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