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#include <linux/aio.h>
  74
  75#include <asm/pgtable.h>
  76#include <asm/pgalloc.h>
  77#include <asm/uaccess.h>
  78#include <asm/mmu_context.h>
  79#include <asm/cacheflush.h>
  80#include <asm/tlbflush.h>
  81
  82#include <trace/events/sched.h>
  83
  84#define CREATE_TRACE_POINTS
  85#include <trace/events/task.h>
  86
  87/*
  88 * Protected counters by write_lock_irq(&tasklist_lock)
  89 */
  90unsigned long total_forks;      /* Handle normal Linux uptimes. */
  91int nr_threads;                 /* The idle threads do not count.. */
  92
  93int max_threads;                /* tunable limit on nr_threads */
  94
  95DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  96
  97__cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  98
  99#ifdef CONFIG_PROVE_RCU
 100int lockdep_tasklist_lock_is_held(void)
 101{
 102        return lockdep_is_held(&tasklist_lock);
 103}
 104EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 105#endif /* #ifdef CONFIG_PROVE_RCU */
 106
 107int nr_processes(void)
 108{
 109        int cpu;
 110        int total = 0;
 111
 112        for_each_possible_cpu(cpu)
 113                total += per_cpu(process_counts, cpu);
 114
 115        return total;
 116}
 117
 118void __weak arch_release_task_struct(struct task_struct *tsk)
 119{
 120}
 121
 122#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 123static struct kmem_cache *task_struct_cachep;
 124
 125static inline struct task_struct *alloc_task_struct_node(int node)
 126{
 127        return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 128}
 129
 130static inline void free_task_struct(struct task_struct *tsk)
 131{
 132        kmem_cache_free(task_struct_cachep, tsk);
 133}
 134#endif
 135
 136void __weak arch_release_thread_info(struct thread_info *ti)
 137{
 138}
 139
 140#ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 141
 142/*
 143 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 144 * kmemcache based allocator.
 145 */
 146# if THREAD_SIZE >= PAGE_SIZE
 147static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 148                                                  int node)
 149{
 150        struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
 151                                             THREAD_SIZE_ORDER);
 152
 153        return page ? page_address(page) : NULL;
 154}
 155
 156static inline void free_thread_info(struct thread_info *ti)
 157{
 158        free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 159}
 160# else
 161static struct kmem_cache *thread_info_cache;
 162
 163static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 164                                                  int node)
 165{
 166        return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 167}
 168
 169static void free_thread_info(struct thread_info *ti)
 170{
 171        kmem_cache_free(thread_info_cache, ti);
 172}
 173
 174void thread_info_cache_init(void)
 175{
 176        thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 177                                              THREAD_SIZE, 0, NULL);
 178        BUG_ON(thread_info_cache == NULL);
 179}
 180# endif
 181#endif
 182
 183/* SLAB cache for signal_struct structures (tsk->signal) */
 184static struct kmem_cache *signal_cachep;
 185
 186/* SLAB cache for sighand_struct structures (tsk->sighand) */
 187struct kmem_cache *sighand_cachep;
 188
 189/* SLAB cache for files_struct structures (tsk->files) */
 190struct kmem_cache *files_cachep;
 191
 192/* SLAB cache for fs_struct structures (tsk->fs) */
 193struct kmem_cache *fs_cachep;
 194
 195/* SLAB cache for vm_area_struct structures */
 196struct kmem_cache *vm_area_cachep;
 197
 198/* SLAB cache for mm_struct structures (tsk->mm) */
 199static struct kmem_cache *mm_cachep;
 200
 201static void account_kernel_stack(struct thread_info *ti, int account)
 202{
 203        struct zone *zone = page_zone(virt_to_page(ti));
 204
 205        mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 206}
 207
 208void free_task(struct task_struct *tsk)
 209{
 210        account_kernel_stack(tsk->stack, -1);
 211        arch_release_thread_info(tsk->stack);
 212        free_thread_info(tsk->stack);
 213        rt_mutex_debug_task_free(tsk);
 214        ftrace_graph_exit_task(tsk);
 215        put_seccomp_filter(tsk);
 216        arch_release_task_struct(tsk);
 217        free_task_struct(tsk);
 218}
 219EXPORT_SYMBOL(free_task);
 220
 221static inline void free_signal_struct(struct signal_struct *sig)
 222{
 223        taskstats_tgid_free(sig);
 224        sched_autogroup_exit(sig);
 225        kmem_cache_free(signal_cachep, sig);
 226}
 227
 228static inline void put_signal_struct(struct signal_struct *sig)
 229{
 230        if (atomic_dec_and_test(&sig->sigcnt))
 231                free_signal_struct(sig);
 232}
 233
 234void __put_task_struct(struct task_struct *tsk)
 235{
 236        WARN_ON(!tsk->exit_state);
 237        WARN_ON(atomic_read(&tsk->usage));
 238        WARN_ON(tsk == current);
 239
 240        security_task_free(tsk);
 241        exit_creds(tsk);
 242        delayacct_tsk_free(tsk);
 243        put_signal_struct(tsk->signal);
 244
 245        if (!profile_handoff_task(tsk))
 246                free_task(tsk);
 247}
 248EXPORT_SYMBOL_GPL(__put_task_struct);
 249
 250void __init __weak arch_task_cache_init(void) { }
 251
 252void __init fork_init(unsigned long mempages)
 253{
 254#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 255#ifndef ARCH_MIN_TASKALIGN
 256#define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 257#endif
 258        /* create a slab on which task_structs can be allocated */
 259        task_struct_cachep =
 260                kmem_cache_create("task_struct", sizeof(struct task_struct),
 261                        ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 262#endif
 263
 264        /* do the arch specific task caches init */
 265        arch_task_cache_init();
 266
 267        /*
 268         * The default maximum number of threads is set to a safe
 269         * value: the thread structures can take up at most half
 270         * of memory.
