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