kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/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  */
  89 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  90 int nr_threads;                 /* The idle threads do not count.. */
  91
  92 int max_threads;                /* tunable limit on nr_threads */
  93
  94 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  95
  96 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  97
  98 #ifdef CONFIG_PROVE_RCU
  99 int lockdep_tasklist_lock_is_held(void)
 100 {
 101         return lockdep_is_held(&tasklist_lock);
 102 }
 103 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 104 #endif /* #ifdef CONFIG_PROVE_RCU */
 105
 106 int 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
 117 void __weak arch_release_task_struct(struct task_struct *tsk)
 118 {
 119 }
 120
 121 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 122 static struct kmem_cache *task_struct_cachep;
 123
 124 static 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
 129 static inline void free_task_struct(struct task_struct *tsk)
 130 {
 131         kmem_cache_free(task_struct_cachep, tsk);
 132 }
 133 #endif
 134
 135 void __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
 146 static 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
 155 static inline void free_thread_info(struct thread_info *ti)
 156 {
 157         free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 158 }
 159 # else
 160 static struct kmem_cache *thread_info_cache;
 161
 162 static 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
 168 static void free_thread_info(struct thread_info *ti)
 169 {
 170         kmem_cache_free(thread_info_cache, ti);
 171 }
 172
 173 void 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) */
 183 static struct kmem_cache *signal_cachep;
 184
 185 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 186 struct kmem_cache *sighand_cachep;
 187
 188 /* SLAB cache for files_struct structures (tsk->files) */
 189 struct kmem_cache *files_cachep;
 190
 191 /* SLAB cache for fs_struct structures (tsk->fs) */
 192 struct kmem_cache *fs_cachep;
 193
 194 /* SLAB cache for vm_area_struct structures */
 195 struct kmem_cache *vm_area_cachep;
 196
 197 /* SLAB cache for mm_struct structures (tsk->mm) */
 198 static struct kmem_cache *mm_cachep;
 199
 200 static 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
 207 void 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 }
 218 EXPORT_SYMBOL(free_task);
 219
 220 static 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
 227 static 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
 233 void __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 }
 247 EXPORT_SYMBOL_GPL(__put_task_struct);
 248
 249 void __init __weak arch_task_cache_init(void) { }
 250
 251 void __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
 285 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 286                                                struct task_struct *src)
 287 {
 288         *dst = *src;
 289         return 0;
 290 }
 291
 292 static 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
 339 free_ti:
 340         free_thread_info(ti);
 341 free_tsk:
 342         free_task_struct(tsk);
 343         return NULL;
 344 }
 345
 346 #ifdef CONFIG_MMU
 347 static 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->f_path.dentry->d_inode;
 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;
 469 out:
 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;
 475 fail_nomem_anon_vma_fork:
 476         mpol_put(pol);
 477 fail_nomem_policy:
 478         kmem_cache_free(vm_area_cachep, tmp);
 479 fail_nomem:
 480         retval = -ENOMEM;
 481         vm_unacct_memory(charge);
 482         goto out;
 483 }
 484
 485 static 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
 493 static 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
 508 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 509
 510 static 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
 522 static 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
 530 static 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
 557 static 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  */
 577 struct 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  */
 595 void __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 }
 604 EXPORT_SYMBOL_GPL(__mmdrop);
 605
 606 /*
 607  * Decrement the use count and release all resources for an mm.
 608  */
 609 void 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 }
 630 EXPORT_SYMBOL_GPL(mmput);
 631
 632 void 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
 641 struct 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
 654 static 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  */
 670 struct 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 }
 685 EXPORT_SYMBOL_GPL(get_task_mm);
 686
 687 struct 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
 707 static 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
 720 static 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  */
 752 void 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  */
 808 struct 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
 849 free_pt:
 850         /* don't put binfmt in mmput, we haven't got module yet */
 851         mm->binfmt = NULL;
 852         mmput(mm);
 853
 854 fail_nomem:
 855         return NULL;
 856
 857 fail_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
 867 static 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
 901 good_mm:
 902         tsk->mm = mm;
 903         tsk->active_mm = mm;
 904         return 0;
 905
 906 fail_nomem:
 907         return retval;
 908 }
 909
 910 static 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
 930 static 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;
 953 out:
 954         return error;
 955 }
 956
 957 static 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
 983 static 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
1000 void __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  */
1012 static 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
1031 static 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
1078 static 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
1087 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1088 {
1089         current->clear_child_tid = tidptr;
1090
1091         return task_pid_vnr(current);
1092 }
1093
1094 static 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
1104 void 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  */
1113 static 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  */
1131 static 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         /*
1145          * Thread groups must share signals as well, and detached threads
1146          * can only be started up within the thread group.
