kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77 #include <linux/sysctl.h>
  78 #include <linux/kcov.h>
  79
  80 #include <asm/pgtable.h>
  81 #include <asm/pgalloc.h>
  82 #include <asm/uaccess.h>
  83 #include <asm/mmu_context.h>
  84 #include <asm/cacheflush.h>
  85 #include <asm/tlbflush.h>
  86
  87 #include <trace/events/sched.h>
  88
  89 #define CREATE_TRACE_POINTS
  90 #include <trace/events/task.h>
  91
  92 /*
  93  * Minimum number of threads to boot the kernel
  94  */
  95 #define MIN_THREADS 20
  96
  97 /*
  98  * Maximum number of threads
  99  */
 100 #define MAX_THREADS FUTEX_TID_MASK
 101
 102 /*
 103  * Protected counters by write_lock_irq(&tasklist_lock)
 104  */
 105 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 106 int nr_threads;                 /* The idle threads do not count.. */
 107
 108 int max_threads;                /* tunable limit on nr_threads */
 109
 110 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 111
 112 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 113
 114 #ifdef CONFIG_PROVE_RCU
 115 int lockdep_tasklist_lock_is_held(void)
 116 {
 117         return lockdep_is_held(&tasklist_lock);
 118 }
 119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 120 #endif /* #ifdef CONFIG_PROVE_RCU */
 121
 122 int nr_processes(void)
 123 {
 124         int cpu;
 125         int total = 0;
 126
 127         for_each_possible_cpu(cpu)
 128                 total += per_cpu(process_counts, cpu);
 129
 130         return total;
 131 }
 132
 133 void __weak arch_release_task_struct(struct task_struct *tsk)
 134 {
 135 }
 136
 137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 138 static struct kmem_cache *task_struct_cachep;
 139
 140 static inline struct task_struct *alloc_task_struct_node(int node)
 141 {
 142         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 143 }
 144
 145 static inline void free_task_struct(struct task_struct *tsk)
 146 {
 147         kmem_cache_free(task_struct_cachep, tsk);
 148 }
 149 #endif
 150
 151 void __weak arch_release_thread_stack(unsigned long *stack)
 152 {
 153 }
 154
 155 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 156
 157 /*
 158  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 159  * kmemcache based allocator.
 160  */
 161 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 162 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 163 {
 164 #ifdef CONFIG_VMAP_STACK
 165         void *stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
 166                                            VMALLOC_START, VMALLOC_END,
 167                                            THREADINFO_GFP | __GFP_HIGHMEM,
 168                                            PAGE_KERNEL,
 169                                            0, node,
 170                                            __builtin_return_address(0));
 171
 172         /*
 173          * We can't call find_vm_area() in interrupt context, and
 174          * free_thread_stack() can be called in interrupt context,
 175          * so cache the vm_struct.
 176          */
 177         if (stack)
 178                 tsk->stack_vm_area = find_vm_area(stack);
 179         return stack;
 180 #else
 181         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 182                                              THREAD_SIZE_ORDER);
 183
 184         return page ? page_address(page) : NULL;
 185 #endif
 186 }
 187
 188 static inline void free_thread_stack(struct task_struct *tsk)
 189 {
 190         if (task_stack_vm_area(tsk))
 191                 vfree(tsk->stack);
 192         else
 193                 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 194 }
 195 # else
 196 static struct kmem_cache *thread_stack_cache;
 197
 198 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 199                                                   int node)
 200 {
 201         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 202 }
 203
 204 static void free_thread_stack(struct task_struct *tsk)
 205 {
 206         kmem_cache_free(thread_stack_cache, tsk->stack);
 207 }
 208
 209 void thread_stack_cache_init(void)
 210 {
 211         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 212                                               THREAD_SIZE, 0, NULL);
 213         BUG_ON(thread_stack_cache == NULL);
 214 }
 215 # endif
 216 #endif
 217
 218 /* SLAB cache for signal_struct structures (tsk->signal) */
 219 static struct kmem_cache *signal_cachep;
 220
 221 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 222 struct kmem_cache *sighand_cachep;
 223
 224 /* SLAB cache for files_struct structures (tsk->files) */
 225 struct kmem_cache *files_cachep;
 226
 227 /* SLAB cache for fs_struct structures (tsk->fs) */
 228 struct kmem_cache *fs_cachep;
 229
 230 /* SLAB cache for vm_area_struct structures */
 231 struct kmem_cache *vm_area_cachep;
 232
 233 /* SLAB cache for mm_struct structures (tsk->mm) */
 234 static struct kmem_cache *mm_cachep;
 235
 236 static void account_kernel_stack(struct task_struct *tsk, int account)
 237 {
 238         void *stack = task_stack_page(tsk);
 239         struct vm_struct *vm = task_stack_vm_area(tsk);
 240
 241         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 242
 243         if (vm) {
 244                 int i;
 245
 246                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 247
 248                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 249                         mod_zone_page_state(page_zone(vm->pages[i]),
 250                                             NR_KERNEL_STACK_KB,
 251                                             PAGE_SIZE / 1024 * account);
 252                 }
 253
 254                 /* All stack pages belong to the same memcg. */
 255                 memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 256                                             account * (THREAD_SIZE / 1024));
 257         } else {
 258                 /*
 259                  * All stack pages are in the same zone and belong to the
 260                  * same memcg.
