kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77
  78 #include <asm/pgtable.h>
  79 #include <asm/pgalloc.h>
  80 #include <asm/uaccess.h>
  81 #include <asm/mmu_context.h>
  82 #include <asm/cacheflush.h>
  83 #include <asm/tlbflush.h>
  84
  85 #include <trace/events/sched.h>
  86
  87 #define CREATE_TRACE_POINTS
  88 #include <trace/events/task.h>
  89
  90 /*
  91  * Protected counters by write_lock_irq(&tasklist_lock)
  92  */
  93 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  94 int nr_threads;                 /* The idle threads do not count.. */
  95
  96 int max_threads;                /* tunable limit on nr_threads */
  97
  98 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  99
 100 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 101
 102 #ifdef CONFIG_PROVE_RCU
 103 int lockdep_tasklist_lock_is_held(void)
 104 {
 105         return lockdep_is_held(&tasklist_lock);
 106 }
 107 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 108 #endif /* #ifdef CONFIG_PROVE_RCU */
 109
 110 int nr_processes(void)
 111 {
 112         int cpu;
 113         int total = 0;
 114
 115         for_each_possible_cpu(cpu)
 116                 total += per_cpu(process_counts, cpu);
 117
 118         return total;
 119 }
 120
 121 void __weak arch_release_task_struct(struct task_struct *tsk)
 122 {
 123 }
 124
 125 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 126 static struct kmem_cache *task_struct_cachep;
 127
 128 static inline struct task_struct *alloc_task_struct_node(int node)
 129 {
 130         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 131 }
 132
 133 static inline void free_task_struct(struct task_struct *tsk)
 134 {
 135         kmem_cache_free(task_struct_cachep, tsk);
 136 }
 137 #endif
 138
 139 void __weak arch_release_thread_info(struct thread_info *ti)
 140 {
 141 }
 142
 143 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 144
 145 /*
 146  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 147  * kmemcache based allocator.
 148  */
 149 # if THREAD_SIZE >= PAGE_SIZE
 150 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 151                                                   int node)
 152 {
 153         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 154                                                   THREAD_SIZE_ORDER);
 155
 156         return page ? page_address(page) : NULL;
 157 }
 158
 159 static inline void free_thread_info(struct thread_info *ti)
 160 {
 161         free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 162 }
 163 # else
 164 static struct kmem_cache *thread_info_cache;
 165
 166 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 167                                                   int node)
 168 {
 169         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 170 }
 171
 172 static void free_thread_info(struct thread_info *ti)
 173 {
 174         kmem_cache_free(thread_info_cache, ti);
 175 }
 176
 177 void thread_info_cache_init(void)
 178 {
 179         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 180                                               THREAD_SIZE, 0, NULL);
 181         BUG_ON(thread_info_cache == NULL);
 182 }
 183 # endif
 184 #endif
 185
 186 /* SLAB cache for signal_struct structures (tsk->signal) */
 187 static struct kmem_cache *signal_cachep;
 188
 189 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 190 struct kmem_cache *sighand_cachep;
 191
 192 /* SLAB cache for files_struct structures (tsk->files) */
 193 struct kmem_cache *files_cachep;
 194
 195 /* SLAB cache for fs_struct structures (tsk->fs) */
 196 struct kmem_cache *fs_cachep;
 197
 198 /* SLAB cache for vm_area_struct structures */
 199 struct kmem_cache *vm_area_cachep;
 200
 201 /* SLAB cache for mm_struct structures (tsk->mm) */
 202 static struct kmem_cache *mm_cachep;
 203
 204 static void account_kernel_stack(struct thread_info *ti, int account)
 205 {
 206         struct zone *zone = page_zone(virt_to_page(ti));
 207
 208         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 209 }
 210
 211 void free_task(struct task_struct *tsk)
 212 {
 213         account_kernel_stack(tsk->stack, -1);
 214         arch_release_thread_info(tsk->stack);
 215         free_thread_info(tsk->stack);
 216         rt_mutex_debug_task_free(tsk);
 217         ftrace_graph_exit_task(tsk);
 218         put_seccomp_filter(tsk);
 219         arch_release_task_struct(tsk);
 220         free_task_struct(tsk);
 221 }
 222 EXPORT_SYMBOL(free_task);
 223
 224 static inline void free_signal_struct(struct signal_struct *sig)
 225 {
 226         taskstats_tgid_free(sig);
 227         sched_autogroup_exit(sig);
 228         kmem_cache_free(signal_cachep, sig);
 229 }
 230
 231 static inline void put_signal_struct(struct signal_struct *sig)
 232 {
 233         if (atomic_dec_and_test(&sig->sigcnt))
 234                 free_signal_struct(sig);
 235 }
 236
 237 void __put_task_struct(struct task_struct *tsk)
 238 {
 239         WARN_ON(!tsk->exit_state);
 240         WARN_ON(atomic_read(&tsk->usage));
 241         WARN_ON(tsk == current);
 242
 243         task_numa_free(tsk);
 244         security_task_free(tsk);
 245         exit_creds(tsk);
 246         delayacct_tsk_free(tsk);
 247         put_signal_struct(tsk->signal);
 248
 249         if (!profile_handoff_task(tsk))
 250                 free_task(tsk);
 251 }
 252 EXPORT_SYMBOL_GPL(__put_task_struct);
 253
 254 void __init __weak arch_task_cache_init(void) { }
 255
 256 void __init fork_init(unsigned long mempages)
 257 {
 258 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 259 #ifndef ARCH_MIN_TASKALIGN
 260 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 261 #endif
 262         /* create a slab on which task_structs can be allocated */
 263         task_struct_cachep =
 264                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 265                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 266 #endif
 267
 268         /* do the arch specific task caches init */
 269         arch_task_cache_init();
 270
 271         /*
 272          * The default maximum number of threads is set to a safe
 273          * value: the thread structures can take up at most half
 274          * of memory.