1147          */
1148         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1149                 return ERR_PTR(-EINVAL);
1150
1151         /*
1152          * Shared signal handlers imply shared VM. By way of the above,
1153          * thread groups also imply shared VM. Blocking this case allows
1154          * for various simplifications in other code.
1155          */
1156         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1157                 return ERR_PTR(-EINVAL);
1158
1159         /*
1160          * Siblings of global init remain as zombies on exit since they are
1161          * not reaped by their parent (swapper). To solve this and to avoid
1162          * multi-rooted process trees, prevent global and container-inits
1163          * from creating siblings.
1164          */
1165         if ((clone_flags & CLONE_PARENT) &&
1166                                 current->signal->flags & SIGNAL_UNKILLABLE)
1167                 return ERR_PTR(-EINVAL);
1168
1169         /*
1170          * If the new process will be in a different pid namespace
1171          * don't allow the creation of threads.
1172          */
1173         if ((clone_flags & (CLONE_VM|CLONE_NEWPID)) &&
1174             (task_active_pid_ns(current) != current->nsproxy->pid_ns))
1175                 return ERR_PTR(-EINVAL);
1176
1177         retval = security_task_create(clone_flags);
1178         if (retval)
1179                 goto fork_out;
1180
1181         retval = -ENOMEM;
1182         p = dup_task_struct(current);
1183         if (!p)
1184                 goto fork_out;
1185
1186         ftrace_graph_init_task(p);
1187         get_seccomp_filter(p);
1188
1189         rt_mutex_init_task(p);
1190
1191 #ifdef CONFIG_PROVE_LOCKING
1192         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1193         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1194 #endif
1195         retval = -EAGAIN;
1196         if (atomic_read(&p->real_cred->user->processes) >=
1197                         task_rlimit(p, RLIMIT_NPROC)) {
1198                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1199                     p->real_cred->user != INIT_USER)
1200                         goto bad_fork_free;
1201         }
1202         current->flags &= ~PF_NPROC_EXCEEDED;
1203
1204         retval = copy_creds(p, clone_flags);
1205         if (retval < 0)
1206                 goto bad_fork_free;
1207
1208         /*
1209          * If multiple threads are within copy_process(), then this check
1210          * triggers too late. This doesn't hurt, the check is only there
1211          * to stop root fork bombs.
1212          */
1213         retval = -EAGAIN;
1214         if (nr_threads >= max_threads)
1215                 goto bad_fork_cleanup_count;
1216
1217         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1218                 goto bad_fork_cleanup_count;
1219
1220         p->did_exec = 0;
1221         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1222         copy_flags(clone_flags, p);
1223         INIT_LIST_HEAD(&p->children);
1224         INIT_LIST_HEAD(&p->sibling);
1225         rcu_copy_process(p);
1226         p->vfork_done = NULL;
1227         spin_lock_init(&p->alloc_lock);
1228
1229         init_sigpending(&p->pending);
1230
1231         p->utime = p->stime = p->gtime = 0;
1232         p->utimescaled = p->stimescaled = 0;
1233 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1234         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1235 #endif
1236 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1237         seqlock_init(&p->vtime_seqlock);
1238         p->vtime_snap = 0;
1239         p->vtime_snap_whence = VTIME_SLEEPING;
1240 #endif
1241
1242 #if defined(SPLIT_RSS_COUNTING)
1243         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1244 #endif
1245
1246         p->default_timer_slack_ns = current->timer_slack_ns;
1247
1248         task_io_accounting_init(&p->ioac);
1249         acct_clear_integrals(p);
1250
1251         posix_cpu_timers_init(p);
1252
1253         do_posix_clock_monotonic_gettime(&p->start_time);
1254         p->real_start_time = p->start_time;
1255         monotonic_to_bootbased(&p->real_start_time);
1256         p->io_context = NULL;
1257         p->audit_context = NULL;
1258         if (clone_flags & CLONE_THREAD)
1259                 threadgroup_change_begin(current);
1260         cgroup_fork(p);
