linux/kernel/fork.c
<<
>>
Prefs
   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/nsproxy.h>
  32#include <linux/capability.h>
  33#include <linux/cpu.h>
  34#include <linux/cgroup.h>
  35#include <linux/security.h>
  36#include <linux/hugetlb.h>
  37#include <linux/seccomp.h>
  38#include <linux/swap.h>
  39#include <linux/syscalls.h>
  40#include <linux/jiffies.h>
  41#include <linux/futex.h>
  42#include <linux/compat.h>
  43#include <linux/kthread.h>
  44#include <linux/task_io_accounting_ops.h>
  45#include <linux/rcupdate.h>
  46#include <linux/ptrace.h>
  47#include <linux/mount.h>
  48#include <linux/audit.h>
  49#include <linux/memcontrol.h>
  50#include <linux/ftrace.h>
  51#include <linux/proc_fs.h>
  52#include <linux/profile.h>
  53#include <linux/rmap.h>
  54#include <linux/ksm.h>
  55#include <linux/acct.h>
  56#include <linux/tsacct_kern.h>
  57#include <linux/cn_proc.h>
  58#include <linux/freezer.h>
  59#include <linux/delayacct.h>
  60#include <linux/taskstats_kern.h>
  61#include <linux/random.h>
  62#include <linux/tty.h>
  63#include <linux/blkdev.h>
  64#include <linux/fs_struct.h>
  65#include <linux/magic.h>
  66#include <linux/perf_event.h>
  67#include <linux/posix-timers.h>
  68#include <linux/user-return-notifier.h>
  69#include <linux/oom.h>
  70#include <linux/khugepaged.h>
  71#include <linux/signalfd.h>
  72#include <linux/uprobes.h>
  73
  74#include <asm/pgtable.h>
  75#include <asm/pgalloc.h>
  76#include <asm/uaccess.h>
  77#include <asm/mmu_context.h>
  78#include <asm/cacheflush.h>
  79#include <asm/tlbflush.h>
  80
  81#include <trace/events/sched.h>
  82
  83#define CREATE_TRACE_POINTS
  84#include <trace/events/task.h>
  85
  86/*
  87 * Protected counters by write_lock_irq(&tasklist_lock)
  88 */
  89unsigned long total_forks;      /* Handle normal Linux uptimes. */
  90int nr_threads;                 /* The idle threads do not count.. */
  91
  92int max_threads;                /* tunable limit on nr_threads */
  93
  94DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  95
  96__cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  97
  98#ifdef CONFIG_PROVE_RCU
  99int lockdep_tasklist_lock_is_held(void)
 100{
 101        return lockdep_is_held(&tasklist_lock);
 102}
 103EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 104#endif /* #ifdef CONFIG_PROVE_RCU */
 105
 106int nr_processes(void)
 107{
 108        int cpu;
 109        int total = 0;
 110
 111        for_each_possible_cpu(cpu)
 112                total += per_cpu(process_counts, cpu);
 113
 114        return total;
 115}
 116
 117void __weak arch_release_task_struct(struct task_struct *tsk)
 118{
 119}
 120
 121#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 122static struct kmem_cache *task_struct_cachep;
 123
 124static inline struct task_struct *alloc_task_struct_node(int node)
 125{
 126        return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 127}
 128
 129static inline void free_task_struct(struct task_struct *tsk)
 130{
 131        kmem_cache_free(task_struct_cachep, tsk);
 132}
 133#endif
 134
 135void __weak arch_release_thread_info(struct thread_info *ti)
 136{
 137}
 138
 139#ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 140
 141/*
 142 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 143 * kmemcache based allocator.
 144 */
 145# if THREAD_SIZE >= PAGE_SIZE
 146static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 147                                                  int node)
 148{
 149        struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
 150                                             THREAD_SIZE_ORDER);
 151
 152        return page ? page_address(page) : NULL;
 153}
 154
 155static inline void free_thread_info(struct thread_info *ti)
 156{
 157        free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 158}
 159# else
 160static struct kmem_cache *thread_info_cache;
 161
 162static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 163                                                  int node)
 164{
 165        return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 166}
 167
 168static void free_thread_info(struct thread_info *ti)
 169{
 170        kmem_cache_free(thread_info_cache, ti);
 171}
 172
 173void thread_info_cache_init(void)
 174{
 175        thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 176                                              THREAD_SIZE, 0, NULL);
 177        BUG_ON(thread_info_cache == NULL);
 178}
 179# endif
 180#endif
 181
 182/* SLAB cache for signal_struct structures (tsk->signal) */
 183static struct kmem_cache *signal_cachep;
 184
 185/* SLAB cache for sighand_struct structures (tsk->sighand) */
 186struct kmem_cache *sighand_cachep;
 187
 188/* SLAB cache for files_struct structures (tsk->files) */
 189struct kmem_cache *files_cachep;
 190
 191/* SLAB cache for fs_struct structures (tsk->fs) */
 192struct kmem_cache *fs_cachep;
 193
 194/* SLAB cache for vm_area_struct structures */
 195struct kmem_cache *vm_area_cachep;
 196
 197/* SLAB cache for mm_struct structures (tsk->mm) */
 198static struct kmem_cache *mm_cachep;
 199
 200static void account_kernel_stack(struct thread_info *ti, int account)
 201{
 202        struct zone *zone = page_zone(virt_to_page(ti));
 203
 204        mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 205}
 206
 207void free_task(struct task_struct *tsk)
 208{
 209        account_kernel_stack(tsk->stack, -1);
 210        arch_release_thread_info(tsk->stack);
 211        free_thread_info(tsk->stack);
 212        rt_mutex_debug_task_free(tsk);
 213        ftrace_graph_exit_task(tsk);
 214        put_seccomp_filter(tsk);
 215        arch_release_task_struct(tsk);
 216        free_task_struct(tsk);
 217}
 218EXPORT_SYMBOL(free_task);
 219
 220static inline void free_signal_struct(struct signal_struct *sig)
 221{
 222        taskstats_tgid_free(sig);
 223        sched_autogroup_exit(sig);
 224        kmem_cache_free(signal_cachep, sig);
 225}
 226
 227static inline void put_signal_struct(struct signal_struct *sig)
 228{
 229        if (atomic_dec_and_test(&sig->sigcnt))
 230                free_signal_struct(sig);
 231}
 232
 233void __put_task_struct(struct task_struct *tsk)
 234{
 235        WARN_ON(!tsk->exit_state);
 236        WARN_ON(atomic_read(&tsk->usage));
 237        WARN_ON(tsk == current);
 238
 239        security_task_free(tsk);
 240        exit_creds(tsk);
 241        delayacct_tsk_free(tsk);
 242        put_signal_struct(tsk->signal);
 243
 244        if (!profile_handoff_task(tsk))
 245                free_task(tsk);
 246}
 247EXPORT_SYMBOL_GPL(__put_task_struct);
 248
 249void __init __weak arch_task_cache_init(void) { }
 250
 251void __init fork_init(unsigned long mempages)
 252{
 253#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 254#ifndef ARCH_MIN_TASKALIGN
 255#define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 256#endif
 257        /* create a slab on which task_structs can be allocated */
 258        task_struct_cachep =
 259                kmem_cache_create("task_struct", sizeof(struct task_struct),
 260                        ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 261#endif
 262
 263        /* do the arch specific task caches init */
 264        arch_task_cache_init();
 265
 266        /*
 267         * The default maximum number of threads is set to a safe
 268         * value: the thread structures can take up at most half
 269         * of memory.
