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/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/tracehook.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
  73 #include <asm/pgtable.h>
  74 #include <asm/pgalloc.h>
  75 #include <asm/uaccess.h>
  76 #include <asm/mmu_context.h>
  77 #include <asm/cacheflush.h>
  78 #include <asm/tlbflush.h>
  79
  80 #include <trace/events/sched.h>
  81
  82 /*
  83  * Protected counters by write_lock_irq(&tasklist_lock)
  84  */
  85 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  86 int nr_threads;                 /* The idle threads do not count.. */
  87
  88 int max_threads;                /* tunable limit on nr_threads */
  89
  90 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  91
  92 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  93
  94 #ifdef CONFIG_PROVE_RCU
  95 int lockdep_tasklist_lock_is_held(void)
  96 {
  97         return lockdep_is_held(&tasklist_lock);
  98 }
  99 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 100 #endif /* #ifdef CONFIG_PROVE_RCU */
 101
 102 int nr_processes(void)
 103 {
 104         int cpu;
 105         int total = 0;
 106
 107         for_each_possible_cpu(cpu)
 108                 total += per_cpu(process_counts, cpu);
 109
 110         return total;
 111 }
 112
 113 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 114 # define alloc_task_struct_node(node)           \
 115                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 116 # define free_task_struct(tsk)                  \
 117                 kmem_cache_free(task_struct_cachep, (tsk))
 118 static struct kmem_cache *task_struct_cachep;
 119 #endif
 120
 121 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 122 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 123                                                   int node)
 124 {
 125 #ifdef CONFIG_DEBUG_STACK_USAGE
 126         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 127 #else
 128         gfp_t mask = GFP_KERNEL;
 129 #endif
 130         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 131
 132         return page ? page_address(page) : NULL;
 133 }
 134
 135 static inline void free_thread_info(struct thread_info *ti)
 136 {
 137         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 138 }
 139 #endif
 140
 141 /* SLAB cache for signal_struct structures (tsk->signal) */
 142 static struct kmem_cache *signal_cachep;
 143
 144 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 145 struct kmem_cache *sighand_cachep;
 146
 147 /* SLAB cache for files_struct structures (tsk->files) */
 148 struct kmem_cache *files_cachep;
 149
 150 /* SLAB cache for fs_struct structures (tsk->fs) */
 151 struct kmem_cache *fs_cachep;
 152
 153 /* SLAB cache for vm_area_struct structures */
 154 struct kmem_cache *vm_area_cachep;
 155
 156 /* SLAB cache for mm_struct structures (tsk->mm) */
 157 static struct kmem_cache *mm_cachep;
 158
 159 static void account_kernel_stack(struct thread_info *ti, int account)
 160 {
 161         struct zone *zone = page_zone(virt_to_page(ti));
 162
 163         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 164 }
 165
 166 void free_task(struct task_struct *tsk)
 167 {
 168         prop_local_destroy_single(&tsk->dirties);
 169         account_kernel_stack(tsk->stack, -1);
 170         free_thread_info(tsk->stack);
 171         rt_mutex_debug_task_free(tsk);
 172         ftrace_graph_exit_task(tsk);
 173         free_task_struct(tsk);
 174 }
 175 EXPORT_SYMBOL(free_task);
 176
 177 static inline void free_signal_struct(struct signal_struct *sig)
 178 {
 179         taskstats_tgid_free(sig);
 180         sched_autogroup_exit(sig);
 181         kmem_cache_free(signal_cachep, sig);
 182 }
 183
 184 static inline void put_signal_struct(struct signal_struct *sig)
 185 {
 186         if (atomic_dec_and_test(&sig->sigcnt))
 187                 free_signal_struct(sig);
 188 }
 189
 190 void __put_task_struct(struct task_struct *tsk)
 191 {
 192         WARN_ON(!tsk->exit_state);
 193         WARN_ON(atomic_read(&tsk->usage));
 194         WARN_ON(tsk == current);
 195
 196         exit_creds(tsk);
 197         delayacct_tsk_free(tsk);
 198         put_signal_struct(tsk->signal);
 199
 200         if (!profile_handoff_task(tsk))
 201                 free_task(tsk);
 202 }
 203 EXPORT_SYMBOL_GPL(__put_task_struct);
 204
 205 /*
 206  * macro override instead of weak attribute alias, to workaround
 207  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 208  */
 209 #ifndef arch_task_cache_init
 210 #define arch_task_cache_init()
 211 #endif
 212
 213 void __init fork_init(unsigned long mempages)
 214 {
 215 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 216 #ifndef ARCH_MIN_TASKALIGN
 217 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 218 #endif
 219         /* create a slab on which task_structs can be allocated */
 220         task_struct_cachep =
 221                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 222                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 223 #endif
 224
 225         /* do the arch specific task caches init */
 226         arch_task_cache_init();
 227
 228         /*
 229          * The default maximum number of threads is set to a safe
 230          * value: the thread structures can take up at most half
 231          * of memory.