 271         */
 272        max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 273
 274        /*
 275         * we need to allow at least 20 threads to boot a system
 276         */
 277        if (max_threads < 20)
 278                max_threads = 20;
 279
 280        init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 281        init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 282        init_task.signal->rlim[RLIMIT_SIGPENDING] =
 283                init_task.signal->rlim[RLIMIT_NPROC];
 284}
 285
 286int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 287                                               struct task_struct *src)
 288{
 289        *dst = *src;
 290        return 0;
 291}
 292
 293static struct task_struct *dup_task_struct(struct task_struct *orig)
 294{
 295        struct task_struct *tsk;
 296        struct thread_info *ti;
 297        unsigned long *stackend;
 298        int node = tsk_fork_get_node(orig);
 299        int err;
 300
 301        tsk = alloc_task_struct_node(node);
 302        if (!tsk)
 303                return NULL;
 304
 305        ti = alloc_thread_info_node(tsk, node);
 306        if (!ti)
 307                goto free_tsk;
 308
 309        err = arch_dup_task_struct(tsk, orig);
 310        if (err)
 311                goto free_ti;
 312
 313        tsk->stack = ti;
 314
 315        setup_thread_stack(tsk, orig);
 316        clear_user_return_notifier(tsk);
 317        clear_tsk_need_resched(tsk);
 318        stackend = end_of_stack(tsk);
 319        *stackend = STACK_END_MAGIC;    /* for overflow detection */
 320
 321#ifdef CONFIG_CC_STACKPROTECTOR
 322        tsk->stack_canary = get_random_int();
 323#endif
 324
 325        /*
 326         * One for us, one for whoever does the "release_task()" (usually
 327         * parent)
 328         */
 329        atomic_set(&tsk->usage, 2);
 330#ifdef CONFIG_BLK_DEV_IO_TRACE
 331        tsk->btrace_seq = 0;
 332#endif
 333        tsk->splice_pipe = NULL;
 334        tsk->task_frag.page = NULL;
 335
 336        account_kernel_stack(ti, 1);
 337
 338        return tsk;
 339
 340free_ti:
 341        free_thread_info(ti);
 342free_tsk:
 343        free_task_struct(tsk);
 344        return NULL;
 345}
 346
 347#ifdef CONFIG_MMU
 348static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 349{
 350        struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 351        struct rb_node **rb_link, *rb_parent;
 352        int retval;
 353        unsigned long charge;
 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->map_count = 0;
 368        cpumask_clear(mm_cpumask(mm));
 369        mm->mm_rb = RB_ROOT;
 370        rb_link = &mm->mm_rb.rb_node;
 371        rb_parent = NULL;
 372        pprev = &mm->mmap;
 373        retval = ksm_fork(mm, oldmm);
 374        if (retval)
 375                goto out;
 376        retval = khugepaged_fork(mm, oldmm);
 377        if (retval)
 378                goto out;
 379
 380        prev = NULL;
 381        for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 382                struct file *file;
 383
 384                if (mpnt->vm_flags & VM_DONTCOPY) {
 385                        vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 386                                                        -vma_pages(mpnt));
 387                        continue;
 388                }
 389                charge = 0;
 390                if (mpnt->vm_flags & VM_ACCOUNT) {
 391                        unsigned long len = vma_pages(mpnt);
 392
 393                        if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 394                                goto fail_nomem;
 395                        charge = len;
 396                }
 397                tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 398                if (!tmp)
 399                        goto fail_nomem;
 400                *tmp = *mpnt;
 401                INIT_LIST_HEAD(&tmp->anon_vma_chain);
 402                retval = vma_dup_policy(mpnt, tmp);
 403                if (retval)
 404                        goto fail_nomem_policy;
 405                tmp->vm_mm = mm;
 406                if (anon_vma_fork(tmp, mpnt))
 407                        goto fail_nomem_anon_vma_fork;
 408                tmp->vm_flags &= ~VM_LOCKED;
 409                tmp->vm_next = tmp->vm_prev = NULL;
 410                file = tmp->vm_file;
 411                if (file) {
 412                        struct inode *inode = file_inode(file);
 413                        struct address_space *mapping = file->f_mapping;
 414
 415                        get_file(file);
 416                        if (tmp->vm_flags & VM_DENYWRITE)
 417                                atomic_dec(&inode->i_writecount);
 418                        mutex_lock(&mapping->i_mmap_mutex);
 419                        if (tmp->vm_flags & VM_SHARED)
 420                                mapping->i_mmap_writable++;
 421                        flush_dcache_mmap_lock(mapping);
 422                        /* insert tmp into the share list, just after mpnt */
 423                        if (unlikely(tmp->vm_flags & VM_NONLINEAR))
 424                                vma_nonlinear_insert(tmp,
 425                                                &mapping->i_mmap_nonlinear);
 426                        else
 427                                vma_interval_tree_insert_after(tmp, mpnt,
 428                                                        &mapping->i_mmap);
 429                        flush_dcache_mmap_unlock(mapping);
 430                        mutex_unlock(&mapping->i_mmap_mutex);
 431                }
 432
 433                /*
 434                 * Clear hugetlb-related page reserves for children. This only
 435                 * affects MAP_PRIVATE mappings. Faults generated by the child
 436                 * are not guaranteed to succeed, even if read-only
 437                 */
 438                if (is_vm_hugetlb_page(tmp))
 439                        reset_vma_resv_huge_pages(tmp);
 440
 441                /*
 442                 * Link in the new vma and copy the page table entries.
 443                 */
 444                *pprev = tmp;
 445                pprev = &tmp->vm_next;
 446                tmp->vm_prev = prev;
 447                prev = tmp;
 448
 449                __vma_link_rb(mm, tmp, rb_link, rb_parent);
 450                rb_link = &tmp->vm_rb.rb_right;
 451                rb_parent = &tmp->vm_rb;
 452
 453                mm->map_count++;
 454                retval = copy_page_range(mm, oldmm, mpnt);
 455
 456                if (tmp->vm_ops && tmp->vm_ops->open)
 457                        tmp->vm_ops->open(tmp);
 458
 459                if (retval)
 460                        goto out;
 461        }
 462        /* a new mm has just been created */
 463        arch_dup_mmap(oldmm, mm);
 464        retval = 0;
 465out:
 466        up_write(&mm->mmap_sem);
 467        flush_tlb_mm(oldmm);
 468        up_write(&oldmm->mmap_sem);
 469        uprobe_end_dup_mmap();
 470        return retval;
 471fail_nomem_anon_vma_fork:
 472        mpol_put(vma_policy(tmp));
 473fail_nomem_policy:
 474        kmem_cache_free(vm_area_cachep, tmp);
 475fail_nomem:
 476        retval = -ENOMEM;
 477        vm_unacct_memory(charge);
 478        goto out;
 479}
 480
 481static inline int mm_alloc_pgd(struct mm_struct *mm)
 482{
 483        mm->pgd = pgd_alloc(mm);
 484        if (unlikely(!mm->pgd))
 485                return -ENOMEM;
 486        return 0;
 487}
 488
 489static inline void mm_free_pgd(struct mm_struct *mm)
 490{
 491        pgd_free(mm, mm->pgd);
 492}
 493#else
 494#define dup_mmap(mm, oldmm)     (0)
 495#define mm_alloc_pgd(mm)        (0)
 496#define mm_free_pgd(mm)
 497#endif /* CONFIG_MMU */
 498
 499__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 500
 501#define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 502#define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 503
 504static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 505
 506static int __init coredump_filter_setup(char *s)
 507{
 508        default_dump_filter =
 509                (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 510                MMF_DUMP_FILTER_MASK;
 511        return 1;
 512}
 513
 514__setup("coredump_filter=", coredump_filter_setup);
 515
 516#include <linux/init_task.h>
 517
 518static void mm_init_aio(struct mm_struct *mm)
 519{
 520#ifdef CONFIG_AIO
 521        spin_lock_init(&mm->ioctx_lock);
 522        mm->ioctx_table = NULL;
 523#endif
 524}
 525
 526static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 527{
 528        atomic_set(&mm->mm_users, 1);
 529        atomic_set(&mm->mm_count, 1);
 530        init_rwsem(&mm->mmap_sem);
 531        INIT_LIST_HEAD(&mm->mmlist);
 532        mm->flags = (current->mm) ?
