 /* Ensure mm->user_ns contains the executable */
1296                 user_ns = old = bprm->mm->user_ns;
1297                 while ((user_ns != &init_user_ns) &&
1298                        !privileged_wrt_inode_uidgid(user_ns, inode))
1299                         user_ns = user_ns->parent;
1300
1301                 if (old != user_ns) {
1302                         bprm->mm->user_ns = get_user_ns(user_ns);
1303                         put_user_ns(old);
1304                 }
1305         }
1306 }
1307 EXPORT_SYMBOL(would_dump);
1308
1309 void setup_new_exec(struct linux_binprm * bprm)
1310 {
1311         arch_pick_mmap_layout(current->mm);
1312
1313         /* This is the point of no return */
1314         current->sas_ss_sp = current->sas_ss_size = 0;
1315
1316         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1317                 set_dumpable(current->mm, SUID_DUMP_USER);
1318         else
1319                 set_dumpable(current->mm, suid_dumpable);
1320
1321         perf_event_exec();
1322         __set_task_comm(current, kbasename(bprm->filename), true);
1323
1324         /* Set the new mm task size. We have to do that late because it may
1325          * depend on TIF_32BIT which is only updated in flush_thread() on
1326          * some architectures like powerpc
1327          */
1328         current->mm->task_size = TASK_SIZE;
1329
1330         /* install the new credentials */
1331         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1332             !gid_eq(bprm->cred->gid, current_egid())) {
1333                 current->pdeath_signal = 0;
1334         } else {
1335                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1336                         set_dumpable(current->mm, suid_dumpable);
1337         }
1338
1339         /* An exec changes our domain. We are no longer part of the thread
1340            group */
1341         current->self_exec_id++;
1342         flush_signal_handlers(current, 0);
1343 }
1344 EXPORT_SYMBOL(setup_new_exec);
1345
1346 /*
1347  * Prepare credentials and lock ->cred_guard_mutex.
1348  * install_exec_creds() commits the new creds and drops the lock.
1349  * Or, if exec fails before, free_bprm() should release ->cred and
1350  * and unlock.
1351  */
1352 int prepare_bprm_creds(struct linux_binprm *bprm)
1353 {
1354         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1355                 return -ERESTARTNOINTR;
1356
1357         bprm->cred = prepare_exec_creds();
1358         if (likely(bprm->cred))
1359                 return 0;
1360
1361         mutex_unlock(&current->signal->cred_guard_mutex);
1362         return -ENOMEM;
1363 }
1364
1365 static void free_bprm(struct linux_binprm *bprm)
1366 {
1367         free_arg_pages(bprm);
1368         if (bprm->cred) {
1369                 mutex_unlock(&current->signal->cred_guard_mutex);
1370                 abort_creds(bprm->cred);
1371         }
1372         if (bprm->file) {
1373                 allow_write_access(bprm->file);
1374                 fput(bprm->file);
1375         }
1376         /* If a binfmt changed the interp, free it. */
1377         if (bprm->interp != bprm->filename)
1378                 kfree(bprm->interp);
1379         kfree(bprm);
1380 }
1381
1382 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1383 {
1384         /* If a binfmt changed the interp, free it first. */
1385         if (bprm->interp != bprm->filename)
1386                 kfree(bprm->interp);
1387         bprm->interp = kstrdup(interp, GFP_KERNEL);
1388         if (!bprm->interp)
1389                 return -ENOMEM;
1390         return 0;
1391 }
1392 EXPORT_SYMBOL(bprm_change_interp);
1393
1394 /*
1395  * install the new credentials for this executable
1396  */
1397 void install_exec_creds(struct linux_binprm *bprm)
1398 {
1399         security_bprm_committing_creds(bprm);
1400
1401         commit_creds(bprm->cred);
1402         bprm->cred = NULL;
1403
1404         /*
1405          * Disable monitoring for regular users
1406          * when executing setuid binaries. Must
1407          * wait until new credentials are committed
1408          * by commit_creds() above
1409          */
1410         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1411                 perf_event_exit_task(current);
1412         /*
1413          * cred_guard_mutex must be held at least to this point to prevent
1414          * ptrace_attach() from altering our determination of the task's
1415          * credentials; any time after this it may be unlocked.
