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/kernel_read_file.h>
  27 #include <linux/slab.h>
  28 #include <linux/file.h>
  29 #include <linux/fdtable.h>
  30 #include <linux/mm.h>
  31 #include <linux/vmacache.h>
  32 #include <linux/stat.h>
  33 #include <linux/fcntl.h>
  34 #include <linux/swap.h>
  35 #include <linux/string.h>
  36 #include <linux/init.h>
  37 #include <linux/sched/mm.h>
  38 #include <linux/sched/coredump.h>
  39 #include <linux/sched/signal.h>
  40 #include <linux/sched/numa_balancing.h>
  41 #include <linux/sched/task.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/perf_event.h>
  44 #include <linux/highmem.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/key.h>
  47 #include <linux/personality.h>
  48 #include <linux/binfmts.h>
  49 #include <linux/utsname.h>
  50 #include <linux/pid_namespace.h>
  51 #include <linux/module.h>
  52 #include <linux/namei.h>
  53 #include <linux/mount.h>
  54 #include <linux/security.h>
  55 #include <linux/syscalls.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/audit.h>
  59 #include <linux/tracehook.h>
  60 #include <linux/kmod.h>
  61 #include <linux/fsnotify.h>
  62 #include <linux/fs_struct.h>
  63 #include <linux/oom.h>
  64 #include <linux/compat.h>
  65 #include <linux/vmalloc.h>
  66 #include <linux/io_uring.h>
  67 #include <linux/syscall_user_dispatch.h>
  68
  69 #include <linux/uaccess.h>
  70 #include <asm/mmu_context.h>
  71 #include <asm/tlb.h>
  72
  73 #include <trace/events/task.h>
  74 #include "internal.h"
  75
  76 #include <trace/events/sched.h>
  77
  78 static int bprm_creds_from_file(struct linux_binprm *bprm);
  79
  80 int suid_dumpable = 0;
  81
  82 static LIST_HEAD(formats);
  83 static DEFINE_RWLOCK(binfmt_lock);
  84
  85 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  86 {
  87         BUG_ON(!fmt);
  88         if (WARN_ON(!fmt->load_binary))
  89                 return;
  90         write_lock(&binfmt_lock);
  91         insert ? list_add(&fmt->lh, &formats) :
  92                  list_add_tail(&fmt->lh, &formats);
  93         write_unlock(&binfmt_lock);
  94 }
  95
  96 EXPORT_SYMBOL(__register_binfmt);
  97
  98 void unregister_binfmt(struct linux_binfmt * fmt)
  99 {
 100         write_lock(&binfmt_lock);
 101         list_del(&fmt->lh);
 102         write_unlock(&binfmt_lock);
 103 }
 104
 105 EXPORT_SYMBOL(unregister_binfmt);
 106
 107 static inline void put_binfmt(struct linux_binfmt * fmt)
 108 {
 109         module_put(fmt->module);
 110 }
 111
 112 bool path_noexec(const struct path *path)
 113 {
 114         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 115                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 116 }
 117
 118 #ifdef CONFIG_USELIB
 119 /*
 120  * Note that a shared library must be both readable and executable due to
 121  * security reasons.
 122  *
 123  * Also note that we take the address to load from from the file itself.
 124  */
 125 SYSCALL_DEFINE1(uselib, const char __user *, library)
 126 {
 127         struct linux_binfmt *fmt;
 128         struct file *file;
 129         struct filename *tmp = getname(library);
 130         int error = PTR_ERR(tmp);
 131         static const struct open_flags uselib_flags = {
 132                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 133                 .acc_mode = MAY_READ | MAY_EXEC,
 134                 .intent = LOOKUP_OPEN,
 135                 .lookup_flags = LOOKUP_FOLLOW,
 136         };
 137
 138         if (IS_ERR(tmp))
 139                 goto out;
 140
 141         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 142         putname(tmp);
 143         error = PTR_ERR(file);
 144         if (IS_ERR(file))
 145                 goto out;
 146
 147         /*
 148          * may_open() has already checked for this, so it should be
 149          * impossible to trip now. But we need to be extra cautious
 150          * and check again at the very end too.
 151          */
 152         error = -EACCES;
 153         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 154                          path_noexec(&file->f_path)))
 155                 goto exit;
 156
 157         fsnotify_open(file);
 158
 159         error = -ENOEXEC;
 160
 161         read_lock(&binfmt_lock);
 162         list_for_each_entry(fmt, &formats, lh) {
 163                 if (!fmt->load_shlib)
 164                         continue;
 165                 if (!try_module_get(fmt->module))
 166                         continue;
 167                 read_unlock(&binfmt_lock);
 168                 error = fmt->load_shlib(file);
 169                 read_lock(&binfmt_lock);
 170                 put_binfmt(fmt);
 171                 if (error != -ENOEXEC)
 172                         break;
 173         }
 174         read_unlock(&binfmt_lock);
 175 exit:
 176         fput(file);
 177 out:
 178         return error;
 179 }
 180 #endif /* #ifdef CONFIG_USELIB */
 181
 182 #ifdef CONFIG_MMU
 183 /*
 184  * The nascent bprm->mm is not visible until exec_mmap() but it can
 185  * use a lot of memory, account these pages in current->mm temporary
 186  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 187  * change the counter back via acct_arg_size(0).
 188  */
 189 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 190 {
 191         struct mm_struct *mm = current->mm;
 192         long diff = (long)(pages - bprm->vma_pages);
 193
 194         if (!mm || !diff)
 195                 return;
 196
 197         bprm->vma_pages = pages;
 198         add_mm_counter(mm, MM_ANONPAGES, diff);
 199 }
 200
 201 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 202                 int write)
 203 {
 204         struct page *page;
 205         int ret;
 206         unsigned int gup_flags = FOLL_FORCE;
 207
 208 #ifdef CONFIG_STACK_GROWSUP
 209         if (write) {
 210                 ret = expand_downwards(bprm->vma, pos);
 211                 if (ret < 0)
 212                         return NULL;
 213         }
 214 #endif
 215
 216         if (write)
 217                 gup_flags |= FOLL_WRITE;
 218
 219         /*
 220          * We are doing an exec().  'current' is the process
 221          * doing the exec and bprm->mm is the new process's mm.
 222          */
 223         ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
 224                         &page, NULL, NULL);
 225         if (ret <= 0)
 226                 return NULL;
 227
 228         if (write)
 229                 acct_arg_size(bprm, vma_pages(bprm->vma));
 230
 231         return page;
 232 }
 233
 234 static void put_arg_page(struct page *page)
 235 {
 236         put_page(page);
 237 }
 238
 239 static void free_arg_pages(struct linux_binprm *bprm)
 240 {
 241 }
 242
 243 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 244                 struct page *page)
 245 {
 246         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 247 }
 248
 249 static int __bprm_mm_init(struct linux_binprm *bprm)
 250 {
 251         int err;
 252         struct vm_area_struct *vma = NULL;
 253         struct mm_struct *mm = bprm->mm;
 254
 255         bprm->vma = vma = vm_area_alloc(mm);
 256         if (!vma)
 257                 return -ENOMEM;
 258         vma_set_anonymous(vma);
 259
 260         if (mmap_write_lock_killable(mm)) {
 261                 err = -EINTR;
 262                 goto err_free;
 263         }
 264
 265         /*
 266          * Place the stack at the largest stack address the architecture
 267          * supports. Later, we'll move this to an appropriate place. We don't
 268          * use STACK_TOP because that can depend on attributes which aren't
 269          * configured yet.