 270         */
 271        max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 272
 273        /*
 274         * we need to allow at least 20 threads to boot a system
 275         */
 276        if (max_threads < 20)
 277                max_threads = 20;
 278
 279        init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 280        init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 281        init_task.signal->rlim[RLIMIT_SIGPENDING] =
 282                init_task.signal->rlim[RLIMIT_NPROC];
 283}
 284
 285int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 286                                               struct task_struct *src)
 287{
 288        *dst = *src;
 289        return 0;
 290}
 291
 292static struct task_struct *dup_task_struct(struct task_struct *orig)
 293{
 294        struct task_struct *tsk;
 295        struct thread_info *ti;
 296        unsigned long *stackend;
 297        int node = tsk_fork_get_node(orig);
 298        int err;
 299
 300        tsk = alloc_task_struct_node(node);
 301        if (!tsk)
 302                return NULL;
 303
 304        ti = alloc_thread_info_node(tsk, node);
 305        if (!ti)
 306                goto free_tsk;
 307
 308        err = arch_dup_task_struct(tsk, orig);
 309        if (err)
 310                goto free_ti;
 311
 312        tsk->stack = ti;
 313
 314        setup_thread_stack(tsk, orig);
 315        clear_user_return_notifier(tsk);
 316        clear_tsk_need_resched(tsk);
 317        stackend = end_of_stack(tsk);
 318        *stackend = STACK_END_MAGIC;    /* for overflow detection */
 319
 320#ifdef CONFIG_CC_STACKPROTECTOR
 321        tsk->stack_canary = get_random_int();
 322#endif
 323
 324        /*
 325         * One for us, one for whoever does the "release_task()" (usually
 326         * parent)
 327         */
 328        atomic_set(&tsk->usage, 2);
 329#ifdef CONFIG_BLK_DEV_IO_TRACE
 330        tsk->btrace_seq = 0;
 331#endif
 332        tsk->splice_pipe = NULL;
 333        tsk->task_frag.page = NULL;
 334
 335        account_kernel_stack(ti, 1);
 336
 337        return tsk;
 338
 339free_ti:
 340        free_thread_info(ti);
 341free_tsk:
 342        free_task_struct(tsk);
 343        return NULL;
 344}
 345
 346#ifdef CONFIG_MMU
 347static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 348{
 349        struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 350        struct rb_node **rb_link, *rb_parent;
 351        int retval;
 352        unsigned long charge;
 353        struct mempolicy *pol;
 354
 355        uprobe_start_dup_mmap();
 356        down_write(&oldmm->mmap_sem);
 357        flush_cache_dup_mm(oldmm);
 358        uprobe_dup_mmap(oldmm, mm);
 359        /*
 360         * Not linked in yet - no deadlock potential:
 361         */
 362        down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 363
 364        mm->locked_vm = 0;
 365        mm->mmap = NULL;
 366        mm->mmap_cache = NULL;
 367        mm->free_area_cache = oldmm->mmap_base;
 368        mm->cached_hole_size = ~0UL;
 369        mm->map_count = 0;
 370        cpumask_clear(mm_cpumask(mm));
 371        mm->mm_rb = RB_ROOT;
 372        rb_link = &mm->mm_rb.rb_node;
 373        rb_parent = NULL;
 374        pprev = &mm->mmap;
 375        retval = ksm_fork(mm, oldmm);
 376        if (retval)
 377                goto out;
 378        retval = khugepaged_fork(mm, oldmm);
 379        if (retval)
 380                goto out;
 381
 382        prev = NULL;
 383        for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 384                struct file *file;
 385
 386                if (mpnt->vm_flags & VM_DONTCOPY) {
 387                        vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 388                                                        -vma_pages(mpnt));
 389                        continue;
 390                }
 391                charge = 0;
 392                if (mpnt->vm_flags & VM_ACCOUNT) {
 393                        unsigned long len = vma_pages(mpnt);
 394
 395                        if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 396                                goto fail_nomem;
 397                        charge = len;
 398                }
 399                tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 400                if (!tmp)
 401                        goto fail_nomem;
 402                *tmp = *mpnt;
 403                INIT_LIST_HEAD(&tmp->anon_vma_chain);
 404                pol = mpol_dup(vma_policy(mpnt));
 405                retval = PTR_ERR(pol);
 406                if (IS_ERR(pol))
 407                        goto fail_nomem_policy;
 408                vma_set_policy(tmp, pol);
 409                tmp->vm_mm = mm;
 410                if (anon_vma_fork(tmp, mpnt))
 411                        goto fail_nomem_anon_vma_fork;
 412                tmp->vm_flags &= ~VM_LOCKED;
 413                tmp->vm_next = tmp->vm_prev = NULL;
 414                file = tmp->vm_file;
 415                if (file) {
 416                        struct inode *inode = file_inode(file);
 417                        struct address_space *mapping = file->f_mapping;
 418
 419                        get_file(file);
 420                        if (tmp->vm_flags & VM_DENYWRITE)
 421                                atomic_dec(&inode->i_writecount);
 422                        mutex_lock(&mapping->i_mmap_mutex);
 423                        if (tmp->vm_flags & VM_SHARED)
 424                                mapping->i_mmap_writable++;
 425                        flush_dcache_mmap_lock(mapping);
 426                        /* insert tmp into the share list, just after mpnt */
 427                        if (unlikely(tmp->vm_flags & VM_NONLINEAR))
 428                                vma_nonlinear_insert(tmp,
 429                                                &mapping->i_mmap_nonlinear);
 430                        else
 431                                vma_interval_tree_insert_after(tmp, mpnt,
 432                                                        &mapping->i_mmap);
 433                        flush_dcache_mmap_unlock(mapping);
 434                        mutex_unlock(&mapping->i_mmap_mutex);
 435                }
 436
 437                /*
 438                 * Clear hugetlb-related page reserves for children. This only
 439                 * affects MAP_PRIVATE mappings. Faults generated by the child
 440                 * are not guaranteed to succeed, even if read-only
 441                 */
 442                if (is_vm_hugetlb_page(tmp))
 443                        reset_vma_resv_huge_pages(tmp);
 444
 445                /*
 446                 * Link in the new vma and copy the page table entries.
 447                 */
 448                *pprev = tmp;
 449                pprev = &tmp->vm_next;
 450                tmp->vm_prev = prev;
 451                prev = tmp;
 452
 453                __vma_link_rb(mm, tmp, rb_link, rb_parent);
 454                rb_link = &tmp->vm_rb.rb_right;
 455                rb_parent = &tmp->vm_rb;
 456
 457                mm->map_count++;
 458                retval = copy_page_range(mm, oldmm, mpnt);
 459
 460                if (tmp->vm_ops && tmp->vm_ops->open)
 461                        tmp->vm_ops->open(tmp);
 462
 463                if (retval)
 464                        goto out;
 465        }
 466        /* a new mm has just been created */
 467        arch_dup_mmap(oldmm, mm);
 468        retval = 0;
 469out:
 470        up_write(&mm->mmap_sem);
 471        flush_tlb_mm(oldmm);
 472        up_write(&oldmm->mmap_sem);
 473        uprobe_end_dup_mmap();
 474        return retval;
 475fail_nomem_anon_vma_fork:
 476        mpol_put(pol);
 477fail_nomem_policy:
 478        kmem_cache_free(vm_area_cachep, tmp);
 479fail_nomem:
 480        retval = -ENOMEM;
 481        vm_unacct_memory(charge);
 482        goto out;
 483}
 484
 485static inline int mm_alloc_pgd(struct mm_struct *mm)
 486{
 487        mm->pgd = pgd_alloc(mm);
 488        if (unlikely(!mm->pgd))
 489                return -ENOMEM;
 490        return 0;
 491}
 492
 493static inline void mm_free_pgd(struct mm_struct *mm)
 494{
 495        pgd_free(mm, mm->pgd);
 496}
 497#else
 498#define dup_mmap(mm, oldmm)     (0)
 499#define mm_alloc_pgd(mm)        (0)
 500#define mm_free_pgd(mm)
 501#endif /* CONFIG_MMU */
 502
 503__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 504
 505#define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 506#define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 507
 508static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 509
 510static int __init coredump_filter_setup(char *s)
 511{
 512        default_dump_filter =
 513                (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 514                MMF_DUMP_FILTER_MASK;
 515        return 1;
 516}
 517
 518__setup("coredump_filter=", coredump_filter_setup);
 519
 520#include <linux/init_task.h>
 521
 522static void mm_init_aio(struct mm_struct *mm)
 523{
 524#ifdef CONFIG_AIO
 525        spin_lock_init(&mm->ioctx_lock);
 526        INIT_HLIST_HEAD(&mm->ioctx_list);
 527#endif
 528}
 529
 530static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 531{
 532        atomic_set(&mm->mm_users, 1);
 533        atomic_set(&mm->mm_count, 1);
 534        init_rwsem(&mm->mmap_sem);
 535        INIT_LIST_HEAD(&mm->mmlist);
 536        mm->flags = (current->mm) ?