 275          */
 276         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 277
 278         /*
 279          * we need to allow at least 20 threads to boot a system
 280          */
 281         if (max_threads < 20)
 282                 max_threads = 20;
 283
 284         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 285         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 286         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 287                 init_task.signal->rlim[RLIMIT_NPROC];
 288 }
 289
 290 int __weak arch_dup_task_struct(struct task_struct *dst,
 291                                                struct task_struct *src)
 292 {
 293         *dst = *src;
 294         return 0;
 295 }
 296
 297 static struct task_struct *dup_task_struct(struct task_struct *orig)
 298 {
 299         struct task_struct *tsk;
 300         struct thread_info *ti;
 301         unsigned long *stackend;
 302         int node = tsk_fork_get_node(orig);
 303         int err;
 304
 305         tsk = alloc_task_struct_node(node);
 306         if (!tsk)
 307                 return NULL;
 308
 309         ti = alloc_thread_info_node(tsk, node);
 310         if (!ti)
 311                 goto free_tsk;
 312
 313         err = arch_dup_task_struct(tsk, orig);
 314         if (err)
 315                 goto free_ti;
 316
 317         tsk->stack = ti;
 318
 319         setup_thread_stack(tsk, orig);
 320         clear_user_return_notifier(tsk);
 321         clear_tsk_need_resched(tsk);
 322         stackend = end_of_stack(tsk);
 323         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 324
 325 #ifdef CONFIG_CC_STACKPROTECTOR
 326         tsk->stack_canary = get_random_int();
 327 #endif
 328
 329         /*
 330          * One for us, one for whoever does the "release_task()" (usually
 331          * parent)
 332          */
 333         atomic_set(&tsk->usage, 2);
 334 #ifdef CONFIG_BLK_DEV_IO_TRACE
 335         tsk->btrace_seq = 0;
 336 #endif
 337         tsk->splice_pipe = NULL;
 338         tsk->task_frag.page = NULL;
 339
 340         account_kernel_stack(ti, 1);
 341
 342         return tsk;
 343
 344 free_ti:
 345         free_thread_info(ti);
 346 free_tsk:
 347         free_task_struct(tsk);
 348         return NULL;
 349 }
 350
 351 #ifdef CONFIG_MMU
 352 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 353 {
 354         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 355         struct rb_node **rb_link, *rb_parent;
 356         int retval;
 357         unsigned long charge;
 358
 359         uprobe_start_dup_mmap();
 360         down_write(&oldmm->mmap_sem);
 361         flush_cache_dup_mm(oldmm);
 362         uprobe_dup_mmap(oldmm, mm);
 363         /*
 364          * Not linked in yet - no deadlock potential:
 365          */
 366         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 367
 368         mm->locked_vm = 0;
 369         mm->mmap = NULL;
 370         mm->vmacache_seqnum = 0;
 371         mm->map_count = 0;
 372         cpumask_clear(mm_cpumask(mm));
 373         mm->mm_rb = RB_ROOT;
 374         rb_link = &mm->mm_rb.rb_node;
 375         rb_parent = NULL;
 376         pprev = &mm->mmap;
 377         retval = ksm_fork(mm, oldmm);
 378         if (retval)
 379                 goto out;
 380         retval = khugepaged_fork(mm, oldmm);
 381         if (retval)
 382                 goto out;
 383
 384         prev = NULL;
 385         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 386                 struct file *file;
 387
 388                 if (mpnt->vm_flags & VM_DONTCOPY) {
 389                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 390                                                         -vma_pages(mpnt));
 391                         continue;
 392                 }
 393                 charge = 0;
 394                 if (mpnt->vm_flags & VM_ACCOUNT) {
 395                         unsigned long len = vma_pages(mpnt);
 396
 397                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 398                                 goto fail_nomem;
 399                         charge = len;
 400                 }
 401                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 402                 if (!tmp)
 403                         goto fail_nomem;
 404                 *tmp = *mpnt;
 405                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 406                 retval = vma_dup_policy(mpnt, tmp);
 407                 if (retval)
 408                         goto fail_nomem_policy;
 409                 tmp->vm_mm = mm;
 410                 if (anon_vma_fork(tmp, mpnt))
 411                         goto fail_nomem_anon_vma_fork;
 412                 tmp->vm_flags &= ~VM_LOCKED;
 413                 tmp->vm_next = tmp->vm_prev = NULL;
 414                 file = tmp->vm_file;
 415                 if (file) {
 416                         struct inode *inode = file_inode(file);
 417                         struct address_space *mapping = file->f_mapping;
 418
 419                         get_file(file);
 420                         if (tmp->vm_flags & VM_DENYWRITE)
 421                                 atomic_dec(&inode->i_writecount);
 422                         mutex_lock(&mapping->i_mmap_mutex);
 423                         if (tmp->vm_flags & VM_SHARED)
 424                                 mapping->i_mmap_writable++;
 425                         flush_dcache_mmap_lock(mapping);
 426                         /* insert tmp into the share list, just after mpnt */
 427                         if (unlikely(tmp->vm_flags & VM_NONLINEAR))
 428                                 vma_nonlinear_insert(tmp,
 429                                                 &mapping->i_mmap_nonlinear);
 430                         else
 431                                 vma_interval_tree_insert_after(tmp, mpnt,
 432                                                         &mapping->i_mmap);
 433                         flush_dcache_mmap_unlock(mapping);
 434                         mutex_unlock(&mapping->i_mmap_mutex);
 435                 }
 436
 437                 /*
 438                  * Clear hugetlb-related page reserves for children. This only
 439                  * affects MAP_PRIVATE mappings. Faults generated by the child
 440                  * are not guaranteed to succeed, even if read-only
 441                  */
 442                 if (is_vm_hugetlb_page(tmp))
 443                         reset_vma_resv_huge_pages(tmp);
 444
 445                 /*
 446                  * Link in the new vma and copy the page table entries.
 447                  */
 448                 *pprev = tmp;
 449                 pprev = &tmp->vm_next;
 450                 tmp->vm_prev = prev;
 451                 prev = tmp;
 452
 453                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 454                 rb_link = &tmp->vm_rb.rb_right;
 455                 rb_parent = &tmp->vm_rb;
 456
 457                 mm->map_count++;
 458                 retval = copy_page_range(mm, oldmm, mpnt);
 459
 460                 if (tmp->vm_ops && tmp->vm_ops->open)
 461                         tmp->vm_ops->open(tmp);
 462
 463                 if (retval)
 464                         goto out;
 465         }
 466         /* a new mm has just been created */
 467         arch_dup_mmap(oldmm, mm);
 468         retval = 0;
 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(vma_policy(tmp));
 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         mm->ioctx_table = NULL;
 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->core_state = NULL;
 537         atomic_long_set(&mm->nr_ptes, 0);
 538         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 539         spin_lock_init(&mm->page_table_lock);
 540         mm_init_aio(mm);
 541         mm_init_owner(mm, p);
 542         clear_tlb_flush_pending(mm);
 543
 544         if (current->mm) {
 545                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 546                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 547         } else {
 548                 mm->flags = default_dump_filter;
 549                 mm->def_flags = 0;
 550         }
 551
 552         if (likely(!mm_alloc_pgd(mm))) {
 553                 mmu_notifier_mm_init(mm);
 554                 return mm;
 555         }
 556
 557         free_mm(mm);
 558         return NULL;
 559 }
 560
 561 static void check_mm(struct mm_struct *mm)
 562 {
 563         int i;
 564
 565         for (i = 0; i < NR_MM_COUNTERS; i++) {
 566                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 567
 568                 if (unlikely(x))
 569                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 570                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 571         }
 572
 573 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 574         VM_BUG_ON(mm->pmd_huge_pte);
 575 #endif
 576 }
 577
 578 /*
 579  * Allocate and initialize an mm_struct.
 580  */
 581 struct mm_struct *mm_alloc(void)
 582 {
 583         struct mm_struct *mm;
 584
 585         mm = allocate_mm();
 586         if (!mm)
 587                 return NULL;
 588
 589         memset(mm, 0, sizeof(*mm));
 590         mm_init_cpumask(mm);
 591         return mm_init(mm, current);
 592 }
 593
 594 /*
 595  * Called when the last reference to the mm
 596  * is dropped: either by a lazy thread or by
 597  * mmput. Free the page directory and the mm.