1261 #ifdef CONFIG_NUMA
1262         p->mempolicy = mpol_dup(p->mempolicy);
1263         if (IS_ERR(p->mempolicy)) {
1264                 retval = PTR_ERR(p->mempolicy);
1265                 p->mempolicy = NULL;
1266                 goto bad_fork_cleanup_cgroup;
1267         }
1268         mpol_fix_fork_child_flag(p);
1269 #endif
1270 #ifdef CONFIG_CPUSETS
1271         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1272         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1273         seqcount_init(&p->mems_allowed_seq);
1274 #endif
1275 #ifdef CONFIG_TRACE_IRQFLAGS
1276         p->irq_events = 0;
1277         p->hardirqs_enabled = 0;
1278         p->hardirq_enable_ip = 0;
1279         p->hardirq_enable_event = 0;
1280         p->hardirq_disable_ip = _THIS_IP_;
1281         p->hardirq_disable_event = 0;
1282         p->softirqs_enabled = 1;
1283         p->softirq_enable_ip = _THIS_IP_;
1284         p->softirq_enable_event = 0;
1285         p->softirq_disable_ip = 0;
1286         p->softirq_disable_event = 0;
1287         p->hardirq_context = 0;
1288         p->softirq_context = 0;
1289 #endif
1290 #ifdef CONFIG_LOCKDEP
1291         p->lockdep_depth = 0; /* no locks held yet */
1292         p->curr_chain_key = 0;
1293         p->lockdep_recursion = 0;
1294 #endif
1295
1296 #ifdef CONFIG_DEBUG_MUTEXES
1297         p->blocked_on = NULL; /* not blocked yet */
1298 #endif
1299 #ifdef CONFIG_MEMCG
1300         p->memcg_batch.do_batch = 0;
1301         p->memcg_batch.memcg = NULL;
1302 #endif
1303
1304         /* Perform scheduler related setup. Assign this task to a CPU. */
1305         sched_fork(p);
1306
1307         retval = perf_event_init_task(p);
1308         if (retval)
1309                 goto bad_fork_cleanup_policy;
1310         retval = audit_alloc(p);
1311         if (retval)
1312                 goto bad_fork_cleanup_policy;
1313         /* copy all the process information */
1314         retval = copy_semundo(clone_flags, p);
1315         if (retval)
1316                 goto bad_fork_cleanup_audit;
1317         retval = copy_files(clone_flags, p);
1318         if (retval)
1319                 goto bad_fork_cleanup_semundo;
1320         retval = copy_fs(clone_flags, p);
1321         if (retval)
1322                 goto bad_fork_cleanup_files;
1323         retval = copy_sighand(clone_flags, p);
1324         if (retval)
1325                 goto bad_fork_cleanup_fs;
1326         retval = copy_signal(clone_flags, p);
1327         if (retval)
1328                 goto bad_fork_cleanup_sighand;
1329         retval = copy_mm(clone_flags, p);
1330         if (retval)
1331                 goto bad_fork_cleanup_signal;
1332         retval = copy_namespaces(clone_flags, p);
1333         if (retval)
1334                 goto bad_fork_cleanup_mm;
1335         retval = copy_io(clone_flags, p);
1336         if (retval)
1337                 goto bad_fork_cleanup_namespaces;
1338         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1339         if (retval)
1340                 goto bad_fork_cleanup_io;
1341
1342         if (pid != &init_struct_pid) {
1343                 retval = -ENOMEM;
1344                 pid = alloc_pid(p->nsproxy->pid_ns);
1345                 if (!pid)
1346                         goto bad_fork_cleanup_io;
1347         }
1348
1349         p->pid = pid_nr(pid);
1350         p->tgid = p->pid;
1351         if (clone_flags & CLONE_THREAD)
1352                 p->tgid = current->tgid;
1353
1354         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1355         /*
1356          * Clear TID on mm_release()?
1357          */
1358         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1359 #ifdef CONFIG_BLOCK
1360         p->plug = NULL;
1361 #endif
1362 #ifdef CONFIG_FUTEX
1363         p->robust_list = NULL;
1364 #ifdef CONFIG_COMPAT
1365         p->compat_robust_list = NULL;
1366 #endif
1367         INIT_LIST_HEAD(&p->pi_state_list);
1368         p->pi_state_cache = NULL;
1369 #endif
1370         uprobe_copy_process(p);
1371         /*
1372          * sigaltstack should be cleared when sharing the same VM
1373          */
1374         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1375                 p->sas_ss_sp = p->sas_ss_size = 0;
1376
1377         /*
1378          * Syscall tracing and stepping should be turned off in the
1379          * child regardless of CLONE_PTRACE.
