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 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 167
 168 /*
 169  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 170  * kmemcache based allocator.
 171  */
 172 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 173
 174 #ifdef CONFIG_VMAP_STACK
 175 /*
 176  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 177  * flush.  Try to minimize the number of calls by caching stacks.
 178  */
 179 #define NR_CACHED_STACKS 2
 180 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 181
 182 static int free_vm_stack_cache(unsigned int cpu)
 183 {
 184         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 185         int i;
 186
 187         for (i = 0; i < NR_CACHED_STACKS; i++) {
 188                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 189
 190                 if (!vm_stack)
 191                         continue;
 192
 193                 vfree(vm_stack->addr);
 194                 cached_vm_stacks[i] = NULL;
 195         }
 196
 197         return 0;
 198 }
 199 #endif
 200
 201 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 202 {
 203 #ifdef CONFIG_VMAP_STACK
 204         void *stack;
 205         int i;
 206
 207         for (i = 0; i < NR_CACHED_STACKS; i++) {
 208                 struct vm_struct *s;
 209
 210                 s = this_cpu_xchg(cached_stacks[i], NULL);
 211
 212                 if (!s)
 213                         continue;
 214
 215                 /* Clear stale pointers from reused stack. */
 216                 memset(s->addr, 0, THREAD_SIZE);
 217
 218                 tsk->stack_vm_area = s;
 219                 tsk->stack = s->addr;
 220                 return s->addr;
 221         }
 222
 223         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 224                                      VMALLOC_START, VMALLOC_END,
 225                                      THREADINFO_GFP,
 226                                      PAGE_KERNEL,
 227                                      0, node, __builtin_return_address(0));
 228
 229         /*
 230          * We can't call find_vm_area() in interrupt context, and
 231          * free_thread_stack() can be called in interrupt context,
 232          * so cache the vm_struct.
 233          */
 234         if (stack) {
 235                 tsk->stack_vm_area = find_vm_area(stack);
 236                 tsk->stack = stack;
 237         }
 238         return stack;
 239 #else
 240         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 241                                              THREAD_SIZE_ORDER);
 242
 243         if (likely(page)) {
 244                 tsk->stack = page_address(page);
 245                 return tsk->stack;
 246         }
 247         return NULL;
 248 #endif
 249 }
 250
 251 static inline void free_thread_stack(struct task_struct *tsk)
 252 {
 253 #ifdef CONFIG_VMAP_STACK
 254         if (task_stack_vm_area(tsk)) {
 255                 int i;
 256
 257                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 258                         if (this_cpu_cmpxchg(cached_stacks[i],
 259                                         NULL, tsk->stack_vm_area) != NULL)
 260                                 continue;
 261
 262                         return;
 263                 }
 264
 265                 vfree_atomic(tsk->stack);
 266                 return;
 267         }
 268 #endif
 269
 270         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 271 }
 272 # else
 273 static struct kmem_cache *thread_stack_cache;
 274
 275 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 276                                                   int node)
 277 {
 278         unsigned long *stack;
 279         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 280         tsk->stack = stack;
 281         return stack;
 282 }
 283
 284 static void free_thread_stack(struct task_struct *tsk)
 285 {
 286         kmem_cache_free(thread_stack_cache, tsk->stack);
 287 }
 288
 289 void thread_stack_cache_init(void)
 290 {
 291         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 292                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 293                                         THREAD_SIZE, NULL);
 294         BUG_ON(thread_stack_cache == NULL);
 295 }
 296 # endif
 297 #endif
 298
 299 /* SLAB cache for signal_struct structures (tsk->signal) */
 300 static struct kmem_cache *signal_cachep;
 301
 302 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 303 struct kmem_cache *sighand_cachep;
 304
 305 /* SLAB cache for files_struct structures (tsk->files) */
 306 struct kmem_cache *files_cachep;
 307
 308 /* SLAB cache for fs_struct structures (tsk->fs) */
 309 struct kmem_cache *fs_cachep;
 310
 311 /* SLAB cache for vm_area_struct structures */
 312 static struct kmem_cache *vm_area_cachep;
 313
 314 /* SLAB cache for mm_struct structures (tsk->mm) */
 315 static struct kmem_cache *mm_cachep;
 316
 317 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 318 {
 319         struct vm_area_struct *vma;
 320
 321         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 322         if (vma)
 323                 vma_init(vma, mm);
 324         return vma;
 325 }
 326
 327 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 328 {
 329         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 330
 331         if (new) {
 332                 *new = *orig;
 333                 INIT_LIST_HEAD(&new->anon_vma_chain);
 334         }
 335         return new;
 336 }
 337
 338 void vm_area_free(struct vm_area_struct *vma)
 339 {
 340         kmem_cache_free(vm_area_cachep, vma);
 341 }
 342
 343 static void account_kernel_stack(struct task_struct *tsk, int account)
 344 {
 345         void *stack = task_stack_page(tsk);
 346         struct vm_struct *vm = task_stack_vm_area(tsk);
 347
 348         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 349
 350         if (vm) {
 351                 int i;
 352
 353                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 354
 355                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 356                         mod_zone_page_state(page_zone(vm->pages[i]),
 357                                             NR_KERNEL_STACK_KB,
 358                                             PAGE_SIZE / 1024 * account);
 359                 }
 360
 361                 /* All stack pages belong to the same memcg. */
 362                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 363                                      account * (THREAD_SIZE / 1024));
 364         } else {
 365                 /*
 366                  * All stack pages are in the same zone and belong to the
 367                  * same memcg.
 368                  */
 369                 struct page *first_page = virt_to_page(stack);
 370
 371                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 372                                     THREAD_SIZE / 1024 * account);
 373
 374                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 375                                      account * (THREAD_SIZE / 1024));
 376         }
 377 }
 378
 379 static void release_task_stack(struct task_struct *tsk)
 380 {
 381         if (WARN_ON(tsk->state != TASK_DEAD))
 382                 return;  /* Better to leak the stack than to free prematurely */
 383
 384         account_kernel_stack(tsk, -1);
 385         free_thread_stack(tsk);
 386         tsk->stack = NULL;
 387 #ifdef CONFIG_VMAP_STACK
 388         tsk->stack_vm_area = NULL;
 389 #endif
 390 }
 391
 392 #ifdef CONFIG_THREAD_INFO_IN_TASK
 393 void put_task_stack(struct task_struct *tsk)
 394 {
 395         if (atomic_dec_and_test(&tsk->stack_refcount))
 396                 release_task_stack(tsk);
 397 }
 398 #endif
 399
 400 void free_task(struct task_struct *tsk)
 401 {
 402 #ifndef CONFIG_THREAD_INFO_IN_TASK
 403         /*
 404          * The task is finally done with both the stack and thread_info,
 405          * so free both.
 406          */
 407         release_task_stack(tsk);
 408 #else
 409         /*
 410          * If the task had a separate stack allocation, it should be gone
 411          * by now.
 412          */
 413         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 414 #endif
 415         rt_mutex_debug_task_free(tsk);
 416         ftrace_graph_exit_task(tsk);
 417         put_seccomp_filter(tsk);
 418         arch_release_task_struct(tsk);
 419         if (tsk->flags & PF_KTHREAD)
 420                 free_kthread_struct(tsk);
 421         free_task_struct(tsk);
 422 }
 423 EXPORT_SYMBOL(free_task);
 424
 425 #ifdef CONFIG_MMU
 426 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 427                                         struct mm_struct *oldmm)
 428 {
 429         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 430         struct rb_node **rb_link, *rb_parent;
 431         int retval;
 432         unsigned long charge;
 433         LIST_HEAD(uf);
 434
 435         uprobe_start_dup_mmap();
 436         if (down_write_killable(&oldmm->mmap_sem)) {
 437                 retval = -EINTR;
 438                 goto fail_uprobe_end;
 439         }
 440         flush_cache_dup_mm(oldmm);
 441         uprobe_dup_mmap(oldmm, mm);
 442         /*
 443          * Not linked in yet - no deadlock potential:
 444          */
 445         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 446
 447         /* No ordering required: file already has been exposed. */
 448         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 449
 450         mm->total_vm = oldmm->total_vm;
 451         mm->data_vm = oldmm->data_vm;
 452         mm->exec_vm = oldmm->exec_vm;
 453         mm->stack_vm = oldmm->stack_vm;
 454
 455         rb_link = &mm->mm_rb.rb_node;
 456         rb_parent = NULL;
 457         pprev = &mm->mmap;
 458         retval = ksm_fork(mm, oldmm);
 459         if (retval)
 460                 goto out;
 461         retval = khugepaged_fork(mm, oldmm);
 462         if (retval)
 463                 goto out;
 464
 465         prev = NULL;
 466         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 467                 struct file *file;
 468
 469                 if (mpnt->vm_flags & VM_DONTCOPY) {
 470                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 471                         continue;
 472                 }
 473                 charge = 0;
 474                 /*
 475                  * Don't duplicate many vmas if we've been oom-killed (for
 476                  * example)
 477                  */
 478                 if (fatal_signal_pending(current)) {
 479                         retval = -EINTR;
 480                         goto out;
 481                 }
 482                 if (mpnt->vm_flags & VM_ACCOUNT) {
 483                         unsigned long len = vma_pages(mpnt);
 484
 485                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 486                                 goto fail_nomem;
 487                         charge = len;
 488                 }
 489                 tmp = vm_area_dup(mpnt);
 490                 if (!tmp)
 491                         goto fail_nomem;
 492                 retval = vma_dup_policy(mpnt, tmp);
 493                 if (retval)
 494                         goto fail_nomem_policy;
 495                 tmp->vm_mm = mm;
 496                 retval = dup_userfaultfd(tmp, &uf);
 497                 if (retval)
 498                         goto fail_nomem_anon_vma_fork;
 499                 if (tmp->vm_flags & VM_WIPEONFORK) {
 500                         /* VM_WIPEONFORK gets a clean slate in the child. */
 501                         tmp->anon_vma = NULL;
 502                         if (anon_vma_prepare(tmp))
 503                                 goto fail_nomem_anon_vma_fork;
 504                 } else if (anon_vma_fork(tmp, mpnt))
 505                         goto fail_nomem_anon_vma_fork;
 506                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 507                 tmp->vm_next = tmp->vm_prev = NULL;
 508                 file = tmp->vm_file;
 509                 if (file) {
 510                         struct inode *inode = file_inode(file);
 511                         struct address_space *mapping = file->f_mapping;
 512
 513                         get_file(file);
 514                         if (tmp->vm_flags & VM_DENYWRITE)
 515                                 atomic_dec(&inode->i_writecount);
 516                         i_mmap_lock_write(mapping);
 517                         if (tmp->vm_flags & VM_SHARED)
 518                                 atomic_inc(&mapping->i_mmap_writable);
 519                         flush_dcache_mmap_lock(mapping);
 520                         /* insert tmp into the share list, just after mpnt */
 521                         vma_interval_tree_insert_after(tmp, mpnt,
 522                                         &mapping->i_mmap);
 523                         flush_dcache_mmap_unlock(mapping);
 524                         i_mmap_unlock_write(mapping);
 525                 }
 526
 527                 /*
 528                  * Clear hugetlb-related page reserves for children. This only
 529                  * affects MAP_PRIVATE mappings. Faults generated by the child
 530                  * are not guaranteed to succeed, even if read-only
 531                  */
 532                 if (is_vm_hugetlb_page(tmp))
 533                         reset_vma_resv_huge_pages(tmp);
 534
 535                 /*
 536                  * Link in the new vma and copy the page table entries.
