kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/sched/autogroup.h>
  16 #include <linux/sched/mm.h>
  17 #include <linux/sched/coredump.h>
  18 #include <linux/sched/user.h>
  19 #include <linux/sched/numa_balancing.h>
  20 #include <linux/sched/stat.h>
  21 #include <linux/sched/task.h>
  22 #include <linux/sched/task_stack.h>
  23 #include <linux/sched/cputime.h>
  24 #include <linux/rtmutex.h>
  25 #include <linux/init.h>
  26 #include <linux/unistd.h>
  27 #include <linux/module.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/completion.h>
  30 #include <linux/personality.h>
  31 #include <linux/mempolicy.h>
  32 #include <linux/sem.h>
  33 #include <linux/file.h>
  34 #include <linux/fdtable.h>
  35 #include <linux/iocontext.h>
  36 #include <linux/key.h>
  37 #include <linux/binfmts.h>
  38 #include <linux/mman.h>
  39 #include <linux/mmu_notifier.h>
  40 #include <linux/hmm.h>
  41 #include <linux/fs.h>
  42 #include <linux/mm.h>
  43 #include <linux/vmacache.h>
  44 #include <linux/nsproxy.h>
  45 #include <linux/capability.h>
  46 #include <linux/cpu.h>
  47 #include <linux/cgroup.h>
  48 #include <linux/security.h>
  49 #include <linux/hugetlb.h>
  50 #include <linux/seccomp.h>
  51 #include <linux/swap.h>
  52 #include <linux/syscalls.h>
  53 #include <linux/jiffies.h>
  54 #include <linux/futex.h>
  55 #include <linux/compat.h>
  56 #include <linux/kthread.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/rcupdate.h>
  59 #include <linux/ptrace.h>
  60 #include <linux/mount.h>
  61 #include <linux/audit.h>
  62 #include <linux/memcontrol.h>
  63 #include <linux/ftrace.h>
  64 #include <linux/proc_fs.h>
  65 #include <linux/profile.h>
  66 #include <linux/rmap.h>
  67 #include <linux/ksm.h>
  68 #include <linux/acct.h>
  69 #include <linux/userfaultfd_k.h>
  70 #include <linux/tsacct_kern.h>
  71 #include <linux/cn_proc.h>
  72 #include <linux/freezer.h>
  73 #include <linux/delayacct.h>
  74 #include <linux/taskstats_kern.h>
  75 #include <linux/random.h>
  76 #include <linux/tty.h>
  77 #include <linux/blkdev.h>
  78 #include <linux/fs_struct.h>
  79 #include <linux/magic.h>
  80 #include <linux/sched/mm.h>
  81 #include <linux/perf_event.h>
  82 #include <linux/posix-timers.h>
  83 #include <linux/user-return-notifier.h>
  84 #include <linux/oom.h>
  85 #include <linux/khugepaged.h>
  86 #include <linux/signalfd.h>
  87 #include <linux/uprobes.h>
  88 #include <linux/aio.h>
  89 #include <linux/compiler.h>
  90 #include <linux/sysctl.h>
  91 #include <linux/kcov.h>
  92 #include <linux/livepatch.h>
  93 #include <linux/thread_info.h>
  94
  95 #include <asm/pgtable.h>
  96 #include <asm/pgalloc.h>
  97 #include <linux/uaccess.h>
  98 #include <asm/mmu_context.h>
  99 #include <asm/cacheflush.h>
 100 #include <asm/tlbflush.h>
 101
 102 #include <trace/events/sched.h>
 103
 104 #define CREATE_TRACE_POINTS
 105 #include <trace/events/task.h>
 106
 107 /*
 108  * Minimum number of threads to boot the kernel
 109  */
 110 #define MIN_THREADS 20
 111
 112 /*
 113  * Maximum number of threads
 114  */
 115 #define MAX_THREADS FUTEX_TID_MASK
 116
 117 /*
 118  * Protected counters by write_lock_irq(&tasklist_lock)
 119  */
 120 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 121 int nr_threads;                 /* The idle threads do not count.. */
 122
 123 int max_threads;                /* tunable limit on nr_threads */
 124
 125 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 126
 127 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 128
 129 #ifdef CONFIG_PROVE_RCU
 130 int lockdep_tasklist_lock_is_held(void)
 131 {
 132         return lockdep_is_held(&tasklist_lock);
 133 }
 134 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 135 #endif /* #ifdef CONFIG_PROVE_RCU */
 136
 137 int nr_processes(void)
 138 {
 139         int cpu;
 140         int total = 0;
 141
 142         for_each_possible_cpu(cpu)
 143                 total += per_cpu(process_counts, cpu);
 144
 145         return total;
 146 }
 147
 148 void __weak arch_release_task_struct(struct task_struct *tsk)
 149 {
 150 }
 151
 152 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 153 static struct kmem_cache *task_struct_cachep;
 154
 155 static inline struct task_struct *alloc_task_struct_node(int node)
 156 {
 157         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 158 }
 159
 160 static inline void free_task_struct(struct task_struct *tsk)
 161 {
 162         kmem_cache_free(task_struct_cachep, tsk);
 163 }
 164 #endif
 165
 166 void __weak arch_release_thread_stack(unsigned long *stack)
 167 {
 168 }
 169
 170 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 171
 172 /*
 173  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 174  * kmemcache based allocator.
 175  */
 176 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 177
 178 #ifdef CONFIG_VMAP_STACK
 179 /*
 180  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 181  * flush.  Try to minimize the number of calls by caching stacks.
 182  */
 183 #define NR_CACHED_STACKS 2
 184 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 185
 186 static int free_vm_stack_cache(unsigned int cpu)
 187 {
 188         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 189         int i;
 190
 191         for (i = 0; i < NR_CACHED_STACKS; i++) {
 192                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 193
 194                 if (!vm_stack)
 195                         continue;
 196
 197                 vfree(vm_stack->addr);
 198                 cached_vm_stacks[i] = NULL;
 199         }
 200
 201         return 0;
 202 }
 203 #endif
 204
 205 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 206 {
 207 #ifdef CONFIG_VMAP_STACK
 208         void *stack;
 209         int i;
 210
 211         for (i = 0; i < NR_CACHED_STACKS; i++) {
 212                 struct vm_struct *s;
 213
 214                 s = this_cpu_xchg(cached_stacks[i], NULL);
 215
 216                 if (!s)
 217                         continue;
 218
 219                 /* Clear stale pointers from reused stack. */
 220                 memset(s->addr, 0, THREAD_SIZE);
 221
 222                 tsk->stack_vm_area = s;
 223                 return s->addr;
 224         }
 225
 226         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 227                                      VMALLOC_START, VMALLOC_END,
 228                                      THREADINFO_GFP,
 229                                      PAGE_KERNEL,
 230                                      0, node, __builtin_return_address(0));
 231
 232         /*
 233          * We can't call find_vm_area() in interrupt context, and
 234          * free_thread_stack() can be called in interrupt context,
 235          * so cache the vm_struct.
 236          */
 237         if (stack)
 238                 tsk->stack_vm_area = find_vm_area(stack);
 239         return stack;
 240 #else
 241         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 242                                              THREAD_SIZE_ORDER);
 243
 244         return page ? page_address(page) : NULL;
 245 #endif
 246 }
 247
 248 static inline void free_thread_stack(struct task_struct *tsk)
 249 {
 250 #ifdef CONFIG_VMAP_STACK
 251         if (task_stack_vm_area(tsk)) {
 252                 int i;
 253
 254                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 255                         if (this_cpu_cmpxchg(cached_stacks[i],
 256                                         NULL, tsk->stack_vm_area) != NULL)
 257                                 continue;
 258
 259                         return;
 260                 }
 261
 262                 vfree_atomic(tsk->stack);
 263                 return;
 264         }
 265 #endif
 266
 267         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 268 }
 269 # else
 270 static struct kmem_cache *thread_stack_cache;
 271
 272 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 273                                                   int node)
 274 {
 275         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 276 }
 277
 278 static void free_thread_stack(struct task_struct *tsk)
 279 {
 280         kmem_cache_free(thread_stack_cache, tsk->stack);
 281 }
 282
 283 void thread_stack_cache_init(void)
 284 {
 285         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 286                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 287                                         THREAD_SIZE, NULL);
 288         BUG_ON(thread_stack_cache == NULL);
 289 }
 290 # endif
 291 #endif
 292
 293 /* SLAB cache for signal_struct structures (tsk->signal) */
 294 static struct kmem_cache *signal_cachep;
 295
 296 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 297 struct kmem_cache *sighand_cachep;
 298
 299 /* SLAB cache for files_struct structures (tsk->files) */
 300 struct kmem_cache *files_cachep;
 301
 302 /* SLAB cache for fs_struct structures (tsk->fs) */
 303 struct kmem_cache *fs_cachep;
 304
 305 /* SLAB cache for vm_area_struct structures */
 306 static struct kmem_cache *vm_area_cachep;
 307
 308 /* SLAB cache for mm_struct structures (tsk->mm) */
 309 static struct kmem_cache *mm_cachep;
 310
 311 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 312 {
 313         struct vm_area_struct *vma;
 314
 315         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 316         if (vma)
 317                 vma_init(vma, mm);
 318         return vma;
 319 }
 320
 321 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 322 {
 323         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 324
 325         if (new) {
 326                 *new = *orig;
 327                 INIT_LIST_HEAD(&new->anon_vma_chain);
 328         }
 329         return new;
 330 }
 331
 332 void vm_area_free(struct vm_area_struct *vma)
 333 {
 334         kmem_cache_free(vm_area_cachep, vma);
 335 }
 336
 337 static void account_kernel_stack(struct task_struct *tsk, int account)
 338 {
 339         void *stack = task_stack_page(tsk);
 340         struct vm_struct *vm = task_stack_vm_area(tsk);
 341
 342         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 343
 344         if (vm) {
 345                 int i;
 346
 347                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 348
 349                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 350                         mod_zone_page_state(page_zone(vm->pages[i]),
 351                                             NR_KERNEL_STACK_KB,
 352                                             PAGE_SIZE / 1024 * account);
 353                 }
 354
 355                 /* All stack pages belong to the same memcg. */
 356                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 357                                      account * (THREAD_SIZE / 1024));
 358         } else {
 359                 /*
 360                  * All stack pages are in the same zone and belong to the
 361                  * same memcg.
