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