 537                  */
 538                 *pprev = tmp;
 539                 pprev = &tmp->vm_next;
 540                 tmp->vm_prev = prev;
 541                 prev = tmp;
 542
 543                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 544                 rb_link = &tmp->vm_rb.rb_right;
 545                 rb_parent = &tmp->vm_rb;
 546
 547                 mm->map_count++;
 548                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 549                         retval = copy_page_range(mm, oldmm, mpnt);
 550
 551                 if (tmp->vm_ops && tmp->vm_ops->open)
 552                         tmp->vm_ops->open(tmp);
 553
 554                 if (retval)
 555                         goto out;
 556         }
 557         /* a new mm has just been created */
 558         retval = arch_dup_mmap(oldmm, mm);
 559 out:
 560         up_write(&mm->mmap_sem);
 561         flush_tlb_mm(oldmm);
 562         up_write(&oldmm->mmap_sem);
 563         dup_userfaultfd_complete(&uf);
 564 fail_uprobe_end:
 565         uprobe_end_dup_mmap();
 566         return retval;
 567 fail_nomem_anon_vma_fork:
 568         mpol_put(vma_policy(tmp));
 569 fail_nomem_policy:
 570         vm_area_free(tmp);
 571 fail_nomem:
 572         retval = -ENOMEM;
 573         vm_unacct_memory(charge);
 574         goto out;
 575 }
 576
 577 static inline int mm_alloc_pgd(struct mm_struct *mm)
 578 {
 579         mm->pgd = pgd_alloc(mm);
 580         if (unlikely(!mm->pgd))
 581                 return -ENOMEM;
 582         return 0;
 583 }
 584
 585 static inline void mm_free_pgd(struct mm_struct *mm)
 586 {
 587         pgd_free(mm, mm->pgd);
 588 }
 589 #else
 590 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 591 {
 592         down_write(&oldmm->mmap_sem);
 593         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 594         up_write(&oldmm->mmap_sem);
 595         return 0;
 596 }
 597 #define mm_alloc_pgd(mm)        (0)
 598 #define mm_free_pgd(mm)
 599 #endif /* CONFIG_MMU */
 600
 601 static void check_mm(struct mm_struct *mm)
 602 {
 603         int i;
 604
 605         for (i = 0; i < NR_MM_COUNTERS; i++) {
 606                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 607
 608                 if (unlikely(x))
 609                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 610                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 611         }
 612
 613         if (mm_pgtables_bytes(mm))
 614                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 615                                 mm_pgtables_bytes(mm));
 616
 617 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 618         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 619 #endif
 620 }
 621
 622 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 623 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 624
 625 /*
 626  * Called when the last reference to the mm
 627  * is dropped: either by a lazy thread or by
 628  * mmput. Free the page directory and the mm.
 629  */
 630 void __mmdrop(struct mm_struct *mm)
 631 {
 632         BUG_ON(mm == &init_mm);
 633         WARN_ON_ONCE(mm == current->mm);
 634         WARN_ON_ONCE(mm == current->active_mm);
 635         mm_free_pgd(mm);
 636         destroy_context(mm);
 637         hmm_mm_destroy(mm);
 638         mmu_notifier_mm_destroy(mm);
 639         check_mm(mm);
 640         put_user_ns(mm->user_ns);
 641         free_mm(mm);
 642 }
 643 EXPORT_SYMBOL_GPL(__mmdrop);
 644
 645 static void mmdrop_async_fn(struct work_struct *work)
 646 {
 647         struct mm_struct *mm;
 648
 649         mm = container_of(work, struct mm_struct, async_put_work);
 650         __mmdrop(mm);
 651 }
 652
 653 static void mmdrop_async(struct mm_struct *mm)
 654 {
 655         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 656                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 657                 schedule_work(&mm->async_put_work);
 658         }
 659 }
 660
 661 static inline void free_signal_struct(struct signal_struct *sig)
 662 {
 663         taskstats_tgid_free(sig);
 664         sched_autogroup_exit(sig);
 665         /*
 666          * __mmdrop is not safe to call from softirq context on x86 due to
 667          * pgd_dtor so postpone it to the async context
 668          */
 669         if (sig->oom_mm)
 670                 mmdrop_async(sig->oom_mm);
 671         kmem_cache_free(signal_cachep, sig);
 672 }
 673
 674 static inline void put_signal_struct(struct signal_struct *sig)
 675 {
 676         if (atomic_dec_and_test(&sig->sigcnt))
 677                 free_signal_struct(sig);
 678 }
 679
 680 void __put_task_struct(struct task_struct *tsk)
 681 {
 682         WARN_ON(!tsk->exit_state);
 683         WARN_ON(atomic_read(&tsk->usage));
 684         WARN_ON(tsk == current);
 685
 686         cgroup_free(tsk);
 687         task_numa_free(tsk, true);
 688         security_task_free(tsk);
 689         exit_creds(tsk);
 690         delayacct_tsk_free(tsk);
 691         put_signal_struct(tsk->signal);
 692
 693         if (!profile_handoff_task(tsk))
 694                 free_task(tsk);
 695 }
 696 EXPORT_SYMBOL_GPL(__put_task_struct);
 697
 698 void __init __weak arch_task_cache_init(void) { }
 699
 700 /*
 701  * set_max_threads
 702  */
 703 static void set_max_threads(unsigned int max_threads_suggested)
 704 {
 705         u64 threads;
 706
 707         /*
 708          * The number of threads shall be limited such that the thread
 709          * structures may only consume a small part of the available memory.
 710          */
 711         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 712                 threads = MAX_THREADS;
 713         else
 714                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 715                                     (u64) THREAD_SIZE * 8UL);
 716
 717         if (threads > max_threads_suggested)
 718                 threads = max_threads_suggested;
 719
 720         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 721 }
 722
 723 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 724 /* Initialized by the architecture: */
 725 int arch_task_struct_size __read_mostly;
 726 #endif
 727
 728 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 729 {
 730         /* Fetch thread_struct whitelist for the architecture. */
 731         arch_thread_struct_whitelist(offset, size);
 732
 733         /*
 734          * Handle zero-sized whitelist or empty thread_struct, otherwise
 735          * adjust offset to position of thread_struct in task_struct.
 736          */
 737         if (unlikely(*size == 0))
 738                 *offset = 0;
 739         else
 740                 *offset += offsetof(struct task_struct, thread);
 741 }
 742
 743 void __init fork_init(void)
 744 {
 745         int i;
 746 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 747 #ifndef ARCH_MIN_TASKALIGN
 748 #define ARCH_MIN_TASKALIGN      0
 749 #endif
 750         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 751         unsigned long useroffset, usersize;
 752
 753         /* create a slab on which task_structs can be allocated */
 754         task_struct_whitelist(&useroffset, &usersize);
 755         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 756                         arch_task_struct_size, align,
 757                         SLAB_PANIC|SLAB_ACCOUNT,
 758                         useroffset, usersize, NULL);
 759 #endif
 760
 761         /* do the arch specific task caches init */
 762         arch_task_cache_init();
 763
 764         set_max_threads(MAX_THREADS);
 765
 766         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 767         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 768         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 769                 init_task.signal->rlim[RLIMIT_NPROC];
 770
 771         for (i = 0; i < UCOUNT_COUNTS; i++) {
 772                 init_user_ns.ucount_max[i] = max_threads/2;
 773         }
 774
 775 #ifdef CONFIG_VMAP_STACK
 776         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 777                           NULL, free_vm_stack_cache);
 778 #endif
 779
 780         lockdep_init_task(&init_task);
 781 }
 782
 783 int __weak arch_dup_task_struct(struct task_struct *dst,
 784                                                struct task_struct *src)
 785 {
 786         *dst = *src;
 787         return 0;
 788 }
 789
 790 void set_task_stack_end_magic(struct task_struct *tsk)
 791 {
 792         unsigned long *stackend;
 793
 794         stackend = end_of_stack(tsk);
 795         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 796 }
 797
 798 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 799 {
 800         struct task_struct *tsk;
 801         unsigned long *stack;
 802         struct vm_struct *stack_vm_area;
 803         int err;
 804
 805         if (node == NUMA_NO_NODE)
 806                 node = tsk_fork_get_node(orig);
 807         tsk = alloc_task_struct_node(node);
 808         if (!tsk)
 809                 return NULL;
 810
 811         stack = alloc_thread_stack_node(tsk, node);
 812         if (!stack)
 813                 goto free_tsk;
 814
 815         stack_vm_area = task_stack_vm_area(tsk);
 816
 817         err = arch_dup_task_struct(tsk, orig);
 818
 819         /*
 820          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 821          * sure they're properly initialized before using any stack-related
 822          * functions again.