 362                  */
 363                 struct page *first_page = virt_to_page(stack);
 364
 365                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 366                                     THREAD_SIZE / 1024 * account);
 367
 368                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 369                                      account * (THREAD_SIZE / 1024));
 370         }
 371 }
 372
 373 static void release_task_stack(struct task_struct *tsk)
 374 {
 375         if (WARN_ON(tsk->state != TASK_DEAD))
 376                 return;  /* Better to leak the stack than to free prematurely */
 377
 378         account_kernel_stack(tsk, -1);
 379         arch_release_thread_stack(tsk->stack);
 380         free_thread_stack(tsk);
 381         tsk->stack = NULL;
 382 #ifdef CONFIG_VMAP_STACK
 383         tsk->stack_vm_area = NULL;
 384 #endif
 385 }
 386
 387 #ifdef CONFIG_THREAD_INFO_IN_TASK
 388 void put_task_stack(struct task_struct *tsk)
 389 {
 390         if (atomic_dec_and_test(&tsk->stack_refcount))
 391                 release_task_stack(tsk);
 392 }
 393 #endif
 394
 395 void free_task(struct task_struct *tsk)
 396 {
 397 #ifndef CONFIG_THREAD_INFO_IN_TASK
 398         /*
 399          * The task is finally done with both the stack and thread_info,
 400          * so free both.
 401          */
 402         release_task_stack(tsk);
 403 #else
 404         /*
 405          * If the task had a separate stack allocation, it should be gone
 406          * by now.
 407          */
 408         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 409 #endif
 410         rt_mutex_debug_task_free(tsk);
 411         ftrace_graph_exit_task(tsk);
 412         put_seccomp_filter(tsk);
 413         arch_release_task_struct(tsk);
 414         if (tsk->flags & PF_KTHREAD)
 415                 free_kthread_struct(tsk);
 416         free_task_struct(tsk);
 417 }
 418 EXPORT_SYMBOL(free_task);
 419
 420 #ifdef CONFIG_MMU
 421 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 422                                         struct mm_struct *oldmm)
 423 {
 424         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 425         struct rb_node **rb_link, *rb_parent;
 426         int retval;
 427         unsigned long charge;
 428         LIST_HEAD(uf);
 429
 430         uprobe_start_dup_mmap();
 431         if (down_write_killable(&oldmm->mmap_sem)) {
 432                 retval = -EINTR;
 433                 goto fail_uprobe_end;
 434         }
 435         flush_cache_dup_mm(oldmm);
 436         uprobe_dup_mmap(oldmm, mm);
 437         /*
 438          * Not linked in yet - no deadlock potential:
 439          */
 440         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 441
 442         /* No ordering required: file already has been exposed. */
 443         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 444
 445         mm->total_vm = oldmm->total_vm;
 446         mm->data_vm = oldmm->data_vm;
 447         mm->exec_vm = oldmm->exec_vm;
 448         mm->stack_vm = oldmm->stack_vm;
 449
 450         rb_link = &mm->mm_rb.rb_node;
 451         rb_parent = NULL;
 452         pprev = &mm->mmap;
 453         retval = ksm_fork(mm, oldmm);
 454         if (retval)
 455                 goto out;
 456         retval = khugepaged_fork(mm, oldmm);
 457         if (retval)
 458                 goto out;
 459
 460         prev = NULL;
 461         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 462                 struct file *file;
 463
 464                 if (mpnt->vm_flags & VM_DONTCOPY) {
 465                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 466                         continue;
 467                 }
 468                 charge = 0;
 469                 /*
 470                  * Don't duplicate many vmas if we've been oom-killed (for
 471                  * example)
 472                  */
 473                 if (fatal_signal_pending(current)) {
 474                         retval = -EINTR;
 475                         goto out;
 476                 }
 477                 if (mpnt->vm_flags & VM_ACCOUNT) {
 478                         unsigned long len = vma_pages(mpnt);
 479
 480                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 481                                 goto fail_nomem;
 482                         charge = len;
 483                 }
 484                 tmp = vm_area_dup(mpnt);
 485                 if (!tmp)
 486                         goto fail_nomem;
 487                 retval = vma_dup_policy(mpnt, tmp);
 488                 if (retval)
 489                         goto fail_nomem_policy;
 490                 tmp->vm_mm = mm;
 491                 retval = dup_userfaultfd(tmp, &uf);
 492                 if (retval)
 493                         goto fail_nomem_anon_vma_fork;
 494                 if (tmp->vm_flags & VM_WIPEONFORK) {
 495                         /* VM_WIPEONFORK gets a clean slate in the child. */
 496                         tmp->anon_vma = NULL;
 497                         if (anon_vma_prepare(tmp))
 498                                 goto fail_nomem_anon_vma_fork;
 499                 } else if (anon_vma_fork(tmp, mpnt))
 500                         goto fail_nomem_anon_vma_fork;
 501                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 502                 tmp->vm_next = tmp->vm_prev = NULL;
 503                 file = tmp->vm_file;
 504                 if (file) {
 505                         struct inode *inode = file_inode(file);
 506                         struct address_space *mapping = file->f_mapping;
 507
 508                         get_file(file);
 509                         if (tmp->vm_flags & VM_DENYWRITE)
 510                                 atomic_dec(&inode->i_writecount);
 511                         i_mmap_lock_write(mapping);
 512                         if (tmp->vm_flags & VM_SHARED)
 513                                 atomic_inc(&mapping->i_mmap_writable);
 514                         flush_dcache_mmap_lock(mapping);
 515                         /* insert tmp into the share list, just after mpnt */
 516                         vma_interval_tree_insert_after(tmp, mpnt,
 517                                         &mapping->i_mmap);
 518                         flush_dcache_mmap_unlock(mapping);
 519                         i_mmap_unlock_write(mapping);
 520                 }
 521
 522                 /*
 523                  * Clear hugetlb-related page reserves for children. This only
 524                  * affects MAP_PRIVATE mappings. Faults generated by the child
 525                  * are not guaranteed to succeed, even if read-only
 526                  */
 527                 if (is_vm_hugetlb_page(tmp))
 528                         reset_vma_resv_huge_pages(tmp);
 529
 530                 /*
 531                  * Link in the new vma and copy the page table entries.
 532                  */
 533                 *pprev = tmp;
 534                 pprev = &tmp->vm_next;
 535                 tmp->vm_prev = prev;
 536                 prev = tmp;
 537
 538                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 539                 rb_link = &tmp->vm_rb.rb_right;
 540                 rb_parent = &tmp->vm_rb;
 541
 542                 mm->map_count++;
 543                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 544                         retval = copy_page_range(mm, oldmm, mpnt);
 545
 546                 if (tmp->vm_ops && tmp->vm_ops->open)
 547                         tmp->vm_ops->open(tmp);
 548
 549                 if (retval)
 550                         goto out;
 551         }
 552         /* a new mm has just been created */
 553         arch_dup_mmap(oldmm, mm);
 554         retval = 0;
 555 out:
 556         up_write(&mm->mmap_sem);
 557         flush_tlb_mm(oldmm);
 558         up_write(&oldmm->mmap_sem);
 559         dup_userfaultfd_complete(&uf);
 560 fail_uprobe_end:
 561         uprobe_end_dup_mmap();
 562         return retval;
 563 fail_nomem_anon_vma_fork:
 564         mpol_put(vma_policy(tmp));
 565 fail_nomem_policy:
 566         vm_area_free(tmp);
 567 fail_nomem:
 568         retval = -ENOMEM;
 569         vm_unacct_memory(charge);
 570         goto out;
 571 }
 572
 573 static inline int mm_alloc_pgd(struct mm_struct *mm)
 574 {
 575         mm->pgd = pgd_alloc(mm);
 576         if (unlikely(!mm->pgd))
 577                 return -ENOMEM;
 578         return 0;
 579 }
 580
 581 static inline void mm_free_pgd(struct mm_struct *mm)
 582 {
 583         pgd_free(mm, mm->pgd);
 584 }
 585 #else
 586 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 587 {
 588         down_write(&oldmm->mmap_sem);
 589         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 590         up_write(&oldmm->mmap_sem);
 591         return 0;
 592 }
 593 #define mm_alloc_pgd(mm)        (0)
 594 #define mm_free_pgd(mm)
 595 #endif /* CONFIG_MMU */
 596
 597 static void check_mm(struct mm_struct *mm)
 598 {
 599         int i;
 600
 601         for (i = 0; i < NR_MM_COUNTERS; i++) {
 602                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 603
 604                 if (unlikely(x))
 605                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 606                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 607         }
 608
 609         if (mm_pgtables_bytes(mm))
 610                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 611                                 mm_pgtables_bytes(mm));
 612
 613 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 614         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 615 #endif
 616 }
 617
 618 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 619 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 620
 621 /*
 622  * Called when the last reference to the mm
 623  * is dropped: either by a lazy thread or by
 624  * mmput. Free the page directory and the mm.
