fs/exec.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/fs/exec.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  * #!-checking implemented by tytso.
  10  */
  11 /*
  12  * Demand-loading implemented 01.12.91 - no need to read anything but
  13  * the header into memory. The inode of the executable is put into
  14  * "current->executable", and page faults do the actual loading. Clean.
  15  *
  16  * Once more I can proudly say that linux stood up to being changed: it
  17  * was less than 2 hours work to get demand-loading completely implemented.
  18  *
  19  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  20  * current->executable is only used by the procfs.  This allows a dispatch
  21  * table to check for several different types  of binary formats.  We keep
  22  * trying until we recognize the file or we run out of supported binary
  23  * formats.
  24  */
  25
  26 #include <linux/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/fdtable.h>
  29 #include <linux/mm.h>
  30 #include <linux/vmacache.h>
  31 #include <linux/stat.h>
  32 #include <linux/fcntl.h>
  33 #include <linux/swap.h>
  34 #include <linux/string.h>
  35 #include <linux/init.h>
  36 #include <linux/sched/mm.h>
  37 #include <linux/sched/coredump.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/sched/numa_balancing.h>
  40 #include <linux/sched/task.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/perf_event.h>
  43 #include <linux/highmem.h>
  44 #include <linux/spinlock.h>
  45 #include <linux/key.h>
  46 #include <linux/personality.h>
  47 #include <linux/binfmts.h>
  48 #include <linux/utsname.h>
  49 #include <linux/pid_namespace.h>
  50 #include <linux/module.h>
  51 #include <linux/namei.h>
  52 #include <linux/mount.h>
  53 #include <linux/security.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/tsacct_kern.h>
  56 #include <linux/cn_proc.h>
  57 #include <linux/audit.h>
  58 #include <linux/tracehook.h>
  59 #include <linux/kmod.h>
  60 #include <linux/fsnotify.h>
  61 #include <linux/fs_struct.h>
  62 #include <linux/oom.h>
  63 #include <linux/compat.h>
  64 #include <linux/vmalloc.h>
  65
  66 #include <linux/uaccess.h>
  67 #include <asm/mmu_context.h>
  68 #include <asm/tlb.h>
  69
  70 #include <trace/events/task.h>
  71 #include "internal.h"
  72
  73 #include <trace/events/sched.h>
  74
  75 int suid_dumpable = 0;
  76
  77 static LIST_HEAD(formats);
  78 static DEFINE_RWLOCK(binfmt_lock);
  79
  80 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  81 {
  82         BUG_ON(!fmt);
  83         if (WARN_ON(!fmt->load_binary))
  84                 return;
  85         write_lock(&binfmt_lock);
  86         insert ? list_add(&fmt->lh, &formats) :
  87                  list_add_tail(&fmt->lh, &formats);
  88         write_unlock(&binfmt_lock);
  89 }
  90
  91 EXPORT_SYMBOL(__register_binfmt);
  92
  93 void unregister_binfmt(struct linux_binfmt * fmt)
  94 {
  95         write_lock(&binfmt_lock);
  96         list_del(&fmt->lh);
  97         write_unlock(&binfmt_lock);
  98 }
  99
 100 EXPORT_SYMBOL(unregister_binfmt);
 101
 102 static inline void put_binfmt(struct linux_binfmt * fmt)
 103 {
 104         module_put(fmt->module);
 105 }
 106
 107 bool path_noexec(const struct path *path)
 108 {
 109         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 110                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 111 }
 112
 113 #ifdef CONFIG_USELIB
 114 /*
 115  * Note that a shared library must be both readable and executable due to
 116  * security reasons.
 117  *
 118  * Also note that we take the address to load from from the file itself.
 119  */
 120 SYSCALL_DEFINE1(uselib, const char __user *, library)
 121 {
 122         struct linux_binfmt *fmt;
 123         struct file *file;
 124         struct filename *tmp = getname(library);
 125         int error = PTR_ERR(tmp);
 126         static const struct open_flags uselib_flags = {
 127                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 128                 .acc_mode = MAY_READ | MAY_EXEC,
 129                 .intent = LOOKUP_OPEN,
 130                 .lookup_flags = LOOKUP_FOLLOW,
 131         };
 132
 133         if (IS_ERR(tmp))
 134                 goto out;
 135
 136         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 137         putname(tmp);
 138         error = PTR_ERR(file);
 139         if (IS_ERR(file))
 140                 goto out;
 141
 142         error = -EINVAL;
 143         if (!S_ISREG(file_inode(file)->i_mode))
 144                 goto exit;
 145
 146         error = -EACCES;
 147         if (path_noexec(&file->f_path))
 148                 goto exit;
 149
 150         fsnotify_open(file);
 151
 152         error = -ENOEXEC;
 153
 154         read_lock(&binfmt_lock);
 155         list_for_each_entry(fmt, &formats, lh) {
 156                 if (!fmt->load_shlib)
 157                         continue;
 158                 if (!try_module_get(fmt->module))
 159                         continue;
 160                 read_unlock(&binfmt_lock);
 161                 error = fmt->load_shlib(file);
 162                 read_lock(&binfmt_lock);
 163                 put_binfmt(fmt);
 164                 if (error != -ENOEXEC)
 165                         break;
 166         }
 167         read_unlock(&binfmt_lock);
 168 exit:
 169         fput(file);
 170 out:
 171         return error;
 172 }
 173 #endif /* #ifdef CONFIG_USELIB */
 174
 175 #ifdef CONFIG_MMU
 176 /*
 177  * The nascent bprm->mm is not visible until exec_mmap() but it can
 178  * use a lot of memory, account these pages in current->mm temporary
 179  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 180  * change the counter back via acct_arg_size(0).
 181  */
 182 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 183 {
 184         struct mm_struct *mm = current->mm;
 185         long diff = (long)(pages - bprm->vma_pages);
 186
 187         if (!mm || !diff)
 188                 return;
 189
 190         bprm->vma_pages = pages;
 191         add_mm_counter(mm, MM_ANONPAGES, diff);
 192 }
 193
 194 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 195                 int write)
 196 {
 197         struct page *page;
 198         int ret;
 199         unsigned int gup_flags = FOLL_FORCE;
 200
 201 #ifdef CONFIG_STACK_GROWSUP
 202         if (write) {
 203                 ret = expand_downwards(bprm->vma, pos);
 204                 if (ret < 0)
 205                         return NULL;
 206         }
 207 #endif
 208
 209         if (write)
 210                 gup_flags |= FOLL_WRITE;
 211
 212         /*
 213          * We are doing an exec().  'current' is the process
 214          * doing the exec and bprm->mm is the new process's mm.
 215          */
 216         ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
 217                         &page, NULL, NULL);
 218         if (ret <= 0)
 219                 return NULL;
 220
 221         if (write)
 222                 acct_arg_size(bprm, vma_pages(bprm->vma));
 223
 224         return page;
 225 }
 226
 227 static void put_arg_page(struct page *page)
 228 {
 229         put_page(page);
 230 }
 231
 232 static void free_arg_pages(struct linux_binprm *bprm)
 233 {
 234 }
 235
 236 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 237                 struct page *page)
 238 {
 239         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 240 }
 241
 242 static int __bprm_mm_init(struct linux_binprm *bprm)
 243 {
 244         int err;
 245         struct vm_area_struct *vma = NULL;
 246         struct mm_struct *mm = bprm->mm;
 247
 248         bprm->vma = vma = vm_area_alloc(mm);
 249         if (!vma)
 250                 return -ENOMEM;
 251         vma_set_anonymous(vma);
 252
 253         if (down_write_killable(&mm->mmap_sem)) {
 254                 err = -EINTR;
 255                 goto err_free;
 256         }
 257
 258         /*
 259          * Place the stack at the largest stack address the architecture
 260          * supports. Later, we'll move this to an appropriate place. We don't
 261          * use STACK_TOP because that can depend on attributes which aren't
 262          * configured yet.