 823          */
 824         tsk->stack = stack;
 825 #ifdef CONFIG_VMAP_STACK
 826         tsk->stack_vm_area = stack_vm_area;
 827 #endif
 828 #ifdef CONFIG_THREAD_INFO_IN_TASK
 829         atomic_set(&tsk->stack_refcount, 1);
 830 #endif
 831
 832         if (err)
 833                 goto free_stack;
 834
 835 #ifdef CONFIG_SECCOMP
 836         /*
 837          * We must handle setting up seccomp filters once we're under
 838          * the sighand lock in case orig has changed between now and
 839          * then. Until then, filter must be NULL to avoid messing up
 840          * the usage counts on the error path calling free_task.
 841          */
 842         tsk->seccomp.filter = NULL;
 843 #endif
 844
 845         setup_thread_stack(tsk, orig);
 846         clear_user_return_notifier(tsk);
 847         clear_tsk_need_resched(tsk);
 848         set_task_stack_end_magic(tsk);
 849
 850 #ifdef CONFIG_STACKPROTECTOR
 851         tsk->stack_canary = get_random_canary();
 852 #endif
 853
 854         /*
 855          * One for us, one for whoever does the "release_task()" (usually
 856          * parent)
 857          */
 858         atomic_set(&tsk->usage, 2);
 859 #ifdef CONFIG_BLK_DEV_IO_TRACE
 860         tsk->btrace_seq = 0;
 861 #endif
 862         tsk->splice_pipe = NULL;
 863         tsk->task_frag.page = NULL;
 864         tsk->wake_q.next = NULL;
 865
 866         account_kernel_stack(tsk, 1);
 867
 868         kcov_task_init(tsk);
 869
 870 #ifdef CONFIG_FAULT_INJECTION
 871         tsk->fail_nth = 0;
 872 #endif
 873
 874 #ifdef CONFIG_BLK_CGROUP
 875         tsk->throttle_queue = NULL;
 876         tsk->use_memdelay = 0;
 877 #endif
 878
 879 #ifdef CONFIG_MEMCG
 880         tsk->active_memcg = NULL;
 881 #endif
 882         return tsk;
 883
 884 free_stack:
 885         free_thread_stack(tsk);
 886 free_tsk:
 887         free_task_struct(tsk);
 888         return NULL;
 889 }
 890
 891 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 892
 893 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 894
 895 static int __init coredump_filter_setup(char *s)
 896 {
 897         default_dump_filter =
 898                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 899                 MMF_DUMP_FILTER_MASK;
 900         return 1;
 901 }
 902
 903 __setup("coredump_filter=", coredump_filter_setup);
 904
 905 #include <linux/init_task.h>
 906
 907 static void mm_init_aio(struct mm_struct *mm)
 908 {
 909 #ifdef CONFIG_AIO
 910         spin_lock_init(&mm->ioctx_lock);
 911         mm->ioctx_table = NULL;
 912 #endif
 913 }
 914
 915 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 916                                            struct task_struct *p)
 917 {
 918 #ifdef CONFIG_MEMCG
 919         if (mm->owner == p)
 920                 WRITE_ONCE(mm->owner, NULL);
 921 #endif
 922 }
 923
 924 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 925 {
 926 #ifdef CONFIG_MEMCG
 927         mm->owner = p;
 928 #endif
 929 }
 930
 931 static void mm_init_uprobes_state(struct mm_struct *mm)
 932 {
 933 #ifdef CONFIG_UPROBES
 934         mm->uprobes_state.xol_area = NULL;
 935 #endif
 936 }
 937
 938 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 939         struct user_namespace *user_ns)
 940 {
 941         mm->mmap = NULL;
 942         mm->mm_rb = RB_ROOT;
 943         mm->vmacache_seqnum = 0;
 944         atomic_set(&mm->mm_users, 1);
 945         atomic_set(&mm->mm_count, 1);
 946         init_rwsem(&mm->mmap_sem);
 947         INIT_LIST_HEAD(&mm->mmlist);
 948         mm->core_state = NULL;
 949         mm_pgtables_bytes_init(mm);
 950         mm->map_count = 0;
 951         mm->locked_vm = 0;
 952         mm->pinned_vm = 0;
 953         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 954         spin_lock_init(&mm->page_table_lock);
 955         spin_lock_init(&mm->arg_lock);
 956         mm_init_cpumask(mm);
 957         mm_init_aio(mm);
 958         mm_init_owner(mm, p);
 959         RCU_INIT_POINTER(mm->exe_file, NULL);
 960         mmu_notifier_mm_init(mm);
 961         hmm_mm_init(mm);
 962         init_tlb_flush_pending(mm);
 963 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 964         mm->pmd_huge_pte = NULL;
 965 #endif
 966         mm_init_uprobes_state(mm);
 967
 968         if (current->mm) {
 969                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 970                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 971         } else {
 972                 mm->flags = default_dump_filter;
 973                 mm->def_flags = 0;
 974         }
 975
 976         if (mm_alloc_pgd(mm))
 977                 goto fail_nopgd;
 978
 979         if (init_new_context(p, mm))
 980                 goto fail_nocontext;
 981
 982         mm->user_ns = get_user_ns(user_ns);
 983         return mm;
 984
 985 fail_nocontext:
 986         mm_free_pgd(mm);
 987 fail_nopgd:
 988         free_mm(mm);
 989         return NULL;
 990 }
 991
 992 /*
 993  * Allocate and initialize an mm_struct.
 994  */
 995 struct mm_struct *mm_alloc(void)
 996 {
 997         struct mm_struct *mm;
 998
 999         mm = allocate_mm();
1000         if (!mm)
1001                 return NULL;
1002
1003         memset(mm, 0, sizeof(*mm));
1004         return mm_init(mm, current, current_user_ns());
1005 }
1006
1007 static inline void __mmput(struct mm_struct *mm)
1008 {
1009         VM_BUG_ON(atomic_read(&mm->mm_users));
1010
1011         uprobe_clear_state(mm);
1012         exit_aio(mm);
1013         ksm_exit(mm);
1014         khugepaged_exit(mm); /* must run before exit_mmap */
1015         exit_mmap(mm);
1016         mm_put_huge_zero_page(mm);
1017         set_mm_exe_file(mm, NULL);
1018         if (!list_empty(&mm->mmlist)) {
1019                 spin_lock(&mmlist_lock);
1020                 list_del(&mm->mmlist);
1021                 spin_unlock(&mmlist_lock);
1022         }
1023         if (mm->binfmt)
1024                 module_put(mm->binfmt->module);
1025         mmdrop(mm);
1026 }
1027
1028 /*
1029  * Decrement the use count and release all resources for an mm.
1030  */
1031 void mmput(struct mm_struct *mm)
1032 {
1033         might_sleep();
1034
1035         if (atomic_dec_and_test(&mm->mm_users))
1036                 __mmput(mm);
1037 }
1038 EXPORT_SYMBOL_GPL(mmput);
1039
1040 #ifdef CONFIG_MMU
1041 static void mmput_async_fn(struct work_struct *work)
1042 {
1043         struct mm_struct *mm = container_of(work, struct mm_struct,
1044                                             async_put_work);
1045
1046         __mmput(mm);
1047 }
1048
1049 void mmput_async(struct mm_struct *mm)
1050 {
1051         if (atomic_dec_and_test(&mm->mm_users)) {
1052                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1053                 schedule_work(&mm->async_put_work);
1054         }
1055 }
1056 #endif
1057
1058 /**
1059  * set_mm_exe_file - change a reference to the mm's executable file
1060  *
1061  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1062  *
1063  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1064  * invocations: in mmput() nobody alive left, in execve task is single
1065  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1066  * mm->exe_file, but does so without using set_mm_exe_file() in order
1067  * to do avoid the need for any locks.
1068  */
1069 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1070 {
1071         struct file *old_exe_file;
1072
1073         /*
1074          * It is safe to dereference the exe_file without RCU as
1075          * this function is only called if nobody else can access
1076          * this mm -- see comment above for justification.
1077          */
1078         old_exe_file = rcu_dereference_raw(mm->exe_file);
1079
1080         if (new_exe_file)
1081                 get_file(new_exe_file);
1082         rcu_assign_pointer(mm->exe_file, new_exe_file);
1083         if (old_exe_file)
1084                 fput(old_exe_file);
1085 }
1086
1087 /**
1088  * get_mm_exe_file - acquire a reference to the mm's executable file
1089  *
1090  * Returns %NULL if mm has no associated executable file.
1091  * User must release file via fput().
1092  */
1093 struct file *get_mm_exe_file(struct mm_struct *mm)
1094 {
1095         struct file *exe_file;
1096
1097         rcu_read_lock();
1098         exe_file = rcu_dereference(mm->exe_file);
1099         if (exe_file && !get_file_rcu(exe_file))
1100                 exe_file = NULL;
1101         rcu_read_unlock();
1102         return exe_file;
1103 }
1104 EXPORT_SYMBOL(get_mm_exe_file);
1105
1106 /**
1107  * get_task_exe_file - acquire a reference to the task's executable file
1108  *
1109  * Returns %NULL if task's mm (if any) has no associated executable file or
1110  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1111  * User must release file via fput().