 625  */
 626 void __mmdrop(struct mm_struct *mm)
 627 {
 628         BUG_ON(mm == &init_mm);
 629         WARN_ON_ONCE(mm == current->mm);
 630         WARN_ON_ONCE(mm == current->active_mm);
 631         mm_free_pgd(mm);
 632         destroy_context(mm);
 633         hmm_mm_destroy(mm);
 634         mmu_notifier_mm_destroy(mm);
 635         check_mm(mm);
 636         put_user_ns(mm->user_ns);
 637         free_mm(mm);
 638 }
 639 EXPORT_SYMBOL_GPL(__mmdrop);
 640
 641 static void mmdrop_async_fn(struct work_struct *work)
 642 {
 643         struct mm_struct *mm;
 644
 645         mm = container_of(work, struct mm_struct, async_put_work);
 646         __mmdrop(mm);
 647 }
 648
 649 static void mmdrop_async(struct mm_struct *mm)
 650 {
 651         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 652                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 653                 schedule_work(&mm->async_put_work);
 654         }
 655 }
 656
 657 static inline void free_signal_struct(struct signal_struct *sig)
 658 {
 659         taskstats_tgid_free(sig);
 660         sched_autogroup_exit(sig);
 661         /*
 662          * __mmdrop is not safe to call from softirq context on x86 due to
 663          * pgd_dtor so postpone it to the async context
 664          */
 665         if (sig->oom_mm)
 666                 mmdrop_async(sig->oom_mm);
 667         kmem_cache_free(signal_cachep, sig);
 668 }
 669
 670 static inline void put_signal_struct(struct signal_struct *sig)
 671 {
 672         if (atomic_dec_and_test(&sig->sigcnt))
 673                 free_signal_struct(sig);
 674 }
 675
 676 void __put_task_struct(struct task_struct *tsk)
 677 {
 678         WARN_ON(!tsk->exit_state);
 679         WARN_ON(atomic_read(&tsk->usage));
 680         WARN_ON(tsk == current);
 681
 682         cgroup_free(tsk);
 683         task_numa_free(tsk);
 684         security_task_free(tsk);
 685         exit_creds(tsk);
 686         delayacct_tsk_free(tsk);
 687         put_signal_struct(tsk->signal);
 688
 689         if (!profile_handoff_task(tsk))
 690                 free_task(tsk);
 691 }
 692 EXPORT_SYMBOL_GPL(__put_task_struct);
 693
 694 void __init __weak arch_task_cache_init(void) { }
 695
 696 /*
 697  * set_max_threads
 698  */
 699 static void set_max_threads(unsigned int max_threads_suggested)
 700 {
 701         u64 threads;
 702
 703         /*
 704          * The number of threads shall be limited such that the thread
 705          * structures may only consume a small part of the available memory.
 706          */
 707         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 708                 threads = MAX_THREADS;
 709         else
 710                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 711                                     (u64) THREAD_SIZE * 8UL);
 712
 713         if (threads > max_threads_suggested)
 714                 threads = max_threads_suggested;
 715
 716         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 717 }
 718
 719 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 720 /* Initialized by the architecture: */
 721 int arch_task_struct_size __read_mostly;
 722 #endif
 723
 724 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 725 {
 726         /* Fetch thread_struct whitelist for the architecture. */
 727         arch_thread_struct_whitelist(offset, size);
 728
 729         /*
 730          * Handle zero-sized whitelist or empty thread_struct, otherwise
 731          * adjust offset to position of thread_struct in task_struct.
 732          */
 733         if (unlikely(*size == 0))
 734                 *offset = 0;
 735         else
 736                 *offset += offsetof(struct task_struct, thread);
 737 }
 738
 739 void __init fork_init(void)
 740 {
 741         int i;
 742 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 743 #ifndef ARCH_MIN_TASKALIGN
 744 #define ARCH_MIN_TASKALIGN      0
 745 #endif
 746         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 747         unsigned long useroffset, usersize;
 748
 749         /* create a slab on which task_structs can be allocated */
 750         task_struct_whitelist(&useroffset, &usersize);
 751         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 752                         arch_task_struct_size, align,
 753                         SLAB_PANIC|SLAB_ACCOUNT,
 754                         useroffset, usersize, NULL);
 755 #endif
 756
 757         /* do the arch specific task caches init */
 758         arch_task_cache_init();
 759
 760         set_max_threads(MAX_THREADS);
 761
 762         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 763         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 764         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 765                 init_task.signal->rlim[RLIMIT_NPROC];
 766
 767         for (i = 0; i < UCOUNT_COUNTS; i++) {
 768                 init_user_ns.ucount_max[i] = max_threads/2;
 769         }
 770
 771 #ifdef CONFIG_VMAP_STACK
 772         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 773                           NULL, free_vm_stack_cache);
 774 #endif
 775
 776         lockdep_init_task(&init_task);
 777 }
 778
 779 int __weak arch_dup_task_struct(struct task_struct *dst,
 780                                                struct task_struct *src)
 781 {
 782         *dst = *src;
 783         return 0;
 784 }
 785
 786 void set_task_stack_end_magic(struct task_struct *tsk)
 787 {
 788         unsigned long *stackend;
 789
 790         stackend = end_of_stack(tsk);
 791         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 792 }
 793
 794 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 795 {
 796         struct task_struct *tsk;
 797         unsigned long *stack;
 798         struct vm_struct *stack_vm_area;
 799         int err;
 800
 801         if (node == NUMA_NO_NODE)
 802                 node = tsk_fork_get_node(orig);
 803         tsk = alloc_task_struct_node(node);
 804         if (!tsk)
 805                 return NULL;
 806
 807         stack = alloc_thread_stack_node(tsk, node);
 808         if (!stack)
 809                 goto free_tsk;
 810
 811         stack_vm_area = task_stack_vm_area(tsk);
 812
 813         err = arch_dup_task_struct(tsk, orig);
 814
 815         /*
 816          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 817          * sure they're properly initialized before using any stack-related
 818          * functions again.
 819          */
 820         tsk->stack = stack;
 821 #ifdef CONFIG_VMAP_STACK
 822         tsk->stack_vm_area = stack_vm_area;
 823 #endif
 824 #ifdef CONFIG_THREAD_INFO_IN_TASK
 825         atomic_set(&tsk->stack_refcount, 1);
 826 #endif
 827
 828         if (err)
 829                 goto free_stack;
 830
 831 #ifdef CONFIG_SECCOMP
 832         /*
 833          * We must handle setting up seccomp filters once we're under
 834          * the sighand lock in case orig has changed between now and
 835          * then. Until then, filter must be NULL to avoid messing up
 836          * the usage counts on the error path calling free_task.
 837          */
 838         tsk->seccomp.filter = NULL;
 839 #endif
 840
 841         setup_thread_stack(tsk, orig);
 842         clear_user_return_notifier(tsk);
 843         clear_tsk_need_resched(tsk);
 844         set_task_stack_end_magic(tsk);
 845
 846 #ifdef CONFIG_STACKPROTECTOR
 847         tsk->stack_canary = get_random_canary();
 848 #endif
 849
 850         /*
 851          * One for us, one for whoever does the "release_task()" (usually
 852          * parent)
 853          */
 854         atomic_set(&tsk->usage, 2);
 855 #ifdef CONFIG_BLK_DEV_IO_TRACE
 856         tsk->btrace_seq = 0;
 857 #endif
 858         tsk->splice_pipe = NULL;
 859         tsk->task_frag.page = NULL;
 860         tsk->wake_q.next = NULL;
 861
 862         account_kernel_stack(tsk, 1);
 863
 864         kcov_task_init(tsk);
 865
 866 #ifdef CONFIG_FAULT_INJECTION
 867         tsk->fail_nth = 0;
 868 #endif
 869
 870 #ifdef CONFIG_BLK_CGROUP
 871         tsk->throttle_queue = NULL;
 872         tsk->use_memdelay = 0;
 873 #endif
 874
 875 #ifdef CONFIG_MEMCG
 876         tsk->active_memcg = NULL;
 877 #endif
 878         return tsk;
 879
 880 free_stack:
 881         free_thread_stack(tsk);
 882 free_tsk:
 883         free_task_struct(tsk);
 884         return NULL;
 885 }
 886
 887 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 888
 889 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 890
 891 static int __init coredump_filter_setup(char *s)
 892 {
 893         default_dump_filter =
 894                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 895                 MMF_DUMP_FILTER_MASK;
 896         return 1;
 897 }
 898
 899 __setup("coredump_filter=", coredump_filter_setup);
 900
 901 #include <linux/init_task.h>
 902
 903 static void mm_init_aio(struct mm_struct *mm)
 904 {
 905 #ifdef CONFIG_AIO
 906         spin_lock_init(&mm->ioctx_lock);
 907         mm->ioctx_table = NULL;
 908 #endif
 909 }
 910
 911 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 912 {
 913 #ifdef CONFIG_MEMCG
 914         mm->owner = p;
 915 #endif
 916 }
 917
 918 static void mm_init_uprobes_state(struct mm_struct *mm)
 919 {
 920 #ifdef CONFIG_UPROBES
 921         mm->uprobes_state.xol_area = NULL;
 922 #endif
 923 }
 924
 925 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 926         struct user_namespace *user_ns)
 927 {
 928         mm->mmap = NULL;
 929         mm->mm_rb = RB_ROOT;
 930         mm->vmacache_seqnum = 0;
 931         atomic_set(&mm->mm_users, 1);
 932         atomic_set(&mm->mm_count, 1);
 933         init_rwsem(&mm->mmap_sem);
 934         INIT_LIST_HEAD(&mm->mmlist);
 935         mm->core_state = NULL;
 936         mm_pgtables_bytes_init(mm);
 937         mm->map_count = 0;
 938         mm->locked_vm = 0;
 939         mm->pinned_vm = 0;
 940         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 941         spin_lock_init(&mm->page_table_lock);
 942         spin_lock_init(&mm->arg_lock);
 943         mm_init_cpumask(mm);
 944         mm_init_aio(mm);
 945         mm_init_owner(mm, p);
 946         RCU_INIT_POINTER(mm->exe_file, NULL);
 947         mmu_notifier_mm_init(mm);
 948         hmm_mm_init(mm);
 949         init_tlb_flush_pending(mm);
 950 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 951         mm->pmd_huge_pte = NULL;
 952 #endif
 953         mm_init_uprobes_state(mm);
 954
 955         if (current->mm) {
 956                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 957                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 958         } else {
 959                 mm->flags = default_dump_filter;
 960                 mm->def_flags = 0;
 961         }
 962
 963         if (mm_alloc_pgd(mm))
 964                 goto fail_nopgd;
 965
 966         if (init_new_context(p, mm))
 967                 goto fail_nocontext;
 968
 969         mm->user_ns = get_user_ns(user_ns);
 970         return mm;
 971
 972 fail_nocontext:
 973         mm_free_pgd(mm);
 974 fail_nopgd:
 975         free_mm(mm);
 976         return NULL;
 977 }
 978
 979 /*
 980  * Allocate and initialize an mm_struct.