1380          */
1381         user_disable_single_step(p);
1382         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1383 #ifdef TIF_SYSCALL_EMU
1384         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1385 #endif
1386         clear_all_latency_tracing(p);
1387
1388         /* ok, now we should be set up.. */
1389         if (clone_flags & CLONE_THREAD)
1390                 p->exit_signal = -1;
1391         else if (clone_flags & CLONE_PARENT)
1392                 p->exit_signal = current->group_leader->exit_signal;
1393         else
1394                 p->exit_signal = (clone_flags & CSIGNAL);
1395
1396         p->pdeath_signal = 0;
1397         p->exit_state = 0;
1398
1399         p->nr_dirtied = 0;
1400         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1401         p->dirty_paused_when = 0;
1402
1403         /*
1404          * Ok, make it visible to the rest of the system.
1405          * We dont wake it up yet.
1406          */
1407         p->group_leader = p;
1408         INIT_LIST_HEAD(&p->thread_group);
1409         p->task_works = NULL;
1410
1411         /* Need tasklist lock for parent etc handling! */
1412         write_lock_irq(&tasklist_lock);
1413
1414         /* CLONE_PARENT re-uses the old parent */
1415         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1416                 p->real_parent = current->real_parent;
1417                 p->parent_exec_id = current->parent_exec_id;
1418         } else {
1419                 p->real_parent = current;
1420                 p->parent_exec_id = current->self_exec_id;
1421         }
1422
1423         spin_lock(&current->sighand->siglock);
1424
1425         /*
1426          * Process group and session signals need to be delivered to just the
1427          * parent before the fork or both the parent and the child after the
1428          * fork. Restart if a signal comes in before we add the new process to
1429          * it's process group.
1430          * A fatal signal pending means that current will exit, so the new
1431          * thread can't slip out of an OOM kill (or normal SIGKILL).
1432         */
1433         recalc_sigpending();
1434         if (signal_pending(current)) {
1435                 spin_unlock(&current->sighand->siglock);
1436                 write_unlock_irq(&tasklist_lock);
1437                 retval = -ERESTARTNOINTR;
1438                 goto bad_fork_free_pid;
1439         }
1440
1441         if (clone_flags & CLONE_THREAD) {
1442                 current->signal->nr_threads++;
1443                 atomic_inc(&current->signal->live);
1444                 atomic_inc(&current->signal->sigcnt);
1445                 p->group_leader = current->group_leader;
1446                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1447         }
1448
1449         if (likely(p->pid)) {
1450                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1451
1452                 if (thread_group_leader(p)) {
1453                         if (is_child_reaper(pid)) {
1454                                 ns_of_pid(pid)->child_reaper = p;
1455                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1456                         }
1457
1458                         p->signal->leader_pid = pid;
1459                         p->signal->tty = tty_kref_get(current->signal->tty);
1460                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1461                         attach_pid(p, PIDTYPE_SID, task_session(current));
1462                         list_add_tail(&p->sibling, &p->real_parent->children);
1463                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1464                         __this_cpu_inc(process_counts);
1465                 }
1466                 attach_pid(p, PIDTYPE_PID, pid);
1467                 nr_threads++;
1468         }
1469
1470         total_forks++;
1471         spin_unlock(&current->sighand->siglock);
1472         write_unlock_irq(&tasklist_lock);
1473         proc_fork_connector(p);
1474         cgroup_post_fork(p);
1475         if (clone_flags & CLONE_THREAD)
1476                 threadgroup_change_end(current);
1477         perf_event_fork(p);
1478
1479         trace_task_newtask(p, clone_flags);
1480
1481         return p;
1482
1483 bad_fork_free_pid:
1484         if (pid != &init_struct_pid)
1485                 free_pid(pid);
1486 bad_fork_cleanup_io:
1487         if (p->io_context)
1488                 exit_io_context(p);
1489 bad_fork_cleanup_namespaces:
1490         exit_task_namespaces(p);
1491 bad_fork_cleanup_mm:
1492         if (p->mm)
1493                 mmput(p->mm);
1494 bad_fork_cleanup_signal:
1495         if (!(clone_flags & CLONE_THREAD))
1496                 free_signal_struct(p->signal);
1497 bad_fork_cleanup_sighand:
1498         __cleanup_sighand(p->sighand);
1499 bad_fork_cleanup_fs:
1500         exit_fs(p); /* blocking */
1501 bad_fork_cleanup_files:
1502         exit_files(p); /* blocking */
1503 bad_fork_cleanup_semundo:
1504         exit_sem(p);
1505 bad_fork_cleanup_audit:
1506         audit_free(p);
1507 bad_fork_cleanup_policy:
1508         perf_event_free_task(p);
1509 #ifdef CONFIG_NUMA
1510         mpol_put(p->mempolicy);
1511 bad_fork_cleanup_cgroup:
1512 #endif
1513         if (clone_flags & CLONE_THREAD)
1514                 threadgroup_change_end(current);
1515         cgroup_exit(p, 0);
1516         delayacct_tsk_free(p);
1517         module_put(task_thread_info(p)->exec_domain->module);
1518 bad_fork_cleanup_count:
1519         atomic_dec(&p->cred->user->processes);
1520         exit_creds(p);
1521 bad_fork_free:
1522         free_task(p);
1523 fork_out:
1524         return ERR_PTR(retval);
1525 }
1526
1527 static inline void init_idle_pids(struct pid_link *links)
1528 {
1529         enum pid_type type;
1530
1531         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1532                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1533                 links[type].pid = &init_struct_pid;
1534         }
1535 }
1536
1537 struct task_struct * __cpuinit fork_idle(int cpu)
1538 {
1539         struct task_struct *task;
1540         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1541         if (!IS_ERR(task)) {
1542                 init_idle_pids(task->pids);
1543                 init_idle(task, cpu);
1544         }
1545
1546         return task;
1547 }
1548
1549 /*
1550  *  Ok, this is the main fork-routine.
1551  *
1552  * It copies the process, and if successful kick-starts
1553  * it and waits for it to finish using the VM if required.
1554  */
1555 long do_fork(unsigned long clone_flags,
1556               unsigned long stack_start,
1557               unsigned long stack_size,
1558               int __user *parent_tidptr,
1559               int __user *child_tidptr)
1560 {
1561         struct task_struct *p;
1562         int trace = 0;
1563         long nr;
1564
1565         /*
1566          * Do some preliminary argument and permissions checking before we
1567          * actually start allocating stuff
1568          */
1569         if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
1570                 if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
1571                         return -EINVAL;
1572         }
1573
1574         /*
1575          * Determine whether and which event to report to ptracer.  When
1576          * called from kernel_thread or CLONE_UNTRACED is explicitly
1577          * requested, no event is reported; otherwise, report if the event
1578          * for the type of forking is enabled.
1579          */
1580         if (!(clone_flags & CLONE_UNTRACED)) {
1581                 if (clone_flags & CLONE_VFORK)
1582                         trace = PTRACE_EVENT_VFORK;
1583                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1584                         trace = PTRACE_EVENT_CLONE;
1585                 else
1586                         trace = PTRACE_EVENT_FORK;
1587
1588                 if (likely(!ptrace_event_enabled(current, trace)))
1589                         trace = 0;
1590         }
1591
1592         p = copy_process(clone_flags, stack_start, stack_size,
1593                          child_tidptr, NULL, trace);
1594         /*
1595          * Do this prior waking up the new thread - the thread pointer
1596          * might get invalid after that point, if the thread exits quickly.
1597          */
1598         if (!IS_ERR(p)) {
1599                 struct completion vfork;
1600
1601                 trace_sched_process_fork(current, p);
1602
1603                 nr = task_pid_vnr(p);
1604
1605                 if (clone_flags & CLONE_PARENT_SETTID)
1606                         put_user(nr, parent_tidptr);
1607
1608                 if (clone_flags & CLONE_VFORK) {
1609                         p->vfork_done = &vfork;
1610                         init_completion(&vfork);
1611                         get_task_struct(p);
1612                 }
1613
1614                 wake_up_new_task(p);
1615
1616                 /* forking complete and child started to run, tell ptracer */
1617                 if (unlikely(trace))
1618                         ptrace_event(trace, nr);
1619
1620                 if (clone_flags & CLONE_VFORK) {
1621                         if (!wait_for_vfork_done(p, &vfork))
1622                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1623                 }
1624         } else {
1625                 nr = PTR_ERR(p);
1626         }
1627         return nr;
1628 }
1629
1630 /*
1631  * Create a kernel thread.