1112  */
1113 struct file *get_task_exe_file(struct task_struct *task)
1114 {
1115         struct file *exe_file = NULL;
1116         struct mm_struct *mm;
1117
1118         task_lock(task);
1119         mm = task->mm;
1120         if (mm) {
1121                 if (!(task->flags & PF_KTHREAD))
1122                         exe_file = get_mm_exe_file(mm);
1123         }
1124         task_unlock(task);
1125         return exe_file;
1126 }
1127 EXPORT_SYMBOL(get_task_exe_file);
1128
1129 /**
1130  * get_task_mm - acquire a reference to the task's mm
1131  *
1132  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1133  * this kernel workthread has transiently adopted a user mm with use_mm,
1134  * to do its AIO) is not set and if so returns a reference to it, after
1135  * bumping up the use count.  User must release the mm via mmput()
1136  * after use.  Typically used by /proc and ptrace.
1137  */
1138 struct mm_struct *get_task_mm(struct task_struct *task)
1139 {
1140         struct mm_struct *mm;
1141
1142         task_lock(task);
1143         mm = task->mm;
1144         if (mm) {
1145                 if (task->flags & PF_KTHREAD)
1146                         mm = NULL;
1147                 else
1148                         mmget(mm);
1149         }
1150         task_unlock(task);
1151         return mm;
1152 }
1153 EXPORT_SYMBOL_GPL(get_task_mm);
1154
1155 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1156 {
1157         struct mm_struct *mm;
1158         int err;
1159
1160         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1161         if (err)
1162                 return ERR_PTR(err);
1163
1164         mm = get_task_mm(task);
1165         if (mm && mm != current->mm &&
1166                         !ptrace_may_access(task, mode)) {
1167                 mmput(mm);
1168                 mm = ERR_PTR(-EACCES);
1169         }
1170         mutex_unlock(&task->signal->cred_guard_mutex);
1171
1172         return mm;
1173 }
1174
1175 static void complete_vfork_done(struct task_struct *tsk)
1176 {
1177         struct completion *vfork;
1178
1179         task_lock(tsk);
1180         vfork = tsk->vfork_done;
1181         if (likely(vfork)) {
1182                 tsk->vfork_done = NULL;
1183                 complete(vfork);
1184         }
1185         task_unlock(tsk);
1186 }
1187
1188 static int wait_for_vfork_done(struct task_struct *child,
1189                                 struct completion *vfork)
1190 {
1191         int killed;
1192
1193         freezer_do_not_count();
1194         killed = wait_for_completion_killable(vfork);
1195         freezer_count();
1196
1197         if (killed) {
1198                 task_lock(child);
1199                 child->vfork_done = NULL;
1200                 task_unlock(child);
1201         }
1202
1203         put_task_struct(child);
1204         return killed;
1205 }
1206
1207 /* Please note the differences between mmput and mm_release.
1208  * mmput is called whenever we stop holding onto a mm_struct,
1209  * error success whatever.
1210  *
1211  * mm_release is called after a mm_struct has been removed
1212  * from the current process.
1213  *
1214  * This difference is important for error handling, when we
1215  * only half set up a mm_struct for a new process and need to restore
1216  * the old one.  Because we mmput the new mm_struct before
1217  * restoring the old one. . .
1218  * Eric Biederman 10 January 1998
1219  */
1220 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1221 {
1222         /* Get rid of any futexes when releasing the mm */
1223 #ifdef CONFIG_FUTEX
1224         if (unlikely(tsk->robust_list)) {
1225                 exit_robust_list(tsk);
1226                 tsk->robust_list = NULL;
1227         }
1228 #ifdef CONFIG_COMPAT
1229         if (unlikely(tsk->compat_robust_list)) {
1230                 compat_exit_robust_list(tsk);
1231                 tsk->compat_robust_list = NULL;
1232         }
1233 #endif
1234         if (unlikely(!list_empty(&tsk->pi_state_list)))
1235                 exit_pi_state_list(tsk);
1236 #endif
1237
1238         uprobe_free_utask(tsk);
1239
1240         /* Get rid of any cached register state */
1241         deactivate_mm(tsk, mm);
1242
1243         /*
1244          * Signal userspace if we're not exiting with a core dump
1245          * because we want to leave the value intact for debugging
1246          * purposes.
1247          */
1248         if (tsk->clear_child_tid) {
1249                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1250                     atomic_read(&mm->mm_users) > 1) {
1251                         /*
1252                          * We don't check the error code - if userspace has
1253                          * not set up a proper pointer then tough luck.
1254                          */
1255                         put_user(0, tsk->clear_child_tid);
1256                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1257                                         1, NULL, NULL, 0, 0);
1258                 }
1259                 tsk->clear_child_tid = NULL;
1260         }
1261
1262         /*
1263          * All done, finally we can wake up parent and return this mm to him.
1264          * Also kthread_stop() uses this completion for synchronization.
1265          */
1266         if (tsk->vfork_done)
1267                 complete_vfork_done(tsk);
1268 }
1269
1270 /*
1271  * Allocate a new mm structure and copy contents from the
1272  * mm structure of the passed in task structure.
1273  */
1274 static struct mm_struct *dup_mm(struct task_struct *tsk)
1275 {
1276         struct mm_struct *mm, *oldmm = current->mm;
1277         int err;
1278
1279         mm = allocate_mm();
1280         if (!mm)
1281                 goto fail_nomem;
1282
1283         memcpy(mm, oldmm, sizeof(*mm));
1284
1285         if (!mm_init(mm, tsk, mm->user_ns))
1286                 goto fail_nomem;
1287
1288         err = dup_mmap(mm, oldmm);
1289         if (err)
1290                 goto free_pt;
1291
1292         mm->hiwater_rss = get_mm_rss(mm);
1293         mm->hiwater_vm = mm->total_vm;
1294
1295         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1296                 goto free_pt;
1297
1298         return mm;
1299
1300 free_pt:
1301         /* don't put binfmt in mmput, we haven't got module yet */
1302         mm->binfmt = NULL;
1303         mm_init_owner(mm, NULL);
1304         mmput(mm);
1305
1306 fail_nomem:
1307         return NULL;
1308 }
1309
1310 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1311 {
1312         struct mm_struct *mm, *oldmm;
1313         int retval;
1314
1315         tsk->min_flt = tsk->maj_flt = 0;
1316         tsk->nvcsw = tsk->nivcsw = 0;
1317 #ifdef CONFIG_DETECT_HUNG_TASK
1318         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1319         tsk->last_switch_time = 0;
1320 #endif
1321
1322         tsk->mm = NULL;
1323         tsk->active_mm = NULL;
1324
1325         /*
1326          * Are we cloning a kernel thread?
1327          *
1328          * We need to steal a active VM for that..
1329          */
1330         oldmm = current->mm;
1331         if (!oldmm)
1332                 return 0;
1333
1334         /* initialize the new vmacache entries */
1335         vmacache_flush(tsk);
1336
1337         if (clone_flags & CLONE_VM) {
1338                 mmget(oldmm);
1339                 mm = oldmm;
1340                 goto good_mm;
1341         }
1342
1343         retval = -ENOMEM;
1344         mm = dup_mm(tsk);
1345         if (!mm)
1346                 goto fail_nomem;
1347
1348 good_mm:
1349         tsk->mm = mm;
1350         tsk->active_mm = mm;
1351         return 0;
1352
1353 fail_nomem:
1354         return retval;
1355 }
1356
1357 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1358 {
1359         struct fs_struct *fs = current->fs;
1360         if (clone_flags & CLONE_FS) {
1361                 /* tsk->fs is already what we want */
1362                 spin_lock(&fs->lock);
1363                 if (fs->in_exec) {
1364                         spin_unlock(&fs->lock);
1365                         return -EAGAIN;
1366                 }
1367                 fs->users++;
1368                 spin_unlock(&fs->lock);
1369                 return 0;
1370         }
1371         tsk->fs = copy_fs_struct(fs);
1372         if (!tsk->fs)
1373                 return -ENOMEM;
1374         return 0;
1375 }
1376
1377 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1378 {
1379         struct files_struct *oldf, *newf;
1380         int error = 0;
1381
1382         /*
1383          * A background process may not have any files ...
1384          */
1385         oldf = current->files;
1386         if (!oldf)
1387                 goto out;
1388
1389         if (clone_flags & CLONE_FILES) {
1390                 atomic_inc(&oldf->count);
1391                 goto out;
1392         }
1393
1394         newf = dup_fd(oldf, &error);
1395         if (!newf)
1396                 goto out;
1397
1398         tsk->files = newf;
1399         error = 0;
1400 out:
1401         return error;
1402 }
1403
1404 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1405 {
1406 #ifdef CONFIG_BLOCK
1407         struct io_context *ioc = current->io_context;
1408         struct io_context *new_ioc;
1409
1410         if (!ioc)
1411                 return 0;
1412         /*
1413          * Share io context with parent, if CLONE_IO is set
1414          */
1415         if (clone_flags & CLONE_IO) {
1416                 ioc_task_link(ioc);
1417                 tsk->io_context = ioc;
1418         } else if (ioprio_valid(ioc->ioprio)) {
1419                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1420                 if (unlikely(!new_ioc))
1421                         return -ENOMEM;
1422
1423                 new_ioc->ioprio = ioc->ioprio;
1424                 put_io_context(new_ioc);
1425         }
1426 #endif
1427         return 0;
1428 }
1429
1430 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1431 {
1432         struct sighand_struct *sig;
1433
1434         if (clone_flags & CLONE_SIGHAND) {
1435                 atomic_inc(&current->sighand->count);
1436                 return 0;
1437         }
1438         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1439         rcu_assign_pointer(tsk->sighand, sig);
1440         if (!sig)
1441                 return -ENOMEM;
1442
1443         atomic_set(&sig->count, 1);
1444         spin_lock_irq(&current->sighand->siglock);
1445         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1446         spin_unlock_irq(&current->sighand->siglock);
1447         return 0;
1448 }
1449
1450 void __cleanup_sighand(struct sighand_struct *sighand)
1451 {
1452         if (atomic_dec_and_test(&sighand->count)) {
1453                 signalfd_cleanup(sighand);
1454                 /*
1455                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1456                  * without an RCU grace period, see __lock_task_sighand().