 981  */
 982 struct mm_struct *mm_alloc(void)
 983 {
 984         struct mm_struct *mm;
 985
 986         mm = allocate_mm();
 987         if (!mm)
 988                 return NULL;
 989
 990         memset(mm, 0, sizeof(*mm));
 991         return mm_init(mm, current, current_user_ns());
 992 }
 993
 994 static inline void __mmput(struct mm_struct *mm)
 995 {
 996         VM_BUG_ON(atomic_read(&mm->mm_users));
 997
 998         uprobe_clear_state(mm);
 999         exit_aio(mm);
1000         ksm_exit(mm);
1001         khugepaged_exit(mm); /* must run before exit_mmap */
1002         exit_mmap(mm);
1003         mm_put_huge_zero_page(mm);
1004         set_mm_exe_file(mm, NULL);
1005         if (!list_empty(&mm->mmlist)) {
1006                 spin_lock(&mmlist_lock);
1007                 list_del(&mm->mmlist);
1008                 spin_unlock(&mmlist_lock);
1009         }
1010         if (mm->binfmt)
1011                 module_put(mm->binfmt->module);
1012         mmdrop(mm);
1013 }
1014
1015 /*
1016  * Decrement the use count and release all resources for an mm.
1017  */
1018 void mmput(struct mm_struct *mm)
1019 {
1020         might_sleep();
1021
1022         if (atomic_dec_and_test(&mm->mm_users))
1023                 __mmput(mm);
1024 }
1025 EXPORT_SYMBOL_GPL(mmput);
1026
1027 #ifdef CONFIG_MMU
1028 static void mmput_async_fn(struct work_struct *work)
1029 {
1030         struct mm_struct *mm = container_of(work, struct mm_struct,
1031                                             async_put_work);
1032
1033         __mmput(mm);
1034 }
1035
1036 void mmput_async(struct mm_struct *mm)
1037 {
1038         if (atomic_dec_and_test(&mm->mm_users)) {
1039                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1040                 schedule_work(&mm->async_put_work);
1041         }
1042 }
1043 #endif
1044
1045 /**
1046  * set_mm_exe_file - change a reference to the mm's executable file
1047  *
1048  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1049  *
1050  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1051  * invocations: in mmput() nobody alive left, in execve task is single
1052  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1053  * mm->exe_file, but does so without using set_mm_exe_file() in order
1054  * to do avoid the need for any locks.
1055  */
1056 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1057 {
1058         struct file *old_exe_file;
1059
1060         /*
1061          * It is safe to dereference the exe_file without RCU as
1062          * this function is only called if nobody else can access
1063          * this mm -- see comment above for justification.
1064          */
1065         old_exe_file = rcu_dereference_raw(mm->exe_file);
1066
1067         if (new_exe_file)
1068                 get_file(new_exe_file);
1069         rcu_assign_pointer(mm->exe_file, new_exe_file);
1070         if (old_exe_file)
1071                 fput(old_exe_file);
1072 }
1073
1074 /**
1075  * get_mm_exe_file - acquire a reference to the mm's executable file
1076  *
1077  * Returns %NULL if mm has no associated executable file.
1078  * User must release file via fput().
1079  */
1080 struct file *get_mm_exe_file(struct mm_struct *mm)
1081 {
1082         struct file *exe_file;
1083
1084         rcu_read_lock();
1085         exe_file = rcu_dereference(mm->exe_file);
1086         if (exe_file && !get_file_rcu(exe_file))
1087                 exe_file = NULL;
1088         rcu_read_unlock();
1089         return exe_file;
1090 }
1091 EXPORT_SYMBOL(get_mm_exe_file);
1092
1093 /**
1094  * get_task_exe_file - acquire a reference to the task's executable file
1095  *
1096  * Returns %NULL if task's mm (if any) has no associated executable file or
1097  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1098  * User must release file via fput().
1099  */
1100 struct file *get_task_exe_file(struct task_struct *task)
1101 {
1102         struct file *exe_file = NULL;
1103         struct mm_struct *mm;
1104
1105         task_lock(task);
1106         mm = task->mm;
1107         if (mm) {
1108                 if (!(task->flags & PF_KTHREAD))
1109                         exe_file = get_mm_exe_file(mm);
1110         }
1111         task_unlock(task);
1112         return exe_file;
1113 }
1114 EXPORT_SYMBOL(get_task_exe_file);
1115
1116 /**
1117  * get_task_mm - acquire a reference to the task's mm
1118  *
1119  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1120  * this kernel workthread has transiently adopted a user mm with use_mm,
1121  * to do its AIO) is not set and if so returns a reference to it, after
1122  * bumping up the use count.  User must release the mm via mmput()
1123  * after use.  Typically used by /proc and ptrace.
1124  */
1125 struct mm_struct *get_task_mm(struct task_struct *task)
1126 {
1127         struct mm_struct *mm;
1128
1129         task_lock(task);
1130         mm = task->mm;
1131         if (mm) {
1132                 if (task->flags & PF_KTHREAD)
1133                         mm = NULL;
1134                 else
1135                         mmget(mm);
1136         }
1137         task_unlock(task);
1138         return mm;
1139 }
1140 EXPORT_SYMBOL_GPL(get_task_mm);
1141
1142 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1143 {
1144         struct mm_struct *mm;
1145         int err;
1146
1147         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1148         if (err)
1149                 return ERR_PTR(err);
1150
1151         mm = get_task_mm(task);
1152         if (mm && mm != current->mm &&
1153                         !ptrace_may_access(task, mode)) {
1154                 mmput(mm);
1155                 mm = ERR_PTR(-EACCES);
1156         }
1157         mutex_unlock(&task->signal->cred_guard_mutex);
1158
1159         return mm;
1160 }
1161
1162 static void complete_vfork_done(struct task_struct *tsk)
1163 {
1164         struct completion *vfork;
1165
1166         task_lock(tsk);
1167         vfork = tsk->vfork_done;
1168         if (likely(vfork)) {
1169                 tsk->vfork_done = NULL;
1170                 complete(vfork);
1171         }
1172         task_unlock(tsk);
1173 }
1174
1175 static int wait_for_vfork_done(struct task_struct *child,
1176                                 struct completion *vfork)
1177 {
1178         int killed;
1179
1180         freezer_do_not_count();
1181         killed = wait_for_completion_killable(vfork);
1182         freezer_count();
1183
1184         if (killed) {
1185                 task_lock(child);
1186                 child->vfork_done = NULL;
1187                 task_unlock(child);
1188         }
1189
1190         put_task_struct(child);
1191         return killed;
1192 }
1193
1194 /* Please note the differences between mmput and mm_release.
1195  * mmput is called whenever we stop holding onto a mm_struct,
1196  * error success whatever.
1197  *
1198  * mm_release is called after a mm_struct has been removed
1199  * from the current process.
1200  *
1201  * This difference is important for error handling, when we
1202  * only half set up a mm_struct for a new process and need to restore
1203  * the old one.  Because we mmput the new mm_struct before
1204  * restoring the old one. . .
1205  * Eric Biederman 10 January 1998
1206  */
1207 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1208 {
1209         /* Get rid of any futexes when releasing the mm */
1210 #ifdef CONFIG_FUTEX
1211         if (unlikely(tsk->robust_list)) {
1212                 exit_robust_list(tsk);
1213                 tsk->robust_list = NULL;
1214         }
1215 #ifdef CONFIG_COMPAT
1216         if (unlikely(tsk->compat_robust_list)) {
1217                 compat_exit_robust_list(tsk);
1218                 tsk->compat_robust_list = NULL;
1219         }
1220 #endif
1221         if (unlikely(!list_empty(&tsk->pi_state_list)))
1222                 exit_pi_state_list(tsk);
1223 #endif
1224
1225         uprobe_free_utask(tsk);
1226
1227         /* Get rid of any cached register state */
1228         deactivate_mm(tsk, mm);
1229
1230         /*
1231          * Signal userspace if we're not exiting with a core dump
1232          * because we want to leave the value intact for debugging
1233          * purposes.
1234          */
1235         if (tsk->clear_child_tid) {
1236                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1237                     atomic_read(&mm->mm_users) > 1) {
1238                         /*
1239                          * We don't check the error code - if userspace has
1240                          * not set up a proper pointer then tough luck.
1241                          */
1242                         put_user(0, tsk->clear_child_tid);
1243                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1244                                         1, NULL, NULL, 0, 0);
1245                 }
1246                 tsk->clear_child_tid = NULL;
1247         }
1248
1249         /*
1250          * All done, finally we can wake up parent and return this mm to him.
1251          * Also kthread_stop() uses this completion for synchronization.
1252          */
1253         if (tsk->vfork_done)
1254                 complete_vfork_done(tsk);
1255 }
1256
1257 /*
1258  * Allocate a new mm structure and copy contents from the
1259  * mm structure of the passed in task structure.
1260  */
1261 static struct mm_struct *dup_mm(struct task_struct *tsk)
1262 {
1263         struct mm_struct *mm, *oldmm = current->mm;
1264         int err;
1265
1266         mm = allocate_mm();
1267         if (!mm)
1268                 goto fail_nomem;
1269
1270         memcpy(mm, oldmm, sizeof(*mm));
1271
1272         if (!mm_init(mm, tsk, mm->user_ns))
1273                 goto fail_nomem;
1274
1275         err = dup_mmap(mm, oldmm);
1276         if (err)
1277                 goto free_pt;
1278
1279         mm->hiwater_rss = get_mm_rss(mm);
1280         mm->hiwater_vm = mm->total_vm;
1281
1282         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1283                 goto free_pt;
1284
1285         return mm;
1286
1287 free_pt:
1288         /* don't put binfmt in mmput, we haven't got module yet */
1289         mm->binfmt = NULL;
1290         mmput(mm);
1291
1292 fail_nomem:
1293         return NULL;
1294 }
1295
1296 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1297 {
1298         struct mm_struct *mm, *oldmm;
1299         int retval;
1300
1301         tsk->min_flt = tsk->maj_flt = 0;
1302         tsk->nvcsw = tsk->nivcsw = 0;
1303 #ifdef CONFIG_DETECT_HUNG_TASK
1304         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1305         tsk->last_switch_time = 0;
1306 #endif
1307
1308         tsk->mm = NULL;
1309         tsk->active_mm = NULL;
1310
1311         /*
1312          * Are we cloning a kernel thread?
1313          *
1314          * We need to steal a active VM for that..