 263          */
 264         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 265         vma->vm_end = STACK_TOP_MAX;
 266         vma->vm_start = vma->vm_end - PAGE_SIZE;
 267         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 268         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 269
 270         err = insert_vm_struct(mm, vma);
 271         if (err)
 272                 goto err;
 273
 274         mm->stack_vm = mm->total_vm = 1;
 275         arch_bprm_mm_init(mm, vma);
 276         up_write(&mm->mmap_sem);
 277         bprm->p = vma->vm_end - sizeof(void *);
 278         return 0;
 279 err:
 280         up_write(&mm->mmap_sem);
 281 err_free:
 282         bprm->vma = NULL;
 283         vm_area_free(vma);
 284         return err;
 285 }
 286
 287 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 288 {
 289         return len <= MAX_ARG_STRLEN;
 290 }
 291
 292 #else
 293
 294 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 295 {
 296 }
 297
 298 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 299                 int write)
 300 {
 301         struct page *page;
 302
 303         page = bprm->page[pos / PAGE_SIZE];
 304         if (!page && write) {
 305                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 306                 if (!page)
 307                         return NULL;
 308                 bprm->page[pos / PAGE_SIZE] = page;
 309         }
 310
 311         return page;
 312 }
 313
 314 static void put_arg_page(struct page *page)
 315 {
 316 }
 317
 318 static void free_arg_page(struct linux_binprm *bprm, int i)
 319 {
 320         if (bprm->page[i]) {
 321                 __free_page(bprm->page[i]);
 322                 bprm->page[i] = NULL;
 323         }
 324 }
 325
 326 static void free_arg_pages(struct linux_binprm *bprm)
 327 {
 328         int i;
 329
 330         for (i = 0; i < MAX_ARG_PAGES; i++)
 331                 free_arg_page(bprm, i);
 332 }
 333
 334 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 335                 struct page *page)
 336 {
 337 }
 338
 339 static int __bprm_mm_init(struct linux_binprm *bprm)
 340 {
 341         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 342         return 0;
 343 }
 344
 345 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 346 {
 347         return len <= bprm->p;
 348 }
 349
 350 #endif /* CONFIG_MMU */
 351
 352 /*
 353  * Create a new mm_struct and populate it with a temporary stack
 354  * vm_area_struct.  We don't have enough context at this point to set the stack
 355  * flags, permissions, and offset, so we use temporary values.  We'll update
 356  * them later in setup_arg_pages().
 357  */
 358 static int bprm_mm_init(struct linux_binprm *bprm)
 359 {
 360         int err;
 361         struct mm_struct *mm = NULL;
 362
 363         bprm->mm = mm = mm_alloc();
 364         err = -ENOMEM;
 365         if (!mm)
 366                 goto err;
 367
 368         /* Save current stack limit for all calculations made during exec. */
 369         task_lock(current->group_leader);
 370         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 371         task_unlock(current->group_leader);
 372
 373         err = __bprm_mm_init(bprm);
 374         if (err)
 375                 goto err;
 376
 377         return 0;
 378
 379 err:
 380         if (mm) {
 381                 bprm->mm = NULL;
 382                 mmdrop(mm);
 383         }
 384
 385         return err;
 386 }
 387
 388 struct user_arg_ptr {
 389 #ifdef CONFIG_COMPAT
 390         bool is_compat;
 391 #endif
 392         union {
 393                 const char __user *const __user *native;
 394 #ifdef CONFIG_COMPAT
 395                 const compat_uptr_t __user *compat;
 396 #endif
 397         } ptr;
 398 };
 399
 400 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 401 {
 402         const char __user *native;
 403
 404 #ifdef CONFIG_COMPAT
 405         if (unlikely(argv.is_compat)) {
 406                 compat_uptr_t compat;
 407
 408                 if (get_user(compat, argv.ptr.compat + nr))
 409                         return ERR_PTR(-EFAULT);
 410
 411                 return compat_ptr(compat);
 412         }
 413 #endif
 414
 415         if (get_user(native, argv.ptr.native + nr))
 416                 return ERR_PTR(-EFAULT);
 417
 418         return native;
 419 }
 420
 421 /*
 422  * count() counts the number of strings in array ARGV.
 423  */
 424 static int count(struct user_arg_ptr argv, int max)
 425 {
 426         int i = 0;
 427
 428         if (argv.ptr.native != NULL) {
 429                 for (;;) {
 430                         const char __user *p = get_user_arg_ptr(argv, i);
 431
 432                         if (!p)
 433                                 break;
 434
 435                         if (IS_ERR(p))
 436                                 return -EFAULT;
 437
 438                         if (i >= max)
 439                                 return -E2BIG;
 440                         ++i;
 441
 442                         if (fatal_signal_pending(current))
 443                                 return -ERESTARTNOHAND;
 444                         cond_resched();
 445                 }
 446         }
 447         return i;
 448 }
 449
 450 static int prepare_arg_pages(struct linux_binprm *bprm,
 451                         struct user_arg_ptr argv, struct user_arg_ptr envp)
 452 {
 453         unsigned long limit, ptr_size;
 454
 455         bprm->argc = count(argv, MAX_ARG_STRINGS);
 456         if (bprm->argc < 0)
 457                 return bprm->argc;
 458
 459         bprm->envc = count(envp, MAX_ARG_STRINGS);
 460         if (bprm->envc < 0)
 461                 return bprm->envc;
 462
 463         /*
 464          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 465          * (whichever is smaller) for the argv+env strings.
 466          * This ensures that:
 467          *  - the remaining binfmt code will not run out of stack space,
 468          *  - the program will have a reasonable amount of stack left
 469          *    to work from.
 470          */
 471         limit = _STK_LIM / 4 * 3;
 472         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 473         /*
 474          * We've historically supported up to 32 pages (ARG_MAX)
 475          * of argument strings even with small stacks
 476          */
 477         limit = max_t(unsigned long, limit, ARG_MAX);
 478         /*
 479          * We must account for the size of all the argv and envp pointers to
 480          * the argv and envp strings, since they will also take up space in
 481          * the stack. They aren't stored until much later when we can't
 482          * signal to the parent that the child has run out of stack space.
 483          * Instead, calculate it here so it's possible to fail gracefully.
 484          */
 485         ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
 486         if (limit <= ptr_size)
 487                 return -E2BIG;
 488         limit -= ptr_size;
 489
 490         bprm->argmin = bprm->p - limit;
 491         return 0;
 492 }
 493
 494 /*
 495  * 'copy_strings()' copies argument/environment strings from the old
 496  * processes's memory to the new process's stack.  The call to get_user_pages()
 497  * ensures the destination page is created and not swapped out.