 261                  */
 262                 struct page *first_page = virt_to_page(stack);
 263
 264                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 265                                     THREAD_SIZE / 1024 * account);
 266
 267                 memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
 268                                             account * (THREAD_SIZE / 1024));
 269         }
 270 }
 271
 272 static void release_task_stack(struct task_struct *tsk)
 273 {
 274         account_kernel_stack(tsk, -1);
 275         arch_release_thread_stack(tsk->stack);
 276         free_thread_stack(tsk);
 277         tsk->stack = NULL;
 278 #ifdef CONFIG_VMAP_STACK
 279         tsk->stack_vm_area = NULL;
 280 #endif
 281 }
 282
 283 #ifdef CONFIG_THREAD_INFO_IN_TASK
 284 void put_task_stack(struct task_struct *tsk)
 285 {
 286         if (atomic_dec_and_test(&tsk->stack_refcount))
 287                 release_task_stack(tsk);
 288 }
 289 #endif
 290
 291 void free_task(struct task_struct *tsk)
 292 {
 293 #ifndef CONFIG_THREAD_INFO_IN_TASK
 294         /*
 295          * The task is finally done with both the stack and thread_info,
 296          * so free both.
 297          */
 298         release_task_stack(tsk);
 299 #else
 300         /*
 301          * If the task had a separate stack allocation, it should be gone
 302          * by now.
 303          */
 304         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 305 #endif
 306         rt_mutex_debug_task_free(tsk);
 307         ftrace_graph_exit_task(tsk);
 308         put_seccomp_filter(tsk);
 309         arch_release_task_struct(tsk);
 310         free_task_struct(tsk);
 311 }
 312 EXPORT_SYMBOL(free_task);
 313
 314 static inline void free_signal_struct(struct signal_struct *sig)
 315 {
 316         taskstats_tgid_free(sig);
 317         sched_autogroup_exit(sig);
 318         kmem_cache_free(signal_cachep, sig);
 319 }
 320
 321 static inline void put_signal_struct(struct signal_struct *sig)
 322 {
 323         if (atomic_dec_and_test(&sig->sigcnt))
 324                 free_signal_struct(sig);
 325 }
 326
 327 void __put_task_struct(struct task_struct *tsk)
 328 {
 329         WARN_ON(!tsk->exit_state);
 330         WARN_ON(atomic_read(&tsk->usage));
 331         WARN_ON(tsk == current);
 332
 333         cgroup_free(tsk);
 334         task_numa_free(tsk);
 335         security_task_free(tsk);
 336         exit_creds(tsk);
 337         delayacct_tsk_free(tsk);
 338         put_signal_struct(tsk->signal);
 339
 340         if (!profile_handoff_task(tsk))
 341                 free_task(tsk);
 342 }
 343 EXPORT_SYMBOL_GPL(__put_task_struct);
 344
 345 void __init __weak arch_task_cache_init(void) { }
 346
 347 /*
 348  * set_max_threads
 349  */
 350 static void set_max_threads(unsigned int max_threads_suggested)
 351 {
 352         u64 threads;
 353
 354         /*
 355          * The number of threads shall be limited such that the thread
 356          * structures may only consume a small part of the available memory.
 357          */
 358         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 359                 threads = MAX_THREADS;
 360         else
 361                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 362                                     (u64) THREAD_SIZE * 8UL);
 363
 364         if (threads > max_threads_suggested)
 365                 threads = max_threads_suggested;
 366
 367         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 368 }
 369
 370 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 371 /* Initialized by the architecture: */
 372 int arch_task_struct_size __read_mostly;
 373 #endif
 374
 375 void __init fork_init(void)
 376 {
 377 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 378 #ifndef ARCH_MIN_TASKALIGN
 379 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 380 #endif
 381         /* create a slab on which task_structs can be allocated */
 382         task_struct_cachep = kmem_cache_create("task_struct",
 383                         arch_task_struct_size, ARCH_MIN_TASKALIGN,
 384                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 385 #endif
 386
 387         /* do the arch specific task caches init */
 388         arch_task_cache_init();
 389
 390         set_max_threads(MAX_THREADS);
 391
 392         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 393         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 394         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 395                 init_task.signal->rlim[RLIMIT_NPROC];
 396 }
 397
 398 int __weak arch_dup_task_struct(struct task_struct *dst,
 399                                                struct task_struct *src)
 400 {
 401         *dst = *src;
 402         return 0;
 403 }
 404
 405 void set_task_stack_end_magic(struct task_struct *tsk)
 406 {
 407         unsigned long *stackend;
 408
 409         stackend = end_of_stack(tsk);
 410         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 411 }
 412
 413 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 414 {
 415         struct task_struct *tsk;
 416         unsigned long *stack;
 417         struct vm_struct *stack_vm_area;
 418         int err;
 419
 420         if (node == NUMA_NO_NODE)
 421                 node = tsk_fork_get_node(orig);
 422         tsk = alloc_task_struct_node(node);
 423         if (!tsk)
 424                 return NULL;
 425
 426         stack = alloc_thread_stack_node(tsk, node);
 427         if (!stack)
 428                 goto free_tsk;
 429
 430         stack_vm_area = task_stack_vm_area(tsk);
 431
 432         err = arch_dup_task_struct(tsk, orig);
 433
 434         /*
 435          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 436          * sure they're properly initialized before using any stack-related
 437          * functions again.
 438          */
 439         tsk->stack = stack;
 440 #ifdef CONFIG_VMAP_STACK
 441         tsk->stack_vm_area = stack_vm_area;
 442 #endif
 443 #ifdef CONFIG_THREAD_INFO_IN_TASK
 444         atomic_set(&tsk->stack_refcount, 1);
 445 #endif
 446
 447         if (err)
 448                 goto free_stack;
 449
 450 #ifdef CONFIG_SECCOMP
 451         /*
 452          * We must handle setting up seccomp filters once we're under
 453          * the sighand lock in case orig has changed between now and
 454          * then. Until then, filter must be NULL to avoid messing up
 455          * the usage counts on the error path calling free_task.