 270          */
 271         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 272         vma->vm_end = STACK_TOP_MAX;
 273         vma->vm_start = vma->vm_end - PAGE_SIZE;
 274         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 275         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 276
 277         err = insert_vm_struct(mm, vma);
 278         if (err)
 279                 goto err;
 280
 281         mm->stack_vm = mm->total_vm = 1;
 282         mmap_write_unlock(mm);
 283         bprm->p = vma->vm_end - sizeof(void *);
 284         return 0;
 285 err:
 286         mmap_write_unlock(mm);
 287 err_free:
 288         bprm->vma = NULL;
 289         vm_area_free(vma);
 290         return err;
 291 }
 292
 293 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 294 {
 295         return len <= MAX_ARG_STRLEN;
 296 }
 297
 298 #else
 299
 300 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 301 {
 302 }
 303
 304 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 305                 int write)
 306 {
 307         struct page *page;
 308
 309         page = bprm->page[pos / PAGE_SIZE];
 310         if (!page && write) {
 311                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 312                 if (!page)
 313                         return NULL;
 314                 bprm->page[pos / PAGE_SIZE] = page;
 315         }
 316
 317         return page;
 318 }
 319
 320 static void put_arg_page(struct page *page)
 321 {
 322 }
 323
 324 static void free_arg_page(struct linux_binprm *bprm, int i)
 325 {
 326         if (bprm->page[i]) {
 327                 __free_page(bprm->page[i]);
 328                 bprm->page[i] = NULL;
 329         }
 330 }
 331
 332 static void free_arg_pages(struct linux_binprm *bprm)
 333 {
 334         int i;
 335
 336         for (i = 0; i < MAX_ARG_PAGES; i++)
 337                 free_arg_page(bprm, i);
 338 }
 339
 340 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 341                 struct page *page)
 342 {
 343 }
 344
 345 static int __bprm_mm_init(struct linux_binprm *bprm)
 346 {
 347         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 348         return 0;
 349 }
 350
 351 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 352 {
 353         return len <= bprm->p;
 354 }
 355
 356 #endif /* CONFIG_MMU */
 357
 358 /*
 359  * Create a new mm_struct and populate it with a temporary stack
 360  * vm_area_struct.  We don't have enough context at this point to set the stack
 361  * flags, permissions, and offset, so we use temporary values.  We'll update
 362  * them later in setup_arg_pages().
 363  */
 364 static int bprm_mm_init(struct linux_binprm *bprm)
 365 {
 366         int err;
 367         struct mm_struct *mm = NULL;
 368
 369         bprm->mm = mm = mm_alloc();
 370         err = -ENOMEM;
 371         if (!mm)
 372                 goto err;
 373
 374         /* Save current stack limit for all calculations made during exec. */
 375         task_lock(current->group_leader);
 376         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 377         task_unlock(current->group_leader);
 378
 379         err = __bprm_mm_init(bprm);
 380         if (err)
 381                 goto err;
 382
 383         return 0;
 384
 385 err:
 386         if (mm) {
 387                 bprm->mm = NULL;
 388                 mmdrop(mm);
 389         }
 390
 391         return err;
 392 }
 393
 394 struct user_arg_ptr {
 395 #ifdef CONFIG_COMPAT
 396         bool is_compat;
 397 #endif
 398         union {
 399                 const char __user *const __user *native;
 400 #ifdef CONFIG_COMPAT
 401                 const compat_uptr_t __user *compat;
 402 #endif
 403         } ptr;
 404 };
 405
 406 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 407 {
 408         const char __user *native;
 409
 410 #ifdef CONFIG_COMPAT
 411         if (unlikely(argv.is_compat)) {
 412                 compat_uptr_t compat;
 413
 414                 if (get_user(compat, argv.ptr.compat + nr))
 415                         return ERR_PTR(-EFAULT);
 416
 417                 return compat_ptr(compat);
 418         }
 419 #endif
 420
 421         if (get_user(native, argv.ptr.native + nr))
 422                 return ERR_PTR(-EFAULT);
 423
 424         return native;
 425 }
 426
 427 /*
 428  * count() counts the number of strings in array ARGV.
 429  */
 430 static int count(struct user_arg_ptr argv, int max)
 431 {
 432         int i = 0;
 433
 434         if (argv.ptr.native != NULL) {
 435                 for (;;) {
 436                         const char __user *p = get_user_arg_ptr(argv, i);
 437
 438                         if (!p)
 439                                 break;
 440
 441                         if (IS_ERR(p))
 442                                 return -EFAULT;
 443
 444                         if (i >= max)
 445                                 return -E2BIG;
 446                         ++i;
 447
 448                         if (fatal_signal_pending(current))
 449                                 return -ERESTARTNOHAND;
 450                         cond_resched();
 451                 }
 452         }
 453         return i;
 454 }
 455
 456 static int count_strings_kernel(const char *const *argv)
 457 {
 458         int i;
 459
 460         if (!argv)
 461                 return 0;
 462
 463         for (i = 0; argv[i]; ++i) {
 464                 if (i >= MAX_ARG_STRINGS)
 465                         return -E2BIG;
 466                 if (fatal_signal_pending(current))
 467                         return -ERESTARTNOHAND;
 468                 cond_resched();
 469         }
 470         return i;
 471 }
 472
 473 static int bprm_stack_limits(struct linux_binprm *bprm)
 474 {
 475         unsigned long limit, ptr_size;
 476
 477         /*
 478          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 479          * (whichever is smaller) for the argv+env strings.
 480          * This ensures that:
 481          *  - the remaining binfmt code will not run out of stack space,
 482          *  - the program will have a reasonable amount of stack left
 483          *    to work from.
 484          */
 485         limit = _STK_LIM / 4 * 3;
 486         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 487         /*
 488          * We've historically supported up to 32 pages (ARG_MAX)
 489          * of argument strings even with small stacks
 490          */
 491         limit = max_t(unsigned long, limit, ARG_MAX);
 492         /*
 493          * We must account for the size of all the argv and envp pointers to
 494          * the argv and envp strings, since they will also take up space in
 495          * the stack. They aren't stored until much later when we can't
 496          * signal to the parent that the child has run out of stack space.
 497          * Instead, calculate it here so it's possible to fail gracefully.
 498          */
 499         ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
 500         if (limit <= ptr_size)
 501                 return -E2BIG;
 502         limit -= ptr_size;
 503
 504         bprm->argmin = bprm->p - limit;
 505         return 0;
 506 }
 507
 508 /*
 509  * 'copy_strings()' copies argument/environment strings from the old
 510  * processes's memory to the new process's stack.  The call to get_user_pages()
 511  * ensures the destination page is created and not swapped out.
 512  */
 513 static int copy_strings(int argc, struct user_arg_ptr argv,
 514                         struct linux_binprm *bprm)
 515 {
 516         struct page *kmapped_page = NULL;
 517         char *kaddr = NULL;
 518         unsigned long kpos = 0;
 519         int ret;
 520
 521         while (argc-- > 0) {
 522                 const char __user *str;
 523                 int len;
 524                 unsigned long pos;
 525
 526                 ret = -EFAULT;
 527                 str = get_user_arg_ptr(argv, argc);
 528                 if (IS_ERR(str))
 529                         goto out;
 530
 531                 len = strnlen_user(str, MAX_ARG_STRLEN);
 532                 if (!len)
 533                         goto out;
 534
 535                 ret = -E2BIG;
 536                 if (!valid_arg_len(bprm, len))
 537                         goto out;
 538
 539                 /* We're going to work our way backwords. */
 540                 pos = bprm->p;
 541                 str += len;
 542                 bprm->p -= len;
 543 #ifdef CONFIG_MMU
 544                 if (bprm->p < bprm->argmin)
 545                         goto out;
 546 #endif
 547
 548                 while (len > 0) {
 549                         int offset, bytes_to_copy;
 550
 551                         if (fatal_signal_pending(current)) {
 552                                 ret = -ERESTARTNOHAND;
 553                                 goto out;
 554                         }
 555                         cond_resched();
 556
 557                         offset = pos % PAGE_SIZE;
 558                         if (offset == 0)
 559                                 offset = PAGE_SIZE;
 560
 561                         bytes_to_copy = offset;
 562                         if (bytes_to_copy > len)
 563                                 bytes_to_copy = len;
 564
 565                         offset -= bytes_to_copy;
 566                         pos -= bytes_to_copy;
 567                         str -= bytes_to_copy;
 568                         len -= bytes_to_copy;
 569
 570                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 571                                 struct page *page;
 572
 573                                 page = get_arg_page(bprm, pos, 1);
 574                                 if (!page) {
 575                                         ret = -E2BIG;
 576                                         goto out;
 577                                 }
 578
 579                                 if (kmapped_page) {
 580                                         flush_kernel_dcache_page(kmapped_page);
 581                                         kunmap(kmapped_page);
 582                                         put_arg_page(kmapped_page);
 583                                 }
 584                                 kmapped_page = page;
 585                                 kaddr = kmap(kmapped_page);
 586                                 kpos = pos & PAGE_MASK;
 587                                 flush_arg_page(bprm, kpos, kmapped_page);
 588                         }
 589                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 590                                 ret = -EFAULT;
 591                                 goto out;
 592                         }
 593                 }
 594         }
 595         ret = 0;
 596 out:
 597         if (kmapped_page) {
 598                 flush_kernel_dcache_page(kmapped_page);
 599                 kunmap(kmapped_page);
 600                 put_arg_page(kmapped_page);
 601         }
 602         return ret;
 603 }
 604
 605 /*
 606  * Copy and argument/environment string from the kernel to the processes stack.