 498  */
 499 static int copy_strings(int argc, struct user_arg_ptr argv,
 500                         struct linux_binprm *bprm)
 501 {
 502         struct page *kmapped_page = NULL;
 503         char *kaddr = NULL;
 504         unsigned long kpos = 0;
 505         int ret;
 506
 507         while (argc-- > 0) {
 508                 const char __user *str;
 509                 int len;
 510                 unsigned long pos;
 511
 512                 ret = -EFAULT;
 513                 str = get_user_arg_ptr(argv, argc);
 514                 if (IS_ERR(str))
 515                         goto out;
 516
 517                 len = strnlen_user(str, MAX_ARG_STRLEN);
 518                 if (!len)
 519                         goto out;
 520
 521                 ret = -E2BIG;
 522                 if (!valid_arg_len(bprm, len))
 523                         goto out;
 524
 525                 /* We're going to work our way backwords. */
 526                 pos = bprm->p;
 527                 str += len;
 528                 bprm->p -= len;
 529 #ifdef CONFIG_MMU
 530                 if (bprm->p < bprm->argmin)
 531                         goto out;
 532 #endif
 533
 534                 while (len > 0) {
 535                         int offset, bytes_to_copy;
 536
 537                         if (fatal_signal_pending(current)) {
 538                                 ret = -ERESTARTNOHAND;
 539                                 goto out;
 540                         }
 541                         cond_resched();
 542
 543                         offset = pos % PAGE_SIZE;
 544                         if (offset == 0)
 545                                 offset = PAGE_SIZE;
 546
 547                         bytes_to_copy = offset;
 548                         if (bytes_to_copy > len)
 549                                 bytes_to_copy = len;
 550
 551                         offset -= bytes_to_copy;
 552                         pos -= bytes_to_copy;
 553                         str -= bytes_to_copy;
 554                         len -= bytes_to_copy;
 555
 556                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 557                                 struct page *page;
 558
 559                                 page = get_arg_page(bprm, pos, 1);
 560                                 if (!page) {
 561                                         ret = -E2BIG;
 562                                         goto out;
 563                                 }
 564
 565                                 if (kmapped_page) {
 566                                         flush_kernel_dcache_page(kmapped_page);
 567                                         kunmap(kmapped_page);
 568                                         put_arg_page(kmapped_page);
 569                                 }
 570                                 kmapped_page = page;
 571                                 kaddr = kmap(kmapped_page);
 572                                 kpos = pos & PAGE_MASK;
 573                                 flush_arg_page(bprm, kpos, kmapped_page);
 574                         }
 575                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 576                                 ret = -EFAULT;
 577                                 goto out;
 578                         }
 579                 }
 580         }
 581         ret = 0;
 582 out:
 583         if (kmapped_page) {
 584                 flush_kernel_dcache_page(kmapped_page);
 585                 kunmap(kmapped_page);
 586                 put_arg_page(kmapped_page);
 587         }
 588         return ret;
 589 }
 590
 591 /*
 592  * Like copy_strings, but get argv and its values from kernel memory.
 593  */
 594 int copy_strings_kernel(int argc, const char *const *__argv,
 595                         struct linux_binprm *bprm)
 596 {
 597         int r;
 598         mm_segment_t oldfs = get_fs();
 599         struct user_arg_ptr argv = {
 600                 .ptr.native = (const char __user *const  __user *)__argv,
 601         };
 602
 603         set_fs(KERNEL_DS);
 604         r = copy_strings(argc, argv, bprm);
 605         set_fs(oldfs);
 606
 607         return r;
 608 }
 609 EXPORT_SYMBOL(copy_strings_kernel);
 610
 611 #ifdef CONFIG_MMU
 612
 613 /*
 614  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 615  * the binfmt code determines where the new stack should reside, we shift it to
 616  * its final location.  The process proceeds as follows:
 617  *
 618  * 1) Use shift to calculate the new vma endpoints.
 619  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 620  *    arguments passed to subsequent functions are consistent.
 621  * 3) Move vma's page tables to the new range.
 622  * 4) Free up any cleared pgd range.
 623  * 5) Shrink the vma to cover only the new range.
 624  */
 625 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 626 {
 627         struct mm_struct *mm = vma->vm_mm;
 628         unsigned long old_start = vma->vm_start;
 629         unsigned long old_end = vma->vm_end;
 630         unsigned long length = old_end - old_start;
 631         unsigned long new_start = old_start - shift;
 632         unsigned long new_end = old_end - shift;
 633         struct mmu_gather tlb;
 634
 635         BUG_ON(new_start > new_end);
 636
 637         /*
 638          * ensure there are no vmas between where we want to go
 639          * and where we are
 640          */
 641         if (vma != find_vma(mm, new_start))
 642                 return -EFAULT;
 643
 644         /*
 645          * cover the whole range: [new_start, old_end)
 646          */
 647         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 648                 return -ENOMEM;
 649
 650         /*
 651          * move the page tables downwards, on failure we rely on
 652          * process cleanup to remove whatever mess we made.
 653          */
 654         if (length != move_page_tables(vma, old_start,
 655                                        vma, new_start, length, false))
 656                 return -ENOMEM;
 657
 658         lru_add_drain();
 659         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 660         if (new_end > old_start) {
 661                 /*
 662                  * when the old and new regions overlap clear from new_end.
 663                  */
 664                 free_pgd_range(&tlb, new_end, old_end, new_end,
 665                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 666         } else {
 667                 /*
 668                  * otherwise, clean from old_start; this is done to not touch
 669                  * the address space in [new_end, old_start) some architectures
 670                  * have constraints on va-space that make this illegal (IA64) -
 671                  * for the others its just a little faster.
 672                  */
 673                 free_pgd_range(&tlb, old_start, old_end, new_end,
 674                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 675         }
 676         tlb_finish_mmu(&tlb, old_start, old_end);
 677
 678         /*
 679          * Shrink the vma to just the new range.  Always succeeds.
 680          */
 681         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 682
 683         return 0;
 684 }
 685
 686 /*
 687  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 688  * the stack is optionally relocated, and some extra space is added.
 689  */
 690 int setup_arg_pages(struct linux_binprm *bprm,
 691                     unsigned long stack_top,
 692                     int executable_stack)
 693 {
 694         unsigned long ret;
 695         unsigned long stack_shift;
 696         struct mm_struct *mm = current->mm;
 697         struct vm_area_struct *vma = bprm->vma;
 698         struct vm_area_struct *prev = NULL;
 699         unsigned long vm_flags;
 700         unsigned long stack_base;
 701         unsigned long stack_size;
 702         unsigned long stack_expand;
 703         unsigned long rlim_stack;
 704
 705 #ifdef CONFIG_STACK_GROWSUP
 706         /* Limit stack size */
 707         stack_base = bprm->rlim_stack.rlim_max;
 708         if (stack_base > STACK_SIZE_MAX)
 709                 stack_base = STACK_SIZE_MAX;
 710
 711         /* Add space for stack randomization. */
 712         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 713
 714         /* Make sure we didn't let the argument array grow too large. */
 715         if (vma->vm_end - vma->vm_start > stack_base)
 716                 return -ENOMEM;
 717
 718         stack_base = PAGE_ALIGN(stack_top - stack_base);
 719
 720         stack_shift = vma->vm_start - stack_base;
 721         mm->arg_start = bprm->p - stack_shift;
 722         bprm->p = vma->vm_end - stack_shift;
 723 #else
 724         stack_top = arch_align_stack(stack_top);
 725         stack_top = PAGE_ALIGN(stack_top);
 726
 727         if (unlikely(stack_top < mmap_min_addr) ||
 728             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 729                 return -ENOMEM;
 730
 731         stack_shift = vma->vm_end - stack_top;
 732
 733         bprm->p -= stack_shift;
 734         mm->arg_start = bprm->p;
 735 #endif
 736
 737         if (bprm->loader)
 738                 bprm->loader -= stack_shift;
 739         bprm->exec -= stack_shift;
 740
 741         if (down_write_killable(&mm->mmap_sem))
 742                 return -EINTR;
 743
 744         vm_flags = VM_STACK_FLAGS;
 745
 746         /*
 747          * Adjust stack execute permissions; explicitly enable for
 748          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 749          * (arch default) otherwise.
 750          */
 751         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 752                 vm_flags |= VM_EXEC;
 753         else if (executable_stack == EXSTACK_DISABLE_X)
 754                 vm_flags &= ~VM_EXEC;
 755         vm_flags |= mm->def_flags;
 756         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 757
 758         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 759                         vm_flags);
 760         if (ret)
 761                 goto out_unlock;
 762         BUG_ON(prev != vma);
 763
 764         /* Move stack pages down in memory. */
 765         if (stack_shift) {
 766                 ret = shift_arg_pages(vma, stack_shift);
 767                 if (ret)
 768                         goto out_unlock;
 769         }
 770
 771         /* mprotect_fixup is overkill to remove the temporary stack flags */
 772         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 773
 774         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 775         stack_size = vma->vm_end - vma->vm_start;
 776         /*
 777          * Align this down to a page boundary as expand_stack
 778          * will align it up.