 537                (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 538        mm->core_state = NULL;
 539        mm->nr_ptes = 0;
 540        memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 541        spin_lock_init(&mm->page_table_lock);
 542        mm->free_area_cache = TASK_UNMAPPED_BASE;
 543        mm->cached_hole_size = ~0UL;
 544        mm_init_aio(mm);
 545        mm_init_owner(mm, p);
 546
 547        if (likely(!mm_alloc_pgd(mm))) {
 548                mm->def_flags = 0;
 549                mmu_notifier_mm_init(mm);
 550                return mm;
 551        }
 552
 553        free_mm(mm);
 554        return NULL;
 555}
 556
 557static void check_mm(struct mm_struct *mm)
 558{
 559        int i;
 560
 561        for (i = 0; i < NR_MM_COUNTERS; i++) {
 562                long x = atomic_long_read(&mm->rss_stat.count[i]);
 563
 564                if (unlikely(x))
 565                        printk(KERN_ALERT "BUG: Bad rss-counter state "
 566                                          "mm:%p idx:%d val:%ld\n", mm, i, x);
 567        }
 568
 569#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 570        VM_BUG_ON(mm->pmd_huge_pte);
 571#endif
 572}
 573
 574/*
 575 * Allocate and initialize an mm_struct.
 576 */
 577struct mm_struct *mm_alloc(void)
 578{
 579        struct mm_struct *mm;
 580
 581        mm = allocate_mm();
 582        if (!mm)
 583                return NULL;
 584
 585        memset(mm, 0, sizeof(*mm));
 586        mm_init_cpumask(mm);
 587        return mm_init(mm, current);
 588}
 589
 590/*
 591 * Called when the last reference to the mm
 592 * is dropped: either by a lazy thread or by
 593 * mmput. Free the page directory and the mm.
 594 */
 595void __mmdrop(struct mm_struct *mm)
 596{
 597        BUG_ON(mm == &init_mm);
 598        mm_free_pgd(mm);
 599        destroy_context(mm);
 600        mmu_notifier_mm_destroy(mm);
 601        check_mm(mm);
 602        free_mm(mm);
 603}
 604EXPORT_SYMBOL_GPL(__mmdrop);
 605
 606/*
 607 * Decrement the use count and release all resources for an mm.
 608 */
 609void mmput(struct mm_struct *mm)
 610{
 611        might_sleep();
 612
 613        if (atomic_dec_and_test(&mm->mm_users)) {
 614                uprobe_clear_state(mm);
 615                exit_aio(mm);
 616                ksm_exit(mm);
 617                khugepaged_exit(mm); /* must run before exit_mmap */
 618                exit_mmap(mm);
 619                set_mm_exe_file(mm, NULL);
 620                if (!list_empty(&mm->mmlist)) {
 621                        spin_lock(&mmlist_lock);
 622                        list_del(&mm->mmlist);
 623                        spin_unlock(&mmlist_lock);
 624                }
 625                if (mm->binfmt)
 626                        module_put(mm->binfmt->module);
 627                mmdrop(mm);
 628        }
 629}
 630EXPORT_SYMBOL_GPL(mmput);
 631
 632void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 633{
 634        if (new_exe_file)
 635                get_file(new_exe_file);
 636        if (mm->exe_file)
 637                fput(mm->exe_file);
 638        mm->exe_file = new_exe_file;
 639}
 640
 641struct file *get_mm_exe_file(struct mm_struct *mm)
 642{
 643        struct file *exe_file;
 644
 645        /* We need mmap_sem to protect against races with removal of exe_file */
 646        down_read(&mm->mmap_sem);
 647        exe_file = mm->exe_file;
 648        if (exe_file)
 649                get_file(exe_file);
 650        up_read(&mm->mmap_sem);
 651        return exe_file;
 652}
 653
 654static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 655{
 656        /* It's safe to write the exe_file pointer without exe_file_lock because
 657         * this is called during fork when the task is not yet in /proc */
 658        newmm->exe_file = get_mm_exe_file(oldmm);
 659}
 660
 661/**
 662 * get_task_mm - acquire a reference to the task's mm
 663 *
 664 * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 665 * this kernel workthread has transiently adopted a user mm with use_mm,
 666 * to do its AIO) is not set and if so returns a reference to it, after
 667 * bumping up the use count.  User must release the mm via mmput()
 668 * after use.  Typically used by /proc and ptrace.
 669 */
 670struct mm_struct *get_task_mm(struct task_struct *task)
 671{
 672        struct mm_struct *mm;
 673
 674        task_lock(task);
 675        mm = task->mm;
 676        if (mm) {
 677                if (task->flags & PF_KTHREAD)
 678                        mm = NULL;
 679                else
 680                        atomic_inc(&mm->mm_users);
 681        }
 682        task_unlock(task);
 683        return mm;
 684}
 685EXPORT_SYMBOL_GPL(get_task_mm);
 686
 687struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 688{
 689        struct mm_struct *mm;
 690        int err;
 691
 692        err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 693        if (err)
 694                return ERR_PTR(err);
 695
 696        mm = get_task_mm(task);
 697        if (mm && mm != current->mm &&
 698                        !ptrace_may_access(task, mode)) {
 699                mmput(mm);
 700                mm = ERR_PTR(-EACCES);
 701        }
 702        mutex_unlock(&task->signal->cred_guard_mutex);
 703
 704        return mm;
 705}
 706
 707static void complete_vfork_done(struct task_struct *tsk)
 708{
 709        struct completion *vfork;
 710
 711        task_lock(tsk);
 712        vfork = tsk->vfork_done;
 713        if (likely(vfork)) {
 714                tsk->vfork_done = NULL;
 715                complete(vfork);
 716        }
 717        task_unlock(tsk);
 718}
 719
 720static int wait_for_vfork_done(struct task_struct *child,
 721                                struct completion *vfork)
 722{
 723        int killed;
 724
 725        freezer_do_not_count();
 726        killed = wait_for_completion_killable(vfork);
 727        freezer_count();
 728
 729        if (killed) {
 730                task_lock(child);
 731                child->vfork_done = NULL;
 732                task_unlock(child);
 733        }
 734
 735        put_task_struct(child);
 736        return killed;
 737}
 738
 739/* Please note the differences between mmput and mm_release.
 740 * mmput is called whenever we stop holding onto a mm_struct,
 741 * error success whatever.
 742 *
 743 * mm_release is called after a mm_struct has been removed
 744 * from the current process.
 745 *
 746 * This difference is important for error handling, when we
 747 * only half set up a mm_struct for a new process and need to restore
 748 * the old one.  Because we mmput the new mm_struct before
 749 * restoring the old one. . .