 533                (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 534        mm->core_state = NULL;
 535        mm->nr_ptes = 0;
 536        memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 537        spin_lock_init(&mm->page_table_lock);
 538        mm_init_aio(mm);
 539        mm_init_owner(mm, p);
 540
 541        if (likely(!mm_alloc_pgd(mm))) {
 542                mm->def_flags = 0;
 543                mmu_notifier_mm_init(mm);
 544                return mm;
 545        }
 546
 547        free_mm(mm);
 548        return NULL;
 549}
 550
 551static void check_mm(struct mm_struct *mm)
 552{
 553        int i;
 554
 555        for (i = 0; i < NR_MM_COUNTERS; i++) {
 556                long x = atomic_long_read(&mm->rss_stat.count[i]);
 557
 558                if (unlikely(x))
 559                        printk(KERN_ALERT "BUG: Bad rss-counter state "
 560                                          "mm:%p idx:%d val:%ld\n", mm, i, x);
 561        }
 562
 563#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 564        VM_BUG_ON(mm->pmd_huge_pte);
 565#endif
 566}
 567
 568/*
 569 * Allocate and initialize an mm_struct.
 570 */
 571struct mm_struct *mm_alloc(void)
 572{
 573        struct mm_struct *mm;
 574
 575        mm = allocate_mm();
 576        if (!mm)
 577                return NULL;
 578
 579        memset(mm, 0, sizeof(*mm));
 580        mm_init_cpumask(mm);
 581        return mm_init(mm, current);
 582}
 583
 584/*
 585 * Called when the last reference to the mm
 586 * is dropped: either by a lazy thread or by
 587 * mmput. Free the page directory and the mm.
 588 */
 589void __mmdrop(struct mm_struct *mm)
 590{
 591        BUG_ON(mm == &init_mm);
 592        mm_free_pgd(mm);
 593        destroy_context(mm);
 594        mmu_notifier_mm_destroy(mm);
 595        check_mm(mm);
 596        free_mm(mm);
 597}
 598EXPORT_SYMBOL_GPL(__mmdrop);
 599
 600/*
 601 * Decrement the use count and release all resources for an mm.
 602 */
 603void mmput(struct mm_struct *mm)
 604{
 605        might_sleep();
 606
 607        if (atomic_dec_and_test(&mm->mm_users)) {
 608                uprobe_clear_state(mm);
 609                exit_aio(mm);
 610                ksm_exit(mm);
 611                khugepaged_exit(mm); /* must run before exit_mmap */
 612                exit_mmap(mm);
 613                set_mm_exe_file(mm, NULL);
 614                if (!list_empty(&mm->mmlist)) {
 615                        spin_lock(&mmlist_lock);
 616                        list_del(&mm->mmlist);
 617                        spin_unlock(&mmlist_lock);
 618                }
 619                if (mm->binfmt)
 620                        module_put(mm->binfmt->module);
 621                mmdrop(mm);
 622        }
 623}
 624EXPORT_SYMBOL_GPL(mmput);
 625
 626void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 627{
 628        if (new_exe_file)
 629                get_file(new_exe_file);
 630        if (mm->exe_file)
 631                fput(mm->exe_file);
 632        mm->exe_file = new_exe_file;
 633}
 634
 635struct file *get_mm_exe_file(struct mm_struct *mm)
 636{
 637        struct file *exe_file;
 638
 639        /* We need mmap_sem to protect against races with removal of exe_file */
 640        down_read(&mm->mmap_sem);
 641        exe_file = mm->exe_file;
 642        if (exe_file)
 643                get_file(exe_file);
 644        up_read(&mm->mmap_sem);
 645        return exe_file;
 646}
 647
 648static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 649{
 650        /* It's safe to write the exe_file pointer without exe_file_lock because
 651         * this is called during fork when the task is not yet in /proc */
 652        newmm->exe_file = get_mm_exe_file(oldmm);
 653}
 654
 655/**
 656 * get_task_mm - acquire a reference to the task's mm
 657 *
 658 * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 659 * this kernel workthread has transiently adopted a user mm with use_mm,
 660 * to do its AIO) is not set and if so returns a reference to it, after
 661 * bumping up the use count.  User must release the mm via mmput()
 662 * after use.  Typically used by /proc and ptrace.
 663 */
 664struct mm_struct *get_task_mm(struct task_struct *task)
 665{
 666        struct mm_struct *mm;
 667
 668        task_lock(task);
 669        mm = task->mm;
 670        if (mm) {
 671                if (task->flags & PF_KTHREAD)
 672                        mm = NULL;
 673                else
 674                        atomic_inc(&mm->mm_users);
 675        }
 676        task_unlock(task);
 677        return mm;
 678}
 679EXPORT_SYMBOL_GPL(get_task_mm);
 680
 681struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 682{
 683        struct mm_struct *mm;
 684        int err;
 685
 686        err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 687        if (err)
 688                return ERR_PTR(err);
 689
 690        mm = get_task_mm(task);
 691        if (mm && mm != current->mm &&
 692                        !ptrace_may_access(task, mode)) {
 693                mmput(mm);
 694                mm = ERR_PTR(-EACCES);
 695        }
 696        mutex_unlock(&task->signal->cred_guard_mutex);
 697
 698        return mm;
 699}
 700
 701static void complete_vfork_done(struct task_struct *tsk)
 702{
 703        struct completion *vfork;
 704
 705        task_lock(tsk);
 706        vfork = tsk->vfork_done;
 707        if (likely(vfork)) {
 708                tsk->vfork_done = NULL;
 709                complete(vfork);
 710        }
 711        task_unlock(tsk);
 712}
 713
 714static int wait_for_vfork_done(struct task_struct *child,
 715                                struct completion *vfork)
 716{
 717        int killed;
 718
 719        freezer_do_not_count();
 720        killed = wait_for_completion_killable(vfork);
 721        freezer_count();
 722
 723        if (killed) {
 724                task_lock(child);
 725                child->vfork_done = NULL;
 726                task_unlock(child);
 727        }
 728
 729        put_task_struct(child);
 730        return killed;
 731}
 732
 733/* Please note the differences between mmput and mm_release.
 734 * mmput is called whenever we stop holding onto a mm_struct,
 735 * error success whatever.
 736 *
 737 * mm_release is called after a mm_struct has been removed
 738 * from the current process.
 739 *
 740 * This difference is important for error handling, when we
 741 * only half set up a mm_struct for a new process and need to restore
 742 * the old one.  Because we mmput the new mm_struct before
 743 * restoring the old one. . .
 744 * Eric Biederman 10 January 1998
 745 */
 746void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 747{
 748        /* Get rid of any futexes when releasing the mm */
 749#ifdef CONFIG_FUTEX
 750        if (unlikely(tsk->robust_list)) {
 751                exit_robust_list(tsk);
 752                tsk->robust_list = NULL;
 753        }
 754#ifdef CONFIG_COMPAT
 755        if (unlikely(tsk->compat_robust_list)) {
 756                compat_exit_robust_list(tsk);
 757                tsk->compat_robust_list = NULL;
 758        }
 759#endif
 760        if (unlikely(!list_empty(&tsk->pi_state_list)))
 761                exit_pi_state_list(tsk);
 762#endif
 763
 764        uprobe_free_utask(tsk);
 765
 766        /* Get rid of any cached register state */
 767        deactivate_mm(tsk, mm);
 768
 769        /*
 770         * If we're exiting normally, clear a user-space tid field if
 771         * requested.  We leave this alone when dying by signal, to leave
 772         * the value intact in a core dump, and to save the unnecessary
 773         * trouble, say, a killed vfork parent shouldn't touch this mm.