1416          */
1417         security_bprm_committed_creds(bprm);
1418         mutex_unlock(&current->signal->cred_guard_mutex);
1419 }
1420 EXPORT_SYMBOL(install_exec_creds);
1421
1422 /*
1423  * determine how safe it is to execute the proposed program
1424  * - the caller must hold ->cred_guard_mutex to protect against
1425  *   PTRACE_ATTACH or seccomp thread-sync
1426  */
1427 static void check_unsafe_exec(struct linux_binprm *bprm)
1428 {
1429         struct task_struct *p = current, *t;
1430         unsigned n_fs;
1431
1432         if (p->ptrace)
1433                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1434
1435         /*
1436          * This isn't strictly necessary, but it makes it harder for LSMs to
1437          * mess up.
1438          */
1439         if (task_no_new_privs(current))
1440                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1441
1442         t = p;
1443         n_fs = 1;
1444         spin_lock(&p->fs->lock);
1445         rcu_read_lock();
1446         while_each_thread(p, t) {
1447                 if (t->fs == p->fs)
1448                         n_fs++;
1449         }
1450         rcu_read_unlock();
1451
1452         if (p->fs->users > n_fs)
1453                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1454         else
1455                 p->fs->in_exec = 1;
1456         spin_unlock(&p->fs->lock);
1457 }
1458
1459 static void bprm_fill_uid(struct linux_binprm *bprm)
1460 {
1461         struct inode *inode;
1462         unsigned int mode;
1463         kuid_t uid;
1464         kgid_t gid;
1465
1466         /*
1467          * Since this can be called multiple times (via prepare_binprm),
1468          * we must clear any previous work done when setting set[ug]id
1469          * bits from any earlier bprm->file uses (for example when run
1470          * first for a setuid script then again for its interpreter).
1471          */
1472         bprm->cred->euid = current_euid();
1473         bprm->cred->egid = current_egid();
1474
1475         if (!mnt_may_suid(bprm->file->f_path.mnt))
1476                 return;
1477
1478         if (task_no_new_privs(current))
1479                 return;
1480
1481         inode = bprm->file->f_path.dentry->d_inode;
1482         mode = READ_ONCE(inode->i_mode);
1483         if (!(mode & (S_ISUID|S_ISGID)))
1484                 return;
1485
1486         /* Be careful if suid/sgid is set */
1487         inode_lock(inode);
1488
1489         /* reload atomically mode/uid/gid now that lock held */
1490         mode = inode->i_mode;
1491         uid = inode->i_uid;
1492         gid = inode->i_gid;
1493         inode_unlock(inode);
1494
1495         /* We ignore suid/sgid if there are no mappings for them in the ns */
1496         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1497                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1498                 return;
1499
1500         if (mode & S_ISUID) {
1501                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1502                 bprm->cred->euid = uid;
1503         }
1504
1505         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1506                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1507                 bprm->cred->egid = gid;
1508         }
1509 }
1510
1511 /*
1512  * Fill the binprm structure from the inode.
1513  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1514  *
1515  * This may be called multiple times for binary chains (scripts for example).
1516  */
1517 int prepare_binprm(struct linux_binprm *bprm)
1518 {
1519         int retval;
1520
1521         bprm_fill_uid(bprm);
1522
1523         /* fill in binprm security blob */
1524         retval = security_bprm_set_creds(bprm);
1525         if (retval)
1526                 return retval;
1527         bprm->cred_prepared = 1;
1528
1529         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1530         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1531 }
1532
1533 EXPORT_SYMBOL(prepare_binprm);
1534
1535 /*
1536  * Arguments are '\0' separated strings found at the location bprm->p
1537  * points to; chop off the first by relocating brpm->p to right after
1538  * the first '\0' encountered.