1315          */
1316         oldmm = current->mm;
1317         if (!oldmm)
1318                 return 0;
1319
1320         /* initialize the new vmacache entries */
1321         vmacache_flush(tsk);
1322
1323         if (clone_flags & CLONE_VM) {
1324                 mmget(oldmm);
1325                 mm = oldmm;
1326                 goto good_mm;
1327         }
1328
1329         retval = -ENOMEM;
1330         mm = dup_mm(tsk);
1331         if (!mm)
1332                 goto fail_nomem;
1333
1334 good_mm:
1335         tsk->mm = mm;
1336         tsk->active_mm = mm;
1337         return 0;
1338
1339 fail_nomem:
1340         return retval;
1341 }
1342
1343 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1344 {
1345         struct fs_struct *fs = current->fs;
1346         if (clone_flags & CLONE_FS) {
1347                 /* tsk->fs is already what we want */
1348                 spin_lock(&fs->lock);
1349                 if (fs->in_exec) {
1350                         spin_unlock(&fs->lock);
1351                         return -EAGAIN;
1352                 }
1353                 fs->users++;
1354                 spin_unlock(&fs->lock);
1355                 return 0;
1356         }
1357         tsk->fs = copy_fs_struct(fs);
1358         if (!tsk->fs)
1359                 return -ENOMEM;
1360         return 0;
1361 }
1362
1363 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1364 {
1365         struct files_struct *oldf, *newf;
1366         int error = 0;
1367
1368         /*
1369          * A background process may not have any files ...
1370          */
1371         oldf = current->files;
1372         if (!oldf)
1373                 goto out;
1374
1375         if (clone_flags & CLONE_FILES) {
1376                 atomic_inc(&oldf->count);
1377                 goto out;
1378         }
1379
1380         newf = dup_fd(oldf, &error);
1381         if (!newf)
1382                 goto out;
1383
1384         tsk->files = newf;
1385         error = 0;
1386 out:
1387         return error;
1388 }
1389
1390 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1391 {
1392 #ifdef CONFIG_BLOCK
1393         struct io_context *ioc = current->io_context;
1394         struct io_context *new_ioc;
1395
1396         if (!ioc)
1397                 return 0;
1398         /*
1399          * Share io context with parent, if CLONE_IO is set
1400          */
1401         if (clone_flags & CLONE_IO) {
1402                 ioc_task_link(ioc);
1403                 tsk->io_context = ioc;
1404         } else if (ioprio_valid(ioc->ioprio)) {
1405                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1406                 if (unlikely(!new_ioc))
1407                         return -ENOMEM;
1408
1409                 new_ioc->ioprio = ioc->ioprio;
1410                 put_io_context(new_ioc);
1411         }
1412 #endif
1413         return 0;
1414 }
1415
1416 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1417 {
1418         struct sighand_struct *sig;
1419
1420         if (clone_flags & CLONE_SIGHAND) {
1421                 atomic_inc(&current->sighand->count);
1422                 return 0;
1423         }
1424         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1425         rcu_assign_pointer(tsk->sighand, sig);
1426         if (!sig)
1427                 return -ENOMEM;
1428
1429         atomic_set(&sig->count, 1);
1430         spin_lock_irq(&current->sighand->siglock);
1431         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1432         spin_unlock_irq(&current->sighand->siglock);
1433         return 0;
1434 }
1435
1436 void __cleanup_sighand(struct sighand_struct *sighand)
1437 {
1438         if (atomic_dec_and_test(&sighand->count)) {
1439                 signalfd_cleanup(sighand);
1440                 /*
1441                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1442                  * without an RCU grace period, see __lock_task_sighand().
1443                  */
1444                 kmem_cache_free(sighand_cachep, sighand);
1445         }
1446 }
1447
1448 #ifdef CONFIG_POSIX_TIMERS
1449 /*
1450  * Initialize POSIX timer handling for a thread group.
1451  */
1452 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1453 {
1454         unsigned long cpu_limit;
1455
1456         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1457         if (cpu_limit != RLIM_INFINITY) {
1458                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1459                 sig->cputimer.running = true;
1460         }
1461
1462         /* The timer lists. */
1463         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1464         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1465         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1466 }
1467 #else
1468 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1469 #endif
1470
1471 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1472 {
1473         struct signal_struct *sig;
1474
1475         if (clone_flags & CLONE_THREAD)
1476                 return 0;
1477
1478         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1479         tsk->signal = sig;
1480         if (!sig)
1481                 return -ENOMEM;
1482
1483         sig->nr_threads = 1;
1484         atomic_set(&sig->live, 1);
1485         atomic_set(&sig->sigcnt, 1);
1486
1487         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1488         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1489         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1490
1491         init_waitqueue_head(&sig->wait_chldexit);
1492         sig->curr_target = tsk;
1493         init_sigpending(&sig->shared_pending);
1494         INIT_HLIST_HEAD(&sig->multiprocess);
1495         seqlock_init(&sig->stats_lock);
1496         prev_cputime_init(&sig->prev_cputime);
1497
1498 #ifdef CONFIG_POSIX_TIMERS
1499         INIT_LIST_HEAD(&sig->posix_timers);
1500         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1501         sig->real_timer.function = it_real_fn;
1502 #endif
1503
1504         task_lock(current->group_leader);
1505         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1506         task_unlock(current->group_leader);
1507
1508         posix_cpu_timers_init_group(sig);
1509
1510         tty_audit_fork(sig);
1511         sched_autogroup_fork(sig);
1512
1513         sig->oom_score_adj = current->signal->oom_score_adj;
1514         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1515
1516         mutex_init(&sig->cred_guard_mutex);
1517
1518         return 0;
1519 }
1520
1521 static void copy_seccomp(struct task_struct *p)
1522 {
1523 #ifdef CONFIG_SECCOMP
1524         /*
1525          * Must be called with sighand->lock held, which is common to
1526          * all threads in the group. Holding cred_guard_mutex is not
1527          * needed because this new task is not yet running and cannot
1528          * be racing exec.
1529          */
1530         assert_spin_locked(&current->sighand->siglock);
1531
1532         /* Ref-count the new filter user, and assign it. */
1533         get_seccomp_filter(current);
1534         p->seccomp = current->seccomp;
1535
1536         /*
1537          * Explicitly enable no_new_privs here in case it got set
1538          * between the task_struct being duplicated and holding the
1539          * sighand lock. The seccomp state and nnp must be in sync.
1540          */
1541         if (task_no_new_privs(current))
1542                 task_set_no_new_privs(p);
1543
1544         /*
1545          * If the parent gained a seccomp mode after copying thread
1546          * flags and between before we held the sighand lock, we have
1547          * to manually enable the seccomp thread flag here.
1548          */
1549         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1550                 set_tsk_thread_flag(p, TIF_SECCOMP);
1551 #endif
1552 }
1553
1554 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1555 {
1556         current->clear_child_tid = tidptr;
1557
1558         return task_pid_vnr(current);
1559 }
1560
1561 static void rt_mutex_init_task(struct task_struct *p)
1562 {
1563         raw_spin_lock_init(&p->pi_lock);
1564 #ifdef CONFIG_RT_MUTEXES
1565         p->pi_waiters = RB_ROOT_CACHED;
1566         p->pi_top_task = NULL;
1567         p->pi_blocked_on = NULL;
1568 #endif
1569 }
1570
1571 #ifdef CONFIG_POSIX_TIMERS
1572 /*
1573  * Initialize POSIX timer handling for a single task.
1574  */
1575 static void posix_cpu_timers_init(struct task_struct *tsk)
1576 {
1577         tsk->cputime_expires.prof_exp = 0;
1578         tsk->cputime_expires.virt_exp = 0;
1579         tsk->cputime_expires.sched_exp = 0;
1580         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1581         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1582         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1583 }
1584 #else
1585 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1586 #endif
1587
1588 static inline void init_task_pid_links(struct task_struct *task)
1589 {
1590         enum pid_type type;
1591
1592         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1593                 INIT_HLIST_NODE(&task->pid_links[type]);
1594         }
1595 }
1596
1597 static inline void
1598 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1599 {
1600         if (type == PIDTYPE_PID)
1601                 task->thread_pid = pid;
1602         else
1603                 task->signal->pids[type] = pid;
1604 }
1605
1606 static inline void rcu_copy_process(struct task_struct *p)
1607 {
1608 #ifdef CONFIG_PREEMPT_RCU
1609         p->rcu_read_lock_nesting = 0;
1610         p->rcu_read_unlock_special.s = 0;
1611         p->rcu_blocked_node = NULL;
1612         INIT_LIST_HEAD(&p->rcu_node_entry);
1613 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1614 #ifdef CONFIG_TASKS_RCU
1615         p->rcu_tasks_holdout = false;
1616         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1617         p->rcu_tasks_idle_cpu = -1;
1618 #endif /* #ifdef CONFIG_TASKS_RCU */
1619 }
1620
1621 /*
1622  * This creates a new process as a copy of the old one,
1623  * but does not actually start it yet.
1624  *
1625  * It copies the registers, and all the appropriate
1626  * parts of the process environment (as per the clone
1627  * flags). The actual kick-off is left to the caller.
1628  */
1629 static __latent_entropy struct task_struct *copy_process(
1630                                         unsigned long clone_flags,
1631                                         unsigned long stack_start,
1632                                         unsigned long stack_size,
1633                                         int __user *child_tidptr,
1634                                         struct pid *pid,
1635                                         int trace,
1636                                         unsigned long tls,
1637                                         int node)
1638 {
1639         int retval;
1640         struct task_struct *p;
1641         struct multiprocess_signals delayed;
1642
1643         /*
1644          * Don't allow sharing the root directory with processes in a different
1645          * namespace
1646          */
1647         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1648                 return ERR_PTR(-EINVAL);
1649
1650         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1651                 return ERR_PTR(-EINVAL);
1652
1653         /*
1654          * Thread groups must share signals as well, and detached threads
1655          * can only be started up within the thread group.
1656          */
1657         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1658                 return ERR_PTR(-EINVAL);
1659
1660         /*
1661          * Shared signal handlers imply shared VM. By way of the above,
1662          * thread groups also imply shared VM. Blocking this case allows
1663          * for various simplifications in other code.
1664          */
1665         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1666                 return ERR_PTR(-EINVAL);
1667
1668         /*
1669          * Siblings of global init remain as zombies on exit since they are
1670          * not reaped by their parent (swapper). To solve this and to avoid
1671          * multi-rooted process trees, prevent global and container-inits
1672          * from creating siblings.