 774         * Userland only wants this done for a sys_exit.
 775         */
 776        if (tsk->clear_child_tid) {
 777                if (!(tsk->flags & PF_SIGNALED) &&
 778                    atomic_read(&mm->mm_users) > 1) {
 779                        /*
 780                         * We don't check the error code - if userspace has
 781                         * not set up a proper pointer then tough luck.
 782                         */
 783                        put_user(0, tsk->clear_child_tid);
 784                        sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 785                                        1, NULL, NULL, 0);
 786                }
 787                tsk->clear_child_tid = NULL;
 788        }
 789
 790        /*
 791         * All done, finally we can wake up parent and return this mm to him.
 792         * Also kthread_stop() uses this completion for synchronization.
 793         */
 794        if (tsk->vfork_done)
 795                complete_vfork_done(tsk);
 796}
 797
 798/*
 799 * Allocate a new mm structure and copy contents from the
 800 * mm structure of the passed in task structure.
 801 */
 802struct mm_struct *dup_mm(struct task_struct *tsk)
 803{
 804        struct mm_struct *mm, *oldmm = current->mm;
 805        int err;
 806
 807        if (!oldmm)
 808                return NULL;
 809
 810        mm = allocate_mm();
 811        if (!mm)
 812                goto fail_nomem;
 813
 814        memcpy(mm, oldmm, sizeof(*mm));
 815        mm_init_cpumask(mm);
 816
 817#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 818        mm->pmd_huge_pte = NULL;
 819#endif
 820#ifdef CONFIG_NUMA_BALANCING
 821        mm->first_nid = NUMA_PTE_SCAN_INIT;
 822#endif
 823        if (!mm_init(mm, tsk))
 824                goto fail_nomem;
 825
 826        if (init_new_context(tsk, mm))
 827                goto fail_nocontext;
 828
 829        dup_mm_exe_file(oldmm, mm);
 830
 831        err = dup_mmap(mm, oldmm);
 832        if (err)
 833                goto free_pt;
 834
 835        mm->hiwater_rss = get_mm_rss(mm);
 836        mm->hiwater_vm = mm->total_vm;
 837
 838        if (mm->binfmt && !try_module_get(mm->binfmt->module))
 839                goto free_pt;
 840
 841        return mm;
 842
 843free_pt:
 844        /* don't put binfmt in mmput, we haven't got module yet */
 845        mm->binfmt = NULL;
 846        mmput(mm);
 847
 848fail_nomem:
 849        return NULL;
 850
 851fail_nocontext:
 852        /*
 853         * If init_new_context() failed, we cannot use mmput() to free the mm
 854         * because it calls destroy_context()
 855         */
 856        mm_free_pgd(mm);
 857        free_mm(mm);
 858        return NULL;
 859}
 860
 861static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 862{
 863        struct mm_struct *mm, *oldmm;
 864        int retval;
 865
 866        tsk->min_flt = tsk->maj_flt = 0;
 867        tsk->nvcsw = tsk->nivcsw = 0;
 868#ifdef CONFIG_DETECT_HUNG_TASK
 869        tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 870#endif
 871
 872        tsk->mm = NULL;
 873        tsk->active_mm = NULL;
 874
 875        /*
 876         * Are we cloning a kernel thread?
 877         *
 878         * We need to steal a active VM for that..
 879         */
 880        oldmm = current->mm;
 881        if (!oldmm)
 882                return 0;
 883
 884        if (clone_flags & CLONE_VM) {
 885                atomic_inc(&oldmm->mm_users);
 886                mm = oldmm;
 887                goto good_mm;
 888        }
 889
 890        retval = -ENOMEM;
 891        mm = dup_mm(tsk);
 892        if (!mm)
 893                goto fail_nomem;
 894
 895good_mm:
 896        tsk->mm = mm;
 897        tsk->active_mm = mm;
 898        return 0;
 899
 900fail_nomem:
 901        return retval;
 902}
 903
 904static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 905{
 906        struct fs_struct *fs = current->fs;
 907        if (clone_flags & CLONE_FS) {
 908                /* tsk->fs is already what we want */
 909                spin_lock(&fs->lock);
 910                if (fs->in_exec) {
 911                        spin_unlock(&fs->lock);
 912                        return -EAGAIN;
 913                }
 914                fs->users++;
 915                spin_unlock(&fs->lock);
 916                return 0;
 917        }
 918        tsk->fs = copy_fs_struct(fs);
 919        if (!tsk->fs)
 920                return -ENOMEM;
 921        return 0;
 922}
 923
 924static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 925{
 926        struct files_struct *oldf, *newf;
 927        int error = 0;
 928
 929        /*
 930         * A background process may not have any files ...
 931         */
 932        oldf = current->files;
 933        if (!oldf)
 934                goto out;
 935
 936        if (clone_flags & CLONE_FILES) {
 937                atomic_inc(&oldf->count);
 938                goto out;
 939        }
 940
 941        newf = dup_fd(oldf, &error);
 942        if (!newf)
 943                goto out;
 944
 945        tsk->files = newf;
 946        error = 0;
 947out:
 948        return error;
 949}
 950
 951static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 952{
 953#ifdef CONFIG_BLOCK
 954        struct io_context *ioc = current->io_context;
 955        struct io_context *new_ioc;
 956
 957        if (!ioc)
 958                return 0;
 959        /*
 960         * Share io context with parent, if CLONE_IO is set
 961         */
 962        if (clone_flags & CLONE_IO) {
 963                ioc_task_link(ioc);
 964                tsk->io_context = ioc;
 965        } else if (ioprio_valid(ioc->ioprio)) {
 966                new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 967                if (unlikely(!new_ioc))
 968                        return -ENOMEM;
 969
 970                new_ioc->ioprio = ioc->ioprio;
 971                put_io_context(new_ioc);
 972        }
 973#endif
 974        return 0;
 975}
 976
 977static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 978{
 979        struct sighand_struct *sig;
 980
 981        if (clone_flags & CLONE_SIGHAND) {
 982                atomic_inc(&current->sighand->count);
 983                return 0;
 984        }
 985        sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 986        rcu_assign_pointer(tsk->sighand, sig);
 987        if (!sig)
 988                return -ENOMEM;
 989        atomic_set(&sig->count, 1);
 990        memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 991        return 0;
 992}
 993
 994void __cleanup_sighand(struct sighand_struct *sighand)
 995{
 996        if (atomic_dec_and_test(&sighand->count)) {
 997                signalfd_cleanup(sighand);
 998                kmem_cache_free(sighand_cachep, sighand);
 999        }
1000}
1001
1002
1003/*
1004 * Initialize POSIX timer handling for a thread group.