 607  */
 608 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
 609 {
 610         int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
 611         unsigned long pos = bprm->p;
 612
 613         if (len == 0)
 614                 return -EFAULT;
 615         if (!valid_arg_len(bprm, len))
 616                 return -E2BIG;
 617
 618         /* We're going to work our way backwards. */
 619         arg += len;
 620         bprm->p -= len;
 621         if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
 622                 return -E2BIG;
 623
 624         while (len > 0) {
 625                 unsigned int bytes_to_copy = min_t(unsigned int, len,
 626                                 min_not_zero(offset_in_page(pos), PAGE_SIZE));
 627                 struct page *page;
 628                 char *kaddr;
 629
 630                 pos -= bytes_to_copy;
 631                 arg -= bytes_to_copy;
 632                 len -= bytes_to_copy;
 633
 634                 page = get_arg_page(bprm, pos, 1);
 635                 if (!page)
 636                         return -E2BIG;
 637                 kaddr = kmap_atomic(page);
 638                 flush_arg_page(bprm, pos & PAGE_MASK, page);
 639                 memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
 640                 flush_kernel_dcache_page(page);
 641                 kunmap_atomic(kaddr);
 642                 put_arg_page(page);
 643         }
 644
 645         return 0;
 646 }
 647 EXPORT_SYMBOL(copy_string_kernel);
 648
 649 static int copy_strings_kernel(int argc, const char *const *argv,
 650                                struct linux_binprm *bprm)
 651 {
 652         while (argc-- > 0) {
 653                 int ret = copy_string_kernel(argv[argc], bprm);
 654                 if (ret < 0)
 655                         return ret;
 656                 if (fatal_signal_pending(current))
 657                         return -ERESTARTNOHAND;
 658                 cond_resched();
 659         }
 660         return 0;
 661 }
 662
 663 #ifdef CONFIG_MMU
 664
 665 /*
 666  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 667  * the binfmt code determines where the new stack should reside, we shift it to
 668  * its final location.  The process proceeds as follows:
 669  *
 670  * 1) Use shift to calculate the new vma endpoints.
 671  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 672  *    arguments passed to subsequent functions are consistent.
 673  * 3) Move vma's page tables to the new range.
 674  * 4) Free up any cleared pgd range.
 675  * 5) Shrink the vma to cover only the new range.
 676  */
 677 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 678 {
 679         struct mm_struct *mm = vma->vm_mm;
 680         unsigned long old_start = vma->vm_start;
 681         unsigned long old_end = vma->vm_end;
 682         unsigned long length = old_end - old_start;
 683         unsigned long new_start = old_start - shift;
 684         unsigned long new_end = old_end - shift;
 685         struct mmu_gather tlb;
 686
 687         BUG_ON(new_start > new_end);
 688
 689         /*
 690          * ensure there are no vmas between where we want to go
 691          * and where we are
 692          */
 693         if (vma != find_vma(mm, new_start))
 694                 return -EFAULT;
 695
 696         /*
 697          * cover the whole range: [new_start, old_end)
 698          */
 699         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 700                 return -ENOMEM;
 701
 702         /*
 703          * move the page tables downwards, on failure we rely on
 704          * process cleanup to remove whatever mess we made.
 705          */
 706         if (length != move_page_tables(vma, old_start,
 707                                        vma, new_start, length, false))
 708                 return -ENOMEM;
 709
 710         lru_add_drain();
 711         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 712         if (new_end > old_start) {
 713                 /*
 714                  * when the old and new regions overlap clear from new_end.
 715                  */
 716                 free_pgd_range(&tlb, new_end, old_end, new_end,
 717                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 718         } else {
 719                 /*
 720                  * otherwise, clean from old_start; this is done to not touch
 721                  * the address space in [new_end, old_start) some architectures
 722                  * have constraints on va-space that make this illegal (IA64) -
 723                  * for the others its just a little faster.
 724                  */
 725                 free_pgd_range(&tlb, old_start, old_end, new_end,
 726                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 727         }
 728         tlb_finish_mmu(&tlb, old_start, old_end);
 729
 730         /*
 731          * Shrink the vma to just the new range.  Always succeeds.
 732          */
 733         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 734
 735         return 0;
 736 }
 737
 738 /*
 739  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 740  * the stack is optionally relocated, and some extra space is added.
 741  */
 742 int setup_arg_pages(struct linux_binprm *bprm,
 743                     unsigned long stack_top,
 744                     int executable_stack)
 745 {
 746         unsigned long ret;
 747         unsigned long stack_shift;
 748         struct mm_struct *mm = current->mm;
 749         struct vm_area_struct *vma = bprm->vma;
 750         struct vm_area_struct *prev = NULL;
 751         unsigned long vm_flags;
 752         unsigned long stack_base;
 753         unsigned long stack_size;
 754         unsigned long stack_expand;
 755         unsigned long rlim_stack;
 756
 757 #ifdef CONFIG_STACK_GROWSUP
 758         /* Limit stack size */
 759         stack_base = bprm->rlim_stack.rlim_max;
 760         if (stack_base > STACK_SIZE_MAX)
 761                 stack_base = STACK_SIZE_MAX;
 762
 763         /* Add space for stack randomization. */
 764         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 765
 766         /* Make sure we didn't let the argument array grow too large. */
 767         if (vma->vm_end - vma->vm_start > stack_base)
 768                 return -ENOMEM;
 769
 770         stack_base = PAGE_ALIGN(stack_top - stack_base);
 771
 772         stack_shift = vma->vm_start - stack_base;
 773         mm->arg_start = bprm->p - stack_shift;
 774         bprm->p = vma->vm_end - stack_shift;
 775 #else
 776         stack_top = arch_align_stack(stack_top);
 777         stack_top = PAGE_ALIGN(stack_top);
 778
 779         if (unlikely(stack_top < mmap_min_addr) ||
 780             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 781                 return -ENOMEM;
 782
 783         stack_shift = vma->vm_end - stack_top;
 784
 785         bprm->p -= stack_shift;
 786         mm->arg_start = bprm->p;
 787 #endif
 788
 789         if (bprm->loader)
 790                 bprm->loader -= stack_shift;
 791         bprm->exec -= stack_shift;
 792
 793         if (mmap_write_lock_killable(mm))
 794                 return -EINTR;
 795
 796         vm_flags = VM_STACK_FLAGS;
 797
 798         /*
 799          * Adjust stack execute permissions; explicitly enable for
 800          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 801          * (arch default) otherwise.
 802          */
 803         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 804                 vm_flags |= VM_EXEC;
 805         else if (executable_stack == EXSTACK_DISABLE_X)
 806                 vm_flags &= ~VM_EXEC;
 807         vm_flags |= mm->def_flags;
 808         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 809
 810         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 811                         vm_flags);
 812         if (ret)
 813                 goto out_unlock;
 814         BUG_ON(prev != vma);
 815
 816         if (unlikely(vm_flags & VM_EXEC)) {
 817                 pr_warn_once("process '%pD4' started with executable stack\n",
 818                              bprm->file);
 819         }
 820
 821         /* Move stack pages down in memory. */
 822         if (stack_shift) {
 823                 ret = shift_arg_pages(vma, stack_shift);
 824                 if (ret)
 825                         goto out_unlock;
 826         }
 827
 828         /* mprotect_fixup is overkill to remove the temporary stack flags */
 829         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 830
 831         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 832         stack_size = vma->vm_end - vma->vm_start;
 833         /*
 834          * Align this down to a page boundary as expand_stack
 835          * will align it up.