1673          */
1674         if ((clone_flags & CLONE_PARENT) &&
1675                                 current->signal->flags & SIGNAL_UNKILLABLE)
1676                 return ERR_PTR(-EINVAL);
1677
1678         /*
1679          * If the new process will be in a different pid or user namespace
1680          * do not allow it to share a thread group with the forking task.
1681          */
1682         if (clone_flags & CLONE_THREAD) {
1683                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1684                     (task_active_pid_ns(current) !=
1685                                 current->nsproxy->pid_ns_for_children))
1686                         return ERR_PTR(-EINVAL);
1687         }
1688
1689         /*
1690          * Force any signals received before this point to be delivered
1691          * before the fork happens.  Collect up signals sent to multiple
1692          * processes that happen during the fork and delay them so that
1693          * they appear to happen after the fork.
1694          */
1695         sigemptyset(&delayed.signal);
1696         INIT_HLIST_NODE(&delayed.node);
1697
1698         spin_lock_irq(&current->sighand->siglock);
1699         if (!(clone_flags & CLONE_THREAD))
1700                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1701         recalc_sigpending();
1702         spin_unlock_irq(&current->sighand->siglock);
1703         retval = -ERESTARTNOINTR;
1704         if (signal_pending(current))
1705                 goto fork_out;
1706
1707         retval = -ENOMEM;
1708         p = dup_task_struct(current, node);
1709         if (!p)
1710                 goto fork_out;
1711
1712         /*
1713          * This _must_ happen before we call free_task(), i.e. before we jump
1714          * to any of the bad_fork_* labels. This is to avoid freeing
1715          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1716          * kernel threads (PF_KTHREAD).
1717          */
1718         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1719         /*
1720          * Clear TID on mm_release()?
1721          */
1722         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1723
1724         ftrace_graph_init_task(p);
1725
1726         rt_mutex_init_task(p);
1727
1728 #ifdef CONFIG_PROVE_LOCKING
1729         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1730         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1731 #endif
1732         retval = -EAGAIN;
1733         if (atomic_read(&p->real_cred->user->processes) >=
1734                         task_rlimit(p, RLIMIT_NPROC)) {
1735                 if (p->real_cred->user != INIT_USER &&
1736                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1737                         goto bad_fork_free;
1738         }
1739         current->flags &= ~PF_NPROC_EXCEEDED;
1740
1741         retval = copy_creds(p, clone_flags);
1742         if (retval < 0)
1743                 goto bad_fork_free;
1744
1745         /*
1746          * If multiple threads are within copy_process(), then this check
1747          * triggers too late. This doesn't hurt, the check is only there
1748          * to stop root fork bombs.
1749          */
1750         retval = -EAGAIN;
1751         if (nr_threads >= max_threads)
1752                 goto bad_fork_cleanup_count;
1753
1754         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1755         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1756         p->flags |= PF_FORKNOEXEC;
1757         INIT_LIST_HEAD(&p->children);
1758         INIT_LIST_HEAD(&p->sibling);
1759         rcu_copy_process(p);
1760         p->vfork_done = NULL;
1761         spin_lock_init(&p->alloc_lock);
1762
1763         init_sigpending(&p->pending);
1764
1765         p->utime = p->stime = p->gtime = 0;
1766 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1767         p->utimescaled = p->stimescaled = 0;
1768 #endif
1769         prev_cputime_init(&p->prev_cputime);
1770
1771 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1772         seqcount_init(&p->vtime.seqcount);
1773         p->vtime.starttime = 0;
1774         p->vtime.state = VTIME_INACTIVE;
1775 #endif
1776
1777 #if defined(SPLIT_RSS_COUNTING)
1778         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1779 #endif
1780
1781         p->default_timer_slack_ns = current->timer_slack_ns;
1782
1783         task_io_accounting_init(&p->ioac);
1784         acct_clear_integrals(p);
1785
1786         posix_cpu_timers_init(p);
1787
1788         p->start_time = ktime_get_ns();
1789         p->real_start_time = ktime_get_boot_ns();
1790         p->io_context = NULL;
1791         audit_set_context(p, NULL);
1792         cgroup_fork(p);
1793 #ifdef CONFIG_NUMA
1794         p->mempolicy = mpol_dup(p->mempolicy);
1795         if (IS_ERR(p->mempolicy)) {
1796                 retval = PTR_ERR(p->mempolicy);
1797                 p->mempolicy = NULL;
1798                 goto bad_fork_cleanup_threadgroup_lock;
1799         }
1800 #endif
1801 #ifdef CONFIG_CPUSETS
1802         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1803         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1804         seqcount_init(&p->mems_allowed_seq);
1805 #endif
1806 #ifdef CONFIG_TRACE_IRQFLAGS
1807         p->irq_events = 0;
1808         p->hardirqs_enabled = 0;
1809         p->hardirq_enable_ip = 0;
1810         p->hardirq_enable_event = 0;
1811         p->hardirq_disable_ip = _THIS_IP_;
1812         p->hardirq_disable_event = 0;
1813         p->softirqs_enabled = 1;
1814         p->softirq_enable_ip = _THIS_IP_;
1815         p->softirq_enable_event = 0;
1816         p->softirq_disable_ip = 0;
1817         p->softirq_disable_event = 0;
1818         p->hardirq_context = 0;
1819         p->softirq_context = 0;
1820 #endif
1821
1822         p->pagefault_disabled = 0;
1823
1824 #ifdef CONFIG_LOCKDEP
1825         p->lockdep_depth = 0; /* no locks held yet */
1826         p->curr_chain_key = 0;
1827         p->lockdep_recursion = 0;
1828         lockdep_init_task(p);
1829 #endif
1830
1831 #ifdef CONFIG_DEBUG_MUTEXES
1832         p->blocked_on = NULL; /* not blocked yet */
1833 #endif
1834 #ifdef CONFIG_BCACHE
1835         p->sequential_io        = 0;
1836         p->sequential_io_avg    = 0;
1837 #endif
1838
1839         /* Perform scheduler related setup. Assign this task to a CPU. */
1840         retval = sched_fork(clone_flags, p);
1841         if (retval)
1842                 goto bad_fork_cleanup_policy;
1843
1844         retval = perf_event_init_task(p);
1845         if (retval)
1846                 goto bad_fork_cleanup_policy;
1847         retval = audit_alloc(p);
1848         if (retval)
1849                 goto bad_fork_cleanup_perf;
1850         /* copy all the process information */
1851         shm_init_task(p);
1852         retval = security_task_alloc(p, clone_flags);
1853         if (retval)
1854                 goto bad_fork_cleanup_audit;
1855         retval = copy_semundo(clone_flags, p);
1856         if (retval)
1857                 goto bad_fork_cleanup_security;
1858         retval = copy_files(clone_flags, p);
1859         if (retval)
1860                 goto bad_fork_cleanup_semundo;
1861         retval = copy_fs(clone_flags, p);
1862         if (retval)
1863                 goto bad_fork_cleanup_files;
1864         retval = copy_sighand(clone_flags, p);
1865         if (retval)
1866                 goto bad_fork_cleanup_fs;
1867         retval = copy_signal(clone_flags, p);
1868         if (retval)
1869                 goto bad_fork_cleanup_sighand;
1870         retval = copy_mm(clone_flags, p);
1871         if (retval)
1872                 goto bad_fork_cleanup_signal;
1873         retval = copy_namespaces(clone_flags, p);
1874         if (retval)
1875                 goto bad_fork_cleanup_mm;
1876         retval = copy_io(clone_flags, p);
1877         if (retval)
1878                 goto bad_fork_cleanup_namespaces;
1879         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1880         if (retval)
1881                 goto bad_fork_cleanup_io;
1882
1883         if (pid != &init_struct_pid) {
1884                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1885                 if (IS_ERR(pid)) {
1886                         retval = PTR_ERR(pid);
1887                         goto bad_fork_cleanup_thread;
1888                 }
1889         }
1890
1891 #ifdef CONFIG_BLOCK
1892         p->plug = NULL;
1893 #endif
1894 #ifdef CONFIG_FUTEX
1895         p->robust_list = NULL;
1896 #ifdef CONFIG_COMPAT
1897         p->compat_robust_list = NULL;
1898 #endif
1899         INIT_LIST_HEAD(&p->pi_state_list);
1900         p->pi_state_cache = NULL;
1901 #endif
1902         /*
1903          * sigaltstack should be cleared when sharing the same VM
1904          */
1905         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1906                 sas_ss_reset(p);
1907
1908         /*
1909          * Syscall tracing and stepping should be turned off in the
1910          * child regardless of CLONE_PTRACE.
1911          */
1912         user_disable_single_step(p);
1913         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1914 #ifdef TIF_SYSCALL_EMU
1915         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1916 #endif
1917         clear_all_latency_tracing(p);
1918
1919         /* ok, now we should be set up.. */
1920         p->pid = pid_nr(pid);
1921         if (clone_flags & CLONE_THREAD) {
1922                 p->exit_signal = -1;
1923                 p->group_leader = current->group_leader;
1924                 p->tgid = current->tgid;
1925         } else {
1926                 if (clone_flags & CLONE_PARENT)
1927                         p->exit_signal = current->group_leader->exit_signal;
1928                 else
1929                         p->exit_signal = (clone_flags & CSIGNAL);
1930                 p->group_leader = p;
1931                 p->tgid = p->pid;
1932         }
1933
1934         p->nr_dirtied = 0;
1935         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1936         p->dirty_paused_when = 0;
1937
1938         p->pdeath_signal = 0;
1939         INIT_LIST_HEAD(&p->thread_group);
1940         p->task_works = NULL;
1941
1942         cgroup_threadgroup_change_begin(current);
1943         /*
1944          * Ensure that the cgroup subsystem policies allow the new process to be
1945          * forked. It should be noted the the new process's css_set can be changed
1946          * between here and cgroup_post_fork() if an organisation operation is in
1947          * progress.
1948          */
1949         retval = cgroup_can_fork(p);
1950         if (retval)
1951                 goto bad_fork_free_pid;
1952
1953         /*
1954          * Make it visible to the rest of the system, but dont wake it up yet.
1955          * Need tasklist lock for parent etc handling!