1539  */
1540 int remove_arg_zero(struct linux_binprm *bprm)
1541 {
1542         int ret = 0;
1543         unsigned long offset;
1544         char *kaddr;
1545         struct page *page;
1546
1547         if (!bprm->argc)
1548                 return 0;
1549
1550         do {
1551                 offset = bprm->p & ~PAGE_MASK;
1552                 page = get_arg_page(bprm, bprm->p, 0);
1553                 if (!page) {
1554                         ret = -EFAULT;
1555                         goto out;
1556                 }
1557                 kaddr = kmap_atomic(page);
1558
1559                 for (; offset < PAGE_SIZE && kaddr[offset];
1560                                 offset++, bprm->p++)
1561                         ;
1562
1563                 kunmap_atomic(kaddr);
1564                 put_arg_page(page);
1565         } while (offset == PAGE_SIZE);
1566
1567         bprm->p++;
1568         bprm->argc--;
1569         ret = 0;
1570
1571 out:
1572         return ret;
1573 }
1574 EXPORT_SYMBOL(remove_arg_zero);
1575
1576 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1577 /*
1578  * cycle the list of binary formats handler, until one recognizes the image
1579  */
1580 int search_binary_handler(struct linux_binprm *bprm)
1581 {
1582         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1583         struct linux_binfmt *fmt;
1584         int retval;
1585
1586         /* This allows 4 levels of binfmt rewrites before failing hard. */
1587         if (bprm->recursion_depth > 5)
1588                 return -ELOOP;
1589
1590         retval = security_bprm_check(bprm);
1591         if (retval)
1592                 return retval;
1593
1594         retval = -ENOENT;
1595  retry:
1596         read_lock(&binfmt_lock);
1597         list_for_each_entry(fmt, &formats, lh) {
1598                 if (!try_module_get(fmt->module))
1599                         continue;
1600                 read_unlock(&binfmt_lock);
1601                 bprm->recursion_depth++;
1602                 retval = fmt->load_binary(bprm);
1603                 read_lock(&binfmt_lock);
1604                 put_binfmt(fmt);
1605                 bprm->recursion_depth--;
1606                 if (retval < 0 && !bprm->mm) {
1607                         /* we got to flush_old_exec() and failed after it */
1608                         read_unlock(&binfmt_lock);
1609                         force_sigsegv(SIGSEGV, current);
1610                         return retval;
1611                 }
1612                 if (retval != -ENOEXEC || !bprm->file) {
1613                         read_unlock(&binfmt_lock);
1614                         return retval;
1615                 }
1616         }
1617         read_unlock(&binfmt_lock);
1618
1619         if (need_retry) {
1620                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1621                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1622                         return retval;
1623                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1624                         return retval;
1625                 need_retry = false;
1626                 goto retry;
1627         }
1628
1629         return retval;
1630 }
1631 EXPORT_SYMBOL(search_binary_handler);
1632
1633 static int exec_binprm(struct linux_binprm *bprm)
1634 {
1635         pid_t old_pid, old_vpid;
1636         int ret;
1637
1638         /* Need to fetch pid before load_binary changes it */
1639         old_pid = current->pid;
1640         rcu_read_lock();
1641         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1642         rcu_read_unlock();
1643
1644         ret = search_binary_handler(bprm);
1645         if (ret >= 0) {
1646                 audit_bprm(bprm);
1647                 trace_sched_process_exec(current, old_pid, bprm);
1648                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1649                 proc_exec_connector(current);
1650         }
1651
1652         return ret;
1653 }
1654
1655 /*
1656  * sys_execve() executes a new program.
1657  */
1658 static int do_execveat_common(int fd, struct filename *filename,
1659                               struct user_arg_ptr argv,
1660                               struct user_arg_ptr envp,
1661                               int flags)
1662 {
1663         char *pathbuf = NULL;
1664         struct linux_binprm *bprm;
1665         struct file *file;
1666         struct files_struct *displaced;
1667         int retval;
1668
1669         if (IS_ERR(filename))
1670                 return PTR_ERR(filename);
1671
1672         /*
1673          * We move the actual failure in case of RLIMIT_NPROC excess from
1674          * set*uid() to execve() because too many poorly written programs
1675          * don't check setuid() return code.  Here we additionally recheck
1676          * whether NPROC limit is still exceeded.
