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