 232          */
 233         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 234
 235         /*
 236          * we need to allow at least 20 threads to boot a system
 237          */
 238         if(max_threads < 20)
 239                 max_threads = 20;
 240
 241         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 242         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 243         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 244                 init_task.signal->rlim[RLIMIT_NPROC];
 245 }
 246
 247 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 248                                                struct task_struct *src)
 249 {
 250         *dst = *src;
 251         return 0;
 252 }
 253
 254 static struct task_struct *dup_task_struct(struct task_struct *orig)
 255 {
 256         struct task_struct *tsk;
 257         struct thread_info *ti;
 258         unsigned long *stackend;
 259         int node = tsk_fork_get_node(orig);
 260         int err;
 261
 262         prepare_to_copy(orig);
 263
 264         tsk = alloc_task_struct_node(node);
 265         if (!tsk)
 266                 return NULL;
 267
 268         ti = alloc_thread_info_node(tsk, node);
 269         if (!ti) {
 270                 free_task_struct(tsk);
 271                 return NULL;
 272         }
 273
 274         err = arch_dup_task_struct(tsk, orig);
 275         if (err)
 276                 goto out;
 277
 278         tsk->stack = ti;
 279
 280         err = prop_local_init_single(&tsk->dirties);
 281         if (err)
 282                 goto out;
 283
 284         setup_thread_stack(tsk, orig);
 285         clear_user_return_notifier(tsk);
 286         clear_tsk_need_resched(tsk);
 287         stackend = end_of_stack(tsk);
 288         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 289
 290 #ifdef CONFIG_CC_STACKPROTECTOR
 291         tsk->stack_canary = get_random_int();
 292 #endif
 293
 294         /* One for us, one for whoever does the "release_task()" (usually parent) */
 295         atomic_set(&tsk->usage,2);
 296         atomic_set(&tsk->fs_excl, 0);
 297 #ifdef CONFIG_BLK_DEV_IO_TRACE
 298         tsk->btrace_seq = 0;
 299 #endif
 300         tsk->splice_pipe = NULL;
 301
 302         account_kernel_stack(ti, 1);
 303
 304         return tsk;
 305
 306 out:
 307         free_thread_info(ti);
 308         free_task_struct(tsk);
 309         return NULL;
 310 }
 311
 312 #ifdef CONFIG_MMU
 313 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 314 {
 315         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 316         struct rb_node **rb_link, *rb_parent;
 317         int retval;
 318         unsigned long charge;
 319         struct mempolicy *pol;
 320
 321         down_write(&oldmm->mmap_sem);
 322         flush_cache_dup_mm(oldmm);
 323         /*
 324          * Not linked in yet - no deadlock potential:
 325          */
 326         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 327
 328         mm->locked_vm = 0;
 329         mm->mmap = NULL;
 330         mm->mmap_cache = NULL;
 331         mm->free_area_cache = oldmm->mmap_base;
 332         mm->cached_hole_size = ~0UL;
 333         mm->map_count = 0;
 334         cpumask_clear(mm_cpumask(mm));
 335         mm->mm_rb = RB_ROOT;
 336         rb_link = &mm->mm_rb.rb_node;
 337         rb_parent = NULL;
 338         pprev = &mm->mmap;
 339         retval = ksm_fork(mm, oldmm);
 340         if (retval)
 341                 goto out;
 342         retval = khugepaged_fork(mm, oldmm);
 343         if (retval)
 344                 goto out;
 345
 346         prev = NULL;
 347         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 348                 struct file *file;
 349
 350                 if (mpnt->vm_flags & VM_DONTCOPY) {
 351                         long pages = vma_pages(mpnt);
 352                         mm->total_vm -= pages;
 353                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 354                                                                 -pages);
 355                         continue;
 356                 }
 357                 charge = 0;
 358                 if (mpnt->vm_flags & VM_ACCOUNT) {
 359                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 360                         if (security_vm_enough_memory(len))
 361                                 goto fail_nomem;
 362                         charge = len;
 363                 }
 364                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 365                 if (!tmp)
 366                         goto fail_nomem;
 367                 *tmp = *mpnt;
 368                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 369                 pol = mpol_dup(vma_policy(mpnt));
 370                 retval = PTR_ERR(pol);
 371                 if (IS_ERR(pol))
 372                         goto fail_nomem_policy;
 373                 vma_set_policy(tmp, pol);
 374                 tmp->vm_mm = mm;
 375                 if (anon_vma_fork(tmp, mpnt))
 376                         goto fail_nomem_anon_vma_fork;
 377                 tmp->vm_flags &= ~VM_LOCKED;
 378                 tmp->vm_next = tmp->vm_prev = NULL;
 379                 file = tmp->vm_file;
 380                 if (file) {
 381                         struct inode *inode = file->f_path.dentry->d_inode;
 382                         struct address_space *mapping = file->f_mapping;
 383
 384                         get_file(file);
 385                         if (tmp->vm_flags & VM_DENYWRITE)
 386                                 atomic_dec(&inode->i_writecount);
 387                         mutex_lock(&mapping->i_mmap_mutex);
 388                         if (tmp->vm_flags & VM_SHARED)
 389                                 mapping->i_mmap_writable++;
 390                         flush_dcache_mmap_lock(mapping);
 391                         /* insert tmp into the share list, just after mpnt */
 392                         vma_prio_tree_add(tmp, mpnt);
 393                         flush_dcache_mmap_unlock(mapping);
 394                         mutex_unlock(&mapping->i_mmap_mutex);
 395                 }
 396
 397                 /*
 398                  * Clear hugetlb-related page reserves for children. This only
 399                  * affects MAP_PRIVATE mappings. Faults generated by the child
 400                  * are not guaranteed to succeed, even if read-only
 401                  */
 402                 if (is_vm_hugetlb_page(tmp))
 403                         reset_vma_resv_huge_pages(tmp);
 404
 405                 /*
 406                  * Link in the new vma and copy the page table entries.
 407                  */
 408                 *pprev = tmp;
 409                 pprev = &tmp->vm_next;
 410                 tmp->vm_prev = prev;
 411                 prev = tmp;
 412
 413                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 414                 rb_link = &tmp->vm_rb.rb_right;
 415                 rb_parent = &tmp->vm_rb;
 416
 417                 mm->map_count++;
 418                 retval = copy_page_range(mm, oldmm, mpnt);
 419
 420                 if (tmp->vm_ops && tmp->vm_ops->open)
 421                         tmp->vm_ops->open(tmp);
 422
 423                 if (retval)
 424                         goto out;
 425         }
 426         /* a new mm has just been created */
 427         arch_dup_mmap(oldmm, mm);
 428         retval = 0;
 429 out:
 430         up_write(&mm->mmap_sem);
 431         flush_tlb_mm(oldmm);
 432         up_write(&oldmm->mmap_sem);
 433         return retval;
 434 fail_nomem_anon_vma_fork:
 435         mpol_put(pol);
 436 fail_nomem_policy:
 437         kmem_cache_free(vm_area_cachep, tmp);
 438 fail_nomem:
 439         retval = -ENOMEM;
 440         vm_unacct_memory(charge);
 441         goto out;
 442 }
 443
 444 static inline int mm_alloc_pgd(struct mm_struct * mm)
 445 {
 446         mm->pgd = pgd_alloc(mm);
 447         if (unlikely(!mm->pgd))
 448                 return -ENOMEM;
 449         return 0;
 450 }
 451
 452 static inline void mm_free_pgd(struct mm_struct * mm)
 453 {
 454         pgd_free(mm, mm->pgd);
 455 }
 456 #else
 457 #define dup_mmap(mm, oldmm)     (0)
 458 #define mm_alloc_pgd(mm)        (0)
 459 #define mm_free_pgd(mm)
 460 #endif /* CONFIG_MMU */
 461
 462 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 463
 464 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 465 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 466
 467 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 468
 469 static int __init coredump_filter_setup(char *s)
 470 {
 471         default_dump_filter =
 472                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 473                 MMF_DUMP_FILTER_MASK;
 474         return 1;
 475 }
 476
 477 __setup("coredump_filter=", coredump_filter_setup);
 478
 479 #include <linux/init_task.h>
 480
 481 static void mm_init_aio(struct mm_struct *mm)
 482 {
 483 #ifdef CONFIG_AIO
 484         spin_lock_init(&mm->ioctx_lock);
 485         INIT_HLIST_HEAD(&mm->ioctx_list);
 486 #endif
 487 }
 488
 489 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 490 {
 491         atomic_set(&mm->mm_users, 1);
 492         atomic_set(&mm->mm_count, 1);
 493         init_rwsem(&mm->mmap_sem);
 494         INIT_LIST_HEAD(&mm->mmlist);
 495         mm->flags = (current->mm) ?