1677          */
1678         if ((current->flags & PF_NPROC_EXCEEDED) &&
1679             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1680                 retval = -EAGAIN;
1681                 goto out_ret;
1682         }
1683
1684         /* We're below the limit (still or again), so we don't want to make
1685          * further execve() calls fail. */
1686         current->flags &= ~PF_NPROC_EXCEEDED;
1687
1688         retval = unshare_files(&displaced);
1689         if (retval)
1690                 goto out_ret;
1691
1692         retval = -ENOMEM;
1693         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1694         if (!bprm)
1695                 goto out_files;
1696
1697         retval = prepare_bprm_creds(bprm);
1698         if (retval)
1699                 goto out_free;
1700
1701         check_unsafe_exec(bprm);
1702         current->in_execve = 1;
1703
1704         file = do_open_execat(fd, filename, flags);
1705         retval = PTR_ERR(file);
1706         if (IS_ERR(file))
1707                 goto out_unmark;
1708
1709         sched_exec();
1710
1711         bprm->file = file;
1712         if (fd == AT_FDCWD || filename->name[0] == '/') {
1713                 bprm->filename = filename->name;
1714         } else {
1715                 if (filename->name[0] == '\0')
1716                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1717                 else
1718                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1719                                             fd, filename->name);
1720                 if (!pathbuf) {
1721                         retval = -ENOMEM;
1722                         goto out_unmark;
1723                 }
1724                 /*
1725                  * Record that a name derived from an O_CLOEXEC fd will be
1726                  * inaccessible after exec. Relies on having exclusive access to
1727                  * current->files (due to unshare_files above).
1728                  */
1729                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1730                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1731                 bprm->filename = pathbuf;
1732         }
1733         bprm->interp = bprm->filename;
1734
1735         retval = bprm_mm_init(bprm);
1736         if (retval)
1737                 goto out_unmark;
1738
1739         bprm->argc = count(argv, MAX_ARG_STRINGS);
1740         if ((retval = bprm->argc) < 0)
1741                 goto out;
1742
1743         bprm->envc = count(envp, MAX_ARG_STRINGS);
1744         if ((retval = bprm->envc) < 0)
1745                 goto out;
1746
1747         retval = prepare_binprm(bprm);
1748         if (retval < 0)
1749                 goto out;
1750
1751         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1752         if (retval < 0)
1753                 goto out;
1754
1755         bprm->exec = bprm->p;
1756         retval = copy_strings(bprm->envc, envp, bprm);
1757         if (retval < 0)
1758                 goto out;
1759
1760         retval = copy_strings(bprm->argc, argv, bprm);
1761         if (retval < 0)
1762                 goto out;
1763
1764         would_dump(bprm, bprm->file);
1765
1766         retval = exec_binprm(bprm);
1767         if (retval < 0)
1768                 goto out;
1769
1770         /* execve succeeded */
1771         current->fs->in_exec = 0;
1772         current->in_execve = 0;
1773         acct_update_integrals(current);
1774         task_numa_free(current);
1775         free_bprm(bprm);
1776         kfree(pathbuf);
1777         putname(filename);
1778         if (displaced)
1779                 put_files_struct(displaced);
1780         return retval;
1781
1782 out:
1783         if (bprm->mm) {
1784                 acct_arg_size(bprm, 0);
1785                 mmput(bprm->mm);
1786         }
1787
1788 out_unmark:
1789         current->fs->in_exec = 0;
1790         current->in_execve = 0;
1791
1792 out_free:
1793         free_bprm(bprm);
1794         kfree(pathbuf);
1795
1796 out_files:
1797         if (displaced)
1798                 reset_files_struct(displaced);
1799 out_ret:
1800         putname(filename);
1801         return retval;
1802 }
1803
1804 int do_execve(struct filename *filename,
1805         const char __user *const __user *__argv,
1806         const char __user *const __user *__envp)