 836          */
 837         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 838 #ifdef CONFIG_STACK_GROWSUP
 839         if (stack_size + stack_expand > rlim_stack)
 840                 stack_base = vma->vm_start + rlim_stack;
 841         else
 842                 stack_base = vma->vm_end + stack_expand;
 843 #else
 844         if (stack_size + stack_expand > rlim_stack)
 845                 stack_base = vma->vm_end - rlim_stack;
 846         else
 847                 stack_base = vma->vm_start - stack_expand;
 848 #endif
 849         current->mm->start_stack = bprm->p;
 850         ret = expand_stack(vma, stack_base);
 851         if (ret)
 852                 ret = -EFAULT;
 853
 854 out_unlock:
 855         mmap_write_unlock(mm);
 856         return ret;
 857 }
 858 EXPORT_SYMBOL(setup_arg_pages);
 859
 860 #else
 861
 862 /*
 863  * Transfer the program arguments and environment from the holding pages
 864  * onto the stack. The provided stack pointer is adjusted accordingly.
 865  */
 866 int transfer_args_to_stack(struct linux_binprm *bprm,
 867                            unsigned long *sp_location)
 868 {
 869         unsigned long index, stop, sp;
 870         int ret = 0;
 871
 872         stop = bprm->p >> PAGE_SHIFT;
 873         sp = *sp_location;
 874
 875         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 876                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 877                 char *src = kmap(bprm->page[index]) + offset;
 878                 sp -= PAGE_SIZE - offset;
 879                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 880                         ret = -EFAULT;
 881                 kunmap(bprm->page[index]);
 882                 if (ret)
 883                         goto out;
 884         }
 885
 886         *sp_location = sp;
 887
 888 out:
 889         return ret;
 890 }
 891 EXPORT_SYMBOL(transfer_args_to_stack);
 892
 893 #endif /* CONFIG_MMU */
 894
 895 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 896 {
 897         struct file *file;
 898         int err;
 899         struct open_flags open_exec_flags = {
 900                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 901                 .acc_mode = MAY_EXEC,
 902                 .intent = LOOKUP_OPEN,
 903                 .lookup_flags = LOOKUP_FOLLOW,
 904         };
 905
 906         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 907                 return ERR_PTR(-EINVAL);
 908         if (flags & AT_SYMLINK_NOFOLLOW)
 909                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 910         if (flags & AT_EMPTY_PATH)
 911                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 912
 913         file = do_filp_open(fd, name, &open_exec_flags);
 914         if (IS_ERR(file))
 915                 goto out;
 916
 917         /*
 918          * may_open() has already checked for this, so it should be
 919          * impossible to trip now. But we need to be extra cautious
 920          * and check again at the very end too.
 921          */
 922         err = -EACCES;
 923         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 924                          path_noexec(&file->f_path)))
 925                 goto exit;
 926
 927         err = deny_write_access(file);
 928         if (err)
 929                 goto exit;
 930
 931         if (name->name[0] != '\0')
 932                 fsnotify_open(file);
 933
 934 out:
 935         return file;
 936
 937 exit:
 938         fput(file);
 939         return ERR_PTR(err);
 940 }
 941
 942 struct file *open_exec(const char *name)
 943 {
 944         struct filename *filename = getname_kernel(name);
 945         struct file *f = ERR_CAST(filename);
 946
 947         if (!IS_ERR(filename)) {
 948                 f = do_open_execat(AT_FDCWD, filename, 0);
 949                 putname(filename);
 950         }
 951         return f;
 952 }
 953 EXPORT_SYMBOL(open_exec);
 954
 955 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
 956     defined(CONFIG_BINFMT_ELF_FDPIC)
 957 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 958 {
 959         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 960         if (res > 0)
 961                 flush_icache_user_range(addr, addr + len);
 962         return res;
 963 }
 964 EXPORT_SYMBOL(read_code);
 965 #endif
 966
 967 /*
 968  * Maps the mm_struct mm into the current task struct.
 969  * On success, this function returns with the mutex
 970  * exec_update_mutex locked.
 971  */
 972 static int exec_mmap(struct mm_struct *mm)
 973 {
 974         struct task_struct *tsk;
 975         struct mm_struct *old_mm, *active_mm;
 976         int ret;
 977
 978         /* Notify parent that we're no longer interested in the old VM */
 979         tsk = current;
 980         old_mm = current->mm;
 981         exec_mm_release(tsk, old_mm);
 982         if (old_mm)
 983                 sync_mm_rss(old_mm);
 984
 985         ret = mutex_lock_killable(&tsk->signal->exec_update_mutex);
 986         if (ret)
 987                 return ret;
 988
 989         if (old_mm) {
 990                 /*
 991                  * Make sure that if there is a core dump in progress
 992                  * for the old mm, we get out and die instead of going
 993                  * through with the exec.  We must hold mmap_lock around
 994                  * checking core_state and changing tsk->mm.
 995                  */
 996                 mmap_read_lock(old_mm);
 997                 if (unlikely(old_mm->core_state)) {
 998                         mmap_read_unlock(old_mm);
 999                         mutex_unlock(&tsk->signal->exec_update_mutex);
1000                         return -EINTR;
1001                 }
1002         }
1003
1004         task_lock(tsk);
1005         membarrier_exec_mmap(mm);
1006
1007         local_irq_disable();
1008         active_mm = tsk->active_mm;
1009         tsk->active_mm = mm;
1010         tsk->mm = mm;
1011         /*
1012          * This prevents preemption while active_mm is being loaded and
1013          * it and mm are being updated, which could cause problems for
1014          * lazy tlb mm refcounting when these are updated by context
1015          * switches. Not all architectures can handle irqs off over
1016          * activate_mm yet.
1017          */
1018         if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1019                 local_irq_enable();
1020         activate_mm(active_mm, mm);
1021         if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1022                 local_irq_enable();
1023         tsk->mm->vmacache_seqnum = 0;
1024         vmacache_flush(tsk);
1025         task_unlock(tsk);
1026         if (old_mm) {
1027                 mmap_read_unlock(old_mm);
1028                 BUG_ON(active_mm != old_mm);
1029                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1030                 mm_update_next_owner(old_mm);
1031                 mmput(old_mm);
1032                 return 0;
1033         }
1034         mmdrop(active_mm);
1035         return 0;
1036 }
1037
1038 static int de_thread(struct task_struct *tsk)
1039 {
1040         struct signal_struct *sig = tsk->signal;
1041         struct sighand_struct *oldsighand = tsk->sighand;
1042         spinlock_t *lock = &oldsighand->siglock;
1043
1044         if (thread_group_empty(tsk))
1045                 goto no_thread_group;
1046
1047         /*
1048          * Kill all other threads in the thread group.
1049          */
1050         spin_lock_irq(lock);
1051         if (signal_group_exit(sig)) {
1052                 /*
1053                  * Another group action in progress, just
1054                  * return so that the signal is processed.