1632  */
1633 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1634 {
1635         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1636                 (unsigned long)arg, NULL, NULL);
1637 }
1638
1639 #ifdef __ARCH_WANT_SYS_FORK
1640 SYSCALL_DEFINE0(fork)
1641 {
1642 #ifdef CONFIG_MMU
1643         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1644 #else
1645         /* can not support in nommu mode */
1646         return(-EINVAL);
1647 #endif
1648 }
1649 #endif
1650
1651 #ifdef __ARCH_WANT_SYS_VFORK
1652 SYSCALL_DEFINE0(vfork)
1653 {
1654         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1655                         0, NULL, NULL);
1656 }
1657 #endif
1658
1659 #ifdef __ARCH_WANT_SYS_CLONE
1660 #ifdef CONFIG_CLONE_BACKWARDS
1661 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1662                  int __user *, parent_tidptr,
1663                  int, tls_val,
1664                  int __user *, child_tidptr)
1665 #elif defined(CONFIG_CLONE_BACKWARDS2)
1666 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1667                  int __user *, parent_tidptr,
1668                  int __user *, child_tidptr,
1669                  int, tls_val)
1670 #else
1671 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1672                  int __user *, parent_tidptr,
1673                  int __user *, child_tidptr,
1674                  int, tls_val)
1675 #endif
1676 {
1677         long ret = do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1678         asmlinkage_protect(5, ret, clone_flags, newsp,
1679                         parent_tidptr, child_tidptr, tls_val);
1680         return ret;
1681 }
1682 #endif
1683
1684 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1685 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1686 #endif
1687
1688 static void sighand_ctor(void *data)
1689 {
1690         struct sighand_struct *sighand = data;
1691
1692         spin_lock_init(&sighand->siglock);
1693         init_waitqueue_head(&sighand->signalfd_wqh);
1694 }
1695
1696 void __init proc_caches_init(void)
1697 {
1698         sighand_cachep = kmem_cache_create("sighand_cache",
1699                         sizeof(struct sighand_struct), 0,
1700                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1701                         SLAB_NOTRACK, sighand_ctor);
1702         signal_cachep = kmem_cache_create("signal_cache",
1703                         sizeof(struct signal_struct), 0,
1704                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1705         files_cachep = kmem_cache_create("files_cache",
1706                         sizeof(struct files_struct), 0,
1707                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1708         fs_cachep = kmem_cache_create("fs_cache",
1709                         sizeof(struct fs_struct), 0,
1710                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1711         /*
1712          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1713          * whole struct cpumask for the OFFSTACK case. We could change
1714          * this to *only* allocate as much of it as required by the
1715          * maximum number of CPU's we can ever have.  The cpumask_allocation
1716          * is at the end of the structure, exactly for that reason.
1717          */
1718         mm_cachep = kmem_cache_create("mm_struct",
1719                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1720                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1721         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1722         mmap_init();
1723         nsproxy_cache_init();
1724 }
1725
1726 /*
1727  * Check constraints on flags passed to the unshare system call.
1728  */
1729 static int check_unshare_flags(unsigned long unshare_flags)
1730 {
1731         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1732                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1733                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1734                                 CLONE_NEWUSER|CLONE_NEWPID))
1735                 return -EINVAL;
1736         /*
1737          * Not implemented, but pretend it works if there is nothing to
1738          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1739          * needs to unshare vm.
1740          */
1741         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1742                 /* FIXME: get_task_mm() increments ->mm_users */
1743                 if (atomic_read(&current->mm->mm_users) > 1)
1744                         return -EINVAL;
1745         }
1746
1747         return 0;
1748 }
1749
1750 /*
1751  * Unshare the filesystem structure if it is being shared
1752  */
1753 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1754 {
1755         struct fs_struct *fs = current->fs;
1756
1757         if (!(unshare_flags & CLONE_FS) || !fs)
1758                 return 0;
1759
1760         /* don't need lock here; in the worst case we'll do useless copy */
1761         if (fs->users == 1)
1762                 return 0;
1763
1764         *new_fsp = copy_fs_struct(fs);
1765         if (!*new_fsp)
1766                 return -ENOMEM;
1767
1768         return 0;
1769 }
1770
1771 /*
1772  * Unshare file descriptor table if it is being shared
1773  */
1774 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1775 {
1776         struct files_struct *fd = current->files;
1777         int error = 0;
1778
1779         if ((unshare_flags & CLONE_FILES) &&
1780             (fd && atomic_read(&fd->count) > 1)) {
1781                 *new_fdp = dup_fd(fd, &error);
1782                 if (!*new_fdp)
1783                         return error;
1784         }
1785
1786         return 0;
1787 }
1788
1789 /*
1790  * unshare allows a process to 'unshare' part of the process
1791  * context which was originally shared using clone.  copy_*
1792  * functions used by do_fork() cannot be used here directly
1793  * because they modify an inactive task_struct that is being
1794  * constructed. Here we are modifying the current, active,
1795  * task_struct.