 779          */
 780         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 781 #ifdef CONFIG_STACK_GROWSUP
 782         if (stack_size + stack_expand > rlim_stack)
 783                 stack_base = vma->vm_start + rlim_stack;
 784         else
 785                 stack_base = vma->vm_end + stack_expand;
 786 #else
 787         if (stack_size + stack_expand > rlim_stack)
 788                 stack_base = vma->vm_end - rlim_stack;
 789         else
 790                 stack_base = vma->vm_start - stack_expand;
 791 #endif
 792         current->mm->start_stack = bprm->p;
 793         ret = expand_stack(vma, stack_base);
 794         if (ret)
 795                 ret = -EFAULT;
 796
 797 out_unlock:
 798         up_write(&mm->mmap_sem);
 799         return ret;
 800 }
 801 EXPORT_SYMBOL(setup_arg_pages);
 802
 803 #else
 804
 805 /*
 806  * Transfer the program arguments and environment from the holding pages
 807  * onto the stack. The provided stack pointer is adjusted accordingly.
 808  */
 809 int transfer_args_to_stack(struct linux_binprm *bprm,
 810                            unsigned long *sp_location)
 811 {
 812         unsigned long index, stop, sp;
 813         int ret = 0;
 814
 815         stop = bprm->p >> PAGE_SHIFT;
 816         sp = *sp_location;
 817
 818         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 819                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 820                 char *src = kmap(bprm->page[index]) + offset;
 821                 sp -= PAGE_SIZE - offset;
 822                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 823                         ret = -EFAULT;
 824                 kunmap(bprm->page[index]);
 825                 if (ret)
 826                         goto out;
 827         }
 828
 829         *sp_location = sp;
 830
 831 out:
 832         return ret;
 833 }
 834 EXPORT_SYMBOL(transfer_args_to_stack);
 835
 836 #endif /* CONFIG_MMU */
 837
 838 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 839 {
 840         struct file *file;
 841         int err;
 842         struct open_flags open_exec_flags = {
 843                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 844                 .acc_mode = MAY_EXEC,
 845                 .intent = LOOKUP_OPEN,
 846                 .lookup_flags = LOOKUP_FOLLOW,
 847         };
 848
 849         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 850                 return ERR_PTR(-EINVAL);
 851         if (flags & AT_SYMLINK_NOFOLLOW)
 852                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 853         if (flags & AT_EMPTY_PATH)
 854                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 855
 856         file = do_filp_open(fd, name, &open_exec_flags);
 857         if (IS_ERR(file))
 858                 goto out;
 859
 860         err = -EACCES;
 861         if (!S_ISREG(file_inode(file)->i_mode))
 862                 goto exit;
 863
 864         if (path_noexec(&file->f_path))
 865                 goto exit;
 866
 867         err = deny_write_access(file);
 868         if (err)
 869                 goto exit;
 870
 871         if (name->name[0] != '\0')
 872                 fsnotify_open(file);
 873
 874 out:
 875         return file;
 876
 877 exit:
 878         fput(file);
 879         return ERR_PTR(err);
 880 }
 881
 882 struct file *open_exec(const char *name)
 883 {
 884         struct filename *filename = getname_kernel(name);
 885         struct file *f = ERR_CAST(filename);
 886
 887         if (!IS_ERR(filename)) {
 888                 f = do_open_execat(AT_FDCWD, filename, 0);
 889                 putname(filename);
 890         }
 891         return f;
 892 }
 893 EXPORT_SYMBOL(open_exec);
 894
 895 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 896                      loff_t max_size, enum kernel_read_file_id id)
 897 {
 898         loff_t i_size, pos;
 899         ssize_t bytes = 0;
 900         int ret;
 901
 902         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 903                 return -EINVAL;
 904
 905         ret = deny_write_access(file);
 906         if (ret)
 907                 return ret;
 908
 909         ret = security_kernel_read_file(file, id);
 910         if (ret)
 911                 goto out;
 912
 913         i_size = i_size_read(file_inode(file));
 914         if (i_size <= 0) {
 915                 ret = -EINVAL;
 916                 goto out;
 917         }
 918         if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
 919                 ret = -EFBIG;
 920                 goto out;
 921         }
 922
 923         if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 924                 *buf = vmalloc(i_size);
 925         if (!*buf) {
 926                 ret = -ENOMEM;
 927                 goto out;
 928         }
 929
 930         pos = 0;
 931         while (pos < i_size) {
 932                 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
 933                 if (bytes < 0) {
 934                         ret = bytes;
 935                         goto out_free;
 936                 }
 937
 938                 if (bytes == 0)
 939                         break;
 940         }
 941
 942         if (pos != i_size) {
 943                 ret = -EIO;
 944                 goto out_free;
 945         }
 946
 947         ret = security_kernel_post_read_file(file, *buf, i_size, id);
 948         if (!ret)
 949                 *size = pos;
 950
 951 out_free:
 952         if (ret < 0) {
 953                 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 954                         vfree(*buf);
 955                         *buf = NULL;
 956                 }
 957         }
 958
 959 out:
 960         allow_write_access(file);
 961         return ret;
 962 }
 963 EXPORT_SYMBOL_GPL(kernel_read_file);
 964
 965 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
 966                                loff_t max_size, enum kernel_read_file_id id)
 967 {
 968         struct file *file;
 969         int ret;
 970
 971         if (!path || !*path)
 972                 return -EINVAL;
 973
 974         file = filp_open(path, O_RDONLY, 0);
 975         if (IS_ERR(file))
 976                 return PTR_ERR(file);
 977
 978         ret = kernel_read_file(file, buf, size, max_size, id);
 979         fput(file);
 980         return ret;
 981 }
 982 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 983
 984 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 985                              enum kernel_read_file_id id)
 986 {
 987         struct fd f = fdget(fd);
 988         int ret = -EBADF;
 989
 990         if (!f.file)
 991                 goto out;
 992
 993         ret = kernel_read_file(f.file, buf, size, max_size, id);
 994 out:
 995         fdput(f);
 996         return ret;
 997 }
 998 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
 999
1000 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1001 {
1002         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1003         if (res > 0)
1004                 flush_icache_range(addr, addr + len);
1005         return res;
1006 }
1007 EXPORT_SYMBOL(read_code);
1008
1009 static int exec_mmap(struct mm_struct *mm)
1010 {
1011         struct task_struct *tsk;
1012         struct mm_struct *old_mm, *active_mm;
1013
1014         /* Notify parent that we're no longer interested in the old VM */
1015         tsk = current;
1016         old_mm = current->mm;
1017         exec_mm_release(tsk, old_mm);
1018
1019         if (old_mm) {
1020                 sync_mm_rss(old_mm);
1021                 /*
1022                  * Make sure that if there is a core dump in progress
1023                  * for the old mm, we get out and die instead of going
1024                  * through with the exec.  We must hold mmap_sem around
1025                  * checking core_state and changing tsk->mm.