 750 * Eric Biederman 10 January 1998
 751 */
 752void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 753{
 754        /* Get rid of any futexes when releasing the mm */
 755#ifdef CONFIG_FUTEX
 756        if (unlikely(tsk->robust_list)) {
 757                exit_robust_list(tsk);
 758                tsk->robust_list = NULL;
 759        }
 760#ifdef CONFIG_COMPAT
 761        if (unlikely(tsk->compat_robust_list)) {
 762                compat_exit_robust_list(tsk);
 763                tsk->compat_robust_list = NULL;
 764        }
 765#endif
 766        if (unlikely(!list_empty(&tsk->pi_state_list)))
 767                exit_pi_state_list(tsk);
 768#endif
 769
 770        uprobe_free_utask(tsk);
 771
 772        /* Get rid of any cached register state */
 773        deactivate_mm(tsk, mm);
 774
 775        /*
 776         * If we're exiting normally, clear a user-space tid field if
 777         * requested.  We leave this alone when dying by signal, to leave
 778         * the value intact in a core dump, and to save the unnecessary
 779         * trouble, say, a killed vfork parent shouldn't touch this mm.
 780         * Userland only wants this done for a sys_exit.
 781         */
 782        if (tsk->clear_child_tid) {
 783                if (!(tsk->flags & PF_SIGNALED) &&
 784                    atomic_read(&mm->mm_users) > 1) {
 785                        /*
 786                         * We don't check the error code - if userspace has
 787                         * not set up a proper pointer then tough luck.
 788                         */
 789                        put_user(0, tsk->clear_child_tid);
 790                        sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 791                                        1, NULL, NULL, 0);
 792                }
 793                tsk->clear_child_tid = NULL;
 794        }
 795
 796        /*
 797         * All done, finally we can wake up parent and return this mm to him.
 798         * Also kthread_stop() uses this completion for synchronization.
 799         */
 800        if (tsk->vfork_done)
 801                complete_vfork_done(tsk);
 802}
 803
 804/*
 805 * Allocate a new mm structure and copy contents from the
 806 * mm structure of the passed in task structure.
 807 */
 808struct mm_struct *dup_mm(struct task_struct *tsk)
 809{
 810        struct mm_struct *mm, *oldmm = current->mm;
 811        int err;
 812
 813        if (!oldmm)
 814                return NULL;
 815
 816        mm = allocate_mm();
 817        if (!mm)
 818                goto fail_nomem;
 819
 820        memcpy(mm, oldmm, sizeof(*mm));
 821        mm_init_cpumask(mm);
 822
 823#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 824        mm->pmd_huge_pte = NULL;
 825#endif
 826#ifdef CONFIG_NUMA_BALANCING
 827        mm->first_nid = NUMA_PTE_SCAN_INIT;
 828#endif
 829        if (!mm_init(mm, tsk))
 830                goto fail_nomem;
 831
 832        if (init_new_context(tsk, mm))
 833                goto fail_nocontext;
 834
 835        dup_mm_exe_file(oldmm, mm);
 836
 837        err = dup_mmap(mm, oldmm);
 838        if (err)
 839                goto free_pt;
 840
 841        mm->hiwater_rss = get_mm_rss(mm);
 842        mm->hiwater_vm = mm->total_vm;
 843
 844        if (mm->binfmt && !try_module_get(mm->binfmt->module))
 845                goto free_pt;
 846
 847        return mm;
 848
 849free_pt:
 850        /* don't put binfmt in mmput, we haven't got module yet */
 851        mm->binfmt = NULL;
 852        mmput(mm);
 853
 854fail_nomem:
 855        return NULL;
 856
 857fail_nocontext:
 858        /*
 859         * If init_new_context() failed, we cannot use mmput() to free the mm
 860         * because it calls destroy_context()
 861         */
 862        mm_free_pgd(mm);
 863        free_mm(mm);
 864        return NULL;
 865}
 866
 867static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 868{
 869        struct mm_struct *mm, *oldmm;
 870        int retval;
 871
 872        tsk->min_flt = tsk->maj_flt = 0;
 873        tsk->nvcsw = tsk->nivcsw = 0;
 874#ifdef CONFIG_DETECT_HUNG_TASK
 875        tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 876#endif
 877
 878        tsk->mm = NULL;
 879        tsk->active_mm = NULL;
 880
 881        /*
 882         * Are we cloning a kernel thread?
 883         *
 884         * We need to steal a active VM for that..
 885         */
 886        oldmm = current->mm;
 887        if (!oldmm)
 888                return 0;
 889
 890        if (clone_flags & CLONE_VM) {
 891                atomic_inc(&oldmm->mm_users);
 892                mm = oldmm;
 893                goto good_mm;
 894        }
 895
 896        retval = -ENOMEM;
 897        mm = dup_mm(tsk);
 898        if (!mm)
 899                goto fail_nomem;
 900
 901good_mm:
 902        tsk->mm = mm;
 903        tsk->active_mm = mm;
 904        return 0;
 905
 906fail_nomem:
 907        return retval;
 908}
 909
 910static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 911{
 912        struct fs_struct *fs = current->fs;
 913        if (clone_flags & CLONE_FS) {
 914                /* tsk->fs is already what we want */
 915                spin_lock(&fs->lock);
 916                if (fs->in_exec) {
 917                        spin_unlock(&fs->lock);
 918                        return -EAGAIN;
 919                }
 920                fs->users++;
 921                spin_unlock(&fs->lock);
 922                return 0;
 923        }
 924        tsk->fs = copy_fs_struct(fs);
 925        if (!tsk->fs)
 926                return -ENOMEM;
 927        return 0;
 928}
 929
 930static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 931{
 932        struct files_struct *oldf, *newf;
 933        int error = 0;
 934
 935        /*
 936         * A background process may not have any files ...
 937         */
 938        oldf = current->files;
 939        if (!oldf)
 940                goto out;
 941
 942        if (clone_flags & CLONE_FILES) {
 943                atomic_inc(&oldf->count);
 944                goto out;
 945        }
 946
 947        newf = dup_fd(oldf, &error);
 948        if (!newf)
 949                goto out;
 950
 951        tsk->files = newf;
 952        error = 0;
 953out:
 954        return error;
 955}
 956
 957static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 958{
 959#ifdef CONFIG_BLOCK
 960        struct io_context *ioc = current->io_context;
 961        struct io_context *new_ioc;
 962
 963        if (!ioc)
 964                return 0;
 965        /*
 966         * Share io context with parent, if CLONE_IO is set
 967         */
 968        if (clone_flags & CLONE_IO) {
 969                ioc_task_link(ioc);
 970                tsk->io_context = ioc;
 971        } else if (ioprio_valid(ioc->ioprio)) {
 972                new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 973                if (unlikely(!new_ioc))
 974                        return -ENOMEM;
 975
 976                new_ioc->ioprio = ioc->ioprio;
 977                put_io_context(new_ioc);
 978        }
 979#endif
 980        return 0;
 981}
 982
 983static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 984{
 985        struct sighand_struct *sig;
 986
 987        if (clone_flags & CLONE_SIGHAND) {
 988                atomic_inc(&current->sighand->count);
 989                return 0;
 990        }
 991        sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 992        rcu_assign_pointer(tsk->sighand, sig);
 993        if (!sig)
 994                return -ENOMEM;
 995        atomic_set(&sig->count, 1);
 996        memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 997        return 0;
 998}
 999
1000void __cleanup_sighand(struct sighand_struct *sighand)
1001{
1002        if (atomic_dec_and_test(&sighand->count)) {
1003                signalfd_cleanup(sighand);
1004                kmem_cache_free(sighand_cachep, sighand);
1005        }
1006}
1007
1008
1009/*
1010 * Initialize POSIX timer handling for a thread group.