 598  */
 599 void __mmdrop(struct mm_struct *mm)
 600 {
 601         BUG_ON(mm == &init_mm);
 602         mm_free_pgd(mm);
 603         destroy_context(mm);
 604         mmu_notifier_mm_destroy(mm);
 605         check_mm(mm);
 606         free_mm(mm);
 607 }
 608 EXPORT_SYMBOL_GPL(__mmdrop);
 609
 610 /*
 611  * Decrement the use count and release all resources for an mm.
 612  */
 613 void mmput(struct mm_struct *mm)
 614 {
 615         might_sleep();
 616
 617         if (atomic_dec_and_test(&mm->mm_users)) {
 618                 uprobe_clear_state(mm);
 619                 exit_aio(mm);
 620                 ksm_exit(mm);
 621                 khugepaged_exit(mm); /* must run before exit_mmap */
 622                 exit_mmap(mm);
 623                 set_mm_exe_file(mm, NULL);
 624                 if (!list_empty(&mm->mmlist)) {
 625                         spin_lock(&mmlist_lock);
 626                         list_del(&mm->mmlist);
 627                         spin_unlock(&mmlist_lock);
 628                 }
 629                 if (mm->binfmt)
 630                         module_put(mm->binfmt->module);
 631                 mmdrop(mm);
 632         }
 633 }
 634 EXPORT_SYMBOL_GPL(mmput);
 635
 636 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 637 {
 638         if (new_exe_file)
 639                 get_file(new_exe_file);
 640         if (mm->exe_file)
 641                 fput(mm->exe_file);
 642         mm->exe_file = new_exe_file;
 643 }
 644
 645 struct file *get_mm_exe_file(struct mm_struct *mm)
 646 {
 647         struct file *exe_file;
 648
 649         /* We need mmap_sem to protect against races with removal of exe_file */
 650         down_read(&mm->mmap_sem);
 651         exe_file = mm->exe_file;
 652         if (exe_file)
 653                 get_file(exe_file);
 654         up_read(&mm->mmap_sem);
 655         return exe_file;
 656 }
 657
 658 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 659 {
 660         /* It's safe to write the exe_file pointer without exe_file_lock because
 661          * this is called during fork when the task is not yet in /proc */
 662         newmm->exe_file = get_mm_exe_file(oldmm);
 663 }
 664
 665 /**
 666  * get_task_mm - acquire a reference to the task's mm
 667  *
 668  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 669  * this kernel workthread has transiently adopted a user mm with use_mm,
 670  * to do its AIO) is not set and if so returns a reference to it, after
 671  * bumping up the use count.  User must release the mm via mmput()
 672  * after use.  Typically used by /proc and ptrace.
 673  */
 674 struct mm_struct *get_task_mm(struct task_struct *task)
 675 {
 676         struct mm_struct *mm;
 677
 678         task_lock(task);
 679         mm = task->mm;
 680         if (mm) {
 681                 if (task->flags & PF_KTHREAD)
 682                         mm = NULL;
 683                 else
 684                         atomic_inc(&mm->mm_users);
 685         }
 686         task_unlock(task);
 687         return mm;
 688 }
 689 EXPORT_SYMBOL_GPL(get_task_mm);
 690
 691 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 692 {
 693         struct mm_struct *mm;
 694         int err;
 695
 696         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 697         if (err)
 698                 return ERR_PTR(err);
 699
 700         mm = get_task_mm(task);
 701         if (mm && mm != current->mm &&
 702                         !ptrace_may_access(task, mode)) {
 703                 mmput(mm);
 704                 mm = ERR_PTR(-EACCES);
 705         }
 706         mutex_unlock(&task->signal->cred_guard_mutex);
 707
 708         return mm;
 709 }
 710
 711 static void complete_vfork_done(struct task_struct *tsk)
 712 {
 713         struct completion *vfork;
 714
 715         task_lock(tsk);
 716         vfork = tsk->vfork_done;
 717         if (likely(vfork)) {
 718                 tsk->vfork_done = NULL;
 719                 complete(vfork);
 720         }
 721         task_unlock(tsk);
 722 }
 723
 724 static int wait_for_vfork_done(struct task_struct *child,
 725                                 struct completion *vfork)
 726 {
 727         int killed;
 728
 729         freezer_do_not_count();
 730         killed = wait_for_completion_killable(vfork);
 731         freezer_count();
 732
 733         if (killed) {
 734                 task_lock(child);
 735                 child->vfork_done = NULL;
 736                 task_unlock(child);
 737         }
 738
 739         put_task_struct(child);
 740         return killed;
 741 }
 742
 743 /* Please note the differences between mmput and mm_release.
 744  * mmput is called whenever we stop holding onto a mm_struct,
 745  * error success whatever.
 746  *
 747  * mm_release is called after a mm_struct has been removed
 748  * from the current process.
 749  *
 750  * This difference is important for error handling, when we
 751  * only half set up a mm_struct for a new process and need to restore
 752  * the old one.  Because we mmput the new mm_struct before
 753  * restoring the old one. . .
 754  * Eric Biederman 10 January 1998
 755  */
 756 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 757 {
 758         /* Get rid of any futexes when releasing the mm */
 759 #ifdef CONFIG_FUTEX
 760         if (unlikely(tsk->robust_list)) {
 761                 exit_robust_list(tsk);
 762                 tsk->robust_list = NULL;
 763         }
 764 #ifdef CONFIG_COMPAT
 765         if (unlikely(tsk->compat_robust_list)) {
 766                 compat_exit_robust_list(tsk);
 767                 tsk->compat_robust_list = NULL;
 768         }
 769 #endif
 770         if (unlikely(!list_empty(&tsk->pi_state_list)))
 771                 exit_pi_state_list(tsk);
 772 #endif
 773
 774         uprobe_free_utask(tsk);
 775
 776         /* Get rid of any cached register state */
 777         deactivate_mm(tsk, mm);
 778
 779         /*
 780          * If we're exiting normally, clear a user-space tid field if
 781          * requested.  We leave this alone when dying by signal, to leave
 782          * the value intact in a core dump, and to save the unnecessary
 783          * trouble, say, a killed vfork parent shouldn't touch this mm.
 784          * Userland only wants this done for a sys_exit.
 785          */
 786         if (tsk->clear_child_tid) {
 787                 if (!(tsk->flags & PF_SIGNALED) &&
 788                     atomic_read(&mm->mm_users) > 1) {
 789                         /*
 790                          * We don't check the error code - if userspace has
 791                          * not set up a proper pointer then tough luck.
 792                          */
 793                         put_user(0, tsk->clear_child_tid);
 794                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 795                                         1, NULL, NULL, 0);
 796                 }
 797                 tsk->clear_child_tid = NULL;
 798         }
 799
 800         /*
 801          * All done, finally we can wake up parent and return this mm to him.
 802          * Also kthread_stop() uses this completion for synchronization.
 803          */
 804         if (tsk->vfork_done)
 805                 complete_vfork_done(tsk);
 806 }
 807
 808 /*
 809  * Allocate a new mm structure and copy contents from the
 810  * mm structure of the passed in task structure.