 456          */
 457         tsk->seccomp.filter = NULL;
 458 #endif
 459
 460         setup_thread_stack(tsk, orig);
 461         clear_user_return_notifier(tsk);
 462         clear_tsk_need_resched(tsk);
 463         set_task_stack_end_magic(tsk);
 464
 465 #ifdef CONFIG_CC_STACKPROTECTOR
 466         tsk->stack_canary = get_random_int();
 467 #endif
 468
 469         /*
 470          * One for us, one for whoever does the "release_task()" (usually
 471          * parent)
 472          */
 473         atomic_set(&tsk->usage, 2);
 474 #ifdef CONFIG_BLK_DEV_IO_TRACE
 475         tsk->btrace_seq = 0;
 476 #endif
 477         tsk->splice_pipe = NULL;
 478         tsk->task_frag.page = NULL;
 479         tsk->wake_q.next = NULL;
 480
 481         account_kernel_stack(tsk, 1);
 482
 483         kcov_task_init(tsk);
 484
 485         return tsk;
 486
 487 free_stack:
 488         free_thread_stack(tsk);
 489 free_tsk:
 490         free_task_struct(tsk);
 491         return NULL;
 492 }
 493
 494 #ifdef CONFIG_MMU
 495 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 496 {
 497         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 498         struct rb_node **rb_link, *rb_parent;
 499         int retval;
 500         unsigned long charge;
 501
 502         uprobe_start_dup_mmap();
 503         if (down_write_killable(&oldmm->mmap_sem)) {
 504                 retval = -EINTR;
 505                 goto fail_uprobe_end;
 506         }
 507         flush_cache_dup_mm(oldmm);
 508         uprobe_dup_mmap(oldmm, mm);
 509         /*
 510          * Not linked in yet - no deadlock potential:
 511          */
 512         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 513
 514         /* No ordering required: file already has been exposed. */
 515         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 516
 517         mm->total_vm = oldmm->total_vm;
 518         mm->data_vm = oldmm->data_vm;
 519         mm->exec_vm = oldmm->exec_vm;
 520         mm->stack_vm = oldmm->stack_vm;
 521
 522         rb_link = &mm->mm_rb.rb_node;
 523         rb_parent = NULL;
 524         pprev = &mm->mmap;
 525         retval = ksm_fork(mm, oldmm);
 526         if (retval)
 527                 goto out;
 528         retval = khugepaged_fork(mm, oldmm);
 529         if (retval)
 530                 goto out;
 531
 532         prev = NULL;
 533         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 534                 struct file *file;
 535
 536                 if (mpnt->vm_flags & VM_DONTCOPY) {
 537                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 538                         continue;
 539                 }
 540                 charge = 0;
 541                 if (mpnt->vm_flags & VM_ACCOUNT) {
 542                         unsigned long len = vma_pages(mpnt);
 543
 544                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 545                                 goto fail_nomem;
 546                         charge = len;
 547                 }
 548                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 549                 if (!tmp)
 550                         goto fail_nomem;
 551                 *tmp = *mpnt;
 552                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 553                 retval = vma_dup_policy(mpnt, tmp);
 554                 if (retval)
 555                         goto fail_nomem_policy;
 556                 tmp->vm_mm = mm;
 557                 if (anon_vma_fork(tmp, mpnt))
 558                         goto fail_nomem_anon_vma_fork;
 559                 tmp->vm_flags &=
 560                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 561                 tmp->vm_next = tmp->vm_prev = NULL;
 562                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 563                 file = tmp->vm_file;
 564                 if (file) {
 565                         struct inode *inode = file_inode(file);
 566                         struct address_space *mapping = file->f_mapping;
 567
 568                         get_file(file);
 569                         if (tmp->vm_flags & VM_DENYWRITE)
 570                                 atomic_dec(&inode->i_writecount);
 571                         i_mmap_lock_write(mapping);
 572                         if (tmp->vm_flags & VM_SHARED)
 573                                 atomic_inc(&mapping->i_mmap_writable);
 574                         flush_dcache_mmap_lock(mapping);
 575                         /* insert tmp into the share list, just after mpnt */
 576                         vma_interval_tree_insert_after(tmp, mpnt,
 577                                         &mapping->i_mmap);
 578                         flush_dcache_mmap_unlock(mapping);
 579                         i_mmap_unlock_write(mapping);
 580                 }
 581
 582                 /*
 583                  * Clear hugetlb-related page reserves for children. This only
 584                  * affects MAP_PRIVATE mappings. Faults generated by the child
 585                  * are not guaranteed to succeed, even if read-only
 586                  */
 587                 if (is_vm_hugetlb_page(tmp))
 588                         reset_vma_resv_huge_pages(tmp);
 589
 590                 /*
 591                  * Link in the new vma and copy the page table entries.