1011 */
1012static void posix_cpu_timers_init_group(struct signal_struct *sig)
1013{
1014        unsigned long cpu_limit;
1015
1016        /* Thread group counters. */
1017        thread_group_cputime_init(sig);
1018
1019        cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1020        if (cpu_limit != RLIM_INFINITY) {
1021                sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1022                sig->cputimer.running = 1;
1023        }
1024
1025        /* The timer lists. */
1026        INIT_LIST_HEAD(&sig->cpu_timers[0]);
1027        INIT_LIST_HEAD(&sig->cpu_timers[1]);
1028        INIT_LIST_HEAD(&sig->cpu_timers[2]);
1029}
1030
1031static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1032{
1033        struct signal_struct *sig;
1034
1035        if (clone_flags & CLONE_THREAD)
1036                return 0;
1037
1038        sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1039        tsk->signal = sig;
1040        if (!sig)
1041                return -ENOMEM;
1042
1043        sig->nr_threads = 1;
1044        atomic_set(&sig->live, 1);
1045        atomic_set(&sig->sigcnt, 1);
1046        init_waitqueue_head(&sig->wait_chldexit);
1047        sig->curr_target = tsk;
1048        init_sigpending(&sig->shared_pending);
1049        INIT_LIST_HEAD(&sig->posix_timers);
1050
1051        hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1052        sig->real_timer.function = it_real_fn;
1053
1054        task_lock(current->group_leader);
1055        memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1056        task_unlock(current->group_leader);
1057
1058        posix_cpu_timers_init_group(sig);
1059
1060        tty_audit_fork(sig);
1061        sched_autogroup_fork(sig);
1062
1063#ifdef CONFIG_CGROUPS
1064        init_rwsem(&sig->group_rwsem);
1065#endif
1066
1067        sig->oom_score_adj = current->signal->oom_score_adj;
1068        sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1069
1070        sig->has_child_subreaper = current->signal->has_child_subreaper ||
1071                                   current->signal->is_child_subreaper;
1072
1073        mutex_init(&sig->cred_guard_mutex);
1074
1075        return 0;
1076}
1077
1078static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1079{
1080        unsigned long new_flags = p->flags;
1081
1082        new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1083        new_flags |= PF_FORKNOEXEC;
1084        p->flags = new_flags;
1085}
1086
1087SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1088{
1089        current->clear_child_tid = tidptr;
1090
1091        return task_pid_vnr(current);
1092}
1093
1094static void rt_mutex_init_task(struct task_struct *p)
1095{
1096        raw_spin_lock_init(&p->pi_lock);
1097#ifdef CONFIG_RT_MUTEXES
1098        plist_head_init(&p->pi_waiters);
1099        p->pi_blocked_on = NULL;
1100#endif
1101}
1102
1103#ifdef CONFIG_MM_OWNER
1104void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1105{
1106        mm->owner = p;
1107}
1108#endif /* CONFIG_MM_OWNER */
1109
1110/*
1111 * Initialize POSIX timer handling for a single task.
1112 */
1113static void posix_cpu_timers_init(struct task_struct *tsk)
1114{
1115        tsk->cputime_expires.prof_exp = 0;
1116        tsk->cputime_expires.virt_exp = 0;
1117        tsk->cputime_expires.sched_exp = 0;
1118        INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1119        INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1120        INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1121}
1122
1123/*
1124 * This creates a new process as a copy of the old one,
1125 * but does not actually start it yet.
1126 *
1127 * It copies the registers, and all the appropriate
1128 * parts of the process environment (as per the clone
1129 * flags). The actual kick-off is left to the caller.
1130 */
1131static struct task_struct *copy_process(unsigned long clone_flags,
1132                                        unsigned long stack_start,
1133                                        unsigned long stack_size,
1134                                        int __user *child_tidptr,
1135                                        struct pid *pid,
1136                                        int trace)
1137{
1138        int retval;
1139        struct task_struct *p;
1140
1141        if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1142                return ERR_PTR(-EINVAL);
1143
1144        if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1145                return ERR_PTR(-EINVAL);
1146
1147        /*
1148         * Thread groups must share signals as well, and detached threads
1149         * can only be started up within the thread group.
1150         */
1151        if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1152                return ERR_PTR(-EINVAL);
1153
1154        /*
1155         * Shared signal handlers imply shared VM. By way of the above,
1156         * thread groups also imply shared VM. Blocking this case allows
1157         * for various simplifications in other code.
1158         */
1159        if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1160                return ERR_PTR(-EINVAL);
1161
1162        /*
1163         * Siblings of global init remain as zombies on exit since they are
1164         * not reaped by their parent (swapper). To solve this and to avoid
1165         * multi-rooted process trees, prevent global and container-inits
1166         * from creating siblings.
1167         */
1168        if ((clone_flags & CLONE_PARENT) &&
1169                                current->signal->flags & SIGNAL_UNKILLABLE)
1170                return ERR_PTR(-EINVAL);
1171
1172        /*
1173         * If the new process will be in a different pid namespace
1174         * don't allow the creation of threads.
1175         */
1176        if ((clone_flags & (CLONE_VM|CLONE_NEWPID)) &&
1177            (task_active_pid_ns(current) != current->nsproxy->pid_ns))
1178                return ERR_PTR(-EINVAL);
1179
1180        retval = security_task_create(clone_flags);
1181        if (retval)
1182                goto fork_out;
1183
1184        retval = -ENOMEM;
1185        p = dup_task_struct(current);
1186        if (!p)
1187                goto fork_out;
1188
1189        ftrace_graph_init_task(p);
1190        get_seccomp_filter(p);
1191
1192        rt_mutex_init_task(p);
1193
1194#ifdef CONFIG_PROVE_LOCKING
1195        DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1196        DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1197#endif
1198        retval = -EAGAIN;
1199        if (atomic_read(&p->real_cred->user->processes) >=
1200                        task_rlimit(p, RLIMIT_NPROC)) {
1201                if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1202                    p->real_cred->user != INIT_USER)
1203                        goto bad_fork_free;
1204        }
1205        current->flags &= ~PF_NPROC_EXCEEDED;
1206
1207        retval = copy_creds(p, clone_flags);
1208        if (retval < 0)
1209                goto bad_fork_free;
1210
1211        /*
1212         * If multiple threads are within copy_process(), then this check
1213         * triggers too late. This doesn't hurt, the check is only there
1214         * to stop root fork bombs.