 811  */
 812 static struct mm_struct *dup_mm(struct task_struct *tsk)
 813 {
 814         struct mm_struct *mm, *oldmm = current->mm;
 815         int err;
 816
 817         mm = allocate_mm();
 818         if (!mm)
 819                 goto fail_nomem;
 820
 821         memcpy(mm, oldmm, sizeof(*mm));
 822         mm_init_cpumask(mm);
 823
 824 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 825         mm->pmd_huge_pte = NULL;
 826 #endif
 827         if (!mm_init(mm, tsk))
 828                 goto fail_nomem;
 829
 830         if (init_new_context(tsk, mm))
 831                 goto fail_nocontext;
 832
 833         dup_mm_exe_file(oldmm, mm);
 834
 835         err = dup_mmap(mm, oldmm);
 836         if (err)
 837                 goto free_pt;
 838
 839         mm->hiwater_rss = get_mm_rss(mm);
 840         mm->hiwater_vm = mm->total_vm;
 841
 842         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 843                 goto free_pt;
 844
 845         return mm;
 846
 847 free_pt:
 848         /* don't put binfmt in mmput, we haven't got module yet */
 849         mm->binfmt = NULL;
 850         mmput(mm);
 851
 852 fail_nomem:
 853         return NULL;
 854
 855 fail_nocontext:
 856         /*
 857          * If init_new_context() failed, we cannot use mmput() to free the mm
 858          * because it calls destroy_context()
 859          */
 860         mm_free_pgd(mm);
 861         free_mm(mm);
 862         return NULL;
 863 }
 864
 865 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 866 {
 867         struct mm_struct *mm, *oldmm;
 868         int retval;
 869
 870         tsk->min_flt = tsk->maj_flt = 0;
 871         tsk->nvcsw = tsk->nivcsw = 0;
 872 #ifdef CONFIG_DETECT_HUNG_TASK
 873         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 874 #endif
 875
 876         tsk->mm = NULL;
 877         tsk->active_mm = NULL;
 878
 879         /*
 880          * Are we cloning a kernel thread?
 881          *
 882          * We need to steal a active VM for that..
 883          */
 884         oldmm = current->mm;
 885         if (!oldmm)
 886                 return 0;
 887
 888         /* initialize the new vmacache entries */
 889         vmacache_flush(tsk);
 890
 891         if (clone_flags & CLONE_VM) {
 892                 atomic_inc(&oldmm->mm_users);
 893                 mm = oldmm;
 894                 goto good_mm;
 895         }
 896
 897         retval = -ENOMEM;
 898         mm = dup_mm(tsk);
 899         if (!mm)
 900                 goto fail_nomem;
 901
 902 good_mm:
 903         tsk->mm = mm;
 904         tsk->active_mm = mm;
 905         return 0;
 906
 907 fail_nomem:
 908         return retval;
 909 }
 910
 911 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 912 {
 913         struct fs_struct *fs = current->fs;
 914         if (clone_flags & CLONE_FS) {
 915                 /* tsk->fs is already what we want */
 916                 spin_lock(&fs->lock);
 917                 if (fs->in_exec) {
 918                         spin_unlock(&fs->lock);
 919                         return -EAGAIN;
 920                 }
 921                 fs->users++;
 922                 spin_unlock(&fs->lock);
 923                 return 0;
 924         }
 925         tsk->fs = copy_fs_struct(fs);
 926         if (!tsk->fs)
 927                 return -ENOMEM;
 928         return 0;
 929 }
 930
 931 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 932 {
 933         struct files_struct *oldf, *newf;
 934         int error = 0;
 935
 936         /*
 937          * A background process may not have any files ...
 938          */
 939         oldf = current->files;
 940         if (!oldf)
 941                 goto out;
 942
 943         if (clone_flags & CLONE_FILES) {
 944                 atomic_inc(&oldf->count);
 945                 goto out;
 946         }
 947
 948         newf = dup_fd(oldf, &error);
 949         if (!newf)
 950                 goto out;
 951
 952         tsk->files = newf;
 953         error = 0;
 954 out:
 955         return error;
 956 }
 957
 958 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 959 {
 960 #ifdef CONFIG_BLOCK
 961         struct io_context *ioc = current->io_context;
 962         struct io_context *new_ioc;
 963
 964         if (!ioc)
 965                 return 0;
 966         /*
 967          * Share io context with parent, if CLONE_IO is set
 968          */
 969         if (clone_flags & CLONE_IO) {
 970                 ioc_task_link(ioc);
 971                 tsk->io_context = ioc;
 972         } else if (ioprio_valid(ioc->ioprio)) {
 973                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 974                 if (unlikely(!new_ioc))
 975                         return -ENOMEM;
 976
 977                 new_ioc->ioprio = ioc->ioprio;
 978                 put_io_context(new_ioc);
 979         }
 980 #endif
 981         return 0;
 982 }
 983
 984 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 985 {
 986         struct sighand_struct *sig;
 987
 988         if (clone_flags & CLONE_SIGHAND) {
 989                 atomic_inc(&current->sighand->count);
 990                 return 0;
 991         }
 992         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 993         rcu_assign_pointer(tsk->sighand, sig);
 994         if (!sig)
 995                 return -ENOMEM;
 996         atomic_set(&sig->count, 1);
 997         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 998         return 0;
 999 }
1000
1001 void __cleanup_sighand(struct sighand_struct *sighand)
1002 {
1003         if (atomic_dec_and_test(&sighand->count)) {
1004                 signalfd_cleanup(sighand);
1005                 kmem_cache_free(sighand_cachep, sighand);
1006         }
1007 }
1008
1009
1010 /*
1011  * Initialize POSIX timer handling for a thread group.
1012  */
1013 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1014 {
1015         unsigned long cpu_limit;
1016
1017         /* Thread group counters. */
1018         thread_group_cputime_init(sig);
1019
1020         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1021         if (cpu_limit != RLIM_INFINITY) {
1022                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1023                 sig->cputimer.running = 1;
1024         }
1025
1026         /* The timer lists. */
1027         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1028         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1029         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1030 }
1031
1032 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1033 {
1034         struct signal_struct *sig;
1035
1036         if (clone_flags & CLONE_THREAD)
1037                 return 0;
1038
1039         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1040         tsk->signal = sig;
1041         if (!sig)
1042                 return -ENOMEM;
1043
1044         sig->nr_threads = 1;
1045         atomic_set(&sig->live, 1);
1046         atomic_set(&sig->sigcnt, 1);
1047
1048         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1049         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1050         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1051
1052         init_waitqueue_head(&sig->wait_chldexit);
1053         sig->curr_target = tsk;
1054         init_sigpending(&sig->shared_pending);
1055         INIT_LIST_HEAD(&sig->posix_timers);
1056
1057         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1058         sig->real_timer.function = it_real_fn;
1059
1060         task_lock(current->group_leader);
1061         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1062         task_unlock(current->group_leader);
1063
1064         posix_cpu_timers_init_group(sig);
1065
1066         tty_audit_fork(sig);
1067         sched_autogroup_fork(sig);
1068
1069 #ifdef CONFIG_CGROUPS
1070         init_rwsem(&sig->group_rwsem);
1071 #endif
1072
1073         sig->oom_score_adj = current->signal->oom_score_adj;
1074         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1075
1076         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1077                                    current->signal->is_child_subreaper;
1078
1079         mutex_init(&sig->cred_guard_mutex);
1080
1081         return 0;
1082 }
1083
1084 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1085 {
1086         current->clear_child_tid = tidptr;
1087
1088         return task_pid_vnr(current);
1089 }
1090
1091 static void rt_mutex_init_task(struct task_struct *p)
1092 {
1093         raw_spin_lock_init(&p->pi_lock);
1094 #ifdef CONFIG_RT_MUTEXES
1095         p->pi_waiters = RB_ROOT;
1096         p->pi_waiters_leftmost = NULL;
1097         p->pi_blocked_on = NULL;
1098 #endif
1099 }
1100
1101 #ifdef CONFIG_MEMCG
1102 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1103 {
1104         mm->owner = p;
1105 }
1106 #endif /* CONFIG_MEMCG */
1107
1108 /*
1109  * Initialize POSIX timer handling for a single task.