1956          */
1957         write_lock_irq(&tasklist_lock);
1958
1959         /* CLONE_PARENT re-uses the old parent */
1960         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1961                 p->real_parent = current->real_parent;
1962                 p->parent_exec_id = current->parent_exec_id;
1963         } else {
1964                 p->real_parent = current;
1965                 p->parent_exec_id = current->self_exec_id;
1966         }
1967
1968         klp_copy_process(p);
1969
1970         spin_lock(&current->sighand->siglock);
1971
1972         /*
1973          * Copy seccomp details explicitly here, in case they were changed
1974          * before holding sighand lock.
1975          */
1976         copy_seccomp(p);
1977
1978         rseq_fork(p, clone_flags);
1979
1980         /* Don't start children in a dying pid namespace */
1981         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
1982                 retval = -ENOMEM;
1983                 goto bad_fork_cancel_cgroup;
1984         }
1985
1986         /* Let kill terminate clone/fork in the middle */
1987         if (fatal_signal_pending(current)) {
1988                 retval = -EINTR;
1989                 goto bad_fork_cancel_cgroup;
1990         }
1991
1992
1993         init_task_pid_links(p);
1994         if (likely(p->pid)) {
1995                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1996
1997                 init_task_pid(p, PIDTYPE_PID, pid);
1998                 if (thread_group_leader(p)) {
1999                         init_task_pid(p, PIDTYPE_TGID, pid);
2000                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2001                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2002
2003                         if (is_child_reaper(pid)) {
2004                                 ns_of_pid(pid)->child_reaper = p;
2005                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2006                         }
2007                         p->signal->shared_pending.signal = delayed.signal;
2008                         p->signal->tty = tty_kref_get(current->signal->tty);
2009                         /*
2010                          * Inherit has_child_subreaper flag under the same
2011                          * tasklist_lock with adding child to the process tree
2012                          * for propagate_has_child_subreaper optimization.
2013                          */
2014                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2015                                                          p->real_parent->signal->is_child_subreaper;
2016                         list_add_tail(&p->sibling, &p->real_parent->children);
2017                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2018                         attach_pid(p, PIDTYPE_TGID);
2019                         attach_pid(p, PIDTYPE_PGID);
2020                         attach_pid(p, PIDTYPE_SID);
2021                         __this_cpu_inc(process_counts);
2022                 } else {
2023                         current->signal->nr_threads++;
2024                         atomic_inc(&current->signal->live);
2025                         atomic_inc(&current->signal->sigcnt);
2026                         task_join_group_stop(p);
2027                         list_add_tail_rcu(&p->thread_group,
2028                                           &p->group_leader->thread_group);
2029                         list_add_tail_rcu(&p->thread_node,
2030                                           &p->signal->thread_head);
2031                 }
2032                 attach_pid(p, PIDTYPE_PID);
2033                 nr_threads++;
2034         }
2035         total_forks++;
2036         hlist_del_init(&delayed.node);
2037         spin_unlock(&current->sighand->siglock);
2038         syscall_tracepoint_update(p);
2039         write_unlock_irq(&tasklist_lock);
2040
2041         proc_fork_connector(p);
2042         cgroup_post_fork(p);
2043         cgroup_threadgroup_change_end(current);
2044         perf_event_fork(p);
2045
2046         trace_task_newtask(p, clone_flags);
2047         uprobe_copy_process(p, clone_flags);
2048
2049         return p;
2050
2051 bad_fork_cancel_cgroup:
2052         spin_unlock(&current->sighand->siglock);
2053         write_unlock_irq(&tasklist_lock);
2054         cgroup_cancel_fork(p);
2055 bad_fork_free_pid:
2056         cgroup_threadgroup_change_end(current);
2057         if (pid != &init_struct_pid)
2058                 free_pid(pid);
2059 bad_fork_cleanup_thread:
2060         exit_thread(p);
2061 bad_fork_cleanup_io:
2062         if (p->io_context)
2063                 exit_io_context(p);
2064 bad_fork_cleanup_namespaces:
2065         exit_task_namespaces(p);
2066 bad_fork_cleanup_mm:
2067         if (p->mm)
2068                 mmput(p->mm);
2069 bad_fork_cleanup_signal:
2070         if (!(clone_flags & CLONE_THREAD))
2071                 free_signal_struct(p->signal);
2072 bad_fork_cleanup_sighand:
2073         __cleanup_sighand(p->sighand);
2074 bad_fork_cleanup_fs:
2075         exit_fs(p); /* blocking */
2076 bad_fork_cleanup_files:
2077         exit_files(p); /* blocking */
2078 bad_fork_cleanup_semundo:
2079         exit_sem(p);
2080 bad_fork_cleanup_security:
2081         security_task_free(p);
2082 bad_fork_cleanup_audit:
2083         audit_free(p);
2084 bad_fork_cleanup_perf:
2085         perf_event_free_task(p);
2086 bad_fork_cleanup_policy:
2087         lockdep_free_task(p);
2088 #ifdef CONFIG_NUMA
2089         mpol_put(p->mempolicy);
2090 bad_fork_cleanup_threadgroup_lock:
2091 #endif
2092         delayacct_tsk_free(p);
2093 bad_fork_cleanup_count:
2094         atomic_dec(&p->cred->user->processes);
2095         exit_creds(p);
2096 bad_fork_free:
2097         p->state = TASK_DEAD;
2098         put_task_stack(p);
2099         free_task(p);
2100 fork_out:
2101         spin_lock_irq(&current->sighand->siglock);
2102         hlist_del_init(&delayed.node);
2103         spin_unlock_irq(&current->sighand->siglock);
2104         return ERR_PTR(retval);
2105 }
2106
2107 static inline void init_idle_pids(struct task_struct *idle)
2108 {
2109         enum pid_type type;
2110
2111         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2112                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2113                 init_task_pid(idle, type, &init_struct_pid);
2114         }
2115 }
2116
2117 struct task_struct *fork_idle(int cpu)
2118 {
2119         struct task_struct *task;
2120         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2121                             cpu_to_node(cpu));
2122         if (!IS_ERR(task)) {
2123                 init_idle_pids(task);
2124                 init_idle(task, cpu);
2125         }
2126
2127         return task;
2128 }
2129
2130 /*
2131  *  Ok, this is the main fork-routine.
2132  *
2133  * It copies the process, and if successful kick-starts
2134  * it and waits for it to finish using the VM if required.
2135  */
2136 long _do_fork(unsigned long clone_flags,
2137               unsigned long stack_start,
2138               unsigned long stack_size,
2139               int __user *parent_tidptr,
2140               int __user *child_tidptr,
2141               unsigned long tls)
2142 {
2143         struct completion vfork;
2144         struct pid *pid;
2145         struct task_struct *p;
2146         int trace = 0;
2147         long nr;
2148
2149         /*
2150          * Determine whether and which event to report to ptracer.  When
2151          * called from kernel_thread or CLONE_UNTRACED is explicitly
2152          * requested, no event is reported; otherwise, report if the event
2153          * for the type of forking is enabled.
2154          */
2155         if (!(clone_flags & CLONE_UNTRACED)) {
2156                 if (clone_flags & CLONE_VFORK)
2157                         trace = PTRACE_EVENT_VFORK;
2158                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2159                         trace = PTRACE_EVENT_CLONE;
2160                 else
2161                         trace = PTRACE_EVENT_FORK;
2162
2163                 if (likely(!ptrace_event_enabled(current, trace)))
2164                         trace = 0;
2165         }
2166
2167         p = copy_process(clone_flags, stack_start, stack_size,
2168                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2169         add_latent_entropy();
2170
2171         if (IS_ERR(p))
2172                 return PTR_ERR(p);
2173
2174         /*
2175          * Do this prior waking up the new thread - the thread pointer
2176          * might get invalid after that point, if the thread exits quickly.
2177          */
2178         trace_sched_process_fork(current, p);
2179
2180         pid = get_task_pid(p, PIDTYPE_PID);
2181         nr = pid_vnr(pid);
2182
2183         if (clone_flags & CLONE_PARENT_SETTID)
2184                 put_user(nr, parent_tidptr);
2185
2186         if (clone_flags & CLONE_VFORK) {
2187                 p->vfork_done = &vfork;
2188                 init_completion(&vfork);
2189                 get_task_struct(p);
2190         }
2191
2192         wake_up_new_task(p);
2193
2194         /* forking complete and child started to run, tell ptracer */
2195         if (unlikely(trace))
2196                 ptrace_event_pid(trace, pid);
2197
2198         if (clone_flags & CLONE_VFORK) {
2199                 if (!wait_for_vfork_done(p, &vfork))
2200                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2201         }
2202
2203         put_pid(pid);
2204         return nr;
2205 }
2206
2207 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2208 /* For compatibility with architectures that call do_fork directly rather than
2209  * using the syscall entry points below. */
2210 long do_fork(unsigned long clone_flags,
2211               unsigned long stack_start,
2212               unsigned long stack_size,
2213               int __user *parent_tidptr,
2214               int __user *child_tidptr)
2215 {
2216         return _do_fork(clone_flags, stack_start, stack_size,
2217                         parent_tidptr, child_tidptr, 0);
2218 }
2219 #endif
2220
2221 /*
2222  * Create a kernel thread.