 592                  */
 593                 *pprev = tmp;
 594                 pprev = &tmp->vm_next;
 595                 tmp->vm_prev = prev;
 596                 prev = tmp;
 597
 598                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 599                 rb_link = &tmp->vm_rb.rb_right;
 600                 rb_parent = &tmp->vm_rb;
 601
 602                 mm->map_count++;
 603                 retval = copy_page_range(mm, oldmm, mpnt);
 604
 605                 if (tmp->vm_ops && tmp->vm_ops->open)
 606                         tmp->vm_ops->open(tmp);
 607
 608                 if (retval)
 609                         goto out;
 610         }
 611         /* a new mm has just been created */
 612         arch_dup_mmap(oldmm, mm);
 613         retval = 0;
 614 out:
 615         up_write(&mm->mmap_sem);
 616         flush_tlb_mm(oldmm);
 617         up_write(&oldmm->mmap_sem);
 618 fail_uprobe_end:
 619         uprobe_end_dup_mmap();
 620         return retval;
 621 fail_nomem_anon_vma_fork:
 622         mpol_put(vma_policy(tmp));
 623 fail_nomem_policy:
 624         kmem_cache_free(vm_area_cachep, tmp);
 625 fail_nomem:
 626         retval = -ENOMEM;
 627         vm_unacct_memory(charge);
 628         goto out;
 629 }
 630
 631 static inline int mm_alloc_pgd(struct mm_struct *mm)
 632 {
 633         mm->pgd = pgd_alloc(mm);
 634         if (unlikely(!mm->pgd))
 635                 return -ENOMEM;
 636         return 0;
 637 }
 638
 639 static inline void mm_free_pgd(struct mm_struct *mm)
 640 {
 641         pgd_free(mm, mm->pgd);
 642 }
 643 #else
 644 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 645 {
 646         down_write(&oldmm->mmap_sem);
 647         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 648         up_write(&oldmm->mmap_sem);
 649         return 0;
 650 }
 651 #define mm_alloc_pgd(mm)        (0)
 652 #define mm_free_pgd(mm)
 653 #endif /* CONFIG_MMU */
 654
 655 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 656
 657 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 658 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 659
 660 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 661
 662 static int __init coredump_filter_setup(char *s)
 663 {
 664         default_dump_filter =
 665                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 666                 MMF_DUMP_FILTER_MASK;
 667         return 1;
 668 }
 669
 670 __setup("coredump_filter=", coredump_filter_setup);
 671
 672 #include <linux/init_task.h>
 673
 674 static void mm_init_aio(struct mm_struct *mm)
 675 {
 676 #ifdef CONFIG_AIO
 677         spin_lock_init(&mm->ioctx_lock);
 678         mm->ioctx_table = NULL;
 679 #endif
 680 }
 681
 682 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 683 {
 684 #ifdef CONFIG_MEMCG
 685         mm->owner = p;
 686 #endif
 687 }
 688
 689 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 690 {
 691         mm->mmap = NULL;
 692         mm->mm_rb = RB_ROOT;
 693         mm->vmacache_seqnum = 0;
 694         atomic_set(&mm->mm_users, 1);
 695         atomic_set(&mm->mm_count, 1);
 696         init_rwsem(&mm->mmap_sem);
 697         INIT_LIST_HEAD(&mm->mmlist);
 698         mm->core_state = NULL;
 699         atomic_long_set(&mm->nr_ptes, 0);
 700         mm_nr_pmds_init(mm);
 701         mm->map_count = 0;
 702         mm->locked_vm = 0;
 703         mm->pinned_vm = 0;
 704         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 705         spin_lock_init(&mm->page_table_lock);
 706         mm_init_cpumask(mm);
 707         mm_init_aio(mm);
 708         mm_init_owner(mm, p);
 709         mmu_notifier_mm_init(mm);
 710         clear_tlb_flush_pending(mm);
 711 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 712         mm->pmd_huge_pte = NULL;
 713 #endif
 714
 715         if (current->mm) {
 716                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 717                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 718         } else {
 719                 mm->flags = default_dump_filter;
 720                 mm->def_flags = 0;
 721         }
 722
 723         if (mm_alloc_pgd(mm))
 724                 goto fail_nopgd;
 725
 726         if (init_new_context(p, mm))
 727                 goto fail_nocontext;
 728
 729         return mm;
 730
 731 fail_nocontext:
 732         mm_free_pgd(mm);
 733 fail_nopgd:
 734         free_mm(mm);
 735         return NULL;
 736 }
 737
 738 static void check_mm(struct mm_struct *mm)
 739 {
 740         int i;
 741
 742         for (i = 0; i < NR_MM_COUNTERS; i++) {
 743                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 744
 745                 if (unlikely(x))
 746                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 747                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 748         }
 749
 750         if (atomic_long_read(&mm->nr_ptes))
 751                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 752                                 atomic_long_read(&mm->nr_ptes));
 753         if (mm_nr_pmds(mm))
 754                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 755                                 mm_nr_pmds(mm));
 756
 757 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 758         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 759 #endif
 760 }
 761
 762 /*
 763  * Allocate and initialize an mm_struct.
 764  */
 765 struct mm_struct *mm_alloc(void)
 766 {
 767         struct mm_struct *mm;
 768
 769         mm = allocate_mm();
 770         if (!mm)
 771                 return NULL;
 772
 773         memset(mm, 0, sizeof(*mm));
 774         return mm_init(mm, current);
 775 }
 776
 777 /*
 778  * Called when the last reference to the mm
 779  * is dropped: either by a lazy thread or by
 780  * mmput. Free the page directory and the mm.
 781  */
 782 void __mmdrop(struct mm_struct *mm)
 783 {
 784         BUG_ON(mm == &init_mm);
 785         mm_free_pgd(mm);
 786         destroy_context(mm);
 787         mmu_notifier_mm_destroy(mm);
 788         check_mm(mm);
 789         free_mm(mm);
 790 }
 791 EXPORT_SYMBOL_GPL(__mmdrop);
 792
 793 static inline void __mmput(struct mm_struct *mm)
 794 {
 795         VM_BUG_ON(atomic_read(&mm->mm_users));
 796
 797         uprobe_clear_state(mm);
 798         exit_aio(mm);
 799         ksm_exit(mm);
 800         khugepaged_exit(mm); /* must run before exit_mmap */
 801         exit_mmap(mm);
 802         set_mm_exe_file(mm, NULL);
 803         if (!list_empty(&mm->mmlist)) {
 804                 spin_lock(&mmlist_lock);
 805                 list_del(&mm->mmlist);
 806                 spin_unlock(&mmlist_lock);
 807         }
 808         if (mm->binfmt)
 809                 module_put(mm->binfmt->module);
 810         mmdrop(mm);
 811 }
 812
 813 /*
 814  * Decrement the use count and release all resources for an mm.
