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