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