1457                  */
1458                 kmem_cache_free(sighand_cachep, sighand);
1459         }
1460 }
1461
1462 #ifdef CONFIG_POSIX_TIMERS
1463 /*
1464  * Initialize POSIX timer handling for a thread group.
1465  */
1466 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1467 {
1468         unsigned long cpu_limit;
1469
1470         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1471         if (cpu_limit != RLIM_INFINITY) {
1472                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1473                 sig->cputimer.running = true;
1474         }
1475
1476         /* The timer lists. */
1477         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1478         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1479         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1480 }
1481 #else
1482 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1483 #endif
1484
1485 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1486 {
1487         struct signal_struct *sig;
1488
1489         if (clone_flags & CLONE_THREAD)
1490                 return 0;
1491
1492         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1493         tsk->signal = sig;
1494         if (!sig)
1495                 return -ENOMEM;
1496
1497         sig->nr_threads = 1;
1498         atomic_set(&sig->live, 1);
1499         atomic_set(&sig->sigcnt, 1);
1500
1501         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1502         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1503         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1504
1505         init_waitqueue_head(&sig->wait_chldexit);
1506         sig->curr_target = tsk;
1507         init_sigpending(&sig->shared_pending);
1508         INIT_HLIST_HEAD(&sig->multiprocess);
1509         seqlock_init(&sig->stats_lock);
1510         prev_cputime_init(&sig->prev_cputime);
1511
1512 #ifdef CONFIG_POSIX_TIMERS
1513         INIT_LIST_HEAD(&sig->posix_timers);
1514         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1515         sig->real_timer.function = it_real_fn;
1516 #endif
1517
1518         task_lock(current->group_leader);
1519         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1520         task_unlock(current->group_leader);
1521
1522         posix_cpu_timers_init_group(sig);
1523
1524         tty_audit_fork(sig);
1525         sched_autogroup_fork(sig);
1526
1527         sig->oom_score_adj = current->signal->oom_score_adj;
1528         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1529
1530         mutex_init(&sig->cred_guard_mutex);
1531
1532         return 0;
1533 }
1534
1535 static void copy_seccomp(struct task_struct *p)
1536 {
1537 #ifdef CONFIG_SECCOMP
1538         /*
1539          * Must be called with sighand->lock held, which is common to
1540          * all threads in the group. Holding cred_guard_mutex is not
1541          * needed because this new task is not yet running and cannot
1542          * be racing exec.
1543          */
1544         assert_spin_locked(&current->sighand->siglock);
1545
1546         /* Ref-count the new filter user, and assign it. */
1547         get_seccomp_filter(current);
1548         p->seccomp = current->seccomp;
1549
1550         /*
1551          * Explicitly enable no_new_privs here in case it got set
1552          * between the task_struct being duplicated and holding the
1553          * sighand lock. The seccomp state and nnp must be in sync.
1554          */
1555         if (task_no_new_privs(current))
1556                 task_set_no_new_privs(p);
1557
1558         /*
1559          * If the parent gained a seccomp mode after copying thread
1560          * flags and between before we held the sighand lock, we have
1561          * to manually enable the seccomp thread flag here.
1562          */
1563         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1564                 set_tsk_thread_flag(p, TIF_SECCOMP);
1565 #endif
1566 }
1567
1568 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1569 {
1570         current->clear_child_tid = tidptr;
1571
1572         return task_pid_vnr(current);
1573 }
1574
1575 static void rt_mutex_init_task(struct task_struct *p)
1576 {
1577         raw_spin_lock_init(&p->pi_lock);
1578 #ifdef CONFIG_RT_MUTEXES
1579         p->pi_waiters = RB_ROOT_CACHED;
1580         p->pi_top_task = NULL;
1581         p->pi_blocked_on = NULL;
1582 #endif
1583 }
1584
1585 #ifdef CONFIG_POSIX_TIMERS
1586 /*
1587  * Initialize POSIX timer handling for a single task.
1588  */
1589 static void posix_cpu_timers_init(struct task_struct *tsk)
1590 {
1591         tsk->cputime_expires.prof_exp = 0;
1592         tsk->cputime_expires.virt_exp = 0;
1593         tsk->cputime_expires.sched_exp = 0;
1594         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1595         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1596         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1597 }
1598 #else
1599 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1600 #endif
1601
1602 static inline void init_task_pid_links(struct task_struct *task)
1603 {
1604         enum pid_type type;
1605
1606         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1607                 INIT_HLIST_NODE(&task->pid_links[type]);
1608         }
1609 }
1610
1611 static inline void
1612 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1613 {
1614         if (type == PIDTYPE_PID)
1615                 task->thread_pid = pid;
1616         else
1617                 task->signal->pids[type] = pid;
1618 }
1619
1620 static inline void rcu_copy_process(struct task_struct *p)
1621 {
1622 #ifdef CONFIG_PREEMPT_RCU
1623         p->rcu_read_lock_nesting = 0;
1624         p->rcu_read_unlock_special.s = 0;
1625         p->rcu_blocked_node = NULL;
1626         INIT_LIST_HEAD(&p->rcu_node_entry);
1627 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1628 #ifdef CONFIG_TASKS_RCU
1629         p->rcu_tasks_holdout = false;
1630         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1631         p->rcu_tasks_idle_cpu = -1;
1632 #endif /* #ifdef CONFIG_TASKS_RCU */
1633 }
1634
1635 static void __delayed_free_task(struct rcu_head *rhp)
1636 {
1637         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1638
1639         free_task(tsk);
1640 }
1641
1642 static __always_inline void delayed_free_task(struct task_struct *tsk)
1643 {
1644         if (IS_ENABLED(CONFIG_MEMCG))
1645                 call_rcu(&tsk->rcu, __delayed_free_task);
1646         else
1647                 free_task(tsk);
1648 }
1649
1650 /*
1651  * This creates a new process as a copy of the old one,
1652  * but does not actually start it yet.
1653  *
1654  * It copies the registers, and all the appropriate
1655  * parts of the process environment (as per the clone
1656  * flags). The actual kick-off is left to the caller.
1657  */
1658 static __latent_entropy struct task_struct *copy_process(
1659                                         unsigned long clone_flags,
1660                                         unsigned long stack_start,
1661                                         unsigned long stack_size,
1662                                         int __user *child_tidptr,
1663                                         struct pid *pid,
1664                                         int trace,
1665                                         unsigned long tls,
1666                                         int node)
1667 {
1668         int retval;
1669         struct task_struct *p;
1670         struct multiprocess_signals delayed;
1671
1672         /*
1673          * Don't allow sharing the root directory with processes in a different
1674          * namespace
1675          */
1676         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1677                 return ERR_PTR(-EINVAL);
1678
1679         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1680                 return ERR_PTR(-EINVAL);
1681
1682         /*
1683          * Thread groups must share signals as well, and detached threads
1684          * can only be started up within the thread group.
1685          */
1686         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1687                 return ERR_PTR(-EINVAL);
1688
1689         /*
1690          * Shared signal handlers imply shared VM. By way of the above,
1691          * thread groups also imply shared VM. Blocking this case allows
1692          * for various simplifications in other code.
1693          */
1694         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1695                 return ERR_PTR(-EINVAL);
1696
1697         /*
1698          * Siblings of global init remain as zombies on exit since they are
1699          * not reaped by their parent (swapper). To solve this and to avoid
1700          * multi-rooted process trees, prevent global and container-inits
1701          * from creating siblings.
1702          */
1703         if ((clone_flags & CLONE_PARENT) &&
1704                                 current->signal->flags & SIGNAL_UNKILLABLE)
1705                 return ERR_PTR(-EINVAL);
1706
1707         /*
1708          * If the new process will be in a different pid or user namespace
1709          * do not allow it to share a thread group with the forking task.
1710          */
1711         if (clone_flags & CLONE_THREAD) {
1712                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1713                     (task_active_pid_ns(current) !=
1714                                 current->nsproxy->pid_ns_for_children))
1715                         return ERR_PTR(-EINVAL);
1716         }
1717
1718         /*
1719          * Force any signals received before this point to be delivered
1720          * before the fork happens.  Collect up signals sent to multiple
1721          * processes that happen during the fork and delay them so that
1722          * they appear to happen after the fork.
1723          */
1724         sigemptyset(&delayed.signal);
1725         INIT_HLIST_NODE(&delayed.node);
1726
1727         spin_lock_irq(&current->sighand->siglock);
1728         if (!(clone_flags & CLONE_THREAD))
1729                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1730         recalc_sigpending();
1731         spin_unlock_irq(&current->sighand->siglock);
1732         retval = -ERESTARTNOINTR;
1733         if (signal_pending(current))
1734                 goto fork_out;
1735
1736         retval = -ENOMEM;
1737         p = dup_task_struct(current, node);
1738         if (!p)
1739                 goto fork_out;
1740
1741         /*
1742          * This _must_ happen before we call free_task(), i.e. before we jump
1743          * to any of the bad_fork_* labels. This is to avoid freeing
1744          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1745          * kernel threads (PF_KTHREAD).
1746          */
1747         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1748         /*
1749          * Clear TID on mm_release()?