1026                  */
1027                 down_read(&old_mm->mmap_sem);
1028                 if (unlikely(old_mm->core_state)) {
1029                         up_read(&old_mm->mmap_sem);
1030                         return -EINTR;
1031                 }
1032         }
1033         task_lock(tsk);
1034         active_mm = tsk->active_mm;
1035         membarrier_exec_mmap(mm);
1036         tsk->mm = mm;
1037         tsk->active_mm = mm;
1038         activate_mm(active_mm, mm);
1039         tsk->mm->vmacache_seqnum = 0;
1040         vmacache_flush(tsk);
1041         task_unlock(tsk);
1042         if (old_mm) {
1043                 up_read(&old_mm->mmap_sem);
1044                 BUG_ON(active_mm != old_mm);
1045                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1046                 mm_update_next_owner(old_mm);
1047                 mmput(old_mm);
1048                 return 0;
1049         }
1050         mmdrop(active_mm);
1051         return 0;
1052 }
1053
1054 /*
1055  * This function makes sure the current process has its own signal table,
1056  * so that flush_signal_handlers can later reset the handlers without
1057  * disturbing other processes.  (Other processes might share the signal
1058  * table via the CLONE_SIGHAND option to clone().)
1059  */
1060 static int de_thread(struct task_struct *tsk)
1061 {
1062         struct signal_struct *sig = tsk->signal;
1063         struct sighand_struct *oldsighand = tsk->sighand;
1064         spinlock_t *lock = &oldsighand->siglock;
1065
1066         if (thread_group_empty(tsk))
1067                 goto no_thread_group;
1068
1069         /*
1070          * Kill all other threads in the thread group.
1071          */
1072         spin_lock_irq(lock);
1073         if (signal_group_exit(sig)) {
1074                 /*
1075                  * Another group action in progress, just
1076                  * return so that the signal is processed.
1077                  */
1078                 spin_unlock_irq(lock);
1079                 return -EAGAIN;
1080         }
1081
1082         sig->group_exit_task = tsk;
1083         sig->notify_count = zap_other_threads(tsk);
1084         if (!thread_group_leader(tsk))
1085                 sig->notify_count--;
1086
1087         while (sig->notify_count) {
1088                 __set_current_state(TASK_KILLABLE);
1089                 spin_unlock_irq(lock);
1090                 schedule();
1091                 if (__fatal_signal_pending(tsk))
1092                         goto killed;
1093                 spin_lock_irq(lock);
1094         }
1095         spin_unlock_irq(lock);
1096
1097         /*
1098          * At this point all other threads have exited, all we have to
1099          * do is to wait for the thread group leader to become inactive,
1100          * and to assume its PID:
1101          */
1102         if (!thread_group_leader(tsk)) {
1103                 struct task_struct *leader = tsk->group_leader;
1104
1105                 for (;;) {
1106                         cgroup_threadgroup_change_begin(tsk);
1107                         write_lock_irq(&tasklist_lock);
1108                         /*
1109                          * Do this under tasklist_lock to ensure that
1110                          * exit_notify() can't miss ->group_exit_task
1111                          */
1112                         sig->notify_count = -1;
1113                         if (likely(leader->exit_state))
1114                                 break;
1115                         __set_current_state(TASK_KILLABLE);
1116                         write_unlock_irq(&tasklist_lock);
1117                         cgroup_threadgroup_change_end(tsk);
1118                         schedule();
1119                         if (__fatal_signal_pending(tsk))
1120                                 goto killed;
1121                 }
1122
1123                 /*
1124                  * The only record we have of the real-time age of a
1125                  * process, regardless of execs it's done, is start_time.
1126                  * All the past CPU time is accumulated in signal_struct
1127                  * from sister threads now dead.  But in this non-leader
1128                  * exec, nothing survives from the original leader thread,
1129                  * whose birth marks the true age of this process now.
1130                  * When we take on its identity by switching to its PID, we
1131                  * also take its birthdate (always earlier than our own).
1132                  */
1133                 tsk->start_time = leader->start_time;
1134                 tsk->start_boottime = leader->start_boottime;
1135
1136                 BUG_ON(!same_thread_group(leader, tsk));
1137                 BUG_ON(has_group_leader_pid(tsk));
1138                 /*
1139                  * An exec() starts a new thread group with the
1140                  * TGID of the previous thread group. Rehash the
1141                  * two threads with a switched PID, and release
1142                  * the former thread group leader:
1143                  */
1144
1145                 /* Become a process group leader with the old leader's pid.
1146                  * The old leader becomes a thread of the this thread group.
1147                  * Note: The old leader also uses this pid until release_task
1148                  *       is called.  Odd but simple and correct.
1149                  */
1150                 tsk->pid = leader->pid;
1151                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1152                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1153                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1154                 transfer_pid(leader, tsk, PIDTYPE_SID);
1155
1156                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1157                 list_replace_init(&leader->sibling, &tsk->sibling);
1158
1159                 tsk->group_leader = tsk;
1160                 leader->group_leader = tsk;
1161
1162                 tsk->exit_signal = SIGCHLD;
1163                 leader->exit_signal = -1;
1164
1165                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1166                 leader->exit_state = EXIT_DEAD;
1167
1168                 /*
1169                  * We are going to release_task()->ptrace_unlink() silently,
1170                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1171                  * the tracer wont't block again waiting for this thread.
1172                  */
1173                 if (unlikely(leader->ptrace))
1174                         __wake_up_parent(leader, leader->parent);
1175                 write_unlock_irq(&tasklist_lock);
1176                 cgroup_threadgroup_change_end(tsk);
1177
1178                 release_task(leader);
1179         }
1180
1181         sig->group_exit_task = NULL;
1182         sig->notify_count = 0;
1183
1184 no_thread_group:
1185         /* we have changed execution domain */
1186         tsk->exit_signal = SIGCHLD;
1187
1188 #ifdef CONFIG_POSIX_TIMERS
1189         exit_itimers(sig);
1190         flush_itimer_signals();
1191 #endif
1192
1193         if (refcount_read(&oldsighand->count) != 1) {
1194                 struct sighand_struct *newsighand;
1195                 /*
1196                  * This ->sighand is shared with the CLONE_SIGHAND
1197                  * but not CLONE_THREAD task, switch to the new one.
1198                  */
1199                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1200                 if (!newsighand)
1201                         return -ENOMEM;
1202
1203                 refcount_set(&newsighand->count, 1);
1204                 memcpy(newsighand->action, oldsighand->action,
1205                        sizeof(newsighand->action));
1206
1207                 write_lock_irq(&tasklist_lock);
1208                 spin_lock(&oldsighand->siglock);
1209                 rcu_assign_pointer(tsk->sighand, newsighand);
1210                 spin_unlock(&oldsighand->siglock);
1211                 write_unlock_irq(&tasklist_lock);
1212
1213                 __cleanup_sighand(oldsighand);
1214         }
1215
1216         BUG_ON(!thread_group_leader(tsk));
1217         return 0;
1218
1219 killed:
1220         /* protects against exit_notify() and __exit_signal() */
1221         read_lock(&tasklist_lock);
1222         sig->group_exit_task = NULL;
1223         sig->notify_count = 0;
1224         read_unlock(&tasklist_lock);
1225         return -EAGAIN;
1226 }
1227
1228 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1229 {
1230         task_lock(tsk);
1231         strncpy(buf, tsk->comm, buf_size);
1232         task_unlock(tsk);
1233         return buf;
1234 }
1235 EXPORT_SYMBOL_GPL(__get_task_comm);
1236
1237 /*
1238  * These functions flushes out all traces of the currently running executable
1239  * so that a new one can be started
1240  */
1241
1242 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1243 {
1244         task_lock(tsk);
1245         trace_task_rename(tsk, buf);
1246         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1247         task_unlock(tsk);
1248         perf_event_comm(tsk, exec);
1249 }
1250
1251 /*
1252  * Calling this is the point of no return. None of the failures will be
1253  * seen by userspace since either the process is already taking a fatal
1254  * signal (via de_thread() or coredump), or will have SEGV raised
1255  * (after exec_mmap()) by search_binary_handlers (see below).