1055                  */
1056                 spin_unlock_irq(lock);
1057                 return -EAGAIN;
1058         }
1059
1060         sig->group_exit_task = tsk;
1061         sig->notify_count = zap_other_threads(tsk);
1062         if (!thread_group_leader(tsk))
1063                 sig->notify_count--;
1064
1065         while (sig->notify_count) {
1066                 __set_current_state(TASK_KILLABLE);
1067                 spin_unlock_irq(lock);
1068                 schedule();
1069                 if (__fatal_signal_pending(tsk))
1070                         goto killed;
1071                 spin_lock_irq(lock);
1072         }
1073         spin_unlock_irq(lock);
1074
1075         /*
1076          * At this point all other threads have exited, all we have to
1077          * do is to wait for the thread group leader to become inactive,
1078          * and to assume its PID:
1079          */
1080         if (!thread_group_leader(tsk)) {
1081                 struct task_struct *leader = tsk->group_leader;
1082
1083                 for (;;) {
1084                         cgroup_threadgroup_change_begin(tsk);
1085                         write_lock_irq(&tasklist_lock);
1086                         /*
1087                          * Do this under tasklist_lock to ensure that
1088                          * exit_notify() can't miss ->group_exit_task
1089                          */
1090                         sig->notify_count = -1;
1091                         if (likely(leader->exit_state))
1092                                 break;
1093                         __set_current_state(TASK_KILLABLE);
1094                         write_unlock_irq(&tasklist_lock);
1095                         cgroup_threadgroup_change_end(tsk);
1096                         schedule();
1097                         if (__fatal_signal_pending(tsk))
1098                                 goto killed;
1099                 }
1100
1101                 /*
1102                  * The only record we have of the real-time age of a
1103                  * process, regardless of execs it's done, is start_time.
1104                  * All the past CPU time is accumulated in signal_struct
1105                  * from sister threads now dead.  But in this non-leader
1106                  * exec, nothing survives from the original leader thread,
1107                  * whose birth marks the true age of this process now.
1108                  * When we take on its identity by switching to its PID, we
1109                  * also take its birthdate (always earlier than our own).
1110                  */
1111                 tsk->start_time = leader->start_time;
1112                 tsk->start_boottime = leader->start_boottime;
1113
1114                 BUG_ON(!same_thread_group(leader, tsk));
1115                 /*
1116                  * An exec() starts a new thread group with the
1117                  * TGID of the previous thread group. Rehash the
1118                  * two threads with a switched PID, and release
1119                  * the former thread group leader:
1120                  */
1121
1122                 /* Become a process group leader with the old leader's pid.
1123                  * The old leader becomes a thread of the this thread group.
1124                  */
1125                 exchange_tids(tsk, leader);
1126                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1127                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1128                 transfer_pid(leader, tsk, PIDTYPE_SID);
1129
1130                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1131                 list_replace_init(&leader->sibling, &tsk->sibling);
1132
1133                 tsk->group_leader = tsk;
1134                 leader->group_leader = tsk;
1135
1136                 tsk->exit_signal = SIGCHLD;
1137                 leader->exit_signal = -1;
1138
1139                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1140                 leader->exit_state = EXIT_DEAD;
1141
1142                 /*
1143                  * We are going to release_task()->ptrace_unlink() silently,
1144                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1145                  * the tracer wont't block again waiting for this thread.
1146                  */
1147                 if (unlikely(leader->ptrace))
1148                         __wake_up_parent(leader, leader->parent);
1149                 write_unlock_irq(&tasklist_lock);
1150                 cgroup_threadgroup_change_end(tsk);
1151
1152                 release_task(leader);
1153         }
1154
1155         sig->group_exit_task = NULL;
1156         sig->notify_count = 0;
1157
1158 no_thread_group:
1159         /* we have changed execution domain */
1160         tsk->exit_signal = SIGCHLD;
1161
1162         BUG_ON(!thread_group_leader(tsk));
1163         return 0;
1164
1165 killed:
1166         /* protects against exit_notify() and __exit_signal() */
1167         read_lock(&tasklist_lock);
1168         sig->group_exit_task = NULL;
1169         sig->notify_count = 0;
1170         read_unlock(&tasklist_lock);
1171         return -EAGAIN;
1172 }
1173
1174
1175 /*
1176  * This function makes sure the current process has its own signal table,
1177  * so that flush_signal_handlers can later reset the handlers without
1178  * disturbing other processes.  (Other processes might share the signal
1179  * table via the CLONE_SIGHAND option to clone().)
1180  */
1181 static int unshare_sighand(struct task_struct *me)
1182 {
1183         struct sighand_struct *oldsighand = me->sighand;
1184
1185         if (refcount_read(&oldsighand->count) != 1) {
1186                 struct sighand_struct *newsighand;
1187                 /*
1188                  * This ->sighand is shared with the CLONE_SIGHAND
1189                  * but not CLONE_THREAD task, switch to the new one.
1190                  */
1191                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1192                 if (!newsighand)
1193                         return -ENOMEM;
1194
1195                 refcount_set(&newsighand->count, 1);
1196                 memcpy(newsighand->action, oldsighand->action,
1197                        sizeof(newsighand->action));
1198
1199                 write_lock_irq(&tasklist_lock);
1200                 spin_lock(&oldsighand->siglock);
1201                 rcu_assign_pointer(me->sighand, newsighand);
1202                 spin_unlock(&oldsighand->siglock);
1203                 write_unlock_irq(&tasklist_lock);
1204
1205                 __cleanup_sighand(oldsighand);
1206         }
1207         return 0;
1208 }
1209
1210 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1211 {
1212         task_lock(tsk);
1213         strncpy(buf, tsk->comm, buf_size);
1214         task_unlock(tsk);
1215         return buf;
1216 }
1217 EXPORT_SYMBOL_GPL(__get_task_comm);
1218
1219 /*
1220  * These functions flushes out all traces of the currently running executable
1221  * so that a new one can be started
1222  */
1223
1224 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1225 {
1226         task_lock(tsk);
1227         trace_task_rename(tsk, buf);
1228         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1229         task_unlock(tsk);
1230         perf_event_comm(tsk, exec);
1231 }
1232
1233 /*
1234  * Calling this is the point of no return. None of the failures will be
1235  * seen by userspace since either the process is already taking a fatal
1236  * signal (via de_thread() or coredump), or will have SEGV raised
1237  * (after exec_mmap()) by search_binary_handler (see below).
1238  */
1239 int begin_new_exec(struct linux_binprm * bprm)
1240 {
1241         struct task_struct *me = current;
1242         int retval;
1243
1244         /* Once we are committed compute the creds */
1245         retval = bprm_creds_from_file(bprm);
1246         if (retval)
1247                 return retval;
1248
1249         /*
1250          * Ensure all future errors are fatal.
1251          */
1252         bprm->point_of_no_return = true;
1253
1254         /*
1255          * Make this the only thread in the thread group.
1256          */
1257         retval = de_thread(me);
1258         if (retval)
1259                 goto out;
1260
1261         /*
1262          * Must be called _before_ exec_mmap() as bprm->mm is
1263          * not visibile until then. This also enables the update
1264          * to be lockless.
1265          */
1266         set_mm_exe_file(bprm->mm, bprm->file);
1267
1268         /* If the binary is not readable then enforce mm->dumpable=0 */
1269         would_dump(bprm, bprm->file);
1270         if (bprm->have_execfd)
1271                 would_dump(bprm, bprm->executable);
1272
1273         /*
1274          * Release all of the old mmap stuff
1275          */
1276         acct_arg_size(bprm, 0);
1277         retval = exec_mmap(bprm->mm);
1278         if (retval)
1279                 goto out;
1280
1281         bprm->mm = NULL;
1282
1283 #ifdef CONFIG_POSIX_TIMERS
1284         exit_itimers(me->signal);
1285         flush_itimer_signals();
1286 #endif
1287
1288         /*
1289          * Make the signal table private.
1290          */
1291         retval = unshare_sighand(me);
1292         if (retval)
1293                 goto out_unlock;
1294
1295         /*
1296          * Ensure that the uaccess routines can actually operate on userspace
1297          * pointers:
1298          */
1299         force_uaccess_begin();
1300
1301         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1302                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1303         flush_thread();
1304         me->personality &= ~bprm->per_clear;
1305
1306         clear_syscall_work_syscall_user_dispatch(me);
1307
1308         /*
1309          * We have to apply CLOEXEC before we change whether the process is
1310          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1311          * trying to access the should-be-closed file descriptors of a process
1312          * undergoing exec(2).