1110  */
1111 static void posix_cpu_timers_init(struct task_struct *tsk)
1112 {
1113         tsk->cputime_expires.prof_exp = 0;
1114         tsk->cputime_expires.virt_exp = 0;
1115         tsk->cputime_expires.sched_exp = 0;
1116         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1117         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1118         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1119 }
1120
1121 static inline void
1122 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1123 {
1124          task->pids[type].pid = pid;
1125 }
1126
1127 /*
1128  * This creates a new process as a copy of the old one,
1129  * but does not actually start it yet.
1130  *
1131  * It copies the registers, and all the appropriate
1132  * parts of the process environment (as per the clone
1133  * flags). The actual kick-off is left to the caller.
1134  */
1135 static struct task_struct *copy_process(unsigned long clone_flags,
1136                                         unsigned long stack_start,
1137                                         unsigned long stack_size,
1138                                         int __user *child_tidptr,
1139                                         struct pid *pid,
1140                                         int trace)
1141 {
1142         int retval;
1143         struct task_struct *p;
1144
1145         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1146                 return ERR_PTR(-EINVAL);
1147
1148         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1149                 return ERR_PTR(-EINVAL);
1150
1151         /*
1152          * Thread groups must share signals as well, and detached threads
1153          * can only be started up within the thread group.
1154          */
1155         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1156                 return ERR_PTR(-EINVAL);
1157
1158         /*
1159          * Shared signal handlers imply shared VM. By way of the above,
1160          * thread groups also imply shared VM. Blocking this case allows
1161          * for various simplifications in other code.
1162          */
1163         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1164                 return ERR_PTR(-EINVAL);
1165
1166         /*
1167          * Siblings of global init remain as zombies on exit since they are
1168          * not reaped by their parent (swapper). To solve this and to avoid
1169          * multi-rooted process trees, prevent global and container-inits
1170          * from creating siblings.
1171          */
1172         if ((clone_flags & CLONE_PARENT) &&
1173                                 current->signal->flags & SIGNAL_UNKILLABLE)
1174                 return ERR_PTR(-EINVAL);
1175
1176         /*
1177          * If the new process will be in a different pid or user namespace
1178          * do not allow it to share a thread group or signal handlers or
1179          * parent with the forking task.
1180          */
1181         if (clone_flags & CLONE_SIGHAND) {
1182                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1183                     (task_active_pid_ns(current) !=
1184                                 current->nsproxy->pid_ns_for_children))
1185                         return ERR_PTR(-EINVAL);
1186         }
1187
1188         retval = security_task_create(clone_flags);
1189         if (retval)
1190                 goto fork_out;
1191
1192         retval = -ENOMEM;
1193         p = dup_task_struct(current);
1194         if (!p)
1195                 goto fork_out;
1196
1197         ftrace_graph_init_task(p);
1198         get_seccomp_filter(p);
1199
1200         rt_mutex_init_task(p);
1201
1202 #ifdef CONFIG_PROVE_LOCKING
1203         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1204         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1205 #endif
1206         retval = -EAGAIN;
1207         if (atomic_read(&p->real_cred->user->processes) >=
1208                         task_rlimit(p, RLIMIT_NPROC)) {
1209                 if (p->real_cred->user != INIT_USER &&
1210                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1211                         goto bad_fork_free;
1212         }
1213         current->flags &= ~PF_NPROC_EXCEEDED;
1214
1215         retval = copy_creds(p, clone_flags);
1216         if (retval < 0)
1217                 goto bad_fork_free;
1218
1219         /*
1220          * If multiple threads are within copy_process(), then this check
1221          * triggers too late. This doesn't hurt, the check is only there
1222          * to stop root fork bombs.
1223          */
1224         retval = -EAGAIN;
1225         if (nr_threads >= max_threads)
1226                 goto bad_fork_cleanup_count;
1227
1228         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1229                 goto bad_fork_cleanup_count;
1230
1231         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1232         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1233         p->flags |= PF_FORKNOEXEC;
1234         INIT_LIST_HEAD(&p->children);
1235         INIT_LIST_HEAD(&p->sibling);
1236         rcu_copy_process(p);
1237         p->vfork_done = NULL;
1238         spin_lock_init(&p->alloc_lock);
1239
1240         init_sigpending(&p->pending);
1241
1242         p->utime = p->stime = p->gtime = 0;
1243         p->utimescaled = p->stimescaled = 0;
1244 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1245         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1246 #endif
1247 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1248         seqlock_init(&p->vtime_seqlock);
1249         p->vtime_snap = 0;
1250         p->vtime_snap_whence = VTIME_SLEEPING;
1251 #endif
1252
1253 #if defined(SPLIT_RSS_COUNTING)
1254         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1255 #endif
1256
1257         p->default_timer_slack_ns = current->timer_slack_ns;
1258
1259         task_io_accounting_init(&p->ioac);
1260         acct_clear_integrals(p);
1261
1262         posix_cpu_timers_init(p);
1263
1264         do_posix_clock_monotonic_gettime(&p->start_time);
1265         p->real_start_time = p->start_time;
1266         monotonic_to_bootbased(&p->real_start_time);
1267         p->io_context = NULL;
1268         p->audit_context = NULL;
1269         if (clone_flags & CLONE_THREAD)
1270                 threadgroup_change_begin(current);
1271         cgroup_fork(p);
1272 #ifdef CONFIG_NUMA
1273         p->mempolicy = mpol_dup(p->mempolicy);
1274         if (IS_ERR(p->mempolicy)) {