1005 */
1006static void posix_cpu_timers_init_group(struct signal_struct *sig)
1007{
1008        unsigned long cpu_limit;
1009
1010        /* Thread group counters. */
1011        thread_group_cputime_init(sig);
1012
1013        cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1014        if (cpu_limit != RLIM_INFINITY) {
1015                sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1016                sig->cputimer.running = 1;
1017        }
1018
1019        /* The timer lists. */
1020        INIT_LIST_HEAD(&sig->cpu_timers[0]);
1021        INIT_LIST_HEAD(&sig->cpu_timers[1]);
1022        INIT_LIST_HEAD(&sig->cpu_timers[2]);
1023}
1024
1025static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1026{
1027        struct signal_struct *sig;
1028
1029        if (clone_flags & CLONE_THREAD)
1030                return 0;
1031
1032        sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1033        tsk->signal = sig;
1034        if (!sig)
1035                return -ENOMEM;
1036
1037        sig->nr_threads = 1;
1038        atomic_set(&sig->live, 1);
1039        atomic_set(&sig->sigcnt, 1);
1040        init_waitqueue_head(&sig->wait_chldexit);
1041        sig->curr_target = tsk;
1042        init_sigpending(&sig->shared_pending);
1043        INIT_LIST_HEAD(&sig->posix_timers);
1044
1045        hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1046        sig->real_timer.function = it_real_fn;
1047
1048        task_lock(current->group_leader);
1049        memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1050        task_unlock(current->group_leader);
1051
1052        posix_cpu_timers_init_group(sig);
1053
1054        tty_audit_fork(sig);
1055        sched_autogroup_fork(sig);
1056
1057#ifdef CONFIG_CGROUPS
1058        init_rwsem(&sig->group_rwsem);
1059#endif
1060
1061        sig->oom_score_adj = current->signal->oom_score_adj;
1062        sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1063
1064        sig->has_child_subreaper = current->signal->has_child_subreaper ||
1065                                   current->signal->is_child_subreaper;
1066
1067        mutex_init(&sig->cred_guard_mutex);
1068
1069        return 0;
1070}
1071
1072static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1073{
1074        unsigned long new_flags = p->flags;
1075
1076        new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1077        new_flags |= PF_FORKNOEXEC;
1078        p->flags = new_flags;
1079}
1080
1081SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1082{
1083        current->clear_child_tid = tidptr;
1084
1085        return task_pid_vnr(current);
1086}
1087
1088static void rt_mutex_init_task(struct task_struct *p)
1089{
1090        raw_spin_lock_init(&p->pi_lock);
1091#ifdef CONFIG_RT_MUTEXES
1092        plist_head_init(&p->pi_waiters);
1093        p->pi_blocked_on = NULL;
1094#endif
1095}
1096
1097#ifdef CONFIG_MM_OWNER
1098void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1099{
1100        mm->owner = p;
1101}
1102#endif /* CONFIG_MM_OWNER */
1103
1104/*
1105 * Initialize POSIX timer handling for a single task.
1106 */
1107static void posix_cpu_timers_init(struct task_struct *tsk)
1108{
1109        tsk->cputime_expires.prof_exp = 0;
1110        tsk->cputime_expires.virt_exp = 0;
1111        tsk->cputime_expires.sched_exp = 0;
1112        INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1113        INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1114        INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1115}
1116
1117static inline void
1118init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1119{
1120         task->pids[type].pid = pid;
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 or user namespace
1174         * do not allow it to share a thread group or signal handlers or
1175         * parent with the forking task.
1176         */
1177        if (clone_flags & (CLONE_SIGHAND | CLONE_PARENT)) {
1178                if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1179                    (task_active_pid_ns(current) !=
1180                                current->nsproxy->pid_ns_for_children))
1181                        return ERR_PTR(-EINVAL);
1182        }
1183
1184        retval = security_task_create(clone_flags);
1185        if (retval)
1186                goto fork_out;
1187
1188        retval = -ENOMEM;
1189        p = dup_task_struct(current);
1190        if (!p)
1191                goto fork_out;
1192
1193        ftrace_graph_init_task(p);
1194        get_seccomp_filter(p);
1195
1196        rt_mutex_init_task(p);
1197
1198#ifdef CONFIG_PROVE_LOCKING
1199        DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1200        DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1201#endif
1202        retval = -EAGAIN;
1203        if (atomic_read(&p->real_cred->user->processes) >=
1204                        task_rlimit(p, RLIMIT_NPROC)) {
1205                if (p->real_cred->user != INIT_USER &&
1206                    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1207                        goto bad_fork_free;
1208        }
1209        current->flags &= ~PF_NPROC_EXCEEDED;
1210
1211        retval = copy_creds(p, clone_flags);
1212        if (retval < 0)
1213                goto bad_fork_free;
1214
1215        /*
1216         * If multiple threads are within copy_process(), then this check
1217         * triggers too late. This doesn't hurt, the check is only there
1218         * to stop root fork bombs.
1219         */
1220        retval = -EAGAIN;
1221        if (nr_threads >= max_threads)
1222                goto bad_fork_cleanup_count;
1223
1224        if (!try_module_get(task_thread_info(p)->exec_domain->module))
1225                goto bad_fork_cleanup_count;
1226
1227        p->did_exec = 0;
1228        delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1229        copy_flags(clone_flags, p);
1230        INIT_LIST_HEAD(&p->children);
1231        INIT_LIST_HEAD(&p->sibling);
1232        rcu_copy_process(p);
1233        p->vfork_done = NULL;
1234        spin_lock_init(&p->alloc_lock);
1235
1236        init_sigpending(&p->pending);
1237
1238        p->utime = p->stime = p->gtime = 0;
1239        p->utimescaled = p->stimescaled = 0;
1240#ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1241        p->prev_cputime.utime = p->prev_cputime.stime = 0;
1242#endif
1243#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1244        seqlock_init(&p->vtime_seqlock);
1245        p->vtime_snap = 0;
1246        p->vtime_snap_whence = VTIME_SLEEPING;
1247#endif
1248
1249#if defined(SPLIT_RSS_COUNTING)
1250        memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1251#endif
1252
1253        p->default_timer_slack_ns = current->timer_slack_ns;
1254
1255        task_io_accounting_init(&p->ioac);
1256        acct_clear_integrals(p);
1257
1258        posix_cpu_timers_init(p);
1259