1750          */
1751         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1752
1753         ftrace_graph_init_task(p);
1754
1755         rt_mutex_init_task(p);
1756
1757 #ifdef CONFIG_PROVE_LOCKING
1758         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1759         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1760 #endif
1761         retval = -EAGAIN;
1762         if (atomic_read(&p->real_cred->user->processes) >=
1763                         task_rlimit(p, RLIMIT_NPROC)) {
1764                 if (p->real_cred->user != INIT_USER &&
1765                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1766                         goto bad_fork_free;
1767         }
1768         current->flags &= ~PF_NPROC_EXCEEDED;
1769
1770         retval = copy_creds(p, clone_flags);
1771         if (retval < 0)
1772                 goto bad_fork_free;
1773
1774         /*
1775          * If multiple threads are within copy_process(), then this check
1776          * triggers too late. This doesn't hurt, the check is only there
1777          * to stop root fork bombs.
1778          */
1779         retval = -EAGAIN;
1780         if (nr_threads >= max_threads)
1781                 goto bad_fork_cleanup_count;
1782
1783         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1784         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1785         p->flags |= PF_FORKNOEXEC;
1786         INIT_LIST_HEAD(&p->children);
1787         INIT_LIST_HEAD(&p->sibling);
1788         rcu_copy_process(p);
1789         p->vfork_done = NULL;
1790         spin_lock_init(&p->alloc_lock);
1791
1792         init_sigpending(&p->pending);
1793
1794         p->utime = p->stime = p->gtime = 0;
1795 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1796         p->utimescaled = p->stimescaled = 0;
1797 #endif
1798         prev_cputime_init(&p->prev_cputime);
1799
1800 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1801         seqcount_init(&p->vtime.seqcount);
1802         p->vtime.starttime = 0;
1803         p->vtime.state = VTIME_INACTIVE;
1804 #endif
1805
1806 #if defined(SPLIT_RSS_COUNTING)
1807         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1808 #endif
1809
1810         p->default_timer_slack_ns = current->timer_slack_ns;
1811
1812         task_io_accounting_init(&p->ioac);
1813         acct_clear_integrals(p);
1814
1815         posix_cpu_timers_init(p);
1816
1817         p->io_context = NULL;
1818         audit_set_context(p, NULL);
1819         cgroup_fork(p);
1820 #ifdef CONFIG_NUMA
1821         p->mempolicy = mpol_dup(p->mempolicy);
1822         if (IS_ERR(p->mempolicy)) {
1823                 retval = PTR_ERR(p->mempolicy);
1824                 p->mempolicy = NULL;
1825                 goto bad_fork_cleanup_threadgroup_lock;
1826         }
1827 #endif
1828 #ifdef CONFIG_CPUSETS
1829         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1830         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1831         seqcount_init(&p->mems_allowed_seq);
1832 #endif
1833 #ifdef CONFIG_TRACE_IRQFLAGS
1834         p->irq_events = 0;
1835         p->hardirqs_enabled = 0;
1836         p->hardirq_enable_ip = 0;
1837         p->hardirq_enable_event = 0;
1838         p->hardirq_disable_ip = _THIS_IP_;
1839         p->hardirq_disable_event = 0;
1840         p->softirqs_enabled = 1;
1841         p->softirq_enable_ip = _THIS_IP_;
1842         p->softirq_enable_event = 0;
1843         p->softirq_disable_ip = 0;
1844         p->softirq_disable_event = 0;
1845         p->hardirq_context = 0;
1846         p->softirq_context = 0;
1847 #endif
1848
1849         p->pagefault_disabled = 0;
1850
1851 #ifdef CONFIG_LOCKDEP
1852         p->lockdep_depth = 0; /* no locks held yet */
1853         p->curr_chain_key = 0;
1854         p->lockdep_recursion = 0;
1855         lockdep_init_task(p);
1856 #endif
1857
1858 #ifdef CONFIG_DEBUG_MUTEXES
1859         p->blocked_on = NULL; /* not blocked yet */
1860 #endif
1861 #ifdef CONFIG_BCACHE
1862         p->sequential_io        = 0;
1863         p->sequential_io_avg    = 0;
1864 #endif
1865
1866         /* Perform scheduler related setup. Assign this task to a CPU. */
1867         retval = sched_fork(clone_flags, p);
1868         if (retval)
1869                 goto bad_fork_cleanup_policy;
1870
1871         retval = perf_event_init_task(p);
1872         if (retval)
1873                 goto bad_fork_cleanup_policy;
1874         retval = audit_alloc(p);
1875         if (retval)
1876                 goto bad_fork_cleanup_perf;
1877         /* copy all the process information */
1878         shm_init_task(p);
1879         retval = security_task_alloc(p, clone_flags);
1880         if (retval)
1881                 goto bad_fork_cleanup_audit;
1882         retval = copy_semundo(clone_flags, p);
1883         if (retval)
1884                 goto bad_fork_cleanup_security;
1885         retval = copy_files(clone_flags, p);
1886         if (retval)
1887                 goto bad_fork_cleanup_semundo;
1888         retval = copy_fs(clone_flags, p);
1889         if (retval)
1890                 goto bad_fork_cleanup_files;
1891         retval = copy_sighand(clone_flags, p);
1892         if (retval)
1893                 goto bad_fork_cleanup_fs;
1894         retval = copy_signal(clone_flags, p);
1895         if (retval)
1896                 goto bad_fork_cleanup_sighand;
1897         retval = copy_mm(clone_flags, p);
1898         if (retval)
1899                 goto bad_fork_cleanup_signal;
1900         retval = copy_namespaces(clone_flags, p);
1901         if (retval)
1902                 goto bad_fork_cleanup_mm;
1903         retval = copy_io(clone_flags, p);
1904         if (retval)
1905                 goto bad_fork_cleanup_namespaces;
1906         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1907         if (retval)
1908                 goto bad_fork_cleanup_io;
1909
1910         if (pid != &init_struct_pid) {
1911                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1912                 if (IS_ERR(pid)) {
1913                         retval = PTR_ERR(pid);
1914                         goto bad_fork_cleanup_thread;
1915                 }
1916         }
1917
1918 #ifdef CONFIG_BLOCK
1919         p->plug = NULL;
1920 #endif
1921 #ifdef CONFIG_FUTEX
1922         p->robust_list = NULL;
1923 #ifdef CONFIG_COMPAT
1924         p->compat_robust_list = NULL;
1925 #endif
1926         INIT_LIST_HEAD(&p->pi_state_list);
1927         p->pi_state_cache = NULL;
1928 #endif
1929         /*
1930          * sigaltstack should be cleared when sharing the same VM
1931          */
1932         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1933                 sas_ss_reset(p);
1934
1935         /*
1936          * Syscall tracing and stepping should be turned off in the
1937          * child regardless of CLONE_PTRACE.
1938          */
1939         user_disable_single_step(p);
1940         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1941 #ifdef TIF_SYSCALL_EMU
1942         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1943 #endif
1944         clear_all_latency_tracing(p);
1945
1946         /* ok, now we should be set up.. */
1947         p->pid = pid_nr(pid);
1948         if (clone_flags & CLONE_THREAD) {
1949                 p->exit_signal = -1;
1950                 p->group_leader = current->group_leader;
1951                 p->tgid = current->tgid;
1952         } else {
1953                 if (clone_flags & CLONE_PARENT)
1954                         p->exit_signal = current->group_leader->exit_signal;
1955                 else
1956                         p->exit_signal = (clone_flags & CSIGNAL);
1957                 p->group_leader = p;
1958                 p->tgid = p->pid;
1959         }
1960
1961         p->nr_dirtied = 0;
1962         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1963         p->dirty_paused_when = 0;
1964
1965         p->pdeath_signal = 0;
1966         INIT_LIST_HEAD(&p->thread_group);
1967         p->task_works = NULL;
1968
1969         cgroup_threadgroup_change_begin(current);
1970         /*
1971          * Ensure that the cgroup subsystem policies allow the new process to be
1972          * forked. It should be noted the the new process's css_set can be changed
1973          * between here and cgroup_post_fork() if an organisation operation is in
1974          * progress.
1975          */
1976         retval = cgroup_can_fork(p);
1977         if (retval)
1978                 goto bad_fork_free_pid;
1979
1980         /*
1981          * From this point on we must avoid any synchronous user-space
1982          * communication until we take the tasklist-lock. In particular, we do
1983          * not want user-space to be able to predict the process start-time by
1984          * stalling fork(2) after we recorded the start_time but before it is
1985          * visible to the system.
1986          */
1987
1988         p->start_time = ktime_get_ns();
1989         p->real_start_time = ktime_get_boot_ns();
1990
1991         /*
1992          * Make it visible to the rest of the system, but dont wake it up yet.
1993          * Need tasklist lock for parent etc handling!
1994          */
1995         write_lock_irq(&tasklist_lock);
1996
1997         /* CLONE_PARENT re-uses the old parent */
1998         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1999                 p->real_parent = current->real_parent;
2000                 p->parent_exec_id = current->parent_exec_id;
2001         } else {
2002                 p->real_parent = current;
2003                 p->parent_exec_id = current->self_exec_id;
2004         }
2005
2006         klp_copy_process(p);
2007
2008         spin_lock(&current->sighand->siglock);
2009
2010         /*
2011          * Copy seccomp details explicitly here, in case they were changed
2012          * before holding sighand lock.