 496                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 497         mm->core_state = NULL;
 498         mm->nr_ptes = 0;
 499         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 500         spin_lock_init(&mm->page_table_lock);
 501         mm->free_area_cache = TASK_UNMAPPED_BASE;
 502         mm->cached_hole_size = ~0UL;
 503         mm_init_aio(mm);
 504         mm_init_owner(mm, p);
 505         atomic_set(&mm->oom_disable_count, 0);
 506
 507         if (likely(!mm_alloc_pgd(mm))) {
 508                 mm->def_flags = 0;
 509                 mmu_notifier_mm_init(mm);
 510                 return mm;
 511         }
 512
 513         free_mm(mm);
 514         return NULL;
 515 }
 516
 517 /*
 518  * Allocate and initialize an mm_struct.
 519  */
 520 struct mm_struct * mm_alloc(void)
 521 {
 522         struct mm_struct * mm;
 523
 524         mm = allocate_mm();
 525         if (!mm)
 526                 return NULL;
 527
 528         memset(mm, 0, sizeof(*mm));
 529         mm_init_cpumask(mm);
 530         return mm_init(mm, current);
 531 }
 532
 533 /*
 534  * Called when the last reference to the mm
 535  * is dropped: either by a lazy thread or by
 536  * mmput. Free the page directory and the mm.
 537  */
 538 void __mmdrop(struct mm_struct *mm)
 539 {
 540         BUG_ON(mm == &init_mm);
 541         mm_free_pgd(mm);
 542         destroy_context(mm);
 543         mmu_notifier_mm_destroy(mm);
 544 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 545         VM_BUG_ON(mm->pmd_huge_pte);
 546 #endif
 547         free_mm(mm);
 548 }
 549 EXPORT_SYMBOL_GPL(__mmdrop);
 550
 551 /*
 552  * Decrement the use count and release all resources for an mm.
 553  */
 554 void mmput(struct mm_struct *mm)
 555 {
 556         might_sleep();
 557
 558         if (atomic_dec_and_test(&mm->mm_users)) {
 559                 exit_aio(mm);
 560                 ksm_exit(mm);
 561                 khugepaged_exit(mm); /* must run before exit_mmap */
 562                 exit_mmap(mm);
 563                 set_mm_exe_file(mm, NULL);
 564                 if (!list_empty(&mm->mmlist)) {
 565                         spin_lock(&mmlist_lock);
 566                         list_del(&mm->mmlist);
 567                         spin_unlock(&mmlist_lock);
 568                 }
 569                 put_swap_token(mm);
 570                 if (mm->binfmt)
 571                         module_put(mm->binfmt->module);
 572                 mmdrop(mm);
 573         }
 574 }
 575 EXPORT_SYMBOL_GPL(mmput);
 576
 577 /*
 578  * We added or removed a vma mapping the executable. The vmas are only mapped
 579  * during exec and are not mapped with the mmap system call.
 580  * Callers must hold down_write() on the mm's mmap_sem for these
 581  */
 582 void added_exe_file_vma(struct mm_struct *mm)
 583 {
 584         mm->num_exe_file_vmas++;
 585 }
 586
 587 void removed_exe_file_vma(struct mm_struct *mm)
 588 {
 589         mm->num_exe_file_vmas--;
 590         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
 591                 fput(mm->exe_file);
 592                 mm->exe_file = NULL;
 593         }
 594
 595 }
 596
 597 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 598 {
 599         if (new_exe_file)
 600                 get_file(new_exe_file);
 601         if (mm->exe_file)
 602                 fput(mm->exe_file);
 603         mm->exe_file = new_exe_file;
 604         mm->num_exe_file_vmas = 0;
 605 }
 606
 607 struct file *get_mm_exe_file(struct mm_struct *mm)
 608 {
 609         struct file *exe_file;
 610
 611         /* We need mmap_sem to protect against races with removal of
 612          * VM_EXECUTABLE vmas */
 613         down_read(&mm->mmap_sem);
 614         exe_file = mm->exe_file;
 615         if (exe_file)
 616                 get_file(exe_file);
 617         up_read(&mm->mmap_sem);
 618         return exe_file;
 619 }
 620
 621 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 622 {
 623         /* It's safe to write the exe_file pointer without exe_file_lock because
 624          * this is called during fork when the task is not yet in /proc */
 625         newmm->exe_file = get_mm_exe_file(oldmm);
 626 }
 627
 628 /**
 629  * get_task_mm - acquire a reference to the task's mm
 630  *
 631  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 632  * this kernel workthread has transiently adopted a user mm with use_mm,
 633  * to do its AIO) is not set and if so returns a reference to it, after
 634  * bumping up the use count.  User must release the mm via mmput()
 635  * after use.  Typically used by /proc and ptrace.
 636  */
 637 struct mm_struct *get_task_mm(struct task_struct *task)
 638 {
 639         struct mm_struct *mm;
 640
 641         task_lock(task);
 642         mm = task->mm;
 643         if (mm) {
 644                 if (task->flags & PF_KTHREAD)
 645                         mm = NULL;
 646                 else
 647                         atomic_inc(&mm->mm_users);
 648         }
 649         task_unlock(task);
 650         return mm;
 651 }
 652 EXPORT_SYMBOL_GPL(get_task_mm);
 653
 654 /* Please note the differences between mmput and mm_release.
 655  * mmput is called whenever we stop holding onto a mm_struct,
 656  * error success whatever.
 657  *
 658  * mm_release is called after a mm_struct has been removed
 659  * from the current process.
 660  *
 661  * This difference is important for error handling, when we
 662  * only half set up a mm_struct for a new process and need to restore
 663  * the old one.  Because we mmput the new mm_struct before
 664  * restoring the old one. . .
 665  * Eric Biederman 10 January 1998
 666  */
 667 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 668 {
 669         struct completion *vfork_done = tsk->vfork_done;
 670
 671         /* Get rid of any futexes when releasing the mm */
 672 #ifdef CONFIG_FUTEX
 673         if (unlikely(tsk->robust_list)) {
 674                 exit_robust_list(tsk);
 675                 tsk->robust_list = NULL;
 676         }
 677 #ifdef CONFIG_COMPAT
 678         if (unlikely(tsk->compat_robust_list)) {
 679                 compat_exit_robust_list(tsk);
 680                 tsk->compat_robust_list = NULL;
 681         }
 682 #endif
 683         if (unlikely(!list_empty(&tsk->pi_state_list)))
 684                 exit_pi_state_list(tsk);
 685 #endif
 686
 687         /* Get rid of any cached register state */
 688         deactivate_mm(tsk, mm);
 689
 690         /* notify parent sleeping on vfork() */
 691         if (vfork_done) {
 692                 tsk->vfork_done = NULL;
 693                 complete(vfork_done);
 694         }
 695
 696         /*
 697          * If we're exiting normally, clear a user-space tid field if
 698          * requested.  We leave this alone when dying by signal, to leave
 699          * the value intact in a core dump, and to save the unnecessary
 700          * trouble otherwise.  Userland only wants this done for a sys_exit.