1260        do_posix_clock_monotonic_gettime(&p->start_time);
1261        p->real_start_time = p->start_time;
1262        monotonic_to_bootbased(&p->real_start_time);
1263        p->io_context = NULL;
1264        p->audit_context = NULL;
1265        if (clone_flags & CLONE_THREAD)
1266                threadgroup_change_begin(current);
1267        cgroup_fork(p);
1268#ifdef CONFIG_NUMA
1269        p->mempolicy = mpol_dup(p->mempolicy);
1270        if (IS_ERR(p->mempolicy)) {
1271                retval = PTR_ERR(p->mempolicy);
1272                p->mempolicy = NULL;
1273                goto bad_fork_cleanup_cgroup;
1274        }
1275        mpol_fix_fork_child_flag(p);
1276#endif
1277#ifdef CONFIG_CPUSETS
1278        p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1279        p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1280        seqcount_init(&p->mems_allowed_seq);
1281#endif
1282#ifdef CONFIG_TRACE_IRQFLAGS
1283        p->irq_events = 0;
1284        p->hardirqs_enabled = 0;
1285        p->hardirq_enable_ip = 0;
1286        p->hardirq_enable_event = 0;
1287        p->hardirq_disable_ip = _THIS_IP_;
1288        p->hardirq_disable_event = 0;
1289        p->softirqs_enabled = 1;
1290        p->softirq_enable_ip = _THIS_IP_;
1291        p->softirq_enable_event = 0;
1292        p->softirq_disable_ip = 0;
1293        p->softirq_disable_event = 0;
1294        p->hardirq_context = 0;
1295        p->softirq_context = 0;
1296#endif
1297#ifdef CONFIG_LOCKDEP
1298        p->lockdep_depth = 0; /* no locks held yet */
1299        p->curr_chain_key = 0;
1300        p->lockdep_recursion = 0;
1301#endif
1302
1303#ifdef CONFIG_DEBUG_MUTEXES
1304        p->blocked_on = NULL; /* not blocked yet */
1305#endif
1306#ifdef CONFIG_MEMCG
1307        p->memcg_batch.do_batch = 0;
1308        p->memcg_batch.memcg = NULL;
1309#endif
1310#ifdef CONFIG_BCACHE
1311        p->sequential_io        = 0;
1312        p->sequential_io_avg    = 0;
1313#endif
1314
1315        /* Perform scheduler related setup. Assign this task to a CPU. */
1316        sched_fork(p);
1317
1318        retval = perf_event_init_task(p);
1319        if (retval)
1320                goto bad_fork_cleanup_policy;
1321        retval = audit_alloc(p);
1322        if (retval)
1323                goto bad_fork_cleanup_policy;
1324        /* copy all the process information */
1325        retval = copy_semundo(clone_flags, p);
1326        if (retval)
1327                goto bad_fork_cleanup_audit;
1328        retval = copy_files(clone_flags, p);
1329        if (retval)
1330                goto bad_fork_cleanup_semundo;
1331        retval = copy_fs(clone_flags, p);
1332        if (retval)
1333                goto bad_fork_cleanup_files;
1334        retval = copy_sighand(clone_flags, p);
1335        if (retval)
1336                goto bad_fork_cleanup_fs;
1337        retval = copy_signal(clone_flags, p);
1338        if (retval)
1339                goto bad_fork_cleanup_sighand;
1340        retval = copy_mm(clone_flags, p);
1341        if (retval)
1342                goto bad_fork_cleanup_signal;
1343        retval = copy_namespaces(clone_flags, p);
1344        if (retval)
1345                goto bad_fork_cleanup_mm;
1346        retval = copy_io(clone_flags, p);
1347        if (retval)
1348                goto bad_fork_cleanup_namespaces;
1349        retval = copy_thread(clone_flags, stack_start, stack_size, p);
1350        if (retval)
1351                goto bad_fork_cleanup_io;
1352
1353        if (pid != &init_struct_pid) {
1354                retval = -ENOMEM;
1355                pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1356                if (!pid)
1357                        goto bad_fork_cleanup_io;
1358        }
1359
1360        p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1361        /*
1362         * Clear TID on mm_release()?
1363         */
1364        p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1365#ifdef CONFIG_BLOCK
1366        p->plug = NULL;
1367#endif
1368#ifdef CONFIG_FUTEX
1369        p->robust_list = NULL;
1370#ifdef CONFIG_COMPAT
1371        p->compat_robust_list = NULL;
1372#endif
1373        INIT_LIST_HEAD(&p->pi_state_list);
1374        p->pi_state_cache = NULL;
1375#endif
1376        uprobe_copy_process(p);
1377        /*
1378         * sigaltstack should be cleared when sharing the same VM
1379         */
1380        if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1381                p->sas_ss_sp = p->sas_ss_size = 0;
1382
1383        /*
1384         * Syscall tracing and stepping should be turned off in the
1385         * child regardless of CLONE_PTRACE.
1386         */
1387        user_disable_single_step(p);
1388        clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1389#ifdef TIF_SYSCALL_EMU
1390        clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1391#endif
1392        clear_all_latency_tracing(p);
1393
1394        /* ok, now we should be set up.. */
1395        p->pid = pid_nr(pid);
1396        if (clone_flags & CLONE_THREAD) {
1397                p->exit_signal = -1;
1398                p->group_leader = current->group_leader;
1399                p->tgid = current->tgid;
1400        } else {
1401                if (clone_flags & CLONE_PARENT)
1402                        p->exit_signal = current->group_leader->exit_signal;
1403                else
1404                        p->exit_signal = (clone_flags & CSIGNAL);
1405                p->group_leader = p;
1406                p->tgid = p->pid;
1407        }
1408
1409        p->pdeath_signal = 0;
1410        p->exit_state = 0;
1411
1412        p->nr_dirtied = 0;
1413        p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1414        p->dirty_paused_when = 0;
1415
1416        INIT_LIST_HEAD(&p->thread_group);
1417        p->task_works = NULL;
1418
1419        /*
1420         * Make it visible to the rest of the system, but dont wake it up yet.
1421         * Need tasklist lock for parent etc handling!
1422         */
1423        write_lock_irq(&tasklist_lock);
1424
1425        /* CLONE_PARENT re-uses the old parent */
1426        if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1427                p->real_parent = current->real_parent;
1428                p->parent_exec_id = current->parent_exec_id;
1429        } else {
1430                p->real_parent = current;
1431                p->parent_exec_id = current->self_exec_id;
1432        }
1433
1434        spin_lock(&current->sighand->siglock);
1435
1436        /*
1437         * Process group and session signals need to be delivered to just the
1438         * parent before the fork or both the parent and the child after the
1439         * fork. Restart if a signal comes in before we add the new process to
1440         * it's process group.
1441         * A fatal signal pending means that current will exit, so the new
1442         * thread can't slip out of an OOM kill (or normal SIGKILL).