1796  */
1797 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1798 {
1799         struct fs_struct *fs, *new_fs = NULL;
1800         struct files_struct *fd, *new_fd = NULL;
1801         struct cred *new_cred = NULL;
1802         struct nsproxy *new_nsproxy = NULL;
1803         int do_sysvsem = 0;
1804         int err;
1805
1806         /*
1807          * If unsharing a user namespace must also unshare the thread.
1808          */
1809         if (unshare_flags & CLONE_NEWUSER)
1810                 unshare_flags |= CLONE_THREAD;
1811         /*
1812          * If unsharing a pid namespace must also unshare the thread.
1813          */
1814         if (unshare_flags & CLONE_NEWPID)
1815                 unshare_flags |= CLONE_THREAD;
1816         /*
1817          * If unsharing a thread from a thread group, must also unshare vm.
1818          */
1819         if (unshare_flags & CLONE_THREAD)
1820                 unshare_flags |= CLONE_VM;
1821         /*
1822          * If unsharing vm, must also unshare signal handlers.
1823          */
1824         if (unshare_flags & CLONE_VM)
1825                 unshare_flags |= CLONE_SIGHAND;
1826         /*
1827          * If unsharing namespace, must also unshare filesystem information.
1828          */
1829         if (unshare_flags & CLONE_NEWNS)
1830                 unshare_flags |= CLONE_FS;
1831
1832         err = check_unshare_flags(unshare_flags);
1833         if (err)
1834                 goto bad_unshare_out;
1835         /*
1836          * CLONE_NEWIPC must also detach from the undolist: after switching
1837          * to a new ipc namespace, the semaphore arrays from the old
1838          * namespace are unreachable.
1839          */
1840         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1841                 do_sysvsem = 1;
1842         err = unshare_fs(unshare_flags, &new_fs);
1843         if (err)
1844                 goto bad_unshare_out;
1845         err = unshare_fd(unshare_flags, &new_fd);
1846         if (err)
1847                 goto bad_unshare_cleanup_fs;
1848         err = unshare_userns(unshare_flags, &new_cred);
1849         if (err)
1850                 goto bad_unshare_cleanup_fd;
1851         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1852                                          new_cred, new_fs);
1853         if (err)
1854                 goto bad_unshare_cleanup_cred;
1855
1856         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1857                 if (do_sysvsem) {
1858                         /*
1859                          * CLONE_SYSVSEM is equivalent to sys_exit().
1860                          */
1861                         exit_sem(current);
1862                 }
1863
1864                 if (new_nsproxy) {
1865                         switch_task_namespaces(current, new_nsproxy);
1866                         new_nsproxy = NULL;
1867                 }
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
1897         if (new_nsproxy)
1898                 put_nsproxy(new_nsproxy);
1899
1900 bad_unshare_cleanup_cred:
1901         if (new_cred)
1902                 put_cred(new_cred);
1903 bad_unshare_cleanup_fd:
1904         if (new_fd)
1905                 put_files_struct(new_fd);
1906
1907 bad_unshare_cleanup_fs:
1908         if (new_fs)
1909                 free_fs_struct(new_fs);
1910
1911 bad_unshare_out:
1912         return err;
1913 }
1914
1915 /*
1916  *      Helper to unshare the files of the current task.
1917  *      We don't want to expose copy_files internals to
1918  *      the exec layer of the kernel.
1919  */
1920
1921 int unshare_files(struct files_struct **displaced)
1922 {
1923         struct task_struct *task = current;
1924         struct files_struct *copy = NULL;
1925         int error;
1926
1927         error = unshare_fd(CLONE_FILES, &copy);
1928         if (error || !copy) {
1929                 *displaced = NULL;
1930                 return error;
1931         }
1932         *displaced = task->files;
1933         task_lock(task);
1934         task->files = copy;
1935         task_unlock(task);
1936         return 0;
1937 }