 701          */
 702         if (tsk->clear_child_tid) {
 703                 if (!(tsk->flags & PF_SIGNALED) &&
 704                     atomic_read(&mm->mm_users) > 1) {
 705                         /*
 706                          * We don't check the error code - if userspace has
 707                          * not set up a proper pointer then tough luck.
 708                          */
 709                         put_user(0, tsk->clear_child_tid);
 710                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 711                                         1, NULL, NULL, 0);
 712                 }
 713                 tsk->clear_child_tid = NULL;
 714         }
 715 }
 716
 717 /*
 718  * Allocate a new mm structure and copy contents from the
 719  * mm structure of the passed in task structure.
 720  */
 721 struct mm_struct *dup_mm(struct task_struct *tsk)
 722 {
 723         struct mm_struct *mm, *oldmm = current->mm;
 724         int err;
 725
 726         if (!oldmm)
 727                 return NULL;
 728
 729         mm = allocate_mm();
 730         if (!mm)
 731                 goto fail_nomem;
 732
 733         memcpy(mm, oldmm, sizeof(*mm));
 734         mm_init_cpumask(mm);
 735
 736         /* Initializing for Swap token stuff */
 737         mm->token_priority = 0;
 738         mm->last_interval = 0;
 739
 740 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 741         mm->pmd_huge_pte = NULL;
 742 #endif
 743
 744         if (!mm_init(mm, tsk))
 745                 goto fail_nomem;
 746
 747         if (init_new_context(tsk, mm))
 748                 goto fail_nocontext;
 749
 750         dup_mm_exe_file(oldmm, mm);
 751
 752         err = dup_mmap(mm, oldmm);
 753         if (err)
 754                 goto free_pt;
 755
 756         mm->hiwater_rss = get_mm_rss(mm);
 757         mm->hiwater_vm = mm->total_vm;
 758
 759         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 760                 goto free_pt;
 761
 762         return mm;
 763
 764 free_pt:
 765         /* don't put binfmt in mmput, we haven't got module yet */
 766         mm->binfmt = NULL;
 767         mmput(mm);
 768
 769 fail_nomem:
 770         return NULL;
 771
 772 fail_nocontext:
 773         /*
 774          * If init_new_context() failed, we cannot use mmput() to free the mm
 775          * because it calls destroy_context()
 776          */
 777         mm_free_pgd(mm);
 778         free_mm(mm);
 779         return NULL;
 780 }
 781
 782 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 783 {
 784         struct mm_struct * mm, *oldmm;
 785         int retval;
 786
 787         tsk->min_flt = tsk->maj_flt = 0;
 788         tsk->nvcsw = tsk->nivcsw = 0;
 789 #ifdef CONFIG_DETECT_HUNG_TASK
 790         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 791 #endif
 792
 793         tsk->mm = NULL;
 794         tsk->active_mm = NULL;
 795
 796         /*
 797          * Are we cloning a kernel thread?
 798          *
 799          * We need to steal a active VM for that..
 800          */
 801         oldmm = current->mm;
 802         if (!oldmm)
 803                 return 0;
 804
 805         if (clone_flags & CLONE_VM) {
 806                 atomic_inc(&oldmm->mm_users);
 807                 mm = oldmm;
 808                 goto good_mm;
 809         }
 810
 811         retval = -ENOMEM;
 812         mm = dup_mm(tsk);
 813         if (!mm)
 814                 goto fail_nomem;
 815
 816 good_mm:
 817         /* Initializing for Swap token stuff */
 818         mm->token_priority = 0;
 819         mm->last_interval = 0;
 820         if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
 821                 atomic_inc(&mm->oom_disable_count);
 822
 823         tsk->mm = mm;
 824         tsk->active_mm = mm;
 825         return 0;
 826
 827 fail_nomem:
 828         return retval;
 829 }
 830
 831 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 832 {
 833         struct fs_struct *fs = current->fs;
 834         if (clone_flags & CLONE_FS) {
 835                 /* tsk->fs is already what we want */
 836                 spin_lock(&fs->lock);
 837                 if (fs->in_exec) {
 838                         spin_unlock(&fs->lock);
 839                         return -EAGAIN;
 840                 }
 841                 fs->users++;
 842                 spin_unlock(&fs->lock);
 843                 return 0;
 844         }
 845         tsk->fs = copy_fs_struct(fs);
 846         if (!tsk->fs)
 847                 return -ENOMEM;
 848         return 0;
 849 }
 850
 851 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 852 {
 853         struct files_struct *oldf, *newf;
 854         int error = 0;
 855
 856         /*
 857          * A background process may not have any files ...
 858          */
 859         oldf = current->files;
 860         if (!oldf)
 861                 goto out;
 862
 863         if (clone_flags & CLONE_FILES) {
 864                 atomic_inc(&oldf->count);
 865                 goto out;
 866         }
 867
 868         newf = dup_fd(oldf, &error);
 869         if (!newf)
 870                 goto out;
 871
 872         tsk->files = newf;
 873         error = 0;
 874 out:
 875         return error;
 876 }
 877
 878 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 879 {
 880 #ifdef CONFIG_BLOCK
 881         struct io_context *ioc = current->io_context;
 882
 883         if (!ioc)
 884                 return 0;
 885         /*
 886          * Share io context with parent, if CLONE_IO is set
 887          */
 888         if (clone_flags & CLONE_IO) {
 889                 tsk->io_context = ioc_task_link(ioc);
 890                 if (unlikely(!tsk->io_context))
 891                         return -ENOMEM;
 892         } else if (ioprio_valid(ioc->ioprio)) {
 893                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 894                 if (unlikely(!tsk->io_context))
 895                         return -ENOMEM;
 896
 897                 tsk->io_context->ioprio = ioc->ioprio;
 898         }
 899 #endif
 900         return 0;
 901 }
 902
 903 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 904 {
 905         struct sighand_struct *sig;
 906
 907         if (clone_flags & CLONE_SIGHAND) {
 908                 atomic_inc(&current->sighand->count);
 909                 return 0;
 910         }
 911         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 912         rcu_assign_pointer(tsk->sighand, sig);
 913         if (!sig)
 914                 return -ENOMEM;
 915         atomic_set(&sig->count, 1);
 916         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 917         return 0;
 918 }
 919
 920 void __cleanup_sighand(struct sighand_struct *sighand)
 921 {
 922         if (atomic_dec_and_test(&sighand->count)) {
 923                 signalfd_cleanup(sighand);
 924                 kmem_cache_free(sighand_cachep, sighand);
 925         }
 926 }
 927
 928
 929 /*
 930  * Initialize POSIX timer handling for a thread group.