 815  */
 816 void mmput(struct mm_struct *mm)
 817 {
 818         might_sleep();
 819
 820         if (atomic_dec_and_test(&mm->mm_users))
 821                 __mmput(mm);
 822 }
 823 EXPORT_SYMBOL_GPL(mmput);
 824
 825 #ifdef CONFIG_MMU
 826 static void mmput_async_fn(struct work_struct *work)
 827 {
 828         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 829         __mmput(mm);
 830 }
 831
 832 void mmput_async(struct mm_struct *mm)
 833 {
 834         if (atomic_dec_and_test(&mm->mm_users)) {
 835                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 836                 schedule_work(&mm->async_put_work);
 837         }
 838 }
 839 #endif
 840
 841 /**
 842  * set_mm_exe_file - change a reference to the mm's executable file
 843  *
 844  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 845  *
 846  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 847  * invocations: in mmput() nobody alive left, in execve task is single
 848  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 849  * mm->exe_file, but does so without using set_mm_exe_file() in order
 850  * to do avoid the need for any locks.
 851  */
 852 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 853 {
 854         struct file *old_exe_file;
 855
 856         /*
 857          * It is safe to dereference the exe_file without RCU as
 858          * this function is only called if nobody else can access
 859          * this mm -- see comment above for justification.
 860          */
 861         old_exe_file = rcu_dereference_raw(mm->exe_file);
 862
 863         if (new_exe_file)
 864                 get_file(new_exe_file);
 865         rcu_assign_pointer(mm->exe_file, new_exe_file);
 866         if (old_exe_file)
 867                 fput(old_exe_file);
 868 }
 869
 870 /**
 871  * get_mm_exe_file - acquire a reference to the mm's executable file
 872  *
 873  * Returns %NULL if mm has no associated executable file.
 874  * User must release file via fput().
 875  */
 876 struct file *get_mm_exe_file(struct mm_struct *mm)
 877 {
 878         struct file *exe_file;
 879
 880         rcu_read_lock();
 881         exe_file = rcu_dereference(mm->exe_file);
 882         if (exe_file && !get_file_rcu(exe_file))
 883                 exe_file = NULL;
 884         rcu_read_unlock();
 885         return exe_file;
 886 }
 887 EXPORT_SYMBOL(get_mm_exe_file);
 888
 889 /**
 890  * get_task_exe_file - acquire a reference to the task's executable file
 891  *
 892  * Returns %NULL if task's mm (if any) has no associated executable file or
 893  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 894  * User must release file via fput().
 895  */
 896 struct file *get_task_exe_file(struct task_struct *task)
 897 {
 898         struct file *exe_file = NULL;
 899         struct mm_struct *mm;
 900
 901         task_lock(task);
 902         mm = task->mm;
 903         if (mm) {
 904                 if (!(task->flags & PF_KTHREAD))
 905                         exe_file = get_mm_exe_file(mm);
 906         }
 907         task_unlock(task);
 908         return exe_file;
 909 }
 910 EXPORT_SYMBOL(get_task_exe_file);
 911
 912 /**
 913  * get_task_mm - acquire a reference to the task's mm
 914  *
 915  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 916  * this kernel workthread has transiently adopted a user mm with use_mm,
 917  * to do its AIO) is not set and if so returns a reference to it, after
 918  * bumping up the use count.  User must release the mm via mmput()
 919  * after use.  Typically used by /proc and ptrace.
 920  */
 921 struct mm_struct *get_task_mm(struct task_struct *task)
 922 {
 923         struct mm_struct *mm;
 924
 925         task_lock(task);
 926         mm = task->mm;
 927         if (mm) {
 928                 if (task->flags & PF_KTHREAD)
 929                         mm = NULL;
 930                 else
 931                         atomic_inc(&mm->mm_users);
 932         }
 933         task_unlock(task);
 934         return mm;
 935 }
 936 EXPORT_SYMBOL_GPL(get_task_mm);
 937
 938 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 939 {
 940         struct mm_struct *mm;
 941         int err;
 942
 943         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 944         if (err)
 945                 return ERR_PTR(err);
 946
 947         mm = get_task_mm(task);
 948         if (mm && mm != current->mm &&
 949                         !ptrace_may_access(task, mode)) {
 950                 mmput(mm);
 951                 mm = ERR_PTR(-EACCES);
 952         }
 953         mutex_unlock(&task->signal->cred_guard_mutex);
 954
 955         return mm;
 956 }
 957
 958 static void complete_vfork_done(struct task_struct *tsk)
 959 {
 960         struct completion *vfork;
 961
 962         task_lock(tsk);
 963         vfork = tsk->vfork_done;
 964         if (likely(vfork)) {
 965                 tsk->vfork_done = NULL;
 966                 complete(vfork);
 967         }
 968         task_unlock(tsk);
 969 }
 970
 971 static int wait_for_vfork_done(struct task_struct *child,
 972                                 struct completion *vfork)
 973 {
 974         int killed;
 975
 976         freezer_do_not_count();
 977         killed = wait_for_completion_killable(vfork);
 978         freezer_count();
 979
 980         if (killed) {
 981                 task_lock(child);
 982                 child->vfork_done = NULL;
 983                 task_unlock(child);
 984         }
 985
 986         put_task_struct(child);
 987         return killed;
 988 }
 989
 990 /* Please note the differences between mmput and mm_release.