1313          */
1314         do_close_on_exec(me->files);
1315
1316         if (bprm->secureexec) {
1317                 /* Make sure parent cannot signal privileged process. */
1318                 me->pdeath_signal = 0;
1319
1320                 /*
1321                  * For secureexec, reset the stack limit to sane default to
1322                  * avoid bad behavior from the prior rlimits. This has to
1323                  * happen before arch_pick_mmap_layout(), which examines
1324                  * RLIMIT_STACK, but after the point of no return to avoid
1325                  * needing to clean up the change on failure.
1326                  */
1327                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1328                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1329         }
1330
1331         me->sas_ss_sp = me->sas_ss_size = 0;
1332
1333         /*
1334          * Figure out dumpability. Note that this checking only of current
1335          * is wrong, but userspace depends on it. This should be testing
1336          * bprm->secureexec instead.
1337          */
1338         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1339             !(uid_eq(current_euid(), current_uid()) &&
1340               gid_eq(current_egid(), current_gid())))
1341                 set_dumpable(current->mm, suid_dumpable);
1342         else
1343                 set_dumpable(current->mm, SUID_DUMP_USER);
1344
1345         perf_event_exec();
1346         __set_task_comm(me, kbasename(bprm->filename), true);
1347
1348         /* An exec changes our domain. We are no longer part of the thread
1349            group */
1350         WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1351         flush_signal_handlers(me, 0);
1352
1353         /*
1354          * install the new credentials for this executable
1355          */
1356         security_bprm_committing_creds(bprm);
1357
1358         commit_creds(bprm->cred);
1359         bprm->cred = NULL;
1360
1361         /*
1362          * Disable monitoring for regular users
1363          * when executing setuid binaries. Must
1364          * wait until new credentials are committed
1365          * by commit_creds() above
1366          */
1367         if (get_dumpable(me->mm) != SUID_DUMP_USER)
1368                 perf_event_exit_task(me);
1369         /*
1370          * cred_guard_mutex must be held at least to this point to prevent
1371          * ptrace_attach() from altering our determination of the task's
1372          * credentials; any time after this it may be unlocked.
1373          */
1374         security_bprm_committed_creds(bprm);
1375
1376         /* Pass the opened binary to the interpreter. */
1377         if (bprm->have_execfd) {
1378                 retval = get_unused_fd_flags(0);
1379                 if (retval < 0)
1380                         goto out_unlock;
1381                 fd_install(retval, bprm->executable);
1382                 bprm->executable = NULL;
1383                 bprm->execfd = retval;
1384         }
1385         return 0;
1386
1387 out_unlock:
1388         mutex_unlock(&me->signal->exec_update_mutex);
1389 out:
1390         return retval;
1391 }
1392 EXPORT_SYMBOL(begin_new_exec);
1393
1394 void would_dump(struct linux_binprm *bprm, struct file *file)
1395 {
1396         struct inode *inode = file_inode(file);
1397         if (inode_permission(inode, MAY_READ) < 0) {
1398                 struct user_namespace *old, *user_ns;
1399                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1400
1401                 /* Ensure mm->user_ns contains the executable */
1402                 user_ns = old = bprm->mm->user_ns;
1403                 while ((user_ns != &init_user_ns) &&
1404                        !privileged_wrt_inode_uidgid(user_ns, inode))
1405                         user_ns = user_ns->parent;
1406
1407                 if (old != user_ns) {
1408                         bprm->mm->user_ns = get_user_ns(user_ns);
1409                         put_user_ns(old);
1410                 }
1411         }
1412 }
1413 EXPORT_SYMBOL(would_dump);
1414
1415 void setup_new_exec(struct linux_binprm * bprm)
1416 {
1417         /* Setup things that can depend upon the personality */
1418         struct task_struct *me = current;
1419
1420         arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1421
1422         arch_setup_new_exec();
1423
1424         /* Set the new mm task size. We have to do that late because it may
1425          * depend on TIF_32BIT which is only updated in flush_thread() on
1426          * some architectures like powerpc
1427          */
1428         me->mm->task_size = TASK_SIZE;
1429         mutex_unlock(&me->signal->exec_update_mutex);
1430         mutex_unlock(&me->signal->cred_guard_mutex);
1431 }
1432 EXPORT_SYMBOL(setup_new_exec);
1433
1434 /* Runs immediately before start_thread() takes over. */
1435 void finalize_exec(struct linux_binprm *bprm)
1436 {
1437         /* Store any stack rlimit changes before starting thread. */
1438         task_lock(current->group_leader);
1439         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1440         task_unlock(current->group_leader);
1441 }
1442 EXPORT_SYMBOL(finalize_exec);
1443
1444 /*
1445  * Prepare credentials and lock ->cred_guard_mutex.
1446  * setup_new_exec() commits the new creds and drops the lock.
1447  * Or, if exec fails before, free_bprm() should release ->cred and
1448  * and unlock.
1449  */
1450 static int prepare_bprm_creds(struct linux_binprm *bprm)
1451 {
1452         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1453                 return -ERESTARTNOINTR;
1454
1455         bprm->cred = prepare_exec_creds();
1456         if (likely(bprm->cred))
1457                 return 0;
1458
1459         mutex_unlock(&current->signal->cred_guard_mutex);
1460         return -ENOMEM;
1461 }
1462
1463 static void free_bprm(struct linux_binprm *bprm)
1464 {
1465         if (bprm->mm) {
1466                 acct_arg_size(bprm, 0);
1467                 mmput(bprm->mm);
1468         }
1469         free_arg_pages(bprm);
1470         if (bprm->cred) {
1471                 mutex_unlock(&current->signal->cred_guard_mutex);
1472                 abort_creds(bprm->cred);
1473         }
1474         if (bprm->file) {
1475                 allow_write_access(bprm->file);
1476                 fput(bprm->file);
1477         }
1478         if (bprm->executable)
1479                 fput(bprm->executable);
1480         /* If a binfmt changed the interp, free it. */
1481         if (bprm->interp != bprm->filename)
1482                 kfree(bprm->interp);
1483         kfree(bprm->fdpath);
1484         kfree(bprm);
1485 }
1486
1487 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1488 {
1489         struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1490         int retval = -ENOMEM;
1491         if (!bprm)
1492                 goto out;
1493
1494         if (fd == AT_FDCWD || filename->name[0] == '/') {
1495                 bprm->filename = filename->name;
1496         } else {
1497                 if (filename->name[0] == '\0')
1498                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1499                 else
1500                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1501                                                   fd, filename->name);
1502                 if (!bprm->fdpath)
1503                         goto out_free;
1504
1505                 bprm->filename = bprm->fdpath;
1506         }
1507         bprm->interp = bprm->filename;
1508
1509         retval = bprm_mm_init(bprm);
1510         if (retval)
1511                 goto out_free;
1512         return bprm;
1513
1514 out_free:
1515         free_bprm(bprm);
1516 out:
1517         return ERR_PTR(retval);
1518 }
1519
1520 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1521 {
1522         /* If a binfmt changed the interp, free it first. */
1523         if (bprm->interp != bprm->filename)
1524                 kfree(bprm->interp);
1525         bprm->interp = kstrdup(interp, GFP_KERNEL);
1526         if (!bprm->interp)
1527                 return -ENOMEM;
1528         return 0;
1529 }
1530 EXPORT_SYMBOL(bprm_change_interp);
1531
1532 /*
1533  * determine how safe it is to execute the proposed program
1534  * - the caller must hold ->cred_guard_mutex to protect against
1535  *   PTRACE_ATTACH or seccomp thread-sync
1536  */
1537 static void check_unsafe_exec(struct linux_binprm *bprm)
1538 {
1539         struct task_struct *p = current, *t;
1540         unsigned n_fs;
1541
1542         if (p->ptrace)
1543                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1544
1545         /*
1546          * This isn't strictly necessary, but it makes it harder for LSMs to
1547          * mess up.