 931  */
 932 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 933 {
 934         unsigned long cpu_limit;
 935
 936         /* Thread group counters. */
 937         thread_group_cputime_init(sig);
 938
 939         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 940         if (cpu_limit != RLIM_INFINITY) {
 941                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 942                 sig->cputimer.running = 1;
 943         }
 944
 945         /* The timer lists. */
 946         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 947         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 948         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 949 }
 950
 951 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 952 {
 953         struct signal_struct *sig;
 954
 955         if (clone_flags & CLONE_THREAD)
 956                 return 0;
 957
 958         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
 959         tsk->signal = sig;
 960         if (!sig)
 961                 return -ENOMEM;
 962
 963         sig->nr_threads = 1;
 964         atomic_set(&sig->live, 1);
 965         atomic_set(&sig->sigcnt, 1);
 966         init_waitqueue_head(&sig->wait_chldexit);
 967         if (clone_flags & CLONE_NEWPID)
 968                 sig->flags |= SIGNAL_UNKILLABLE;
 969         sig->curr_target = tsk;
 970         init_sigpending(&sig->shared_pending);
 971         INIT_LIST_HEAD(&sig->posix_timers);
 972
 973         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 974         sig->real_timer.function = it_real_fn;
 975
 976         task_lock(current->group_leader);
 977         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 978         task_unlock(current->group_leader);
 979
 980         posix_cpu_timers_init_group(sig);
 981
 982         tty_audit_fork(sig);
 983         sched_autogroup_fork(sig);
 984
 985 #ifdef CONFIG_CGROUPS
 986         init_rwsem(&sig->threadgroup_fork_lock);
 987 #endif
 988 #ifdef CONFIG_CPUSETS
 989         seqcount_init(&tsk->mems_allowed_seq);
 990 #endif
 991
 992         sig->oom_adj = current->signal->oom_adj;
 993         sig->oom_score_adj = current->signal->oom_score_adj;
 994         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
 995
 996         mutex_init(&sig->cred_guard_mutex);
 997
 998         return 0;
 999 }
1000
1001 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1002 {
1003         unsigned long new_flags = p->flags;
1004
1005         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1006         new_flags |= PF_FORKNOEXEC;
1007         new_flags |= PF_STARTING;
1008         p->flags = new_flags;
1009         clear_freeze_flag(p);
1010 }
1011
1012 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1013 {
1014         current->clear_child_tid = tidptr;
1015
1016         return task_pid_vnr(current);
1017 }
1018
1019 static void rt_mutex_init_task(struct task_struct *p)
1020 {
1021         raw_spin_lock_init(&p->pi_lock);
1022 #ifdef CONFIG_RT_MUTEXES
1023         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
1024         p->pi_blocked_on = NULL;
1025 #endif
1026 }
1027
1028 #ifdef CONFIG_MM_OWNER
1029 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1030 {
1031         mm->owner = p;
1032 }
1033 #endif /* CONFIG_MM_OWNER */
1034
1035 /*
1036  * Initialize POSIX timer handling for a single task.
1037  */
1038 static void posix_cpu_timers_init(struct task_struct *tsk)
1039 {
1040         tsk->cputime_expires.prof_exp = cputime_zero;
1041         tsk->cputime_expires.virt_exp = cputime_zero;
1042         tsk->cputime_expires.sched_exp = 0;
1043         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1044         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1045         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1046 }
1047
1048 /*
1049  * This creates a new process as a copy of the old one,
1050  * but does not actually start it yet.
1051  *
1052  * It copies the registers, and all the appropriate
1053  * parts of the process environment (as per the clone
1054  * flags). The actual kick-off is left to the caller.
1055  */
1056 static struct task_struct *copy_process(unsigned long clone_flags,
1057                                         unsigned long stack_start,
1058                                         struct pt_regs *regs,
1059                                         unsigned long stack_size,
1060                                         int __user *child_tidptr,
1061                                         struct pid *pid,
1062                                         int trace)
1063 {
1064         int retval;
1065         struct task_struct *p;
1066         int cgroup_callbacks_done = 0;
1067
1068         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1069                 return ERR_PTR(-EINVAL);
1070
1071         /*
1072          * Thread groups must share signals as well, and detached threads
1073          * can only be started up within the thread group.
1074          */
1075         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1076                 return ERR_PTR(-EINVAL);
1077
1078         /*
1079          * Shared signal handlers imply shared VM. By way of the above,
1080          * thread groups also imply shared VM. Blocking this case allows
1081          * for various simplifications in other code.
1082          */
1083         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1084                 return ERR_PTR(-EINVAL);
1085
1086         /*
1087          * Siblings of global init remain as zombies on exit since they are
1088          * not reaped by their parent (swapper). To solve this and to avoid
1089          * multi-rooted process trees, prevent global and container-inits
1090          * from creating siblings.
1091          */
1092         if ((clone_flags & CLONE_PARENT) &&
1093                                 current->signal->flags & SIGNAL_UNKILLABLE)
1094                 return ERR_PTR(-EINVAL);
1095
1096         retval = security_task_create(clone_flags);
1097         if (retval)
1098                 goto fork_out;
1099
1100         retval = -ENOMEM;
1101         p = dup_task_struct(current);
1102         if (!p)
1103                 goto fork_out;
1104
1105         ftrace_graph_init_task(p);
1106
1107         rt_mutex_init_task(p);
1108
1109 #ifdef CONFIG_PROVE_LOCKING
1110         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1111         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1112 #endif
1113         retval = -EAGAIN;
1114         if (atomic_read(&p->real_cred->user->processes) >=
1115                         task_rlimit(p, RLIMIT_NPROC)) {
1116                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1117                     p->real_cred->user != INIT_USER)
1118                         goto bad_fork_free;
1119         }
1120
1121         retval = copy_creds(p, clone_flags);
1122         if (retval < 0)
1123                 goto bad_fork_free;
1124
1125         /*
1126          * If multiple threads are within copy_process(), then this check
1127          * triggers too late. This doesn't hurt, the check is only there
1128          * to stop root fork bombs.