1275                 retval = PTR_ERR(p->mempolicy);
1276                 p->mempolicy = NULL;
1277                 goto bad_fork_cleanup_threadgroup_lock;
1278         }
1279 #endif
1280 #ifdef CONFIG_CPUSETS
1281         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1282         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1283         seqcount_init(&p->mems_allowed_seq);
1284 #endif
1285 #ifdef CONFIG_TRACE_IRQFLAGS
1286         p->irq_events = 0;
1287         p->hardirqs_enabled = 0;
1288         p->hardirq_enable_ip = 0;
1289         p->hardirq_enable_event = 0;
1290         p->hardirq_disable_ip = _THIS_IP_;
1291         p->hardirq_disable_event = 0;
1292         p->softirqs_enabled = 1;
1293         p->softirq_enable_ip = _THIS_IP_;
1294         p->softirq_enable_event = 0;
1295         p->softirq_disable_ip = 0;
1296         p->softirq_disable_event = 0;
1297         p->hardirq_context = 0;
1298         p->softirq_context = 0;
1299 #endif
1300 #ifdef CONFIG_LOCKDEP
1301         p->lockdep_depth = 0; /* no locks held yet */
1302         p->curr_chain_key = 0;
1303         p->lockdep_recursion = 0;
1304 #endif
1305
1306 #ifdef CONFIG_DEBUG_MUTEXES
1307         p->blocked_on = NULL; /* not blocked yet */
1308 #endif
1309 #ifdef CONFIG_MEMCG
1310         p->memcg_batch.do_batch = 0;
1311         p->memcg_batch.memcg = NULL;
1312 #endif
1313 #ifdef CONFIG_BCACHE
1314         p->sequential_io        = 0;
1315         p->sequential_io_avg    = 0;
1316 #endif
1317
1318         /* Perform scheduler related setup. Assign this task to a CPU. */
1319         retval = sched_fork(clone_flags, p);
1320         if (retval)
1321                 goto bad_fork_cleanup_policy;
1322
1323         retval = perf_event_init_task(p);
1324         if (retval)
1325                 goto bad_fork_cleanup_policy;
1326         retval = audit_alloc(p);
1327         if (retval)
1328                 goto bad_fork_cleanup_policy;
1329         /* copy all the process information */
1330         retval = copy_semundo(clone_flags, p);
1331         if (retval)
1332                 goto bad_fork_cleanup_audit;
1333         retval = copy_files(clone_flags, p);
1334         if (retval)
1335                 goto bad_fork_cleanup_semundo;
1336         retval = copy_fs(clone_flags, p);
1337         if (retval)
1338                 goto bad_fork_cleanup_files;
1339         retval = copy_sighand(clone_flags, p);
1340         if (retval)
1341                 goto bad_fork_cleanup_fs;
1342         retval = copy_signal(clone_flags, p);
1343         if (retval)
1344                 goto bad_fork_cleanup_sighand;
1345         retval = copy_mm(clone_flags, p);
1346         if (retval)
1347                 goto bad_fork_cleanup_signal;
1348         retval = copy_namespaces(clone_flags, p);
1349         if (retval)
1350                 goto bad_fork_cleanup_mm;
1351         retval = copy_io(clone_flags, p);
1352         if (retval)
1353                 goto bad_fork_cleanup_namespaces;
1354         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1355         if (retval)
1356                 goto bad_fork_cleanup_io;
1357
1358         if (pid != &init_struct_pid) {
1359                 retval = -ENOMEM;
1360                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1361                 if (!pid)
1362                         goto bad_fork_cleanup_io;
1363         }
1364
1365         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1366         /*
1367          * Clear TID on mm_release()?
1368          */
1369         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1370 #ifdef CONFIG_BLOCK
1371         p->plug = NULL;
1372 #endif
1373 #ifdef CONFIG_FUTEX
1374         p->robust_list = NULL;
1375 #ifdef CONFIG_COMPAT
1376         p->compat_robust_list = NULL;
1377 #endif
1378         INIT_LIST_HEAD(&p->pi_state_list);
1379         p->pi_state_cache = NULL;
1380 #endif
1381         /*
1382          * sigaltstack should be cleared when sharing the same VM
1383          */
1384         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1385                 p->sas_ss_sp = p->sas_ss_size = 0;
1386
1387         /*
1388          * Syscall tracing and stepping should be turned off in the
1389          * child regardless of CLONE_PTRACE.
1390          */
1391         user_disable_single_step(p);
1392         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1393 #ifdef TIF_SYSCALL_EMU
1394         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1395 #endif
1396         clear_all_latency_tracing(p);
1397
1398         /* ok, now we should be set up.. */
1399         p->pid = pid_nr(pid);
1400         if (clone_flags & CLONE_THREAD) {
1401                 p->exit_signal = -1;
1402                 p->group_leader = current->group_leader;
1403                 p->tgid = current->tgid;
1404         } else {
1405                 if (clone_flags & CLONE_PARENT)
1406                         p->exit_signal = current->group_leader->exit_signal;
1407                 else
1408                         p->exit_signal = (clone_flags & CSIGNAL);
1409                 p->group_leader = p;
1410                 p->tgid = p->pid;
1411         }
1412
1413         p->nr_dirtied = 0;
1414         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1415         p->dirty_paused_when = 0;
1416
1417         p->pdeath_signal = 0;
1418         INIT_LIST_HEAD(&p->thread_group);
1419         p->task_works = NULL;
1420
1421         /*
1422          * Make it visible to the rest of the system, but dont wake it up yet.
1423          * Need tasklist lock for parent etc handling!
1424          */
1425         write_lock_irq(&tasklist_lock);
1426
1427         /* CLONE_PARENT re-uses the old parent */
1428         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1429                 p->real_parent = current->real_parent;
1430                 p->parent_exec_id = current->parent_exec_id;
1431         } else {
1432                 p->real_parent = current;
1433                 p->parent_exec_id = current->self_exec_id;
1434         }
1435
1436         spin_lock(&current->sighand->siglock);
1437
1438         /*
1439          * Process group and session signals need to be delivered to just the
1440          * parent before the fork or both the parent and the child after the
1441          * fork. Restart if a signal comes in before we add the new process to
1442          * it's process group.
1443          * A fatal signal pending means that current will exit, so the new
1444          * thread can't slip out of an OOM kill (or normal SIGKILL).