1548          */
1549         if (task_no_new_privs(current))
1550                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1551
1552         t = p;
1553         n_fs = 1;
1554         spin_lock(&p->fs->lock);
1555         rcu_read_lock();
1556         while_each_thread(p, t) {
1557                 if (t->fs == p->fs)
1558                         n_fs++;
1559         }
1560         rcu_read_unlock();
1561
1562         if (p->fs->users > n_fs)
1563                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1564         else
1565                 p->fs->in_exec = 1;
1566         spin_unlock(&p->fs->lock);
1567 }
1568
1569 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1570 {
1571         /* Handle suid and sgid on files */
1572         struct inode *inode;
1573         unsigned int mode;
1574         kuid_t uid;
1575         kgid_t gid;
1576
1577         if (!mnt_may_suid(file->f_path.mnt))
1578                 return;
1579
1580         if (task_no_new_privs(current))
1581                 return;
1582
1583         inode = file->f_path.dentry->d_inode;
1584         mode = READ_ONCE(inode->i_mode);
1585         if (!(mode & (S_ISUID|S_ISGID)))
1586                 return;
1587
1588         /* Be careful if suid/sgid is set */
1589         inode_lock(inode);
1590
1591         /* reload atomically mode/uid/gid now that lock held */
1592         mode = inode->i_mode;
1593         uid = inode->i_uid;
1594         gid = inode->i_gid;
1595         inode_unlock(inode);
1596
1597         /* We ignore suid/sgid if there are no mappings for them in the ns */
1598         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1599                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1600                 return;
1601
1602         if (mode & S_ISUID) {
1603                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1604                 bprm->cred->euid = uid;
1605         }
1606
1607         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1608                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1609                 bprm->cred->egid = gid;
1610         }
1611 }
1612
1613 /*
1614  * Compute brpm->cred based upon the final binary.
1615  */
1616 static int bprm_creds_from_file(struct linux_binprm *bprm)
1617 {
1618         /* Compute creds based on which file? */
1619         struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1620
1621         bprm_fill_uid(bprm, file);
1622         return security_bprm_creds_from_file(bprm, file);
1623 }
1624
1625 /*
1626  * Fill the binprm structure from the inode.
1627  * Read the first BINPRM_BUF_SIZE bytes
1628  *
1629  * This may be called multiple times for binary chains (scripts for example).
1630  */
1631 static int prepare_binprm(struct linux_binprm *bprm)
1632 {
1633         loff_t pos = 0;
1634
1635         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1636         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1637 }
1638
1639 /*
1640  * Arguments are '\0' separated strings found at the location bprm->p
1641  * points to; chop off the first by relocating brpm->p to right after
1642  * the first '\0' encountered.
1643  */
1644 int remove_arg_zero(struct linux_binprm *bprm)
1645 {
1646         int ret = 0;
1647         unsigned long offset;
1648         char *kaddr;
1649         struct page *page;
1650
1651         if (!bprm->argc)
1652                 return 0;
1653
1654         do {
1655                 offset = bprm->p & ~PAGE_MASK;
1656                 page = get_arg_page(bprm, bprm->p, 0);
1657                 if (!page) {
1658                         ret = -EFAULT;
1659                         goto out;
1660                 }
1661                 kaddr = kmap_atomic(page);
1662
1663                 for (; offset < PAGE_SIZE && kaddr[offset];
1664                                 offset++, bprm->p++)
1665                         ;
1666
1667                 kunmap_atomic(kaddr);
1668                 put_arg_page(page);
1669         } while (offset == PAGE_SIZE);
1670
1671         bprm->p++;
1672         bprm->argc--;
1673         ret = 0;
1674
1675 out:
1676         return ret;
1677 }
1678 EXPORT_SYMBOL(remove_arg_zero);
1679
1680 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1681 /*
1682  * cycle the list of binary formats handler, until one recognizes the image
1683  */
1684 static int search_binary_handler(struct linux_binprm *bprm)
1685 {
1686         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1687         struct linux_binfmt *fmt;
1688         int retval;
1689
1690         retval = prepare_binprm(bprm);
1691         if (retval < 0)
1692                 return retval;
1693
1694         retval = security_bprm_check(bprm);
1695         if (retval)
1696                 return retval;
1697
1698         retval = -ENOENT;
1699  retry:
1700         read_lock(&binfmt_lock);
1701         list_for_each_entry(fmt, &formats, lh) {
1702                 if (!try_module_get(fmt->module))
1703                         continue;
1704                 read_unlock(&binfmt_lock);
1705
1706                 retval = fmt->load_binary(bprm);
1707
1708                 read_lock(&binfmt_lock);
1709                 put_binfmt(fmt);
1710                 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1711                         read_unlock(&binfmt_lock);
1712                         return retval;
1713                 }
1714         }
1715         read_unlock(&binfmt_lock);
1716
1717         if (need_retry) {
1718                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1719                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1720                         return retval;
1721                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1722                         return retval;
1723                 need_retry = false;
1724                 goto retry;
1725         }
1726
1727         return retval;
1728 }
1729
1730 static int exec_binprm(struct linux_binprm *bprm)
1731 {
1732         pid_t old_pid, old_vpid;
1733         int ret, depth;
1734
1735         /* Need to fetch pid before load_binary changes it */
1736         old_pid = current->pid;
1737         rcu_read_lock();
1738         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1739         rcu_read_unlock();
1740
1741         /* This allows 4 levels of binfmt rewrites before failing hard. */
1742         for (depth = 0;; depth++) {
1743                 struct file *exec;
1744                 if (depth > 5)
1745                         return -ELOOP;
1746
1747                 ret = search_binary_handler(bprm);
1748                 if (ret < 0)
1749                         return ret;
1750                 if (!bprm->interpreter)
1751                         break;
1752
1753                 exec = bprm->file;
1754                 bprm->file = bprm->interpreter;
1755                 bprm->interpreter = NULL;
1756
1757                 allow_write_access(exec);
1758                 if (unlikely(bprm->have_execfd)) {
1759                         if (bprm->executable) {
1760                                 fput(exec);
1761                                 return -ENOEXEC;
1762                         }
1763                         bprm->executable = exec;
1764                 } else
1765                         fput(exec);
1766         }
1767
1768         audit_bprm(bprm);
1769         trace_sched_process_exec(current, old_pid, bprm);
1770         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1771         proc_exec_connector(current);
1772         return 0;
1773 }
1774
1775 /*
1776  * sys_execve() executes a new program.
1777  */
1778 static int bprm_execve(struct linux_binprm *bprm,
1779                        int fd, struct filename *filename, int flags)
1780 {
1781         struct file *file;
1782         struct files_struct *displaced;
1783         int retval;
1784
1785         /*
1786          * Cancel any io_uring activity across execve
1787          */
1788         io_uring_task_cancel();
1789
1790         retval = unshare_files(&displaced);
1791         if (retval)
1792                 return retval;
1793
1794         retval = prepare_bprm_creds(bprm);
1795         if (retval)
1796                 goto out_files;
1797
1798         check_unsafe_exec(bprm);
1799         current->in_execve = 1;
1800
1801         file = do_open_execat(fd, filename, flags);
1802         retval = PTR_ERR(file);
1803         if (IS_ERR(file))
1804                 goto out_unmark;
1805
1806         sched_exec();
1807
1808         bprm->file = file;
1809         /*
1810          * Record that a name derived from an O_CLOEXEC fd will be
1811          * inaccessible after exec. Relies on having exclusive access to
1812          * current->files (due to unshare_files above).
1813          */
1814         if (bprm->fdpath &&
1815             close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1816                 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1817
1818         /* Set the unchanging part of bprm->cred */
1819         retval = security_bprm_creds_for_exec(bprm);
1820         if (retval)
1821                 goto out;
1822
1823         retval = exec_binprm(bprm);
1824         if (retval < 0)
1825                 goto out;
1826
1827         /* execve succeeded */
1828         current->fs->in_exec = 0;
1829         current->in_execve = 0;
1830         rseq_execve(current);
1831         acct_update_integrals(current);
1832         task_numa_free(current, false);
1833         if (displaced)
1834                 put_files_struct(displaced);
1835         return retval;
1836
1837 out:
1838         /*
1839          * If past the point of no return ensure the the code never
1840          * returns to the userspace process.  Use an existing fatal
1841          * signal if present otherwise terminate the process with
1842          * SIGSEGV.