1443        */
1444        recalc_sigpending();
1445        if (signal_pending(current)) {
1446                spin_unlock(&current->sighand->siglock);
1447                write_unlock_irq(&tasklist_lock);
1448                retval = -ERESTARTNOINTR;
1449                goto bad_fork_free_pid;
1450        }
1451
1452        if (likely(p->pid)) {
1453                ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1454
1455                init_task_pid(p, PIDTYPE_PID, pid);
1456                if (thread_group_leader(p)) {
1457                        init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1458                        init_task_pid(p, PIDTYPE_SID, task_session(current));
1459
1460                        if (is_child_reaper(pid)) {
1461                                ns_of_pid(pid)->child_reaper = p;
1462                                p->signal->flags |= SIGNAL_UNKILLABLE;
1463                        }
1464
1465                        p->signal->leader_pid = pid;
1466                        p->signal->tty = tty_kref_get(current->signal->tty);
1467                        list_add_tail(&p->sibling, &p->real_parent->children);
1468                        list_add_tail_rcu(&p->tasks, &init_task.tasks);
1469                        attach_pid(p, PIDTYPE_PGID);
1470                        attach_pid(p, PIDTYPE_SID);
1471                        __this_cpu_inc(process_counts);
1472                } else {
1473                        current->signal->nr_threads++;
1474                        atomic_inc(&current->signal->live);
1475                        atomic_inc(&current->signal->sigcnt);
1476                        list_add_tail_rcu(&p->thread_group,
1477                                          &p->group_leader->thread_group);
1478                }
1479                attach_pid(p, PIDTYPE_PID);
1480                nr_threads++;
1481        }
1482
1483        total_forks++;
1484        spin_unlock(&current->sighand->siglock);
1485        write_unlock_irq(&tasklist_lock);
1486        proc_fork_connector(p);
1487        cgroup_post_fork(p);
1488        if (clone_flags & CLONE_THREAD)
1489                threadgroup_change_end(current);
1490        perf_event_fork(p);
1491
1492        trace_task_newtask(p, clone_flags);
1493
1494        return p;
1495
1496bad_fork_free_pid:
1497        if (pid != &init_struct_pid)
1498                free_pid(pid);
1499bad_fork_cleanup_io:
1500        if (p->io_context)
1501                exit_io_context(p);
1502bad_fork_cleanup_namespaces:
1503        exit_task_namespaces(p);
1504bad_fork_cleanup_mm:
1505        if (p->mm)
1506                mmput(p->mm);
1507bad_fork_cleanup_signal:
1508        if (!(clone_flags & CLONE_THREAD))
1509                free_signal_struct(p->signal);
1510bad_fork_cleanup_sighand:
1511        __cleanup_sighand(p->sighand);
1512bad_fork_cleanup_fs:
1513        exit_fs(p); /* blocking */
1514bad_fork_cleanup_files:
1515        exit_files(p); /* blocking */
1516bad_fork_cleanup_semundo:
1517        exit_sem(p);
1518bad_fork_cleanup_audit:
1519        audit_free(p);
1520bad_fork_cleanup_policy:
1521        perf_event_free_task(p);
1522#ifdef CONFIG_NUMA
1523        mpol_put(p->mempolicy);
1524bad_fork_cleanup_cgroup:
1525#endif
1526        if (clone_flags & CLONE_THREAD)
1527                threadgroup_change_end(current);
1528        cgroup_exit(p, 0);
1529        delayacct_tsk_free(p);
1530        module_put(task_thread_info(p)->exec_domain->module);
1531bad_fork_cleanup_count:
1532        atomic_dec(&p->cred->user->processes);
1533        exit_creds(p);
1534bad_fork_free:
1535        free_task(p);
1536fork_out:
1537        return ERR_PTR(retval);
1538}
1539
1540static inline void init_idle_pids(struct pid_link *links)
1541{
1542        enum pid_type type;
1543
1544        for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1545                INIT_HLIST_NODE(&links[type].node); /* not really needed */
1546                links[type].pid = &init_struct_pid;
1547        }
1548}
1549
1550struct task_struct *fork_idle(int cpu)
1551{
1552        struct task_struct *task;
1553        task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1554        if (!IS_ERR(task)) {
1555                init_idle_pids(task->pids);
1556                init_idle(task, cpu);
1557        }
1558
1559        return task;
1560}
1561
1562/*
1563 *  Ok, this is the main fork-routine.
1564 *
1565 * It copies the process, and if successful kick-starts
1566 * it and waits for it to finish using the VM if required.
1567 */
1568long do_fork(unsigned long clone_flags,
1569              unsigned long stack_start,
1570              unsigned long stack_size,
1571              int __user *parent_tidptr,
1572              int __user *child_tidptr)
1573{
1574        struct task_struct *p;
1575        int trace = 0;
1576        long nr;
1577
1578        /*
1579         * Determine whether and which event to report to ptracer.  When
1580         * called from kernel_thread or CLONE_UNTRACED is explicitly
1581         * requested, no event is reported; otherwise, report if the event
1582         * for the type of forking is enabled.
1583         */
1584        if (!(clone_flags & CLONE_UNTRACED)) {
1585                if (clone_flags & CLONE_VFORK)
1586                        trace = PTRACE_EVENT_VFORK;
1587                else if ((clone_flags & CSIGNAL) != SIGCHLD)
1588                        trace = PTRACE_EVENT_CLONE;
1589                else
1590                        trace = PTRACE_EVENT_FORK;
1591
1592                if (likely(!ptrace_event_enabled(current, trace)))
1593                        trace = 0;
1594        }
1595
1596        p = copy_process(clone_flags, stack_start, stack_size,
1597                         child_tidptr, NULL, trace);
1598        /*
1599         * Do this prior waking up the new thread - the thread pointer
1600         * might get invalid after that point, if the thread exits quickly.
1601         */
1602        if (!IS_ERR(p)) {
1603                struct completion vfork;
1604
1605                trace_sched_process_fork(current, p);
1606
1607                nr = task_pid_vnr(p);
1608
1609                if (clone_flags & CLONE_PARENT_SETTID)
1610                        put_user(nr, parent_tidptr);
1611
1612                if (clone_flags & CLONE_VFORK) {
1613                        p->vfork_done = &vfork;
1614                        init_completion(&vfork);
1615                        get_task_struct(p);
1616                }
1617
1618                wake_up_new_task(p);
1619
1620                /* forking complete and child started to run, tell ptracer */
1621                if (unlikely(trace))
1622                        ptrace_event(trace, nr);
1623
1624                if (clone_flags & CLONE_VFORK) {
1625                        if (!wait_for_vfork_done(p, &vfork))
1626                                ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1627                }
1628        } else {
1629                nr = PTR_ERR(p);
1630        }
1631        return nr;
1632}
1633
1634/*
1635 * Create a kernel thread.
1636 */
1637pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1638{
1639        return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1640                (unsigned long)arg, NULL, NULL);
1641}
1642
1643#ifdef __ARCH_WANT_SYS_FORK
1644SYSCALL_DEFINE0(fork)
1645{
1646#ifdef CONFIG_MMU
1647        return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1648#else
1649        /* can not support in nommu mode */
1650        return(-EINVAL);
1651#endif
1652}
1653#endif
1654
1655#ifdef __ARCH_WANT_SYS_VFORK
1656SYSCALL_DEFINE0(vfork)
1657{
1658        return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0, 
1659                        0, NULL, NULL);
1660}
1661#endif
1662
1663#ifdef __ARCH_WANT_SYS_CLONE
1664#ifdef CONFIG_CLONE_BACKWARDS
1665SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1666                 int __user *, parent_tidptr,
1667                 int, tls_val,
1668                 int __user *, child_tidptr)
1669#elif defined(CONFIG_CLONE_BACKWARDS2)
1670SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1671                 int __user *, parent_tidptr,
1672                 int __user *, child_tidptr,
1673                 int, tls_val)
1674#elif defined(CONFIG_CLONE_BACKWARDS3)
1675SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1676                int, stack_size,
1677                int __user *, parent_tidptr,
1678                int __user *, child_tidptr,
1679                int, tls_val)
1680#else
1681SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1682                 int __user *, parent_tidptr,
1683                 int __user *, child_tidptr,
1684                 int, tls_val)
1685#endif
1686{
1687        return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1688}
1689#endif
1690
1691#ifndef ARCH_MIN_MMSTRUCT_ALIGN
1692#define ARCH_MIN_MMSTRUCT_ALIGN 0
1693#endif
1694
1695static void sighand_ctor(void *data)
1696{
1697        struct sighand_struct *sighand = data;
1698
1699        spin_lock_init(&sighand->siglock);
1700        init_waitqueue_head(&sighand->signalfd_wqh);
1701}
1702
1703void __init proc_caches_init(void)
1704{
1705        sighand_cachep = kmem_cache_create("sighand_cache",
1706                        sizeof(struct sighand_struct), 0,
1707                        SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1708                        SLAB_NOTRACK, sighand_ctor);
1709        signal_cachep = kmem_cache_create("signal_cache",
1710                        sizeof(struct signal_struct), 0,
1711                        SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1712        files_cachep = kmem_cache_create("files_cache",
1713                        sizeof(struct files_struct), 0,
1714                        SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1715        fs_cachep = kmem_cache_create("fs_cache",
1716                        sizeof(struct fs_struct), 0,
1717                        SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1718        /*
1719         * FIXME! The "sizeof(struct mm_struct)" currently includes the
1720         * whole struct cpumask for the OFFSTACK case. We could change
1721         * this to *only* allocate as much of it as required by the
1722         * maximum number of CPU's we can ever have.  The cpumask_allocation
1723         * is at the end of the structure, exactly for that reason.