 991  * mmput is called whenever we stop holding onto a mm_struct,
 992  * error success whatever.
 993  *
 994  * mm_release is called after a mm_struct has been removed
 995  * from the current process.
 996  *
 997  * This difference is important for error handling, when we
 998  * only half set up a mm_struct for a new process and need to restore
 999  * the old one.  Because we mmput the new mm_struct before
1000  * restoring the old one. . .
1001  * Eric Biederman 10 January 1998
1002  */
1003 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1004 {
1005         /* Get rid of any futexes when releasing the mm */
1006 #ifdef CONFIG_FUTEX
1007         if (unlikely(tsk->robust_list)) {
1008                 exit_robust_list(tsk);
1009                 tsk->robust_list = NULL;
1010         }
1011 #ifdef CONFIG_COMPAT
1012         if (unlikely(tsk->compat_robust_list)) {
1013                 compat_exit_robust_list(tsk);
1014                 tsk->compat_robust_list = NULL;
1015         }
1016 #endif
1017         if (unlikely(!list_empty(&tsk->pi_state_list)))
1018                 exit_pi_state_list(tsk);
1019 #endif
1020
1021         uprobe_free_utask(tsk);
1022
1023         /* Get rid of any cached register state */
1024         deactivate_mm(tsk, mm);
1025
1026         /*
1027          * Signal userspace if we're not exiting with a core dump
1028          * because we want to leave the value intact for debugging
1029          * purposes.
1030          */
1031         if (tsk->clear_child_tid) {
1032                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1033                     atomic_read(&mm->mm_users) > 1) {
1034                         /*
1035                          * We don't check the error code - if userspace has
1036                          * not set up a proper pointer then tough luck.
1037                          */
1038                         put_user(0, tsk->clear_child_tid);
1039                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1040                                         1, NULL, NULL, 0);
1041                 }
1042                 tsk->clear_child_tid = NULL;
1043         }
1044
1045         /*
1046          * All done, finally we can wake up parent and return this mm to him.
1047          * Also kthread_stop() uses this completion for synchronization.
1048          */
1049         if (tsk->vfork_done)
1050                 complete_vfork_done(tsk);
1051 }
1052
1053 /*
1054  * Allocate a new mm structure and copy contents from the
1055  * mm structure of the passed in task structure.
1056  */
1057 static struct mm_struct *dup_mm(struct task_struct *tsk)
1058 {
1059         struct mm_struct *mm, *oldmm = current->mm;
1060         int err;
1061
1062         mm = allocate_mm();
1063         if (!mm)
1064                 goto fail_nomem;
1065
1066         memcpy(mm, oldmm, sizeof(*mm));
1067
1068         if (!mm_init(mm, tsk))
1069                 goto fail_nomem;
1070
1071         err = dup_mmap(mm, oldmm);
1072         if (err)
1073                 goto free_pt;
1074
1075         mm->hiwater_rss = get_mm_rss(mm);
1076         mm->hiwater_vm = mm->total_vm;
1077
1078         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1079                 goto free_pt;
1080
1081         return mm;
1082
1083 free_pt:
1084         /* don't put binfmt in mmput, we haven't got module yet */
1085         mm->binfmt = NULL;
1086         mmput(mm);
1087
1088 fail_nomem:
1089         return NULL;
1090 }
1091
1092 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1093 {
1094         struct mm_struct *mm, *oldmm;
1095         int retval;
1096
1097         tsk->min_flt = tsk->maj_flt = 0;
1098         tsk->nvcsw = tsk->nivcsw = 0;
1099 #ifdef CONFIG_DETECT_HUNG_TASK
1100         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1101 #endif
1102
1103         tsk->mm = NULL;
1104         tsk->active_mm = NULL;
1105
1106         /*
1107          * Are we cloning a kernel thread?
1108          *
1109          * We need to steal a active VM for that..
1110          */
1111         oldmm = current->mm;
1112         if (!oldmm)
1113                 return 0;
1114
1115         /* initialize the new vmacache entries */
1116         vmacache_flush(tsk);
1117
1118         if (clone_flags & CLONE_VM) {
1119                 atomic_inc(&oldmm->mm_users);
1120                 mm = oldmm;
1121                 goto good_mm;
1122         }
1123
1124         retval = -ENOMEM;
1125         mm = dup_mm(tsk);
1126         if (!mm)
1127                 goto fail_nomem;
1128
1129 good_mm:
1130         tsk->mm = mm;
1131         tsk->active_mm = mm;
1132         return 0;
1133
1134 fail_nomem:
1135         return retval;
1136 }
1137
1138 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1139 {
1140         struct fs_struct *fs = current->fs;
1141         if (clone_flags & CLONE_FS) {
1142                 /* tsk->fs is already what we want */
1143                 spin_lock(&fs->lock);
1144                 if (fs->in_exec) {
1145                         spin_unlock(&fs->lock);
1146                         return -EAGAIN;
1147                 }
1148                 fs->users++;
1149                 spin_unlock(&fs->lock);
1150                 return 0;
1151         }
1152         tsk->fs = copy_fs_struct(fs);
1153         if (!tsk->fs)
1154                 return -ENOMEM;
1155         return 0;
1156 }
1157
1158 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1159 {
1160         struct files_struct *oldf, *newf;
1161         int error = 0;
1162
1163         /*
1164          * A background process may not have any files ...
