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