1129          */
1130         retval = -EAGAIN;
1131         if (nr_threads >= max_threads)
1132                 goto bad_fork_cleanup_count;
1133
1134         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1135                 goto bad_fork_cleanup_count;
1136
1137         p->did_exec = 0;
1138         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1139         copy_flags(clone_flags, p);
1140         INIT_LIST_HEAD(&p->children);
1141         INIT_LIST_HEAD(&p->sibling);
1142         rcu_copy_process(p);
1143         p->vfork_done = NULL;
1144         spin_lock_init(&p->alloc_lock);
1145
1146         init_sigpending(&p->pending);
1147
1148         p->utime = cputime_zero;
1149         p->stime = cputime_zero;
1150         p->gtime = cputime_zero;
1151         p->utimescaled = cputime_zero;
1152         p->stimescaled = cputime_zero;
1153 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1154         p->prev_utime = cputime_zero;
1155         p->prev_stime = cputime_zero;
1156 #endif
1157 #if defined(SPLIT_RSS_COUNTING)
1158         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1159 #endif
1160
1161         p->default_timer_slack_ns = current->timer_slack_ns;
1162
1163         task_io_accounting_init(&p->ioac);
1164         acct_clear_integrals(p);
1165
1166         posix_cpu_timers_init(p);
1167
1168         do_posix_clock_monotonic_gettime(&p->start_time);
1169         p->real_start_time = p->start_time;
1170         monotonic_to_bootbased(&p->real_start_time);
1171         p->io_context = NULL;
1172         p->audit_context = NULL;
1173         if (clone_flags & CLONE_THREAD)
1174                 threadgroup_fork_read_lock(current);
1175         cgroup_fork(p);
1176 #ifdef CONFIG_NUMA
1177         p->mempolicy = mpol_dup(p->mempolicy);
1178         if (IS_ERR(p->mempolicy)) {
1179                 retval = PTR_ERR(p->mempolicy);
1180                 p->mempolicy = NULL;
1181                 goto bad_fork_cleanup_cgroup;
1182         }
1183         mpol_fix_fork_child_flag(p);
1184 #endif
1185 #ifdef CONFIG_TRACE_IRQFLAGS
1186         p->irq_events = 0;
1187 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1188         p->hardirqs_enabled = 1;
1189 #else
1190         p->hardirqs_enabled = 0;
1191 #endif
1192         p->hardirq_enable_ip = 0;
1193         p->hardirq_enable_event = 0;
1194         p->hardirq_disable_ip = _THIS_IP_;
1195         p->hardirq_disable_event = 0;
1196         p->softirqs_enabled = 1;
1197         p->softirq_enable_ip = _THIS_IP_;
1198         p->softirq_enable_event = 0;
1199         p->softirq_disable_ip = 0;
1200         p->softirq_disable_event = 0;
1201         p->hardirq_context = 0;
1202         p->softirq_context = 0;
1203 #endif
1204 #ifdef CONFIG_LOCKDEP
1205         p->lockdep_depth = 0; /* no locks held yet */
1206         p->curr_chain_key = 0;
1207         p->lockdep_recursion = 0;
1208 #endif
1209
1210 #ifdef CONFIG_DEBUG_MUTEXES
1211         p->blocked_on = NULL; /* not blocked yet */
1212 #endif
1213 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1214         p->memcg_batch.do_batch = 0;
1215         p->memcg_batch.memcg = NULL;
1216 #endif
1217
1218         /* Perform scheduler related setup. Assign this task to a CPU. */
1219         sched_fork(p);
1220
1221         retval = perf_event_init_task(p);
1222         if (retval)
1223                 goto bad_fork_cleanup_policy;
1224
1225         if ((retval = audit_alloc(p)))
1226                 goto bad_fork_cleanup_policy;
1227         /* copy all the process information */
1228         if ((retval = copy_semundo(clone_flags, p)))
1229                 goto bad_fork_cleanup_audit;
1230         if ((retval = copy_files(clone_flags, p)))
1231                 goto bad_fork_cleanup_semundo;
1232         if ((retval = copy_fs(clone_flags, p)))
1233                 goto bad_fork_cleanup_files;
1234         if ((retval = copy_sighand(clone_flags, p)))
1235                 goto bad_fork_cleanup_fs;
1236         if ((retval = copy_signal(clone_flags, p)))
1237                 goto bad_fork_cleanup_sighand;
1238         if ((retval = copy_mm(clone_flags, p)))
1239                 goto bad_fork_cleanup_signal;
1240         if ((retval = copy_namespaces(clone_flags, p)))
1241                 goto bad_fork_cleanup_mm;
1242         if ((retval = copy_io(clone_flags, p)))
1243                 goto bad_fork_cleanup_namespaces;
1244         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1245         if (retval)
1246                 goto bad_fork_cleanup_io;
1247
1248         if (pid != &init_struct_pid) {
1249                 retval = -ENOMEM;
1250                 pid = alloc_pid(p->nsproxy->pid_ns);
1251                 if (!pid)
1252                         goto bad_fork_cleanup_io;
1253         }
1254
1255         p->pid = pid_nr(pid);
1256         p->tgid = p->pid;
1257         if (clone_flags & CLONE_THREAD)
1258                 p->tgid = current->tgid;
1259
1260         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1261         /*
1262          * Clear TID on mm_release()?
1263          */
1264         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1265 #ifdef CONFIG_BLOCK
1266         p->plug = NULL;
1267 #endif
1268 #ifdef CONFIG_FUTEX
1269         p->robust_list = NULL;
1270 #ifdef CONFIG_COMPAT
1271         p->compat_robust_list = NULL;
1272 #endif
1273         INIT_LIST_HEAD(&p->pi_state_list);
1274         p->pi_state_cache = NULL;
1275 #endif
1276         /*
1277          * sigaltstack should be cleared when sharing the same VM
1278          */
1279         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1280                 p->sas_ss_sp = p->sas_ss_size = 0;
1281
1282         /*
1283          * Syscall tracing and stepping should be turned off in the
1284          * child regardless of CLONE_PTRACE.
1285          */
1286         user_disable_single_step(p);
1287         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1288 #ifdef TIF_SYSCALL_EMU
1289         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1290 #endif
1291         clear_all_latency_tracing(p);
1292
1293         /* ok, now we should be set up.. */
1294         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1295         p->pdeath_signal = 0;
1296         p->exit_state = 0;
1297
1298         /*
1299          * Ok, make it visible to the rest of the system.
1300          * We dont wake it up yet.
1301          */
1302         p->group_leader = p;
1303         INIT_LIST_HEAD(&p->thread_group);
1304
1305         /* Now that the task is set up, run cgroup callbacks if
1306          * necessary. We need to run them before the task is visible
1307          * on the tasklist. */
1308         cgroup_fork_callbacks(p);
1309         cgroup_callbacks_done = 1;
1310
1311         /* Need tasklist lock for parent etc handling! */
1312         write_lock_irq(&tasklist_lock);
1313
1314         /* CLONE_PARENT re-uses the old parent */
1315         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1316                 p->real_parent = current->real_parent;
1317                 p->parent_exec_id = current->parent_exec_id;
1318         } else {
1319                 p->real_parent = current;
1320                 p->parent_exec_id = current->self_exec_id;
1321         }
1322
1323         spin_lock(&current->sighand->siglock);
1324
1325         /*
1326          * Process group and session signals need to be delivered to just the
1327          * parent before the fork or both the parent and the child after the
1328          * fork. Restart if a signal comes in before we add the new process to
1329          * it's process group.
1330          * A fatal signal pending means that current will exit, so the new
1331          * thread can't slip out of an OOM kill (or normal SIGKILL).
