kernel/fork.c

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