1843          */
1844         if (bprm->point_of_no_return && !fatal_signal_pending(current))
1845                 force_sigsegv(SIGSEGV);
1846
1847 out_unmark:
1848         current->fs->in_exec = 0;
1849         current->in_execve = 0;
1850
1851 out_files:
1852         if (displaced)
1853                 reset_files_struct(displaced);
1854
1855         return retval;
1856 }
1857
1858 static int do_execveat_common(int fd, struct filename *filename,
1859                               struct user_arg_ptr argv,
1860                               struct user_arg_ptr envp,
1861                               int flags)
1862 {
1863         struct linux_binprm *bprm;
1864         int retval;
1865
1866         if (IS_ERR(filename))
1867                 return PTR_ERR(filename);
1868
1869         /*
1870          * We move the actual failure in case of RLIMIT_NPROC excess from
1871          * set*uid() to execve() because too many poorly written programs
1872          * don't check setuid() return code.  Here we additionally recheck
1873          * whether NPROC limit is still exceeded.
1874          */
1875         if ((current->flags & PF_NPROC_EXCEEDED) &&
1876             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1877                 retval = -EAGAIN;
1878                 goto out_ret;
1879         }
1880
1881         /* We're below the limit (still or again), so we don't want to make
1882          * further execve() calls fail. */
1883         current->flags &= ~PF_NPROC_EXCEEDED;
1884
1885         bprm = alloc_bprm(fd, filename);
1886         if (IS_ERR(bprm)) {
1887                 retval = PTR_ERR(bprm);
1888                 goto out_ret;
1889         }
1890
1891         retval = count(argv, MAX_ARG_STRINGS);
1892         if (retval < 0)
1893                 goto out_free;
1894         bprm->argc = retval;
1895
1896         retval = count(envp, MAX_ARG_STRINGS);
1897         if (retval < 0)
1898                 goto out_free;
1899         bprm->envc = retval;
1900
1901         retval = bprm_stack_limits(bprm);
1902         if (retval < 0)
1903                 goto out_free;
1904
1905         retval = copy_string_kernel(bprm->filename, bprm);
1906         if (retval < 0)
1907                 goto out_free;
1908         bprm->exec = bprm->p;
1909
1910         retval = copy_strings(bprm->envc, envp, bprm);
1911         if (retval < 0)
1912                 goto out_free;
1913
1914         retval = copy_strings(bprm->argc, argv, bprm);
1915         if (retval < 0)
1916                 goto out_free;
1917
1918         retval = bprm_execve(bprm, fd, filename, flags);
1919 out_free:
1920         free_bprm(bprm);
1921
1922 out_ret:
1923         putname(filename);
1924         return retval;
1925 }
1926
1927 int kernel_execve(const char *kernel_filename,
1928                   const char *const *argv, const char *const *envp)
1929 {
1930         struct filename *filename;
1931         struct linux_binprm *bprm;
1932         int fd = AT_FDCWD;
1933         int retval;
1934
1935         filename = getname_kernel(kernel_filename);
1936         if (IS_ERR(filename))
1937                 return PTR_ERR(filename);
1938
1939         bprm = alloc_bprm(fd, filename);
1940         if (IS_ERR(bprm)) {
1941                 retval = PTR_ERR(bprm);
1942                 goto out_ret;
1943         }
1944
1945         retval = count_strings_kernel(argv);
1946         if (retval < 0)
1947                 goto out_free;
1948         bprm->argc = retval;
1949
1950         retval = count_strings_kernel(envp);
1951         if (retval < 0)
1952                 goto out_free;
1953         bprm->envc = retval;
1954
1955         retval = bprm_stack_limits(bprm);
1956         if (retval < 0)
1957                 goto out_free;
1958
1959         retval = copy_string_kernel(bprm->filename, bprm);
1960         if (retval < 0)
1961                 goto out_free;
1962         bprm->exec = bprm->p;
1963
1964         retval = copy_strings_kernel(bprm->envc, envp, bprm);
1965         if (retval < 0)
1966                 goto out_free;
1967
1968         retval = copy_strings_kernel(bprm->argc, argv, bprm);
1969         if (retval < 0)
1970                 goto out_free;
1971
1972         retval = bprm_execve(bprm, fd, filename, 0);
1973 out_free:
1974         free_bprm(bprm);
1975 out_ret:
1976         putname(filename);
1977         return retval;
1978 }
1979
1980 static int do_execve(struct filename *filename,
1981         const char __user *const __user *__argv,
1982         const char __user *const __user *__envp)
1983 {
1984         struct user_arg_ptr argv = { .ptr.native = __argv };
1985         struct user_arg_ptr envp = { .ptr.native = __envp };
1986         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1987 }
1988
1989 static int do_execveat(int fd, struct filename *filename,
1990                 const char __user *const __user *__argv,
1991                 const char __user *const __user *__envp,
1992                 int flags)
1993 {
1994         struct user_arg_ptr argv = { .ptr.native = __argv };
1995         struct user_arg_ptr envp = { .ptr.native = __envp };
1996
1997         return do_execveat_common(fd, filename, argv, envp, flags);
1998 }
1999
2000 #ifdef CONFIG_COMPAT
2001 static int compat_do_execve(struct filename *filename,
2002         const compat_uptr_t __user *__argv,
2003         const compat_uptr_t __user *__envp)
2004 {
2005         struct user_arg_ptr argv = {
2006                 .is_compat = true,
2007                 .ptr.compat = __argv,
2008         };
2009         struct user_arg_ptr envp = {
2010                 .is_compat = true,
2011                 .ptr.compat = __envp,
2012         };
2013         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2014 }
2015
2016 static int compat_do_execveat(int fd, struct filename *filename,
2017                               const compat_uptr_t __user *__argv,
2018                               const compat_uptr_t __user *__envp,
2019                               int flags)
2020 {
2021         struct user_arg_ptr argv = {
2022                 .is_compat = true,
2023                 .ptr.compat = __argv,
2024         };
2025         struct user_arg_ptr envp = {
2026                 .is_compat = true,
2027                 .ptr.compat = __envp,
2028         };
2029         return do_execveat_common(fd, filename, argv, envp, flags);
2030 }
2031 #endif
2032
2033 void set_binfmt(struct linux_binfmt *new)
2034 {
2035         struct mm_struct *mm = current->mm;
2036
2037         if (mm->binfmt)
2038                 module_put(mm->binfmt->module);
2039
2040         mm->binfmt = new;
2041         if (new)
2042                 __module_get(new->module);
2043 }
2044 EXPORT_SYMBOL(set_binfmt);
2045
2046 /*
2047  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2048  */
2049 void set_dumpable(struct mm_struct *mm, int value)
2050 {
2051         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2052                 return;
2053
2054         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2055 }
2056
2057 SYSCALL_DEFINE3(execve,
2058                 const char __user *, filename,
2059                 const char __user *const __user *, argv,
2060                 const char __user *const __user *, envp)
2061 {
2062         return do_execve(getname(filename), argv, envp);
2063 }
2064
2065 SYSCALL_DEFINE5(execveat,
2066                 int, fd, const char __user *, filename,
2067                 const char __user *const __user *, argv,
2068                 const char __user *const __user *, envp,
2069                 int, flags)
2070 {
2071         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2072
2073         return do_execveat(fd,
2074                            getname_flags(filename, lookup_flags, NULL),
2075                            argv, envp, flags);
2076 }
2077
2078 #ifdef CONFIG_COMPAT
2079 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2080         const compat_uptr_t __user *, argv,
2081         const compat_uptr_t __user *, envp)
2082 {
2083         return compat_do_execve(getname(filename), argv, envp);
2084 }
2085
2086 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2087                        const char __user *, filename,
2088                        const compat_uptr_t __user *, argv,
2089                        const compat_uptr_t __user *, envp,
2090                        int,  flags)
2091 {
2092         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2093
2094         return compat_do_execveat(fd,
2095                                   getname_flags(filename, lookup_flags, NULL),
2096                                   argv, envp, flags);
2097 }
2098 #endif