1215         */
1216        retval = -EAGAIN;
1217        if (nr_threads >= max_threads)
1218                goto bad_fork_cleanup_count;
1219
1220        if (!try_module_get(task_thread_info(p)->exec_domain->module))
1221                goto bad_fork_cleanup_count;
1222
1223        p->did_exec = 0;
1224        delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1225        copy_flags(clone_flags, p);
1226        INIT_LIST_HEAD(&p->children);
1227        INIT_LIST_HEAD(&p->sibling);
1228        rcu_copy_process(p);
1229        p->vfork_done = NULL;
1230        spin_lock_init(&p->alloc_lock);
1231
1232        init_sigpending(&p->pending);
1233
1234        p->utime = p->stime = p->gtime = 0;
1235        p->utimescaled = p->stimescaled = 0;
1236#ifndef CONFIG_VIRT_CPU_ACCOUNTING
1237        p->prev_cputime.utime = p->prev_cputime.stime = 0;
1238#endif
1239#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1240        seqlock_init(&p->vtime_seqlock);
1241        p->vtime_snap = 0;
1242        p->vtime_snap_whence = VTIME_SLEEPING;
1243#endif
1244
1245#if defined(SPLIT_RSS_COUNTING)
1246        memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1247#endif
1248
1249        p->default_timer_slack_ns = current->timer_slack_ns;
1250
1251        task_io_accounting_init(&p->ioac);
1252        acct_clear_integrals(p);
1253
1254        posix_cpu_timers_init(p);
1255
1256        do_posix_clock_monotonic_gettime(&p->start_time);
1257        p->real_start_time = p->start_time;
1258        monotonic_to_bootbased(&p->real_start_time);
1259        p->io_context = NULL;
1260        p->audit_context = NULL;
1261        if (clone_flags & CLONE_THREAD)
1262                threadgroup_change_begin(current);
1263        cgroup_fork(p);
1264#ifdef CONFIG_NUMA
1265        p->mempolicy = mpol_dup(p->mempolicy);
1266        if (IS_ERR(p->mempolicy)) {
1267                retval = PTR_ERR(p->mempolicy);
1268                p->mempolicy = NULL;
1269                goto bad_fork_cleanup_cgroup;
1270        }
1271        mpol_fix_fork_child_flag(p);
1272#endif
1273#ifdef CONFIG_CPUSETS
1274        p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1275        p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1276        seqcount_init(&p->mems_allowed_seq);
1277#endif
1278#ifdef CONFIG_TRACE_IRQFLAGS
1279        p->irq_events = 0;
1280        p->hardirqs_enabled = 0;
1281        p->hardirq_enable_ip = 0;
1282        p->hardirq_enable_event = 0;
1283        p->hardirq_disable_ip = _THIS_IP_;
1284        p->hardirq_disable_event = 0;
1285        p->softirqs_enabled = 1;
1286        p->softirq_enable_ip = _THIS_IP_;
1287        p->softirq_enable_event = 0;
1288        p->softirq_disable_ip = 0;
1289        p->softirq_disable_event = 0;
1290        p->hardirq_context = 0;
1291        p->softirq_context = 0;
1292#endif
1293#ifdef CONFIG_LOCKDEP
1294        p->lockdep_depth = 0; /* no locks held yet */
1295        p->curr_chain_key = 0;
1296        p->lockdep_recursion = 0;
1297#endif
1298
1299#ifdef CONFIG_DEBUG_MUTEXES
1300        p->blocked_on = NULL; /* not blocked yet */
1301#endif
1302#ifdef CONFIG_MEMCG
1303        p->memcg_batch.do_batch = 0;
1304        p->memcg_batch.memcg = NULL;
1305#endif
1306
1307        /* Perform scheduler related setup. Assign this task to a CPU. */
1308        sched_fork(p);
1309
1310        retval = perf_event_init_task(p);
1311        if (retval)
1312                goto bad_fork_cleanup_policy;
1313        retval = audit_alloc(p);
1314        if (retval)
1315                goto bad_fork_cleanup_policy;
1316        /* copy all the process information */
1317        retval = copy_semundo(clone_flags, p);
1318        if (retval)
1319                goto bad_fork_cleanup_audit;
1320        retval = copy_files(clone_flags, p);
1321        if (retval)
1322                goto bad_fork_cleanup_semundo;
1323        retval = copy_fs(clone_flags, p);
1324        if (retval)
1325                goto bad_fork_cleanup_files;
1326        retval = copy_sighand(clone_flags, p);
1327        if (retval)
1328                goto bad_fork_cleanup_fs;
1329        retval = copy_signal(clone_flags, p);
1330        if (retval)
1331                goto bad_fork_cleanup_sighand;
1332        retval = copy_mm(clone_flags, p);
1333        if (retval)
1334                goto bad_fork_cleanup_signal;
1335        retval = copy_namespaces(clone_flags, p);
1336        if (retval)
1337                goto bad_fork_cleanup_mm;
1338        retval = copy_io(clone_flags, p);
1339        if (retval)
1340                goto bad_fork_cleanup_namespaces;
1341        retval = copy_thread(clone_flags, stack_start, stack_size, p);
1342        if (retval)
1343                goto bad_fork_cleanup_io;
1344
1345        if (pid != &init_struct_pid) {
1346                retval = -ENOMEM;
1347                pid = alloc_pid(p->nsproxy->pid_ns);
1348                if (!pid)
1349                        goto bad_fork_cleanup_io;
1350        }
1351
1352        p->pid = pid_nr(pid);
1353        p->tgid = p->pid;
1354        if (clone_flags & CLONE_THREAD)
1355                p->tgid = current->tgid;
1356
1357        p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1358        /*
1359         * Clear TID on mm_release()?
1360         */
1361        p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1362#ifdef CONFIG_BLOCK
1363        p->plug = NULL;
1364#endif
1365#ifdef CONFIG_FUTEX
1366        p->robust_list = NULL;
1367#ifdef CONFIG_COMPAT
1368        p->compat_robust_list = NULL;
1369#endif
1370        INIT_LIST_HEAD(&p->pi_state_list);
1371        p->pi_state_cache = NULL;
1372#endif
1373        uprobe_copy_process(p);
1374        /*
1375         * sigaltstack should be cleared when sharing the same VM
1376         */
1377        if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1378                p->sas_ss_sp = p->sas_ss_size = 0;
1379
1380        /*
1381         * Syscall tracing and stepping should be turned off in the
1382         * child regardless of CLONE_PTRACE.
1383         */
1384        user_disable_single_step(p);
1385        clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1386#ifdef TIF_SYSCALL_EMU
1387        clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1388#endif
1389        clear_all_latency_tracing(p);
1390
1391        /* ok, now we should be set up.. */
1392        if (clone_flags & CLONE_THREAD)
1393                p->exit_signal = -1;
1394        else if (clone_flags & CLONE_PARENT)
1395                p->exit_signal = current->group_leader->exit_signal;
1396        else
1397                p->exit_signal = (clone_flags & CSIGNAL);
1398
1399        p->pdeath_signal = 0;
1400        p->exit_state = 0;
1401
1402        p->nr_dirtied = 0;
1403        p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1404        p->dirty_paused_when = 0;
1405
1406        /*
1407         * Ok, make it visible to the rest of the system.
1408         * We dont wake it up yet.
1409         */
1410        p->group_leader = p;
1411        INIT_LIST_HEAD(&p->thread_group);
1412        p->task_works = NULL;
1413
1414        /* Need tasklist lock for parent etc handling! */
1415        write_lock_irq(&tasklist_lock);
1416
1417        /* CLONE_PARENT re-uses the old parent */
1418        if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1419                p->real_parent = current->real_parent;
1420                p->parent_exec_id = current->parent_exec_id;
1421        } else {
1422                p->real_parent = current;
1423                p->parent_exec_id = current->self_exec_id;
1424        }
1425
1426        spin_lock(&current->sighand->siglock);
1427
1428        /*
1429         * Process group and session signals need to be delivered to just the
1430         * parent before the fork or both the parent and the child after the
1431         * fork. Restart if a signal comes in before we add the new process to
1432         * it's process group.
1433         * A fatal signal pending means that current will exit, so the new
1434         * thread can't slip out of an OOM kill (or normal SIGKILL).