1724         */
1725        mm_cachep = kmem_cache_create("mm_struct",
1726                        sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1727                        SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1728        vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1729        mmap_init();
1730        nsproxy_cache_init();
1731}
1732
1733/*
1734 * Check constraints on flags passed to the unshare system call.
1735 */
1736static int check_unshare_flags(unsigned long unshare_flags)
1737{
1738        if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1739                                CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1740                                CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1741                                CLONE_NEWUSER|CLONE_NEWPID))
1742                return -EINVAL;
1743        /*
1744         * Not implemented, but pretend it works if there is nothing to
1745         * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1746         * needs to unshare vm.
1747         */
1748        if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1749                /* FIXME: get_task_mm() increments ->mm_users */
1750                if (atomic_read(&current->mm->mm_users) > 1)
1751                        return -EINVAL;
1752        }
1753
1754        return 0;
1755}
1756
1757/*
1758 * Unshare the filesystem structure if it is being shared
1759 */
1760static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1761{
1762        struct fs_struct *fs = current->fs;
1763
1764        if (!(unshare_flags & CLONE_FS) || !fs)
1765                return 0;
1766
1767        /* don't need lock here; in the worst case we'll do useless copy */
1768        if (fs->users == 1)
1769                return 0;
1770
1771        *new_fsp = copy_fs_struct(fs);
1772        if (!*new_fsp)
1773                return -ENOMEM;
1774
1775        return 0;
1776}
1777
1778/*
1779 * Unshare file descriptor table if it is being shared
1780 */
1781static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1782{
1783        struct files_struct *fd = current->files;
1784        int error = 0;
1785
1786        if ((unshare_flags & CLONE_FILES) &&
1787            (fd && atomic_read(&fd->count) > 1)) {
1788                *new_fdp = dup_fd(fd, &error);
1789                if (!*new_fdp)
1790                        return error;
1791        }
1792
1793        return 0;
1794}
1795
1796/*
1797 * unshare allows a process to 'unshare' part of the process
1798 * context which was originally shared using clone.  copy_*
1799 * functions used by do_fork() cannot be used here directly
1800 * because they modify an inactive task_struct that is being
1801 * constructed. Here we are modifying the current, active,
1802 * task_struct.
1803 */
1804SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1805{
1806        struct fs_struct *fs, *new_fs = NULL;
1807        struct files_struct *fd, *new_fd = NULL;
1808        struct cred *new_cred = NULL;
1809        struct nsproxy *new_nsproxy = NULL;
1810        int do_sysvsem = 0;
1811        int err;
1812
1813        /*
1814         * If unsharing a user namespace must also unshare the thread.
1815         */
1816        if (unshare_flags & CLONE_NEWUSER)
1817                unshare_flags |= CLONE_THREAD | CLONE_FS;
1818        /*
1819         * If unsharing a thread from a thread group, must also unshare vm.
1820         */
1821        if (unshare_flags & CLONE_THREAD)
1822                unshare_flags |= CLONE_VM;
1823        /*
1824         * If unsharing vm, must also unshare signal handlers.
1825         */
1826        if (unshare_flags & CLONE_VM)
1827                unshare_flags |= CLONE_SIGHAND;
1828        /*
1829         * If unsharing namespace, must also unshare filesystem information.
1830         */
1831        if (unshare_flags & CLONE_NEWNS)
1832                unshare_flags |= CLONE_FS;
1833
1834        err = check_unshare_flags(unshare_flags);
1835        if (err)
1836                goto bad_unshare_out;
1837        /*
1838         * CLONE_NEWIPC must also detach from the undolist: after switching
1839         * to a new ipc namespace, the semaphore arrays from the old
1840         * namespace are unreachable.
1841         */
1842        if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1843                do_sysvsem = 1;
1844        err = unshare_fs(unshare_flags, &new_fs);
1845        if (err)
1846                goto bad_unshare_out;
1847        err = unshare_fd(unshare_flags, &new_fd);
1848        if (err)
1849                goto bad_unshare_cleanup_fs;
1850        err = unshare_userns(unshare_flags, &new_cred);
1851        if (err)
1852                goto bad_unshare_cleanup_fd;
1853        err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1854                                         new_cred, new_fs);
1855        if (err)
1856                goto bad_unshare_cleanup_cred;
1857
1858        if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1859                if (do_sysvsem) {
1860                        /*
1861                         * CLONE_SYSVSEM is equivalent to sys_exit().
1862                         */
1863                        exit_sem(current);
1864                }
1865
1866                if (new_nsproxy)
1867                        switch_task_namespaces(current, new_nsproxy);
1868
1869                task_lock(current);
1870
1871                if (new_fs) {
1872                        fs = current->fs;
1873                        spin_lock(&fs->lock);
1874                        current->fs = new_fs;
1875                        if (--fs->users)
1876                                new_fs = NULL;
1877                        else
1878                                new_fs = fs;
1879                        spin_unlock(&fs->lock);
1880                }
1881
1882                if (new_fd) {
1883                        fd = current->files;
1884                        current->files = new_fd;
1885                        new_fd = fd;
1886                }
1887
1888                task_unlock(current);
1889
1890                if (new_cred) {
1891                        /* Install the new user namespace */
1892                        commit_creds(new_cred);
1893                        new_cred = NULL;
1894                }
1895        }
1896
1897bad_unshare_cleanup_cred:
1898        if (new_cred)
1899                put_cred(new_cred);
1900bad_unshare_cleanup_fd:
1901        if (new_fd)
1902                put_files_struct(new_fd);
1903
1904bad_unshare_cleanup_fs:
1905        if (new_fs)
1906                free_fs_struct(new_fs);
1907
1908bad_unshare_out:
1909        return err;
1910}
1911
1912/*
1913 *      Helper to unshare the files of the current task.
1914 *      We don't want to expose copy_files internals to
1915 *      the exec layer of the kernel.
1916 */
1917
1918int unshare_files(struct files_struct **displaced)
1919{
1920        struct task_struct *task = current;
1921        struct files_struct *copy = NULL;
1922        int error;
1923
1924        error = unshare_fd(CLONE_FILES, &copy);
1925        if (error || !copy) {
1926                *displaced = NULL;
1927                return error;
1928        }
1929        *displaced = task->files;
1930        task_lock(task);
1931        task->files = copy;
1932        task_unlock(task);
1933        return 0;
1934}
1935
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.