1807 {
1808         struct user_arg_ptr argv = { .ptr.native = __argv };
1809         struct user_arg_ptr envp = { .ptr.native = __envp };
1810         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1811 }
1812
1813 int do_execveat(int fd, struct filename *filename,
1814                 const char __user *const __user *__argv,
1815                 const char __user *const __user *__envp,
1816                 int flags)
1817 {
1818         struct user_arg_ptr argv = { .ptr.native = __argv };
1819         struct user_arg_ptr envp = { .ptr.native = __envp };
1820
1821         return do_execveat_common(fd, filename, argv, envp, flags);
1822 }
1823
1824 #ifdef CONFIG_COMPAT
1825 static int compat_do_execve(struct filename *filename,
1826         const compat_uptr_t __user *__argv,
1827         const compat_uptr_t __user *__envp)
1828 {
1829         struct user_arg_ptr argv = {
1830                 .is_compat = true,
1831                 .ptr.compat = __argv,
1832         };
1833         struct user_arg_ptr envp = {
1834                 .is_compat = true,
1835                 .ptr.compat = __envp,
1836         };
1837         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1838 }
1839
1840 static int compat_do_execveat(int fd, struct filename *filename,
1841                               const compat_uptr_t __user *__argv,
1842                               const compat_uptr_t __user *__envp,
1843                               int flags)
1844 {
1845         struct user_arg_ptr argv = {
1846                 .is_compat = true,
1847                 .ptr.compat = __argv,
1848         };
1849         struct user_arg_ptr envp = {
1850                 .is_compat = true,
1851                 .ptr.compat = __envp,
1852         };
1853         return do_execveat_common(fd, filename, argv, envp, flags);
1854 }
1855 #endif
1856
1857 void set_binfmt(struct linux_binfmt *new)
1858 {
1859         struct mm_struct *mm = current->mm;
1860
1861         if (mm->binfmt)
1862                 module_put(mm->binfmt->module);
1863
1864         mm->binfmt = new;
1865         if (new)
1866                 __module_get(new->module);
1867 }
1868 EXPORT_SYMBOL(set_binfmt);
1869
1870 /*
1871  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1872  */
1873 void set_dumpable(struct mm_struct *mm, int value)
1874 {
1875         unsigned long old, new;
1876
1877         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1878                 return;
1879
1880         do {
1881                 old = ACCESS_ONCE(mm->flags);
1882                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1883         } while (cmpxchg(&mm->flags, old, new) != old);
1884 }
1885
1886 SYSCALL_DEFINE3(execve,
1887                 const char __user *, filename,
1888                 const char __user *const __user *, argv,
1889                 const char __user *const __user *, envp)
1890 {
1891         return do_execve(getname(filename), argv, envp);
1892 }
1893
1894 SYSCALL_DEFINE5(execveat,
1895                 int, fd, const char __user *, filename,
1896                 const char __user *const __user *, argv,
1897                 const char __user *const __user *, envp,
1898                 int, flags)
1899 {
1900         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1901
1902         return do_execveat(fd,
1903                            getname_flags(filename, lookup_flags, NULL),
1904                            argv, envp, flags);
1905 }
1906
1907 #ifdef CONFIG_COMPAT
1908 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1909         const compat_uptr_t __user *, argv,
1910         const compat_uptr_t __user *, envp)
1911 {
1912         return compat_do_execve(getname(filename), argv, envp);
1913 }
1914
1915 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1916                        const char __user *, filename,
1917                        const compat_uptr_t __user *, argv,
1918                        const compat_uptr_t __user *, envp,
1919                        int,  flags)
1920 {
1921         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1922
1923         return compat_do_execveat(fd,
1924                                   getname_flags(filename, lookup_flags, NULL),
1925                                   argv, envp, flags);
1926 }
1927 #endif