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