2013          */
2014         copy_seccomp(p);
2015
2016         rseq_fork(p, clone_flags);
2017
2018         /* Don't start children in a dying pid namespace */
2019         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2020                 retval = -ENOMEM;
2021                 goto bad_fork_cancel_cgroup;
2022         }
2023
2024         /* Let kill terminate clone/fork in the middle */
2025         if (fatal_signal_pending(current)) {
2026                 retval = -EINTR;
2027                 goto bad_fork_cancel_cgroup;
2028         }
2029
2030
2031         init_task_pid_links(p);
2032         if (likely(p->pid)) {
2033                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2034
2035                 init_task_pid(p, PIDTYPE_PID, pid);
2036                 if (thread_group_leader(p)) {
2037                         init_task_pid(p, PIDTYPE_TGID, pid);
2038                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2039                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2040
2041                         if (is_child_reaper(pid)) {
2042                                 ns_of_pid(pid)->child_reaper = p;
2043                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2044                         }
2045                         p->signal->shared_pending.signal = delayed.signal;
2046                         p->signal->tty = tty_kref_get(current->signal->tty);
2047                         /*
2048                          * Inherit has_child_subreaper flag under the same
2049                          * tasklist_lock with adding child to the process tree
2050                          * for propagate_has_child_subreaper optimization.
2051                          */
2052                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2053                                                          p->real_parent->signal->is_child_subreaper;
2054                         list_add_tail(&p->sibling, &p->real_parent->children);
2055                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2056                         attach_pid(p, PIDTYPE_TGID);
2057                         attach_pid(p, PIDTYPE_PGID);
2058                         attach_pid(p, PIDTYPE_SID);
2059                         __this_cpu_inc(process_counts);
2060                 } else {
2061                         current->signal->nr_threads++;
2062                         atomic_inc(&current->signal->live);
2063                         atomic_inc(&current->signal->sigcnt);
2064                         task_join_group_stop(p);
2065                         list_add_tail_rcu(&p->thread_group,
2066                                           &p->group_leader->thread_group);
2067                         list_add_tail_rcu(&p->thread_node,
2068                                           &p->signal->thread_head);
2069                 }
2070                 attach_pid(p, PIDTYPE_PID);
2071                 nr_threads++;
2072         }
2073         total_forks++;
2074         hlist_del_init(&delayed.node);
2075         spin_unlock(&current->sighand->siglock);
2076         syscall_tracepoint_update(p);
2077         write_unlock_irq(&tasklist_lock);
2078
2079         proc_fork_connector(p);
2080         cgroup_post_fork(p);
2081         cgroup_threadgroup_change_end(current);
2082         perf_event_fork(p);
2083
2084         trace_task_newtask(p, clone_flags);
2085         uprobe_copy_process(p, clone_flags);
2086
2087         return p;
2088
2089 bad_fork_cancel_cgroup:
2090         spin_unlock(&current->sighand->siglock);
2091         write_unlock_irq(&tasklist_lock);
2092         cgroup_cancel_fork(p);
2093 bad_fork_free_pid:
2094         cgroup_threadgroup_change_end(current);
2095         if (pid != &init_struct_pid)
2096                 free_pid(pid);
2097 bad_fork_cleanup_thread:
2098         exit_thread(p);
2099 bad_fork_cleanup_io:
2100         if (p->io_context)
2101                 exit_io_context(p);
2102 bad_fork_cleanup_namespaces:
2103         exit_task_namespaces(p);
2104 bad_fork_cleanup_mm:
2105         if (p->mm) {
2106                 mm_clear_owner(p->mm, p);
2107                 mmput(p->mm);
2108         }
2109 bad_fork_cleanup_signal:
2110         if (!(clone_flags & CLONE_THREAD))
2111                 free_signal_struct(p->signal);
2112 bad_fork_cleanup_sighand:
2113         __cleanup_sighand(p->sighand);
2114 bad_fork_cleanup_fs:
2115         exit_fs(p); /* blocking */
2116 bad_fork_cleanup_files:
2117         exit_files(p); /* blocking */
2118 bad_fork_cleanup_semundo:
2119         exit_sem(p);
2120 bad_fork_cleanup_security:
2121         security_task_free(p);
2122 bad_fork_cleanup_audit:
2123         audit_free(p);
2124 bad_fork_cleanup_perf:
2125         perf_event_free_task(p);
2126 bad_fork_cleanup_policy:
2127         lockdep_free_task(p);
2128 #ifdef CONFIG_NUMA
2129         mpol_put(p->mempolicy);
2130 bad_fork_cleanup_threadgroup_lock:
2131 #endif
2132         delayacct_tsk_free(p);
2133 bad_fork_cleanup_count:
2134         atomic_dec(&p->cred->user->processes);
2135         exit_creds(p);
2136 bad_fork_free:
2137         p->state = TASK_DEAD;
2138         put_task_stack(p);
2139         delayed_free_task(p);
2140 fork_out:
2141         spin_lock_irq(&current->sighand->siglock);
2142         hlist_del_init(&delayed.node);
2143         spin_unlock_irq(&current->sighand->siglock);
2144         return ERR_PTR(retval);
2145 }
2146
2147 static inline void init_idle_pids(struct task_struct *idle)
2148 {
2149         enum pid_type type;
2150
2151         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2152                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2153                 init_task_pid(idle, type, &init_struct_pid);
2154         }
2155 }
2156
2157 struct task_struct *fork_idle(int cpu)
2158 {
2159         struct task_struct *task;
2160         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2161                             cpu_to_node(cpu));
2162         if (!IS_ERR(task)) {
2163                 init_idle_pids(task);
2164                 init_idle(task, cpu);
2165         }
2166
2167         return task;
2168 }
2169
2170 /*
2171  *  Ok, this is the main fork-routine.
2172  *
2173  * It copies the process, and if successful kick-starts
2174  * it and waits for it to finish using the VM if required.
2175  */
2176 long _do_fork(unsigned long clone_flags,
2177               unsigned long stack_start,
2178               unsigned long stack_size,
2179               int __user *parent_tidptr,
2180               int __user *child_tidptr,
2181               unsigned long tls)
2182 {
2183         struct completion vfork;
2184         struct pid *pid;
2185         struct task_struct *p;
2186         int trace = 0;
2187         long nr;
2188
2189         /*
2190          * Determine whether and which event to report to ptracer.  When
2191          * called from kernel_thread or CLONE_UNTRACED is explicitly
2192          * requested, no event is reported; otherwise, report if the event
2193          * for the type of forking is enabled.
2194          */
2195         if (!(clone_flags & CLONE_UNTRACED)) {
2196                 if (clone_flags & CLONE_VFORK)
2197                         trace = PTRACE_EVENT_VFORK;
2198                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2199                         trace = PTRACE_EVENT_CLONE;
2200                 else
2201                         trace = PTRACE_EVENT_FORK;
2202
2203                 if (likely(!ptrace_event_enabled(current, trace)))
2204                         trace = 0;
2205         }
2206
2207         p = copy_process(clone_flags, stack_start, stack_size,
2208                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2209         add_latent_entropy();
2210
2211         if (IS_ERR(p))
2212                 return PTR_ERR(p);
2213
2214         /*
2215          * Do this prior waking up the new thread - the thread pointer
2216          * might get invalid after that point, if the thread exits quickly.
2217          */
2218         trace_sched_process_fork(current, p);
2219
2220         pid = get_task_pid(p, PIDTYPE_PID);
2221         nr = pid_vnr(pid);
2222
2223         if (clone_flags & CLONE_PARENT_SETTID)
2224                 put_user(nr, parent_tidptr);
2225
2226         if (clone_flags & CLONE_VFORK) {
2227                 p->vfork_done = &vfork;
2228                 init_completion(&vfork);
2229                 get_task_struct(p);
2230         }
2231
2232         wake_up_new_task(p);
2233
2234         /* forking complete and child started to run, tell ptracer */
2235         if (unlikely(trace))
2236                 ptrace_event_pid(trace, pid);
2237
2238         if (clone_flags & CLONE_VFORK) {
2239                 if (!wait_for_vfork_done(p, &vfork))
2240                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2241         }
2242
2243         put_pid(pid);
2244         return nr;
2245 }
2246
2247 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2248 /* For compatibility with architectures that call do_fork directly rather than
2249  * using the syscall entry points below. */
2250 long do_fork(unsigned long clone_flags,
2251               unsigned long stack_start,
2252               unsigned long stack_size,
2253               int __user *parent_tidptr,
2254               int __user *child_tidptr)
2255 {
2256         return _do_fork(clone_flags, stack_start, stack_size,
2257                         parent_tidptr, child_tidptr, 0);
2258 }
2259 #endif
2260
2261 /*
2262  * Create a kernel thread.
