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