1256  */
1257 int flush_old_exec(struct linux_binprm * bprm)
1258 {
1259         int retval;
1260
1261         /*
1262          * Make sure we have a private signal table and that
1263          * we are unassociated from the previous thread group.
1264          */
1265         retval = de_thread(current);
1266         if (retval)
1267                 goto out;
1268
1269         /*
1270          * Must be called _before_ exec_mmap() as bprm->mm is
1271          * not visibile until then. This also enables the update
1272          * to be lockless.
1273          */
1274         set_mm_exe_file(bprm->mm, bprm->file);
1275
1276         /*
1277          * Release all of the old mmap stuff
1278          */
1279         acct_arg_size(bprm, 0);
1280         retval = exec_mmap(bprm->mm);
1281         if (retval)
1282                 goto out;
1283
1284         /*
1285          * After clearing bprm->mm (to mark that current is using the
1286          * prepared mm now), we have nothing left of the original
1287          * process. If anything from here on returns an error, the check
1288          * in search_binary_handler() will SEGV current.
1289          */
1290         bprm->mm = NULL;
1291
1292         set_fs(USER_DS);
1293         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1294                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1295         flush_thread();
1296         current->personality &= ~bprm->per_clear;
1297
1298         /*
1299          * We have to apply CLOEXEC before we change whether the process is
1300          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1301          * trying to access the should-be-closed file descriptors of a process
1302          * undergoing exec(2).
1303          */
1304         do_close_on_exec(current->files);
1305         return 0;
1306
1307 out:
1308         return retval;
1309 }
1310 EXPORT_SYMBOL(flush_old_exec);
1311
1312 void would_dump(struct linux_binprm *bprm, struct file *file)
1313 {
1314         struct inode *inode = file_inode(file);
1315         if (inode_permission(inode, MAY_READ) < 0) {
1316                 struct user_namespace *old, *user_ns;
1317                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1318
1319                 /* Ensure mm->user_ns contains the executable */
1320                 user_ns = old = bprm->mm->user_ns;
1321                 while ((user_ns != &init_user_ns) &&
1322                        !privileged_wrt_inode_uidgid(user_ns, inode))
1323                         user_ns = user_ns->parent;
1324
1325                 if (old != user_ns) {
1326                         bprm->mm->user_ns = get_user_ns(user_ns);
1327                         put_user_ns(old);
1328                 }
1329         }
1330 }
1331 EXPORT_SYMBOL(would_dump);
1332
1333 void setup_new_exec(struct linux_binprm * bprm)
1334 {
1335         /*
1336          * Once here, prepare_binrpm() will not be called any more, so
1337          * the final state of setuid/setgid/fscaps can be merged into the
1338          * secureexec flag.
1339          */
1340         bprm->secureexec |= bprm->cap_elevated;
1341
1342         if (bprm->secureexec) {
1343                 /* Make sure parent cannot signal privileged process. */
1344                 current->pdeath_signal = 0;
1345
1346                 /*
1347                  * For secureexec, reset the stack limit to sane default to
1348                  * avoid bad behavior from the prior rlimits. This has to
1349                  * happen before arch_pick_mmap_layout(), which examines
1350                  * RLIMIT_STACK, but after the point of no return to avoid
1351                  * needing to clean up the change on failure.
1352                  */
1353                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1354                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1355         }
1356
1357         arch_pick_mmap_layout(current->mm, &bprm->rlim_stack);
1358
1359         current->sas_ss_sp = current->sas_ss_size = 0;
1360
1361         /*
1362          * Figure out dumpability. Note that this checking only of current
1363          * is wrong, but userspace depends on it. This should be testing
1364          * bprm->secureexec instead.
1365          */
1366         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1367             !(uid_eq(current_euid(), current_uid()) &&
1368               gid_eq(current_egid(), current_gid())))
1369                 set_dumpable(current->mm, suid_dumpable);
1370         else
1371                 set_dumpable(current->mm, SUID_DUMP_USER);
1372
1373         arch_setup_new_exec();
1374         perf_event_exec();
1375         __set_task_comm(current, kbasename(bprm->filename), true);
1376
1377         /* Set the new mm task size. We have to do that late because it may
1378          * depend on TIF_32BIT which is only updated in flush_thread() on
1379          * some architectures like powerpc
1380          */
1381         current->mm->task_size = TASK_SIZE;
1382
1383         /* An exec changes our domain. We are no longer part of the thread
1384            group */
1385         current->self_exec_id++;
1386         flush_signal_handlers(current, 0);
1387 }
1388 EXPORT_SYMBOL(setup_new_exec);
1389
1390 /* Runs immediately before start_thread() takes over. */
1391 void finalize_exec(struct linux_binprm *bprm)
1392 {
1393         /* Store any stack rlimit changes before starting thread. */
1394         task_lock(current->group_leader);
1395         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1396         task_unlock(current->group_leader);
1397 }
1398 EXPORT_SYMBOL(finalize_exec);
1399
1400 /*
1401  * Prepare credentials and lock ->cred_guard_mutex.
1402  * install_exec_creds() commits the new creds and drops the lock.
1403  * Or, if exec fails before, free_bprm() should release ->cred and
1404  * and unlock.
1405  */
1406 static int prepare_bprm_creds(struct linux_binprm *bprm)
1407 {
1408         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1409                 return -ERESTARTNOINTR;
1410
1411         bprm->cred = prepare_exec_creds();
1412         if (likely(bprm->cred))
1413                 return 0;
1414
1415         mutex_unlock(&current->signal->cred_guard_mutex);
1416         return -ENOMEM;
1417 }
1418
1419 static void free_bprm(struct linux_binprm *bprm)
1420 {
1421         free_arg_pages(bprm);
1422         if (bprm->cred) {
1423                 mutex_unlock(&current->signal->cred_guard_mutex);
1424                 abort_creds(bprm->cred);
1425         }
1426         if (bprm->file) {
1427                 allow_write_access(bprm->file);
1428                 fput(bprm->file);
1429         }
1430         /* If a binfmt changed the interp, free it. */
1431         if (bprm->interp != bprm->filename)
1432                 kfree(bprm->interp);
1433         kfree(bprm);
1434 }
1435
1436 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1437 {
1438         /* If a binfmt changed the interp, free it first. */
1439         if (bprm->interp != bprm->filename)
1440                 kfree(bprm->interp);
1441         bprm->interp = kstrdup(interp, GFP_KERNEL);
1442         if (!bprm->interp)
1443                 return -ENOMEM;
1444         return 0;
1445 }
1446 EXPORT_SYMBOL(bprm_change_interp);
1447
1448 /*
1449  * install the new credentials for this executable
1450  */
1451 void install_exec_creds(struct linux_binprm *bprm)
1452 {
1453         security_bprm_committing_creds(bprm);
1454
1455         commit_creds(bprm->cred);
1456         bprm->cred = NULL;
1457
1458         /*
1459          * Disable monitoring for regular users
1460          * when executing setuid binaries. Must
1461          * wait until new credentials are committed
1462          * by commit_creds() above
1463          */
1464         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1465                 perf_event_exit_task(current);
1466         /*
1467          * cred_guard_mutex must be held at least to this point to prevent
1468          * ptrace_attach() from altering our determination of the task's
1469          * credentials; any time after this it may be unlocked.
1470          */
1471         security_bprm_committed_creds(bprm);
1472         mutex_unlock(&current->signal->cred_guard_mutex);
1473 }
1474 EXPORT_SYMBOL(install_exec_creds);
1475
1476 /*
1477  * determine how safe it is to execute the proposed program
1478  * - the caller must hold ->cred_guard_mutex to protect against
1479  *   PTRACE_ATTACH or seccomp thread-sync
1480  */
1481 static void check_unsafe_exec(struct linux_binprm *bprm)
1482 {
1483         struct task_struct *p = current, *t;
1484         unsigned n_fs;
1485
1486         if (p->ptrace)
1487                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1488
1489         /*
1490          * This isn't strictly necessary, but it makes it harder for LSMs to
1491          * mess up.