1445         */
1446         recalc_sigpending();
1447         if (signal_pending(current)) {
1448                 spin_unlock(&current->sighand->siglock);
1449                 write_unlock_irq(&tasklist_lock);
1450                 retval = -ERESTARTNOINTR;
1451                 goto bad_fork_free_pid;
1452         }
1453
1454         if (likely(p->pid)) {
1455                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1456
1457                 init_task_pid(p, PIDTYPE_PID, pid);
1458                 if (thread_group_leader(p)) {
1459                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1460                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1461
1462                         if (is_child_reaper(pid)) {
1463                                 ns_of_pid(pid)->child_reaper = p;
1464                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1465                         }
1466
1467                         p->signal->leader_pid = pid;
1468                         p->signal->tty = tty_kref_get(current->signal->tty);
1469                         list_add_tail(&p->sibling, &p->real_parent->children);
1470                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1471                         attach_pid(p, PIDTYPE_PGID);
1472                         attach_pid(p, PIDTYPE_SID);
1473                         __this_cpu_inc(process_counts);
1474                 } else {
1475                         current->signal->nr_threads++;
1476                         atomic_inc(&current->signal->live);
1477                         atomic_inc(&current->signal->sigcnt);
1478                         list_add_tail_rcu(&p->thread_group,
1479                                           &p->group_leader->thread_group);
1480                         list_add_tail_rcu(&p->thread_node,
1481                                           &p->signal->thread_head);
1482                 }
1483                 attach_pid(p, PIDTYPE_PID);
1484                 nr_threads++;
1485         }
1486
1487         total_forks++;
1488         spin_unlock(&current->sighand->siglock);
1489         syscall_tracepoint_update(p);
1490         write_unlock_irq(&tasklist_lock);
1491
1492         proc_fork_connector(p);
1493         cgroup_post_fork(p);
1494         if (clone_flags & CLONE_THREAD)
1495                 threadgroup_change_end(current);
1496         perf_event_fork(p);
1497
1498         trace_task_newtask(p, clone_flags);
1499         uprobe_copy_process(p, clone_flags);
1500
1501         return p;
1502
1503 bad_fork_free_pid:
1504         if (pid != &init_struct_pid)
1505                 free_pid(pid);
1506 bad_fork_cleanup_io:
1507         if (p->io_context)
1508                 exit_io_context(p);
1509 bad_fork_cleanup_namespaces:
1510         exit_task_namespaces(p);
1511 bad_fork_cleanup_mm:
1512         if (p->mm)
1513                 mmput(p->mm);
1514 bad_fork_cleanup_signal:
1515         if (!(clone_flags & CLONE_THREAD))
1516                 free_signal_struct(p->signal);
1517 bad_fork_cleanup_sighand:
1518         __cleanup_sighand(p->sighand);
1519 bad_fork_cleanup_fs:
1520         exit_fs(p); /* blocking */
1521 bad_fork_cleanup_files:
1522         exit_files(p); /* blocking */
1523 bad_fork_cleanup_semundo:
1524         exit_sem(p);
1525 bad_fork_cleanup_audit:
1526         audit_free(p);
1527 bad_fork_cleanup_policy:
1528         perf_event_free_task(p);
1529 #ifdef CONFIG_NUMA
1530         mpol_put(p->mempolicy);
1531 bad_fork_cleanup_threadgroup_lock:
1532 #endif
1533         if (clone_flags & CLONE_THREAD)
1534                 threadgroup_change_end(current);
1535         delayacct_tsk_free(p);
1536         module_put(task_thread_info(p)->exec_domain->module);
1537 bad_fork_cleanup_count:
1538         atomic_dec(&p->cred->user->processes);
1539         exit_creds(p);
1540 bad_fork_free:
1541         free_task(p);
1542 fork_out:
1543         return ERR_PTR(retval);
1544 }
1545
1546 static inline void init_idle_pids(struct pid_link *links)
1547 {
1548         enum pid_type type;
1549
1550         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1551                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1552                 links[type].pid = &init_struct_pid;
1553         }
1554 }
1555
1556 struct task_struct *fork_idle(int cpu)
1557 {
1558         struct task_struct *task;
1559         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1560         if (!IS_ERR(task)) {
1561                 init_idle_pids(task->pids);
1562                 init_idle(task, cpu);
1563         }
1564
1565         return task;
1566 }
1567
1568 /*
1569  *  Ok, this is the main fork-routine.
1570  *
1571  * It copies the process, and if successful kick-starts
1572  * it and waits for it to finish using the VM if required.
1573  */
1574 long do_fork(unsigned long clone_flags,
1575               unsigned long stack_start,
1576               unsigned long stack_size,
1577               int __user *parent_tidptr,
1578               int __user *child_tidptr)
1579 {
1580         struct task_struct *p;
1581         int trace = 0;
1582         long nr;
1583
1584         /*
1585          * Determine whether and which event to report to ptracer.  When
1586          * called from kernel_thread or CLONE_UNTRACED is explicitly
1587          * requested, no event is reported; otherwise, report if the event
1588          * for the type of forking is enabled.
1589          */
1590         if (!(clone_flags & CLONE_UNTRACED)) {
1591                 if (clone_flags & CLONE_VFORK)
1592                         trace = PTRACE_EVENT_VFORK;
1593                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1594                         trace = PTRACE_EVENT_CLONE;
1595                 else
1596                         trace = PTRACE_EVENT_FORK;
1597
1598                 if (likely(!ptrace_event_enabled(current, trace)))
1599                         trace = 0;
1600         }
1601
1602         p = copy_process(clone_flags, stack_start, stack_size,
1603                          child_tidptr, NULL, trace);
1604         /*
1605          * Do this prior waking up the new thread - the thread pointer
1606          * might get invalid after that point, if the thread exits quickly.
1607          */
1608         if (!IS_ERR(p)) {
1609                 struct completion vfork;
1610                 struct pid *pid;
1611
1612                 trace_sched_process_fork(current, p);
1613
1614                 pid = get_task_pid(p, PIDTYPE_PID);
1615                 nr = pid_vnr(pid);
1616
1617                 if (clone_flags & CLONE_PARENT_SETTID)
1618                         put_user(nr, parent_tidptr);
1619
1620                 if (clone_flags & CLONE_VFORK) {
1621                         p->vfork_done = &vfork;
1622                         init_completion(&vfork);
1623                         get_task_struct(p);
1624                 }
1625
1626                 wake_up_new_task(p);
1627
1628                 /* forking complete and child started to run, tell ptracer */
1629                 if (unlikely(trace))
1630                         ptrace_event_pid(trace, pid);
1631
1632                 if (clone_flags & CLONE_VFORK) {
1633                         if (!wait_for_vfork_done(p, &vfork))
1634                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1635                 }
1636
1637                 put_pid(pid);
1638         } else {
1639                 nr = PTR_ERR(p);
1640         }
1641         return nr;
1642 }
1643
1644 /*
1645  * Create a kernel thread.
1646  */
1647 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1648 {
1649         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1650                 (unsigned long)arg, NULL, NULL);
1651 }
1652
1653 #ifdef __ARCH_WANT_SYS_FORK
1654 SYSCALL_DEFINE0(fork)
1655 {
1656 #ifdef CONFIG_MMU
1657         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1658 #else
1659         /* can not support in nommu mode */
1660         return -EINVAL;
1661 #endif
1662 }
1663 #endif
1664
1665 #ifdef __ARCH_WANT_SYS_VFORK
1666 SYSCALL_DEFINE0(vfork)
1667 {
1668         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1669                         0, NULL, NULL);
1670 }
1671 #endif
1672
1673 #ifdef __ARCH_WANT_SYS_CLONE
1674 #ifdef CONFIG_CLONE_BACKWARDS
1675 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1676                  int __user *, parent_tidptr,
1677                  int, tls_val,
1678                  int __user *, child_tidptr)
1679 #elif defined(CONFIG_CLONE_BACKWARDS2)
1680 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1681                  int __user *, parent_tidptr,
1682                  int __user *, child_tidptr,
1683                  int, tls_val)
1684 #elif defined(CONFIG_CLONE_BACKWARDS3)
1685 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1686                 int, stack_size,
1687                 int __user *, parent_tidptr,
1688                 int __user *, child_tidptr,
1689                 int, tls_val)
1690 #else
1691 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1692                  int __user *, parent_tidptr,
1693                  int __user *, child_tidptr,
1694                  int, tls_val)
1695 #endif
1696 {
1697         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1698 }
1699 #endif
1700
1701 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1702 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1703 #endif
1704
1705 static void sighand_ctor(void *data)
1706 {
1707         struct sighand_struct *sighand = data;
1708
1709         spin_lock_init(&sighand->siglock);
1710         init_waitqueue_head(&sighand->signalfd_wqh);
1711 }
1712
1713 void __init proc_caches_init(void)
1714 {
1715         sighand_cachep = kmem_cache_create("sighand_cache",
1716                         sizeof(struct sighand_struct), 0,
1717                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1718                         SLAB_NOTRACK, sighand_ctor);
1719         signal_cachep = kmem_cache_create("signal_cache",
1720                         sizeof(struct signal_struct), 0,
1721                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1722         files_cachep = kmem_cache_create("files_cache",
1723                         sizeof(struct files_struct), 0,
1724                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1725         fs_cachep = kmem_cache_create("fs_cache",
1726                         sizeof(struct fs_struct), 0,
1727                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1728         /*
1729          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1730          * whole struct cpumask for the OFFSTACK case. We could change
1731          * this to *only* allocate as much of it as required by the
1732          * maximum number of CPU's we can ever have.  The cpumask_allocation
1733          * is at the end of the structure, exactly for that reason.