1165          */
1166         oldf = current->files;
1167         if (!oldf)
1168                 goto out;
1169
1170         if (clone_flags & CLONE_FILES) {
1171                 atomic_inc(&oldf->count);
1172                 goto out;
1173         }
1174
1175         newf = dup_fd(oldf, &error);
1176         if (!newf)
1177                 goto out;
1178
1179         tsk->files = newf;
1180         error = 0;
1181 out:
1182         return error;
1183 }
1184
1185 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1186 {
1187 #ifdef CONFIG_BLOCK
1188         struct io_context *ioc = current->io_context;
1189         struct io_context *new_ioc;
1190
1191         if (!ioc)
1192                 return 0;
1193         /*
1194          * Share io context with parent, if CLONE_IO is set
1195          */
1196         if (clone_flags & CLONE_IO) {
1197                 ioc_task_link(ioc);
1198                 tsk->io_context = ioc;
1199         } else if (ioprio_valid(ioc->ioprio)) {
1200                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1201                 if (unlikely(!new_ioc))
1202                         return -ENOMEM;
1203
1204                 new_ioc->ioprio = ioc->ioprio;
1205                 put_io_context(new_ioc);
1206         }
1207 #endif
1208         return 0;
1209 }
1210
1211 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1212 {
1213         struct sighand_struct *sig;
1214
1215         if (clone_flags & CLONE_SIGHAND) {
1216                 atomic_inc(&current->sighand->count);
1217                 return 0;
1218         }
1219         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1220         rcu_assign_pointer(tsk->sighand, sig);
1221         if (!sig)
1222                 return -ENOMEM;
1223
1224         atomic_set(&sig->count, 1);
1225         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1226         return 0;
1227 }
1228
1229 void __cleanup_sighand(struct sighand_struct *sighand)
1230 {
1231         if (atomic_dec_and_test(&sighand->count)) {
1232                 signalfd_cleanup(sighand);
1233                 /*
1234                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1235                  * without an RCU grace period, see __lock_task_sighand().
1236                  */
1237                 kmem_cache_free(sighand_cachep, sighand);
1238         }
1239 }
1240
1241 /*
1242  * Initialize POSIX timer handling for a thread group.
1243  */
1244 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1245 {
1246         unsigned long cpu_limit;
1247
1248         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1249         if (cpu_limit != RLIM_INFINITY) {
1250                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1251                 sig->cputimer.running = true;
1252         }
1253
1254         /* The timer lists. */
1255         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1256         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1257         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1258 }
1259
1260 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1261 {
1262         struct signal_struct *sig;
1263
1264         if (clone_flags & CLONE_THREAD)
1265                 return 0;
1266
1267         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1268         tsk->signal = sig;
1269         if (!sig)
1270                 return -ENOMEM;
1271
1272         sig->nr_threads = 1;
1273         atomic_set(&sig->live, 1);
1274         atomic_set(&sig->sigcnt, 1);
1275
1276         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1277         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1278         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1279
1280         init_waitqueue_head(&sig->wait_chldexit);
1281         sig->curr_target = tsk;
1282         init_sigpending(&sig->shared_pending);
1283         INIT_LIST_HEAD(&sig->posix_timers);
1284         seqlock_init(&sig->stats_lock);
1285         prev_cputime_init(&sig->prev_cputime);
1286
1287         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1288         sig->real_timer.function = it_real_fn;
1289
1290         task_lock(current->group_leader);
1291         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1292         task_unlock(current->group_leader);
1293
1294         posix_cpu_timers_init_group(sig);
1295
1296         tty_audit_fork(sig);
1297         sched_autogroup_fork(sig);
1298
1299         sig->oom_score_adj = current->signal->oom_score_adj;
1300         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1301
1302         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1303                                    current->signal->is_child_subreaper;
1304
1305         mutex_init(&sig->cred_guard_mutex);
1306
1307         return 0;
1308 }
1309
1310 static void copy_seccomp(struct task_struct *p)
1311 {
1312 #ifdef CONFIG_SECCOMP
1313         /*
1314          * Must be called with sighand->lock held, which is common to
1315          * all threads in the group. Holding cred_guard_mutex is not
1316          * needed because this new task is not yet running and cannot
1317          * be racing exec.
1318          */
1319         assert_spin_locked(&current->sighand->siglock);
1320
1321         /* Ref-count the new filter user, and assign it. */
1322         get_seccomp_filter(current);
1323         p->seccomp = current->seccomp;
1324
1325         /*
1326          * Explicitly enable no_new_privs here in case it got set
1327          * between the task_struct being duplicated and holding the
1328          * sighand lock. The seccomp state and nnp must be in sync.
1329          */
1330         if (task_no_new_privs(current))
1331                 task_set_no_new_privs(p);
1332
1333         /*
1334          * If the parent gained a seccomp mode after copying thread
1335          * flags and between before we held the sighand lock, we have
1336          * to manually enable the seccomp thread flag here.
1337          */
1338         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1339                 set_tsk_thread_flag(p, TIF_SECCOMP);
1340 #endif
1341 }
1342
1343 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1344 {
1345         current->clear_child_tid = tidptr;
1346
1347         return task_pid_vnr(current);
1348 }
1349
1350 static void rt_mutex_init_task(struct task_struct *p)
1351 {
1352         raw_spin_lock_init(&p->pi_lock);
1353 #ifdef CONFIG_RT_MUTEXES
1354         p->pi_waiters = RB_ROOT;
1355         p->pi_waiters_leftmost = NULL;
1356         p->pi_blocked_on = NULL;
1357 #endif
1358 }
1359
1360 /*
1361  * Initialize POSIX timer handling for a single task.