1492          */
1493         if (task_no_new_privs(current))
1494                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1495
1496         t = p;
1497         n_fs = 1;
1498         spin_lock(&p->fs->lock);
1499         rcu_read_lock();
1500         while_each_thread(p, t) {
1501                 if (t->fs == p->fs)
1502                         n_fs++;
1503         }
1504         rcu_read_unlock();
1505
1506         if (p->fs->users > n_fs)
1507                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1508         else
1509                 p->fs->in_exec = 1;
1510         spin_unlock(&p->fs->lock);
1511 }
1512
1513 static void bprm_fill_uid(struct linux_binprm *bprm)
1514 {
1515         struct inode *inode;
1516         unsigned int mode;
1517         kuid_t uid;
1518         kgid_t gid;
1519
1520         /*
1521          * Since this can be called multiple times (via prepare_binprm),
1522          * we must clear any previous work done when setting set[ug]id
1523          * bits from any earlier bprm->file uses (for example when run
1524          * first for a setuid script then again for its interpreter).
1525          */
1526         bprm->cred->euid = current_euid();
1527         bprm->cred->egid = current_egid();
1528
1529         if (!mnt_may_suid(bprm->file->f_path.mnt))
1530                 return;
1531
1532         if (task_no_new_privs(current))
1533                 return;
1534
1535         inode = bprm->file->f_path.dentry->d_inode;
1536         mode = READ_ONCE(inode->i_mode);
1537         if (!(mode & (S_ISUID|S_ISGID)))
1538                 return;
1539
1540         /* Be careful if suid/sgid is set */
1541         inode_lock(inode);
1542
1543         /* reload atomically mode/uid/gid now that lock held */
1544         mode = inode->i_mode;
1545         uid = inode->i_uid;
1546         gid = inode->i_gid;
1547         inode_unlock(inode);
1548
1549         /* We ignore suid/sgid if there are no mappings for them in the ns */
1550         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1551                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1552                 return;
1553
1554         if (mode & S_ISUID) {
1555                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1556                 bprm->cred->euid = uid;
1557         }
1558
1559         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1560                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1561                 bprm->cred->egid = gid;
1562         }
1563 }
1564
1565 /*
1566  * Fill the binprm structure from the inode.
1567  * Check permissions, then read the first BINPRM_BUF_SIZE bytes
1568  *
1569  * This may be called multiple times for binary chains (scripts for example).
1570  */
1571 int prepare_binprm(struct linux_binprm *bprm)
1572 {
1573         int retval;
1574         loff_t pos = 0;
1575
1576         bprm_fill_uid(bprm);
1577
1578         /* fill in binprm security blob */
1579         retval = security_bprm_set_creds(bprm);
1580         if (retval)
1581                 return retval;
1582         bprm->called_set_creds = 1;
1583
1584         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1585         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1586 }
1587
1588 EXPORT_SYMBOL(prepare_binprm);
1589
1590 /*
1591  * Arguments are '\0' separated strings found at the location bprm->p
1592  * points to; chop off the first by relocating brpm->p to right after
1593  * the first '\0' encountered.
1594  */
1595 int remove_arg_zero(struct linux_binprm *bprm)
1596 {
1597         int ret = 0;
1598         unsigned long offset;
1599         char *kaddr;
1600         struct page *page;
1601
1602         if (!bprm->argc)
1603                 return 0;
1604
1605         do {
1606                 offset = bprm->p & ~PAGE_MASK;
1607                 page = get_arg_page(bprm, bprm->p, 0);
1608                 if (!page) {
1609                         ret = -EFAULT;
1610                         goto out;
1611                 }
1612                 kaddr = kmap_atomic(page);
1613
1614                 for (; offset < PAGE_SIZE && kaddr[offset];
1615                                 offset++, bprm->p++)
1616                         ;
1617
1618                 kunmap_atomic(kaddr);
1619                 put_arg_page(page);
1620         } while (offset == PAGE_SIZE);
1621
1622         bprm->p++;
1623         bprm->argc--;
1624         ret = 0;
1625
1626 out:
1627         return ret;
1628 }
1629 EXPORT_SYMBOL(remove_arg_zero);
1630
1631 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1632 /*
1633  * cycle the list of binary formats handler, until one recognizes the image
1634  */
1635 int search_binary_handler(struct linux_binprm *bprm)
1636 {
1637         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1638         struct linux_binfmt *fmt;
1639         int retval;
1640
1641         /* This allows 4 levels of binfmt rewrites before failing hard. */
1642         if (bprm->recursion_depth > 5)
1643                 return -ELOOP;
1644
1645         retval = security_bprm_check(bprm);
1646         if (retval)
1647                 return retval;
1648
1649         retval = -ENOENT;
1650  retry:
1651         read_lock(&binfmt_lock);
1652         list_for_each_entry(fmt, &formats, lh) {
1653                 if (!try_module_get(fmt->module))
1654                         continue;
1655                 read_unlock(&binfmt_lock);
1656
1657                 bprm->recursion_depth++;
1658                 retval = fmt->load_binary(bprm);
1659                 bprm->recursion_depth--;
1660
1661                 read_lock(&binfmt_lock);
1662                 put_binfmt(fmt);
1663                 if (retval < 0 && !bprm->mm) {
1664                         /* we got to flush_old_exec() and failed after it */
1665                         read_unlock(&binfmt_lock);
1666                         force_sigsegv(SIGSEGV);
1667                         return retval;
1668                 }
1669                 if (retval != -ENOEXEC || !bprm->file) {
1670                         read_unlock(&binfmt_lock);
1671                         return retval;
1672                 }
1673         }
1674         read_unlock(&binfmt_lock);
1675
1676         if (need_retry) {
1677                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1678                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1679                         return retval;
1680                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1681                         return retval;
1682                 need_retry = false;
1683                 goto retry;
1684         }
1685
1686         return retval;
1687 }
1688 EXPORT_SYMBOL(search_binary_handler);
1689
1690 static int exec_binprm(struct linux_binprm *bprm)
1691 {
1692         pid_t old_pid, old_vpid;
1693         int ret;
1694
1695         /* Need to fetch pid before load_binary changes it */
1696         old_pid = current->pid;
1697         rcu_read_lock();
1698         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1699         rcu_read_unlock();
1700
1701         ret = search_binary_handler(bprm);
1702         if (ret >= 0) {
1703                 audit_bprm(bprm);
1704                 trace_sched_process_exec(current, old_pid, bprm);
1705                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1706                 proc_exec_connector(current);
1707         }
1708
1709         return ret;
1710 }
1711
1712 /*
1713  * sys_execve() executes a new program.
1714  */
1715 static int __do_execve_file(int fd, struct filename *filename,
1716                             struct user_arg_ptr argv,
1717                             struct user_arg_ptr envp,
1718                             int flags, struct file *file)
1719 {
1720         char *pathbuf = NULL;
1721         struct linux_binprm *bprm;
1722         struct files_struct *displaced;
1723         int retval;
1724
1725         if (IS_ERR(filename))
1726                 return PTR_ERR(filename);
1727
1728         /*
1729          * We move the actual failure in case of RLIMIT_NPROC excess from
1730          * set*uid() to execve() because too many poorly written programs
1731          * don't check setuid() return code.  Here we additionally recheck
1732          * whether NPROC limit is still exceeded.