1734          */
1735         mm_cachep = kmem_cache_create("mm_struct",
1736                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1737                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1738         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1739         mmap_init();
1740         nsproxy_cache_init();
1741 }
1742
1743 /*
1744  * Check constraints on flags passed to the unshare system call.
1745  */
1746 static int check_unshare_flags(unsigned long unshare_flags)
1747 {
1748         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1749                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1750                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1751                                 CLONE_NEWUSER|CLONE_NEWPID))
1752                 return -EINVAL;
1753         /*
1754          * Not implemented, but pretend it works if there is nothing to
1755          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1756          * needs to unshare vm.
1757          */
1758         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1759                 /* FIXME: get_task_mm() increments ->mm_users */
1760                 if (atomic_read(&current->mm->mm_users) > 1)
1761                         return -EINVAL;
1762         }
1763
1764         return 0;
1765 }
1766
1767 /*
1768  * Unshare the filesystem structure if it is being shared
1769  */
1770 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1771 {
1772         struct fs_struct *fs = current->fs;
1773
1774         if (!(unshare_flags & CLONE_FS) || !fs)
1775                 return 0;
1776
1777         /* don't need lock here; in the worst case we'll do useless copy */
1778         if (fs->users == 1)
1779                 return 0;
1780
1781         *new_fsp = copy_fs_struct(fs);
1782         if (!*new_fsp)
1783                 return -ENOMEM;
1784
1785         return 0;
1786 }
1787
1788 /*
1789  * Unshare file descriptor table if it is being shared
1790  */
1791 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1792 {
1793         struct files_struct *fd = current->files;
1794         int error = 0;
1795
1796         if ((unshare_flags & CLONE_FILES) &&
1797             (fd && atomic_read(&fd->count) > 1)) {
1798                 *new_fdp = dup_fd(fd, &error);
1799                 if (!*new_fdp)
1800                         return error;
1801         }
1802
1803         return 0;
1804 }
1805
1806 /*
1807  * unshare allows a process to 'unshare' part of the process
1808  * context which was originally shared using clone.  copy_*
1809  * functions used by do_fork() cannot be used here directly
1810  * because they modify an inactive task_struct that is being
1811  * constructed. Here we are modifying the current, active,
1812  * task_struct.
1813  */
1814 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1815 {
1816         struct fs_struct *fs, *new_fs = NULL;
1817         struct files_struct *fd, *new_fd = NULL;
1818         struct cred *new_cred = NULL;
1819         struct nsproxy *new_nsproxy = NULL;
1820         int do_sysvsem = 0;
1821         int err;
1822
1823         /*
1824          * If unsharing a user namespace must also unshare the thread.
1825          */
1826         if (unshare_flags & CLONE_NEWUSER)
1827                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1828         /*
1829          * If unsharing a thread from a thread group, must also unshare vm.
1830          */
1831         if (unshare_flags & CLONE_THREAD)
1832                 unshare_flags |= CLONE_VM;
1833         /*
1834          * If unsharing vm, must also unshare signal handlers.
1835          */
1836         if (unshare_flags & CLONE_VM)
1837                 unshare_flags |= CLONE_SIGHAND;
1838         /*
1839          * If unsharing namespace, must also unshare filesystem information.
1840          */
1841         if (unshare_flags & CLONE_NEWNS)
1842                 unshare_flags |= CLONE_FS;
1843
1844         err = check_unshare_flags(unshare_flags);
1845         if (err)
1846                 goto bad_unshare_out;
1847         /*
1848          * CLONE_NEWIPC must also detach from the undolist: after switching
1849          * to a new ipc namespace, the semaphore arrays from the old
1850          * namespace are unreachable.
1851          */
1852         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1853                 do_sysvsem = 1;
1854         err = unshare_fs(unshare_flags, &new_fs);
1855         if (err)
1856                 goto bad_unshare_out;
1857         err = unshare_fd(unshare_flags, &new_fd);
1858         if (err)
1859                 goto bad_unshare_cleanup_fs;
1860         err = unshare_userns(unshare_flags, &new_cred);
1861         if (err)
1862                 goto bad_unshare_cleanup_fd;
1863         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1864                                          new_cred, new_fs);
1865         if (err)
1866                 goto bad_unshare_cleanup_cred;
1867
1868         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1869                 if (do_sysvsem) {
1870                         /*
1871                          * CLONE_SYSVSEM is equivalent to sys_exit().
1872                          */
1873                         exit_sem(current);
1874                 }
1875
1876                 if (new_nsproxy)
1877                         switch_task_namespaces(current, new_nsproxy);
1878
1879                 task_lock(current);
1880
1881                 if (new_fs) {
1882                         fs = current->fs;
1883                         spin_lock(&fs->lock);
1884                         current->fs = new_fs;
1885                         if (--fs->users)
1886                                 new_fs = NULL;
1887                         else
1888                                 new_fs = fs;
1889                         spin_unlock(&fs->lock);
1890                 }
1891
1892                 if (new_fd) {
1893                         fd = current->files;
1894                         current->files = new_fd;
1895                         new_fd = fd;
1896                 }
1897
1898                 task_unlock(current);
1899
1900                 if (new_cred) {
1901                         /* Install the new user namespace */
1902                         commit_creds(new_cred);
1903                         new_cred = NULL;
1904                 }
1905         }
1906
1907 bad_unshare_cleanup_cred:
1908         if (new_cred)
1909                 put_cred(new_cred);
1910 bad_unshare_cleanup_fd:
1911         if (new_fd)
1912                 put_files_struct(new_fd);
1913
1914 bad_unshare_cleanup_fs:
1915         if (new_fs)
1916                 free_fs_struct(new_fs);
1917
1918 bad_unshare_out:
1919         return err;
1920 }
1921
1922 /*
1923  *      Helper to unshare the files of the current task.
1924  *      We don't want to expose copy_files internals to
1925  *      the exec layer of the kernel.
1926  */
1927
1928 int unshare_files(struct files_struct **displaced)
1929 {
1930         struct task_struct *task = current;
1931         struct files_struct *copy = NULL;
1932         int error;
1933
1934         error = unshare_fd(CLONE_FILES, &copy);
1935         if (error || !copy) {
1936                 *displaced = NULL;
1937                 return error;
1938         }
1939         *displaced = task->files;
1940         task_lock(task);
1941         task->files = copy;
1942         task_unlock(task);
1943         return 0;
1944 }