2263  */
2264 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2265 {
2266         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2267                 (unsigned long)arg, NULL, NULL, 0);
2268 }
2269
2270 #ifdef __ARCH_WANT_SYS_FORK
2271 SYSCALL_DEFINE0(fork)
2272 {
2273 #ifdef CONFIG_MMU
2274         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2275 #else
2276         /* can not support in nommu mode */
2277         return -EINVAL;
2278 #endif
2279 }
2280 #endif
2281
2282 #ifdef __ARCH_WANT_SYS_VFORK
2283 SYSCALL_DEFINE0(vfork)
2284 {
2285         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2286                         0, NULL, NULL, 0);
2287 }
2288 #endif
2289
2290 #ifdef __ARCH_WANT_SYS_CLONE
2291 #ifdef CONFIG_CLONE_BACKWARDS
2292 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2293                  int __user *, parent_tidptr,
2294                  unsigned long, tls,
2295                  int __user *, child_tidptr)
2296 #elif defined(CONFIG_CLONE_BACKWARDS2)
2297 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2298                  int __user *, parent_tidptr,
2299                  int __user *, child_tidptr,
2300                  unsigned long, tls)
2301 #elif defined(CONFIG_CLONE_BACKWARDS3)
2302 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2303                 int, stack_size,
2304                 int __user *, parent_tidptr,
2305                 int __user *, child_tidptr,
2306                 unsigned long, tls)
2307 #else
2308 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2309                  int __user *, parent_tidptr,
2310                  int __user *, child_tidptr,
2311                  unsigned long, tls)
2312 #endif
2313 {
2314         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2315 }
2316 #endif
2317
2318 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2319 {
2320         struct task_struct *leader, *parent, *child;
2321         int res;
2322
2323         read_lock(&tasklist_lock);
2324         leader = top = top->group_leader;
2325 down:
2326         for_each_thread(leader, parent) {
2327                 list_for_each_entry(child, &parent->children, sibling) {
2328                         res = visitor(child, data);
2329                         if (res) {
2330                                 if (res < 0)
2331                                         goto out;
2332                                 leader = child;
2333                                 goto down;
2334                         }
2335 up:
2336                         ;
2337                 }
2338         }
2339
2340         if (leader != top) {
2341                 child = leader;
2342                 parent = child->real_parent;
2343                 leader = parent->group_leader;
2344                 goto up;
2345         }
2346 out:
2347         read_unlock(&tasklist_lock);
2348 }
2349
2350 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2351 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2352 #endif
2353
2354 static void sighand_ctor(void *data)
2355 {
2356         struct sighand_struct *sighand = data;
2357
2358         spin_lock_init(&sighand->siglock);
2359         init_waitqueue_head(&sighand->signalfd_wqh);
2360 }
2361
2362 void __init proc_caches_init(void)
2363 {
2364         unsigned int mm_size;
2365
2366         sighand_cachep = kmem_cache_create("sighand_cache",
2367                         sizeof(struct sighand_struct), 0,
2368                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2369                         SLAB_ACCOUNT, sighand_ctor);
2370         signal_cachep = kmem_cache_create("signal_cache",
2371                         sizeof(struct signal_struct), 0,
2372                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2373                         NULL);
2374         files_cachep = kmem_cache_create("files_cache",
2375                         sizeof(struct files_struct), 0,
2376                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2377                         NULL);
2378         fs_cachep = kmem_cache_create("fs_cache",
2379                         sizeof(struct fs_struct), 0,
2380                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2381                         NULL);
2382
2383         /*
2384          * The mm_cpumask is located at the end of mm_struct, and is
2385          * dynamically sized based on the maximum CPU number this system
2386          * can have, taking hotplug into account (nr_cpu_ids).
2387          */
2388         mm_size = sizeof(struct mm_struct) + cpumask_size();
2389
2390         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2391                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2392                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2393                         offsetof(struct mm_struct, saved_auxv),
2394                         sizeof_field(struct mm_struct, saved_auxv),
2395                         NULL);
2396         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2397         mmap_init();
2398         nsproxy_cache_init();
2399 }
2400
2401 /*
2402  * Check constraints on flags passed to the unshare system call.
2403  */
2404 static int check_unshare_flags(unsigned long unshare_flags)
2405 {
2406         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2407                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2408                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2409                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2410                 return -EINVAL;
2411         /*
2412          * Not implemented, but pretend it works if there is nothing
2413          * to unshare.  Note that unsharing the address space or the
2414          * signal handlers also need to unshare the signal queues (aka
2415          * CLONE_THREAD).
2416          */
2417         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2418                 if (!thread_group_empty(current))
2419                         return -EINVAL;
2420         }
2421         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2422                 if (atomic_read(&current->sighand->count) > 1)
2423                         return -EINVAL;
2424         }
2425         if (unshare_flags & CLONE_VM) {
2426                 if (!current_is_single_threaded())
2427                         return -EINVAL;
2428         }
2429
2430         return 0;
2431 }
2432
2433 /*
2434  * Unshare the filesystem structure if it is being shared
2435  */
2436 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2437 {
2438         struct fs_struct *fs = current->fs;
2439
2440         if (!(unshare_flags & CLONE_FS) || !fs)
2441                 return 0;
2442
2443         /* don't need lock here; in the worst case we'll do useless copy */
2444         if (fs->users == 1)
2445                 return 0;
2446
2447         *new_fsp = copy_fs_struct(fs);
2448         if (!*new_fsp)
2449                 return -ENOMEM;
2450
2451         return 0;
2452 }
2453
2454 /*
2455  * Unshare file descriptor table if it is being shared
2456  */
2457 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2458 {
2459         struct files_struct *fd = current->files;
2460         int error = 0;
2461
2462         if ((unshare_flags & CLONE_FILES) &&
2463             (fd && atomic_read(&fd->count) > 1)) {
2464                 *new_fdp = dup_fd(fd, &error);
2465                 if (!*new_fdp)
2466                         return error;
2467         }
2468
2469         return 0;
2470 }
2471
2472 /*
2473  * unshare allows a process to 'unshare' part of the process
2474  * context which was originally shared using clone.  copy_*
2475  * functions used by do_fork() cannot be used here directly
2476  * because they modify an inactive task_struct that is being
2477  * constructed. Here we are modifying the current, active,
2478  * task_struct.
2479  */
2480 int ksys_unshare(unsigned long unshare_flags)
2481 {
2482         struct fs_struct *fs, *new_fs = NULL;
2483         struct files_struct *fd, *new_fd = NULL;
2484         struct cred *new_cred = NULL;
2485         struct nsproxy *new_nsproxy = NULL;
2486         int do_sysvsem = 0;
2487         int err;
2488
2489         /*
2490          * If unsharing a user namespace must also unshare the thread group
2491          * and unshare the filesystem root and working directories.
2492          */
2493         if (unshare_flags & CLONE_NEWUSER)
2494                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2495         /*
2496          * If unsharing vm, must also unshare signal handlers.
2497          */
2498         if (unshare_flags & CLONE_VM)
2499                 unshare_flags |= CLONE_SIGHAND;
2500         /*
2501          * If unsharing a signal handlers, must also unshare the signal queues.
2502          */
2503         if (unshare_flags & CLONE_SIGHAND)
2504                 unshare_flags |= CLONE_THREAD;
2505         /*
2506          * If unsharing namespace, must also unshare filesystem information.
2507          */
2508         if (unshare_flags & CLONE_NEWNS)
2509                 unshare_flags |= CLONE_FS;
2510
2511         err = check_unshare_flags(unshare_flags);
2512         if (err)
2513                 goto bad_unshare_out;
2514         /*
2515          * CLONE_NEWIPC must also detach from the undolist: after switching
2516          * to a new ipc namespace, the semaphore arrays from the old
2517          * namespace are unreachable.
2518          */
2519         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2520                 do_sysvsem = 1;
2521         err = unshare_fs(unshare_flags, &new_fs);
2522         if (err)
2523                 goto bad_unshare_out;
2524         err = unshare_fd(unshare_flags, &new_fd);
2525         if (err)
2526                 goto bad_unshare_cleanup_fs;
2527         err = unshare_userns(unshare_flags, &new_cred);
2528         if (err)
2529                 goto bad_unshare_cleanup_fd;
2530         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2531                                          new_cred, new_fs);
2532         if (err)
2533                 goto bad_unshare_cleanup_cred;
2534
2535         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2536                 if (do_sysvsem) {
2537                         /*
2538                          * CLONE_SYSVSEM is equivalent to sys_exit().
2539                          */
2540                         exit_sem(current);
2541                 }
2542                 if (unshare_flags & CLONE_NEWIPC) {
2543                         /* Orphan segments in old ns (see sem above). */
2544                         exit_shm(current);
2545                         shm_init_task(current);
2546                 }
2547
2548                 if (new_nsproxy)
2549                         switch_task_namespaces(current, new_nsproxy);
2550
2551                 task_lock(current);
2552
2553                 if (new_fs) {
2554                         fs = current->fs;
2555                         spin_lock(&fs->lock);
2556                         current->fs = new_fs;
2557                         if (--fs->users)
2558                                 new_fs = NULL;
2559                         else
2560                                 new_fs = fs;
2561                         spin_unlock(&fs->lock);
2562                 }
2563
2564                 if (new_fd) {
2565                         fd = current->files;
2566                         current->files = new_fd;
2567                         new_fd = fd;
2568                 }
2569
2570                 task_unlock(current);
2571
2572                 if (new_cred) {
2573                         /* Install the new user namespace */
2574                         commit_creds(new_cred);
2575                         new_cred = NULL;
2576                 }
2577         }
2578
2579         perf_event_namespaces(current);
2580
2581 bad_unshare_cleanup_cred:
2582         if (new_cred)
2583                 put_cred(new_cred);
2584 bad_unshare_cleanup_fd:
2585         if (new_fd)
2586                 put_files_struct(new_fd);
2587
2588 bad_unshare_cleanup_fs:
2589         if (new_fs)
2590                 free_fs_struct(new_fs);
2591
2592 bad_unshare_out:
2593         return err;
2594 }
2595
2596 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2597 {
2598         return ksys_unshare(unshare_flags);
2599 }
2600
2601 /*
2602  *      Helper to unshare the files of the current task.
2603  *      We don't want to expose copy_files internals to
2604  *      the exec layer of the kernel.
2605  */
2606
2607 int unshare_files(struct files_struct **displaced)
2608 {
2609         struct task_struct *task = current;
2610         struct files_struct *copy = NULL;
2611         int error;
2612
2613         error = unshare_fd(CLONE_FILES, &copy);
2614         if (error || !copy) {
2615                 *displaced = NULL;
2616                 return error;
2617         }
2618         *displaced = task->files;
2619         task_lock(task);
2620         task->files = copy;
2621         task_unlock(task);
2622         return 0;
2623 }
2624
2625 int sysctl_max_threads(struct ctl_table *table, int write,
2626                        void __user *buffer, size_t *lenp, loff_t *ppos)
2627 {
2628         struct ctl_table t;
2629         int ret;
2630         int threads = max_threads;
2631         int min = 1;
2632         int max = MAX_THREADS;
2633
2634         t = *table;
2635         t.data = &threads;
2636         t.extra1 = &min;
2637         t.extra2 = &max;
2638
2639         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2640         if (ret || !write)
2641                 return ret;
2642
2643         max_threads = threads;
2644
2645         return 0;
2646 }