1332          */
1333         recalc_sigpending();
1334         if (signal_pending(current)) {
1335                 spin_unlock(&current->sighand->siglock);
1336                 write_unlock_irq(&tasklist_lock);
1337                 retval = -ERESTARTNOINTR;
1338                 goto bad_fork_free_pid;
1339         }
1340
1341         if (clone_flags & CLONE_THREAD) {
1342                 current->signal->nr_threads++;
1343                 atomic_inc(&current->signal->live);
1344                 atomic_inc(&current->signal->sigcnt);
1345                 p->group_leader = current->group_leader;
1346                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1347         }
1348
1349         if (likely(p->pid)) {
1350                 tracehook_finish_clone(p, clone_flags, trace);
1351
1352                 if (thread_group_leader(p)) {
1353                         if (is_child_reaper(pid))
1354                                 p->nsproxy->pid_ns->child_reaper = p;
1355
1356                         p->signal->leader_pid = pid;
1357                         p->signal->tty = tty_kref_get(current->signal->tty);
1358                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1359                         attach_pid(p, PIDTYPE_SID, task_session(current));
1360                         list_add_tail(&p->sibling, &p->real_parent->children);
1361                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1362                         __this_cpu_inc(process_counts);
1363                 }
1364                 attach_pid(p, PIDTYPE_PID, pid);
1365                 nr_threads++;
1366         }
1367
1368         total_forks++;
1369         spin_unlock(&current->sighand->siglock);
1370         write_unlock_irq(&tasklist_lock);
1371         proc_fork_connector(p);
1372         cgroup_post_fork(p);
1373         if (clone_flags & CLONE_THREAD)
1374                 threadgroup_fork_read_unlock(current);
1375         perf_event_fork(p);
1376         return p;
1377
1378 bad_fork_free_pid:
1379         if (pid != &init_struct_pid)
1380                 free_pid(pid);
1381 bad_fork_cleanup_io:
1382         if (p->io_context)
1383                 exit_io_context(p);
1384 bad_fork_cleanup_namespaces:
1385         if (unlikely(clone_flags & CLONE_NEWPID))
1386                 pid_ns_release_proc(p->nsproxy->pid_ns);
1387         exit_task_namespaces(p);
1388 bad_fork_cleanup_mm:
1389         if (p->mm) {
1390                 task_lock(p);
1391                 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1392                         atomic_dec(&p->mm->oom_disable_count);
1393                 task_unlock(p);
1394                 mmput(p->mm);
1395         }
1396 bad_fork_cleanup_signal:
1397         if (!(clone_flags & CLONE_THREAD))
1398                 free_signal_struct(p->signal);
1399 bad_fork_cleanup_sighand:
1400         __cleanup_sighand(p->sighand);
1401 bad_fork_cleanup_fs:
1402         exit_fs(p); /* blocking */
1403 bad_fork_cleanup_files:
1404         exit_files(p); /* blocking */
1405 bad_fork_cleanup_semundo:
1406         exit_sem(p);
1407 bad_fork_cleanup_audit:
1408         audit_free(p);
1409 bad_fork_cleanup_policy:
1410         perf_event_free_task(p);
1411 #ifdef CONFIG_NUMA
1412         mpol_put(p->mempolicy);
1413 bad_fork_cleanup_cgroup:
1414 #endif
1415         if (clone_flags & CLONE_THREAD)
1416                 threadgroup_fork_read_unlock(current);
1417         cgroup_exit(p, cgroup_callbacks_done);
1418         delayacct_tsk_free(p);
1419         module_put(task_thread_info(p)->exec_domain->module);
1420 bad_fork_cleanup_count:
1421         atomic_dec(&p->cred->user->processes);
1422         exit_creds(p);
1423 bad_fork_free:
1424         free_task(p);
1425 fork_out:
1426         return ERR_PTR(retval);
1427 }
1428
1429 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1430 {
1431         memset(regs, 0, sizeof(struct pt_regs));
1432         return regs;
1433 }
1434
1435 static inline void init_idle_pids(struct pid_link *links)
1436 {
1437         enum pid_type type;
1438
1439         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1440                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1441                 links[type].pid = &init_struct_pid;
1442         }
1443 }
1444
1445 struct task_struct * __cpuinit fork_idle(int cpu)
1446 {
1447         struct task_struct *task;
1448         struct pt_regs regs;
1449
1450         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1451                             &init_struct_pid, 0);
1452         if (!IS_ERR(task)) {
1453                 init_idle_pids(task->pids);
1454                 init_idle(task, cpu);
1455         }
1456
1457         return task;
1458 }
1459
1460 /*
1461  *  Ok, this is the main fork-routine.
1462  *
1463  * It copies the process, and if successful kick-starts
1464  * it and waits for it to finish using the VM if required.
1465  */
1466 long do_fork(unsigned long clone_flags,
1467               unsigned long stack_start,
1468               struct pt_regs *regs,
1469               unsigned long stack_size,
1470               int __user *parent_tidptr,
1471               int __user *child_tidptr)
1472 {
1473         struct task_struct *p;
1474         int trace = 0;
1475         long nr;
1476
1477         /*
1478          * Do some preliminary argument and permissions checking before we
1479          * actually start allocating stuff
1480          */
1481         if (clone_flags & CLONE_NEWUSER) {
1482                 if (clone_flags & CLONE_THREAD)
1483                         return -EINVAL;
1484                 /* hopefully this check will go away when userns support is
1485                  * complete
1486                  */
1487                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1488                                 !capable(CAP_SETGID))
1489                         return -EPERM;
1490         }
1491
1492         /*
1493          * When called from kernel_thread, don't do user tracing stuff.
1494          */
1495         if (likely(user_mode(regs)))
1496                 trace = tracehook_prepare_clone(clone_flags);
1497
1498         p = copy_process(clone_flags, stack_start, regs, stack_size,
1499                          child_tidptr, NULL, trace);
1500         /*
1501          * Do this prior waking up the new thread - the thread pointer
1502          * might get invalid after that point, if the thread exits quickly.
1503          */
1504         if (!IS_ERR(p)) {
1505                 struct completion vfork;
1506
1507                 trace_sched_process_fork(current, p);
1508
1509                 nr = task_pid_vnr(p);
1510
1511                 if (clone_flags & CLONE_PARENT_SETTID)
1512                         put_user(nr, parent_tidptr);
1513
1514                 if (clone_flags & CLONE_VFORK) {
1515                         p->vfork_done = &vfork;
1516                         init_completion(&vfork);
1517                 }
1518
1519                 audit_finish_fork(p);
1520                 tracehook_report_clone(regs, clone_flags, nr, p);
1521
1522                 /*
1523                  * We set PF_STARTING at creation in case tracing wants to
1524                  * use this to distinguish a fully live task from one that
1525                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1526                  * clear it and set the child going.