1733          */
1734         if ((current->flags & PF_NPROC_EXCEEDED) &&
1735             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1736                 retval = -EAGAIN;
1737                 goto out_ret;
1738         }
1739
1740         /* We're below the limit (still or again), so we don't want to make
1741          * further execve() calls fail. */
1742         current->flags &= ~PF_NPROC_EXCEEDED;
1743
1744         retval = unshare_files(&displaced);
1745         if (retval)
1746                 goto out_ret;
1747
1748         retval = -ENOMEM;
1749         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1750         if (!bprm)
1751                 goto out_files;
1752
1753         retval = prepare_bprm_creds(bprm);
1754         if (retval)
1755                 goto out_free;
1756
1757         check_unsafe_exec(bprm);
1758         current->in_execve = 1;
1759
1760         if (!file)
1761                 file = do_open_execat(fd, filename, flags);
1762         retval = PTR_ERR(file);
1763         if (IS_ERR(file))
1764                 goto out_unmark;
1765
1766         sched_exec();
1767
1768         bprm->file = file;
1769         if (!filename) {
1770                 bprm->filename = "none";
1771         } else if (fd == AT_FDCWD || filename->name[0] == '/') {
1772                 bprm->filename = filename->name;
1773         } else {
1774                 if (filename->name[0] == '\0')
1775                         pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1776                 else
1777                         pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1778                                             fd, filename->name);
1779                 if (!pathbuf) {
1780                         retval = -ENOMEM;
1781                         goto out_unmark;
1782                 }
1783                 /*
1784                  * Record that a name derived from an O_CLOEXEC fd will be
1785                  * inaccessible after exec. Relies on having exclusive access to
1786                  * current->files (due to unshare_files above).
1787                  */
1788                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1789                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1790                 bprm->filename = pathbuf;
1791         }
1792         bprm->interp = bprm->filename;
1793
1794         retval = bprm_mm_init(bprm);
1795         if (retval)
1796                 goto out_unmark;
1797
1798         retval = prepare_arg_pages(bprm, argv, envp);
1799         if (retval < 0)
1800                 goto out;
1801
1802         retval = prepare_binprm(bprm);
1803         if (retval < 0)
1804                 goto out;
1805
1806         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1807         if (retval < 0)
1808                 goto out;
1809
1810         bprm->exec = bprm->p;
1811         retval = copy_strings(bprm->envc, envp, bprm);
1812         if (retval < 0)
1813                 goto out;
1814
1815         retval = copy_strings(bprm->argc, argv, bprm);
1816         if (retval < 0)
1817                 goto out;
1818
1819         would_dump(bprm, bprm->file);
1820
1821         retval = exec_binprm(bprm);
1822         if (retval < 0)
1823                 goto out;
1824
1825         /* execve succeeded */
1826         current->fs->in_exec = 0;
1827         current->in_execve = 0;
1828         rseq_execve(current);
1829         acct_update_integrals(current);
1830         task_numa_free(current, false);
1831         free_bprm(bprm);
1832         kfree(pathbuf);
1833         if (filename)
1834                 putname(filename);
1835         if (displaced)
1836                 put_files_struct(displaced);
1837         return retval;
1838
1839 out:
1840         if (bprm->mm) {
1841                 acct_arg_size(bprm, 0);
1842                 mmput(bprm->mm);
1843         }
1844
1845 out_unmark:
1846         current->fs->in_exec = 0;
1847         current->in_execve = 0;
1848
1849 out_free:
1850         free_bprm(bprm);
1851         kfree(pathbuf);
1852
1853 out_files:
1854         if (displaced)
1855                 reset_files_struct(displaced);
1856 out_ret:
1857         if (filename)
1858                 putname(filename);
1859         return retval;
1860 }
1861
1862 static int do_execveat_common(int fd, struct filename *filename,
1863                               struct user_arg_ptr argv,
1864                               struct user_arg_ptr envp,
1865                               int flags)
1866 {
1867         return __do_execve_file(fd, filename, argv, envp, flags, NULL);
1868 }
1869
1870 int do_execve_file(struct file *file, void *__argv, void *__envp)
1871 {
1872         struct user_arg_ptr argv = { .ptr.native = __argv };
1873         struct user_arg_ptr envp = { .ptr.native = __envp };
1874
1875         return __do_execve_file(AT_FDCWD, NULL, argv, envp, 0, file);
1876 }
1877
1878 int do_execve(struct filename *filename,
1879         const char __user *const __user *__argv,
1880         const char __user *const __user *__envp)
1881 {
1882         struct user_arg_ptr argv = { .ptr.native = __argv };
1883         struct user_arg_ptr envp = { .ptr.native = __envp };
1884         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1885 }
1886
1887 int do_execveat(int fd, struct filename *filename,
1888                 const char __user *const __user *__argv,
1889                 const char __user *const __user *__envp,
1890                 int flags)
1891 {
1892         struct user_arg_ptr argv = { .ptr.native = __argv };
1893         struct user_arg_ptr envp = { .ptr.native = __envp };
1894
1895         return do_execveat_common(fd, filename, argv, envp, flags);
1896 }
1897
1898 #ifdef CONFIG_COMPAT
1899 static int compat_do_execve(struct filename *filename,
1900         const compat_uptr_t __user *__argv,
1901         const compat_uptr_t __user *__envp)
1902 {
1903         struct user_arg_ptr argv = {
1904                 .is_compat = true,
1905                 .ptr.compat = __argv,
1906         };
1907         struct user_arg_ptr envp = {
1908                 .is_compat = true,
1909                 .ptr.compat = __envp,
1910         };
1911         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1912 }
1913
1914 static int compat_do_execveat(int fd, struct filename *filename,
1915                               const compat_uptr_t __user *__argv,
1916                               const compat_uptr_t __user *__envp,
1917                               int flags)
1918 {
1919         struct user_arg_ptr argv = {
1920                 .is_compat = true,
1921                 .ptr.compat = __argv,
1922         };
1923         struct user_arg_ptr envp = {
1924                 .is_compat = true,
1925                 .ptr.compat = __envp,
1926         };
1927         return do_execveat_common(fd, filename, argv, envp, flags);
1928 }
1929 #endif
1930
1931 void set_binfmt(struct linux_binfmt *new)
1932 {
1933         struct mm_struct *mm = current->mm;
1934
1935         if (mm->binfmt)
1936                 module_put(mm->binfmt->module);
1937
1938         mm->binfmt = new;
1939         if (new)
1940                 __module_get(new->module);
1941 }
1942 EXPORT_SYMBOL(set_binfmt);
1943
1944 /*
1945  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1946  */
1947 void set_dumpable(struct mm_struct *mm, int value)
1948 {
1949         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1950                 return;
1951
1952         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
1953 }
1954
1955 SYSCALL_DEFINE3(execve,
1956                 const char __user *, filename,
1957                 const char __user *const __user *, argv,
1958                 const char __user *const __user *, envp)
1959 {
1960         return do_execve(getname(filename), argv, envp);
1961 }
1962
1963 SYSCALL_DEFINE5(execveat,
1964                 int, fd, const char __user *, filename,
1965                 const char __user *const __user *, argv,
1966                 const char __user *const __user *, envp,
1967                 int, flags)
1968 {
1969         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1970
1971         return do_execveat(fd,
1972                            getname_flags(filename, lookup_flags, NULL),
1973                            argv, envp, flags);
1974 }
1975
1976 #ifdef CONFIG_COMPAT
1977 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1978         const compat_uptr_t __user *, argv,
1979         const compat_uptr_t __user *, envp)
1980 {
1981         return compat_do_execve(getname(filename), argv, envp);
1982 }
1983
1984 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1985                        const char __user *, filename,
1986                        const compat_uptr_t __user *, argv,
1987                        const compat_uptr_t __user *, envp,
1988                        int,  flags)
1989 {
1990         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1991
1992         return compat_do_execveat(fd,
1993                                   getname_flags(filename, lookup_flags, NULL),
1994                                   argv, envp, flags);
1995 }
1996 #endif