1435        */
1436        recalc_sigpending();
1437        if (signal_pending(current)) {
1438                spin_unlock(&current->sighand->siglock);
1439                write_unlock_irq(&tasklist_lock);
1440                retval = -ERESTARTNOINTR;
1441                goto bad_fork_free_pid;
1442        }
1443
1444        if (clone_flags & CLONE_THREAD) {
1445                current->signal->nr_threads++;
1446                atomic_inc(&current->signal->live);
1447                atomic_inc(&current->signal->sigcnt);
1448                p->group_leader = current->group_leader;
1449                list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1450        }
1451
1452        if (likely(p->pid)) {
1453                ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1454
1455                if (thread_group_leader(p)) {
1456                        if (is_child_reaper(pid)) {
1457                                ns_of_pid(pid)->child_reaper = p;
1458                                p->signal->flags |= SIGNAL_UNKILLABLE;
1459                        }
1460
1461                        p->signal->leader_pid = pid;
1462                        p->signal->tty = tty_kref_get(current->signal->tty);
1463                        attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1464                        attach_pid(p, PIDTYPE_SID, task_session(current));
1465                        list_add_tail(&p->sibling, &p->real_parent->children);
1466                        list_add_tail_rcu(&p->tasks, &init_task.tasks);
1467                        __this_cpu_inc(process_counts);
1468                }
1469                attach_pid(p, PIDTYPE_PID, pid);
1470                nr_threads++;
1471        }
1472
1473        total_forks++;
1474        spin_unlock(&current->sighand->siglock);
1475        write_unlock_irq(&tasklist_lock);
1476        proc_fork_connector(p);
1477        cgroup_post_fork(p);
1478        if (clone_flags & CLONE_THREAD)
1479                threadgroup_change_end(current);
1480        perf_event_fork(p);
1481
1482        trace_task_newtask(p, clone_flags);
1483
1484        return p;
1485
1486bad_fork_free_pid:
1487        if (pid != &init_struct_pid)
1488                free_pid(pid);
1489bad_fork_cleanup_io:
1490        if (p->io_context)
1491                exit_io_context(p);
1492bad_fork_cleanup_namespaces:
1493        exit_task_namespaces(p);
1494bad_fork_cleanup_mm:
1495        if (p->mm)
1496                mmput(p->mm);
1497bad_fork_cleanup_signal:
1498        if (!(clone_flags & CLONE_THREAD))
1499                free_signal_struct(p->signal);
1500bad_fork_cleanup_sighand:
1501        __cleanup_sighand(p->sighand);
1502bad_fork_cleanup_fs:
1503        exit_fs(p); /* blocking */
1504bad_fork_cleanup_files:
1505        exit_files(p); /* blocking */
1506bad_fork_cleanup_semundo:
1507        exit_sem(p);
1508bad_fork_cleanup_audit:
1509        audit_free(p);
1510bad_fork_cleanup_policy:
1511        perf_event_free_task(p);
1512#ifdef CONFIG_NUMA
1513        mpol_put(p->mempolicy);
1514bad_fork_cleanup_cgroup:
1515#endif
1516        if (clone_flags & CLONE_THREAD)
1517                threadgroup_change_end(current);
1518        cgroup_exit(p, 0);
1519        delayacct_tsk_free(p);
1520        module_put(task_thread_info(p)->exec_domain->module);
1521bad_fork_cleanup_count:
1522        atomic_dec(&p->cred->user->processes);
1523        exit_creds(p);
1524bad_fork_free:
1525        free_task(p);
1526fork_out:
1527        return ERR_PTR(retval);
1528}
1529
1530static inline void init_idle_pids(struct pid_link *links)
1531{
1532        enum pid_type type;
1533
1534        for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1535                INIT_HLIST_NODE(&links[type].node); /* not really needed */
1536                links[type].pid = &init_struct_pid;
1537        }
1538}
1539
1540struct task_struct * __cpuinit fork_idle(int cpu)
1541{
1542        struct task_struct *task;
1543        task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1544        if (!IS_ERR(task)) {
1545                init_idle_pids(task->pids);
1546                init_idle(task, cpu);
1547        }
1548
1549        return task;
1550}
1551
1552/*
1553 *  Ok, this is the main fork-routine.
1554 *
1555 * It copies the process, and if successful kick-starts
1556 * it and waits for it to finish using the VM if required.
1557 */
1558long do_fork(unsigned long clone_flags,
1559              unsigned long stack_start,
1560              unsigned long stack_size,
1561              int __user *parent_tidptr,
1562              int __user *child_tidptr)
1563{
1564        struct task_struct *p;
1565        int trace = 0;
1566        long nr;
1567
1568        /*
1569         * Do some preliminary argument and permissions checking before we
1570         * actually start allocating stuff
1571         */
1572        if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
1573                if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
1574                        return -EINVAL;
1575        }
1576
1577        /*
1578         * Determine whether and which event to report to ptracer.  When
1579         * called from kernel_thread or CLONE_UNTRACED is explicitly
1580         * requested, no event is reported; otherwise, report if the event
1581         * for the type of forking is enabled.
1582         */
1583        if (!(clone_flags & CLONE_UNTRACED)) {
1584                if (clone_flags & CLONE_VFORK)
1585                        trace = PTRACE_EVENT_VFORK;
1586                else if ((clone_flags & CSIGNAL) != SIGCHLD)
1587                        trace = PTRACE_EVENT_CLONE;
1588                else
1589                        trace = PTRACE_EVENT_FORK;
1590
1591                if (likely(!ptrace_event_enabled(current, trace)))
1592                        trace = 0;
1593        }
1594
1595        p = copy_process(clone_flags, stack_start, stack_size,
1596                         child_tidptr, NULL, trace);
1597        /*
1598         * Do this prior waking up the new thread - the thread pointer
1599         * might get invalid after that point, if the thread exits quickly.
1600         */
1601        if (!IS_ERR(p)) {
1602                struct completion vfork;
1603
1604                trace_sched_process_fork(current, p);
1605
1606                nr = task_pid_vnr(p);
1607
1608                if (clone_flags & CLONE_PARENT_SETTID)
1609                        put_user(nr, parent_tidptr);
1610
1611                if (clone_flags & CLONE_VFORK) {
1612                        p->vfork_done = &vfork;
1613                        init_completion(&vfork);
1614                        get_task_struct(p);
1615                }
1616
1617                wake_up_new_task(p);
1618
1619                /* forking complete and child started to run, tell ptracer */
1620                if (unlikely(trace))
1621                        ptrace_event(trace, nr);
1622
1623                if (clone_flags & CLONE_VFORK) {
1624                        if (!wait_for_vfork_done(p, &vfork))
1625                                ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1626                }
1627        } else {
1628                nr = PTR_ERR(p);
1629        }
1630        return nr;
1631}
1632
1633/*
1634 * Create a kernel thread.
1635 */
1636pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1637{
1638        return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1639                (unsigned long)arg, NULL, NULL);
1640}
1641
1642#ifdef __ARCH_WANT_SYS_FORK
1643SYSCALL_DEFINE0(fork)
1644{
1645#ifdef CONFIG_MMU
1646        return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1647#else
1648        /* can not support in nommu mode */
1649        return(-EINVAL);
1650#endif
1651}
1652#endif
1653
1654#ifdef __ARCH_WANT_SYS_VFORK
1655SYSCALL_DEFINE0(vfork)
1656{
1657        return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0, 
1658                        0, NULL, NULL);
1659}
1660#endif
1661
1662#ifdef __ARCH_WANT_SYS_CLONE
1663#ifdef CONFIG_CLONE_BACKWARDS
1664SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1665                 int __user *, parent_tidptr,
1666                 int, tls_val,
1667                 int __user *, child_tidptr)
1668#elif defined(CONFIG_CLONE_BACKWARDS2)
1669SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1670                 int __user *, parent_tidptr,
1671                 int __user *, child_tidptr,
1672                 int, tls_val)
1673#else
1674SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1675                 int __user *, parent_tidptr,
1676                 int __user *, child_tidptr,
1677                 int, tls_val)
1678#endif
1679{
1680        long ret = do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1681        asmlinkage_protect(5, ret, clone_flags, newsp,
1682                        parent_tidptr, child_tidptr, tls_val);
1683        return ret;
1684}
1685#endif
1686
1687#ifndef ARCH_MIN_MMSTRUCT_ALIGN
1688#define ARCH_MIN_MMSTRUCT_ALIGN 0
1689#endif
1690
1691static void sighand_ctor(void *data)
1692{
1693        struct sighand_struct *sighand = data;
1694
1695        spin_lock_init(&sighand->siglock);
1696        init_waitqueue_head(&sighand->signalfd_wqh);
1697}
1698
1699void __init proc_caches_init(void)
1700{
1701        sighand_cachep = kmem_cache_create("sighand_cache",
1702                        sizeof(struct sighand_struct), 0,
1703                        SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1704                        SLAB_NOTRACK, sighand_ctor);
1705        signal_cachep = kmem_cache_create("signal_cache",
1706                        sizeof(struct signal_struct), 0,
1707                        SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1708        files_cachep = kmem_cache_create("files_cache",
1709                        sizeof(struct files_struct), 0,
1710                        SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1711        fs_cachep = kmem_cache_create("fs_cache",
1712                        sizeof(struct fs_struct), 0,
1713                        SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1714        /*
1715         * FIXME! The "sizeof(struct mm_struct)" currently includes the
1716         * whole struct cpumask for the OFFSTACK case. We could change
1717         * this to *only* allocate as much of it as required by the
1718         * maximum number of CPU's we can ever have.  The cpumask_allocation
1719         * is at the end of the structure, exactly for that reason.