1527                  */
1528                 p->flags &= ~PF_STARTING;
1529
1530                 wake_up_new_task(p);
1531
1532                 tracehook_report_clone_complete(trace, regs,
1533                                                 clone_flags, nr, p);
1534
1535                 if (clone_flags & CLONE_VFORK) {
1536                         freezer_do_not_count();
1537                         wait_for_completion(&vfork);
1538                         freezer_count();
1539                         tracehook_report_vfork_done(p, nr);
1540                 }
1541         } else {
1542                 nr = PTR_ERR(p);
1543         }
1544         return nr;
1545 }
1546
1547 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1548 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1549 #endif
1550
1551 static void sighand_ctor(void *data)
1552 {
1553         struct sighand_struct *sighand = data;
1554
1555         spin_lock_init(&sighand->siglock);
1556         init_waitqueue_head(&sighand->signalfd_wqh);
1557 }
1558
1559 void __init proc_caches_init(void)
1560 {
1561         sighand_cachep = kmem_cache_create("sighand_cache",
1562                         sizeof(struct sighand_struct), 0,
1563                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1564                         SLAB_NOTRACK, sighand_ctor);
1565         signal_cachep = kmem_cache_create("signal_cache",
1566                         sizeof(struct signal_struct), 0,
1567                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1568         files_cachep = kmem_cache_create("files_cache",
1569                         sizeof(struct files_struct), 0,
1570                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1571         fs_cachep = kmem_cache_create("fs_cache",
1572                         sizeof(struct fs_struct), 0,
1573                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1574         /*
1575          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1576          * whole struct cpumask for the OFFSTACK case. We could change
1577          * this to *only* allocate as much of it as required by the
1578          * maximum number of CPU's we can ever have.  The cpumask_allocation
1579          * is at the end of the structure, exactly for that reason.
1580          */
1581         mm_cachep = kmem_cache_create("mm_struct",
1582                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1583                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1584         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1585         mmap_init();
1586 }
1587
1588 /*
1589  * Check constraints on flags passed to the unshare system call.
1590  */
1591 static int check_unshare_flags(unsigned long unshare_flags)
1592 {
1593         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1594                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1595                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1596                 return -EINVAL;
1597         /*
1598          * Not implemented, but pretend it works if there is nothing to
1599          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1600          * needs to unshare vm.
1601          */
1602         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1603                 /* FIXME: get_task_mm() increments ->mm_users */
1604                 if (atomic_read(&current->mm->mm_users) > 1)
1605                         return -EINVAL;
1606         }
1607
1608         return 0;
1609 }
1610
1611 /*
1612  * Unshare the filesystem structure if it is being shared
1613  */
1614 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1615 {
1616         struct fs_struct *fs = current->fs;
1617
1618         if (!(unshare_flags & CLONE_FS) || !fs)
1619                 return 0;
1620
1621         /* don't need lock here; in the worst case we'll do useless copy */
1622         if (fs->users == 1)
1623                 return 0;
1624
1625         *new_fsp = copy_fs_struct(fs);
1626         if (!*new_fsp)
1627                 return -ENOMEM;
1628
1629         return 0;
1630 }
1631
1632 /*
1633  * Unshare file descriptor table if it is being shared
1634  */
1635 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1636 {
1637         struct files_struct *fd = current->files;
1638         int error = 0;
1639
1640         if ((unshare_flags & CLONE_FILES) &&
1641             (fd && atomic_read(&fd->count) > 1)) {
1642                 *new_fdp = dup_fd(fd, &error);
1643                 if (!*new_fdp)
1644                         return error;
1645         }
1646
1647         return 0;
1648 }
1649
1650 /*
1651  * unshare allows a process to 'unshare' part of the process
1652  * context which was originally shared using clone.  copy_*
1653  * functions used by do_fork() cannot be used here directly
1654  * because they modify an inactive task_struct that is being
1655  * constructed. Here we are modifying the current, active,
1656  * task_struct.
1657  */
1658 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1659 {
1660         struct fs_struct *fs, *new_fs = NULL;
1661         struct files_struct *fd, *new_fd = NULL;
1662         struct nsproxy *new_nsproxy = NULL;
1663         int do_sysvsem = 0;
1664         int err;
1665
1666         err = check_unshare_flags(unshare_flags);
1667         if (err)
1668                 goto bad_unshare_out;
1669
1670         /*
1671          * If unsharing namespace, must also unshare filesystem information.
1672          */
1673         if (unshare_flags & CLONE_NEWNS)
1674                 unshare_flags |= CLONE_FS;
1675         /*
1676          * CLONE_NEWIPC must also detach from the undolist: after switching
1677          * to a new ipc namespace, the semaphore arrays from the old
1678          * namespace are unreachable.
1679          */
1680         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1681                 do_sysvsem = 1;
1682         if ((err = unshare_fs(unshare_flags, &new_fs)))
1683                 goto bad_unshare_out;
1684         if ((err = unshare_fd(unshare_flags, &new_fd)))
1685                 goto bad_unshare_cleanup_fs;
1686         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1687                         new_fs)))
1688                 goto bad_unshare_cleanup_fd;
1689
1690         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1691                 if (do_sysvsem) {
1692                         /*
1693                          * CLONE_SYSVSEM is equivalent to sys_exit().
1694                          */
1695                         exit_sem(current);
1696                 }
1697
1698                 if (new_nsproxy) {
1699                         switch_task_namespaces(current, new_nsproxy);
1700                         new_nsproxy = NULL;
1701                 }
1702
1703                 task_lock(current);
1704
1705                 if (new_fs) {
1706                         fs = current->fs;
1707                         spin_lock(&fs->lock);
1708                         current->fs = new_fs;
1709                         if (--fs->users)
1710                                 new_fs = NULL;
1711                         else
1712                                 new_fs = fs;
1713                         spin_unlock(&fs->lock);
1714                 }
1715
1716                 if (new_fd) {
1717                         fd = current->files;
1718                         current->files = new_fd;
1719                         new_fd = fd;
1720                 }
1721
1722                 task_unlock(current);
1723         }
1724
1725         if (new_nsproxy)
1726                 put_nsproxy(new_nsproxy);
1727
1728 bad_unshare_cleanup_fd:
1729         if (new_fd)
1730                 put_files_struct(new_fd);
1731
1732 bad_unshare_cleanup_fs:
1733         if (new_fs)
1734                 free_fs_struct(new_fs);
1735
1736 bad_unshare_out:
1737         return err;
1738 }
1739
1740 /*
1741  *      Helper to unshare the files of the current task.
1742  *      We don't want to expose copy_files internals to
1743  *      the exec layer of the kernel.
1744  */
1745
1746 int unshare_files(struct files_struct **displaced)
1747 {
1748         struct task_struct *task = current;
1749         struct files_struct *copy = NULL;
1750         int error;
1751
1752         error = unshare_fd(CLONE_FILES, &copy);
1753         if (error || !copy) {
1754                 *displaced = NULL;
1755                 return error;
1756         }
1757         *displaced = task->files;
1758         task_lock(task);
1759         task->files = copy;
1760         task_unlock(task);
1761         return 0;
1762 }