2223  */
2224 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2225 {
2226         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2227                 (unsigned long)arg, NULL, NULL, 0);
2228 }
2229
2230 #ifdef __ARCH_WANT_SYS_FORK
2231 SYSCALL_DEFINE0(fork)
2232 {
2233 #ifdef CONFIG_MMU
2234         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2235 #else
2236         /* can not support in nommu mode */
2237         return -EINVAL;
2238 #endif
2239 }
2240 #endif
2241
2242 #ifdef __ARCH_WANT_SYS_VFORK
2243 SYSCALL_DEFINE0(vfork)
2244 {
2245         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2246                         0, NULL, NULL, 0);
2247 }
2248 #endif
2249
2250 #ifdef __ARCH_WANT_SYS_CLONE
2251 #ifdef CONFIG_CLONE_BACKWARDS
2252 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2253                  int __user *, parent_tidptr,
2254                  unsigned long, tls,
2255                  int __user *, child_tidptr)
2256 #elif defined(CONFIG_CLONE_BACKWARDS2)
2257 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2258                  int __user *, parent_tidptr,
2259                  int __user *, child_tidptr,
2260                  unsigned long, tls)
2261 #elif defined(CONFIG_CLONE_BACKWARDS3)
2262 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2263                 int, stack_size,
2264                 int __user *, parent_tidptr,
2265                 int __user *, child_tidptr,
2266                 unsigned long, tls)
2267 #else
2268 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2269                  int __user *, parent_tidptr,
2270                  int __user *, child_tidptr,
2271                  unsigned long, tls)
2272 #endif
2273 {
2274         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2275 }
2276 #endif
2277
2278 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2279 {
2280         struct task_struct *leader, *parent, *child;
2281         int res;
2282
2283         read_lock(&tasklist_lock);
2284         leader = top = top->group_leader;
2285 down:
2286         for_each_thread(leader, parent) {
2287                 list_for_each_entry(child, &parent->children, sibling) {
2288                         res = visitor(child, data);
2289                         if (res) {
2290                                 if (res < 0)
2291                                         goto out;
2292                                 leader = child;
2293                                 goto down;
2294                         }
2295 up:
2296                         ;
2297                 }
2298         }
2299
2300         if (leader != top) {
2301                 child = leader;
2302                 parent = child->real_parent;
2303                 leader = parent->group_leader;
2304                 goto up;
2305         }
2306 out:
2307         read_unlock(&tasklist_lock);
2308 }
2309
2310 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2311 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2312 #endif
2313
2314 static void sighand_ctor(void *data)
2315 {
2316         struct sighand_struct *sighand = data;
2317
2318         spin_lock_init(&sighand->siglock);
2319         init_waitqueue_head(&sighand->signalfd_wqh);
2320 }
2321
2322 void __init proc_caches_init(void)
2323 {
2324         unsigned int mm_size;
2325
2326         sighand_cachep = kmem_cache_create("sighand_cache",
2327                         sizeof(struct sighand_struct), 0,
2328                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2329                         SLAB_ACCOUNT, sighand_ctor);
2330         signal_cachep = kmem_cache_create("signal_cache",
2331                         sizeof(struct signal_struct), 0,
2332                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2333                         NULL);
2334         files_cachep = kmem_cache_create("files_cache",
2335                         sizeof(struct files_struct), 0,
2336                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2337                         NULL);
2338         fs_cachep = kmem_cache_create("fs_cache",
2339                         sizeof(struct fs_struct), 0,
2340                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2341                         NULL);
2342
2343         /*
2344          * The mm_cpumask is located at the end of mm_struct, and is
2345          * dynamically sized based on the maximum CPU number this system
2346          * can have, taking hotplug into account (nr_cpu_ids).
2347          */
2348         mm_size = sizeof(struct mm_struct) + cpumask_size();
2349
2350         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2351                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2352                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2353                         offsetof(struct mm_struct, saved_auxv),
2354                         sizeof_field(struct mm_struct, saved_auxv),
2355                         NULL);
2356         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2357         mmap_init();
2358         nsproxy_cache_init();
2359 }
2360
2361 /*
2362  * Check constraints on flags passed to the unshare system call.
2363  */
2364 static int check_unshare_flags(unsigned long unshare_flags)
2365 {
2366         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2367                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2368                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2369                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2370                 return -EINVAL;
2371         /*
2372          * Not implemented, but pretend it works if there is nothing
2373          * to unshare.  Note that unsharing the address space or the
2374          * signal handlers also need to unshare the signal queues (aka
2375          * CLONE_THREAD).
2376          */
2377         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2378                 if (!thread_group_empty(current))
2379                         return -EINVAL;
2380         }
2381         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2382                 if (atomic_read(&current->sighand->count) > 1)
2383                         return -EINVAL;
2384         }
2385         if (unshare_flags & CLONE_VM) {
2386                 if (!current_is_single_threaded())
2387                         return -EINVAL;
2388         }
2389
2390         return 0;
2391 }
2392
2393 /*
2394  * Unshare the filesystem structure if it is being shared
2395  */
2396 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2397 {
2398         struct fs_struct *fs = current->fs;
2399
2400         if (!(unshare_flags & CLONE_FS) || !fs)
2401                 return 0;
2402
2403         /* don't need lock here; in the worst case we'll do useless copy */
2404         if (fs->users == 1)
2405                 return 0;
2406
2407         *new_fsp = copy_fs_struct(fs);
2408         if (!*new_fsp)
2409                 return -ENOMEM;
2410
2411         return 0;
2412 }
2413
2414 /*
2415  * Unshare file descriptor table if it is being shared
2416  */
2417 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2418 {
2419         struct files_struct *fd = current->files;
2420         int error = 0;
2421
2422         if ((unshare_flags & CLONE_FILES) &&
2423             (fd && atomic_read(&fd->count) > 1)) {
2424                 *new_fdp = dup_fd(fd, &error);
2425                 if (!*new_fdp)
2426                         return error;
2427         }
2428
2429         return 0;
2430 }
2431
2432 /*
2433  * unshare allows a process to 'unshare' part of the process
2434  * context which was originally shared using clone.  copy_*
2435  * functions used by do_fork() cannot be used here directly
2436  * because they modify an inactive task_struct that is being
2437  * constructed. Here we are modifying the current, active,
2438  * task_struct.
2439  */
2440 int ksys_unshare(unsigned long unshare_flags)
2441 {
2442         struct fs_struct *fs, *new_fs = NULL;
2443         struct files_struct *fd, *new_fd = NULL;
2444         struct cred *new_cred = NULL;
2445         struct nsproxy *new_nsproxy = NULL;
2446         int do_sysvsem = 0;
2447         int err;
2448
2449         /*
2450          * If unsharing a user namespace must also unshare the thread group
2451          * and unshare the filesystem root and working directories.
2452          */
2453         if (unshare_flags & CLONE_NEWUSER)
2454                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2455         /*
2456          * If unsharing vm, must also unshare signal handlers.
2457          */
2458         if (unshare_flags & CLONE_VM)
2459                 unshare_flags |= CLONE_SIGHAND;
2460         /*
2461          * If unsharing a signal handlers, must also unshare the signal queues.
2462          */
2463         if (unshare_flags & CLONE_SIGHAND)
2464                 unshare_flags |= CLONE_THREAD;
2465         /*
2466          * If unsharing namespace, must also unshare filesystem information.
2467          */
2468         if (unshare_flags & CLONE_NEWNS)
2469                 unshare_flags |= CLONE_FS;
2470
2471         err = check_unshare_flags(unshare_flags);
2472         if (err)
2473                 goto bad_unshare_out;
2474         /*
2475          * CLONE_NEWIPC must also detach from the undolist: after switching
2476          * to a new ipc namespace, the semaphore arrays from the old
2477          * namespace are unreachable.
2478          */
2479         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2480                 do_sysvsem = 1;
2481         err = unshare_fs(unshare_flags, &new_fs);
2482         if (err)
2483                 goto bad_unshare_out;
2484         err = unshare_fd(unshare_flags, &new_fd);
2485         if (err)
2486                 goto bad_unshare_cleanup_fs;
2487         err = unshare_userns(unshare_flags, &new_cred);
2488         if (err)
2489                 goto bad_unshare_cleanup_fd;
2490         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2491                                          new_cred, new_fs);
2492         if (err)
2493                 goto bad_unshare_cleanup_cred;
2494
2495         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2496                 if (do_sysvsem) {
2497                         /*
2498                          * CLONE_SYSVSEM is equivalent to sys_exit().
2499                          */
2500                         exit_sem(current);
2501                 }
2502                 if (unshare_flags & CLONE_NEWIPC) {
2503                         /* Orphan segments in old ns (see sem above). */
2504                         exit_shm(current);
2505                         shm_init_task(current);
2506                 }
2507
2508                 if (new_nsproxy)
2509                         switch_task_namespaces(current, new_nsproxy);
2510
2511                 task_lock(current);
2512
2513                 if (new_fs) {
2514                         fs = current->fs;
2515                         spin_lock(&fs->lock);
2516                         current->fs = new_fs;
2517                         if (--fs->users)
2518                                 new_fs = NULL;
2519                         else
2520                                 new_fs = fs;
2521                         spin_unlock(&fs->lock);
2522                 }
2523
2524                 if (new_fd) {
2525                         fd = current->files;
2526                         current->files = new_fd;
2527                         new_fd = fd;
2528                 }
2529
2530                 task_unlock(current);
2531
2532                 if (new_cred) {
2533                         /* Install the new user namespace */
2534                         commit_creds(new_cred);
2535                         new_cred = NULL;
2536                 }
2537         }
2538
2539         perf_event_namespaces(current);
2540
2541 bad_unshare_cleanup_cred:
2542         if (new_cred)
2543                 put_cred(new_cred);
2544 bad_unshare_cleanup_fd:
2545         if (new_fd)
2546                 put_files_struct(new_fd);
2547
2548 bad_unshare_cleanup_fs:
2549         if (new_fs)
2550                 free_fs_struct(new_fs);
2551
2552 bad_unshare_out:
2553         return err;
2554 }
2555
2556 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2557 {
2558         return ksys_unshare(unshare_flags);
2559 }
2560
2561 /*
2562  *      Helper to unshare the files of the current task.
2563  *      We don't want to expose copy_files internals to
2564  *      the exec layer of the kernel.
2565  */
2566
2567 int unshare_files(struct files_struct **displaced)
2568 {
2569         struct task_struct *task = current;
2570         struct files_struct *copy = NULL;
2571         int error;
2572
2573         error = unshare_fd(CLONE_FILES, &copy);
2574         if (error || !copy) {
2575                 *displaced = NULL;
2576                 return error;
2577         }
2578         *displaced = task->files;
2579         task_lock(task);
2580         task->files = copy;
2581         task_unlock(task);
2582         return 0;
2583 }
2584
2585 int sysctl_max_threads(struct ctl_table *table, int write,
2586                        void __user *buffer, size_t *lenp, loff_t *ppos)
2587 {
2588         struct ctl_table t;
2589         int ret;
2590         int threads = max_threads;
2591         int min = MIN_THREADS;
2592         int max = MAX_THREADS;
2593
2594         t = *table;
2595         t.data = &threads;
2596         t.extra1 = &min;
2597         t.extra2 = &max;
2598
2599         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2600         if (ret || !write)
2601                 return ret;
2602
2603         set_max_threads(threads);
2604
2605         return 0;
2606 }