1720         */
1721        mm_cachep = kmem_cache_create("mm_struct",
1722                        sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1723                        SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1724        vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1725        mmap_init();
1726        nsproxy_cache_init();
1727}
1728
1729/*
1730 * Check constraints on flags passed to the unshare system call.
1731 */
1732static int check_unshare_flags(unsigned long unshare_flags)
1733{
1734        if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1735                                CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1736                                CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1737                                CLONE_NEWUSER|CLONE_NEWPID))
1738                return -EINVAL;
1739        /*
1740         * Not implemented, but pretend it works if there is nothing to
1741         * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1742         * needs to unshare vm.
1743         */
1744        if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1745                /* FIXME: get_task_mm() increments ->mm_users */
1746                if (atomic_read(&current->mm->mm_users) > 1)
1747                        return -EINVAL;
1748        }
1749
1750        return 0;
1751}
1752
1753/*
1754 * Unshare the filesystem structure if it is being shared
1755 */
1756static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1757{
1758        struct fs_struct *fs = current->fs;
1759
1760        if (!(unshare_flags & CLONE_FS) || !fs)
1761                return 0;
1762
1763        /* don't need lock here; in the worst case we'll do useless copy */
1764        if (fs->users == 1)
1765                return 0;
1766
1767        *new_fsp = copy_fs_struct(fs);
1768        if (!*new_fsp)
1769                return -ENOMEM;
1770
1771        return 0;
1772}
1773
1774/*
1775 * Unshare file descriptor table if it is being shared
1776 */
1777static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1778{
1779        struct files_struct *fd = current->files;
1780        int error = 0;
1781
1782        if ((unshare_flags & CLONE_FILES) &&
1783            (fd && atomic_read(&fd->count) > 1)) {
1784                *new_fdp = dup_fd(fd, &error);
1785                if (!*new_fdp)
1786                        return error;
1787        }
1788
1789        return 0;
1790}
1791
1792/*
1793 * unshare allows a process to 'unshare' part of the process
1794 * context which was originally shared using clone.  copy_*
1795 * functions used by do_fork() cannot be used here directly
1796 * because they modify an inactive task_struct that is being
1797 * constructed. Here we are modifying the current, active,
1798 * task_struct.
1799 */
1800SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1801{
1802        struct fs_struct *fs, *new_fs = NULL;
1803        struct files_struct *fd, *new_fd = NULL;
1804        struct cred *new_cred = NULL;
1805        struct nsproxy *new_nsproxy = NULL;
1806        int do_sysvsem = 0;
1807        int err;
1808
1809        /*
1810         * If unsharing a user namespace must also unshare the thread.
1811         */
1812        if (unshare_flags & CLONE_NEWUSER)
1813                unshare_flags |= CLONE_THREAD | CLONE_FS;
1814        /*
1815         * If unsharing a pid namespace must also unshare the thread.
1816         */
1817        if (unshare_flags & CLONE_NEWPID)
1818                unshare_flags |= CLONE_THREAD;
1819        /*
1820         * If unsharing a thread from a thread group, must also unshare vm.
1821         */
1822        if (unshare_flags & CLONE_THREAD)
1823                unshare_flags |= CLONE_VM;
1824        /*
1825         * If unsharing vm, must also unshare signal handlers.
1826         */
1827        if (unshare_flags & CLONE_VM)
1828                unshare_flags |= CLONE_SIGHAND;
1829        /*
1830         * If unsharing namespace, must also unshare filesystem information.
1831         */
1832        if (unshare_flags & CLONE_NEWNS)
1833                unshare_flags |= CLONE_FS;
1834
1835        err = check_unshare_flags(unshare_flags);
1836        if (err)
1837                goto bad_unshare_out;
1838        /*
1839         * CLONE_NEWIPC must also detach from the undolist: after switching
1840         * to a new ipc namespace, the semaphore arrays from the old
1841         * namespace are unreachable.
1842         */
1843        if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1844                do_sysvsem = 1;
1845        err = unshare_fs(unshare_flags, &new_fs);
1846        if (err)
1847                goto bad_unshare_out;
1848        err = unshare_fd(unshare_flags, &new_fd);
1849        if (err)
1850                goto bad_unshare_cleanup_fs;
1851        err = unshare_userns(unshare_flags, &new_cred);
1852        if (err)
1853                goto bad_unshare_cleanup_fd;
1854        err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1855                                         new_cred, new_fs);
1856        if (err)
1857                goto bad_unshare_cleanup_cred;
1858
1859        if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1860                if (do_sysvsem) {
1861                        /*
1862                         * CLONE_SYSVSEM is equivalent to sys_exit().
1863                         */
1864                        exit_sem(current);
1865                }
1866
1867                if (new_nsproxy)
1868                        switch_task_namespaces(current, new_nsproxy);
1869
1870                task_lock(current);
1871
1872                if (new_fs) {
1873                        fs = current->fs;
1874                        spin_lock(&fs->lock);
1875                        current->fs = new_fs;
1876                        if (--fs->users)
1877                                new_fs = NULL;
1878                        else
1879                                new_fs = fs;
1880                        spin_unlock(&fs->lock);
1881                }
1882
1883                if (new_fd) {
1884                        fd = current->files;
1885                        current->files = new_fd;
1886                        new_fd = fd;
1887                }
1888
1889                task_unlock(current);
1890
1891                if (new_cred) {
1892                        /* Install the new user namespace */
1893                        commit_creds(new_cred);
1894                        new_cred = NULL;
1895                }
1896        }
1897
1898bad_unshare_cleanup_cred:
1899        if (new_cred)
1900                put_cred(new_cred);
1901bad_unshare_cleanup_fd:
1902        if (new_fd)
1903                put_files_struct(new_fd);
1904
1905bad_unshare_cleanup_fs:
1906        if (new_fs)
1907                free_fs_struct(new_fs);
1908
1909bad_unshare_out:
1910        return err;
1911}
1912
1913/*
1914 *      Helper to unshare the files of the current task.
1915 *      We don't want to expose copy_files internals to
1916 *      the exec layer of the kernel.
1917 */
1918
1919int unshare_files(struct files_struct **displaced)
1920{
1921        struct task_struct *task = current;
1922        struct files_struct *copy = NULL;
1923        int error;
1924
1925        error = unshare_fd(CLONE_FILES, &copy);
1926        if (error || !copy) {
1927                *displaced = NULL;
1928                return error;
1929        }
1930        *displaced = task->files;
1931        task_lock(task);
1932        task->files = copy;
1933        task_unlock(task);
1934        return 0;
1935}
1936
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.