1362  */
1363 static void posix_cpu_timers_init(struct task_struct *tsk)
1364 {
1365         tsk->cputime_expires.prof_exp = 0;
1366         tsk->cputime_expires.virt_exp = 0;
1367         tsk->cputime_expires.sched_exp = 0;
1368         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1369         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1370         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1371 }
1372
1373 static inline void
1374 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1375 {
1376          task->pids[type].pid = pid;
1377 }
1378
1379 /*
1380  * This creates a new process as a copy of the old one,
1381  * but does not actually start it yet.
1382  *
1383  * It copies the registers, and all the appropriate
1384  * parts of the process environment (as per the clone
1385  * flags). The actual kick-off is left to the caller.
1386  */
1387 static struct task_struct *copy_process(unsigned long clone_flags,
1388                                         unsigned long stack_start,
1389                                         unsigned long stack_size,
1390                                         int __user *child_tidptr,
1391                                         struct pid *pid,
1392                                         int trace,
1393                                         unsigned long tls,
1394                                         int node)
1395 {
1396         int retval;
1397         struct task_struct *p;
1398
1399         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1400                 return ERR_PTR(-EINVAL);
1401
1402         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1403                 return ERR_PTR(-EINVAL);
1404
1405         /*
1406          * Thread groups must share signals as well, and detached threads
1407          * can only be started up within the thread group.
1408          */
1409         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1410                 return ERR_PTR(-EINVAL);
1411
1412         /*
1413          * Shared signal handlers imply shared VM. By way of the above,
1414          * thread groups also imply shared VM. Blocking this case allows
1415          * for various simplifications in other code.
1416          */
1417         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1418                 return ERR_PTR(-EINVAL);
1419
1420         /*
1421          * Siblings of global init remain as zombies on exit since they are
1422          * not reaped by their parent (swapper). To solve this and to avoid
1423          * multi-rooted process trees, prevent global and container-inits
1424          * from creating siblings.
1425          */
1426         if ((clone_flags & CLONE_PARENT) &&
1427                                 current->signal->flags & SIGNAL_UNKILLABLE)
1428                 return ERR_PTR(-EINVAL);
1429
1430         /*
1431          * If the new process will be in a different pid or user namespace
1432          * do not allow it to share a thread group with the forking task.
1433          */
1434         if (clone_flags & CLONE_THREAD) {
1435                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1436                     (task_active_pid_ns(current) !=
1437                                 current->nsproxy->pid_ns_for_children))
1438                         return ERR_PTR(-EINVAL);
1439         }
1440
1441         retval = security_task_create(clone_flags);
1442         if (retval)
1443                 goto fork_out;
1444
1445         retval = -ENOMEM;
1446         p = dup_task_struct(current, node);
1447         if (!p)
1448                 goto fork_out;
1449
1450         ftrace_graph_init_task(p);
1451
1452         rt_mutex_init_task(p);
1453
1454 #ifdef CONFIG_PROVE_LOCKING
1455         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1456         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1457 #endif
1458         retval = -EAGAIN;
1459         if (atomic_read(&p->real_cred->user->processes) >=
1460                         task_rlimit(p, RLIMIT_NPROC)) {
1461                 if (p->real_cred->user != INIT_USER &&
1462                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1463                         goto bad_fork_free;
1464         }
1465         current->flags &= ~PF_NPROC_EXCEEDED;
1466
1467         retval = copy_creds(p, clone_flags);
1468         if (retval < 0)
1469                 goto bad_fork_free;
1470
1471         /*
1472          * If multiple threads are within copy_process(), then this check
1473          * triggers too late. This doesn't hurt, the check is only there
1474          * to stop root fork bombs.
1475          */
1476         retval = -EAGAIN;
1477         if (nr_threads >= max_threads)
1478                 goto bad_fork_cleanup_count;
1479
1480         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1481         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1482         p->flags |= PF_FORKNOEXEC;
1483         INIT_LIST_HEAD(&p->children);
1484         INIT_LIST_HEAD(&p->sibling);
1485         rcu_copy_process(p);
1486         p->vfork_done = NULL;
1487         spin_lock_init(&p->alloc_lock);
1488
1489         init_sigpending(&p->pending);
1490
1491         p->utime = p->stime = p->gtime = 0;
1492         p->utimescaled = p->stimescaled = 0;
1493         prev_cputime_init(&p->prev_cputime);
1494
1495 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1496         seqcount_init(&p->vtime_seqcount);
1497         p->vtime_snap = 0;
1498         p->vtime_snap_whence = VTIME_INACTIVE;
1499 #endif
1500
1501 #if defined(SPLIT_RSS_COUNTING)
1502         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1503 #endif
1504
1505         p->default_timer_slack_ns = current->timer_slack_ns;
1506
1507         task_io_accounting_init(&p->ioac);
1508         acct_clear_integrals(p);
1509
1510         posix_cpu_timers_init(p);
1511
1512         p->start_time = ktime_get_ns();
1513         p->real_start_time = ktime_get_boot_ns();
1514         p->io_context = NULL;
1515         p->audit_context = NULL;
1516         cgroup_fork(p);
1517 #ifdef CONFIG_NUMA
1518         p->mempolicy = mpol_dup(p->mempolicy);
1519         if (IS_ERR(p->mempolicy)) {
1520                 retval = PTR_ERR(p->mempolicy);
1521                 p->mempolicy = NULL;
1522                 goto bad_fork_cleanup_threadgroup_lock;
1523         }
1524 #endif
1525 #ifdef CONFIG_CPUSETS
1526         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1527         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1528         seqcount_init(&p->mems_allowed_seq);
1529 #endif
1530 #ifdef CONFIG_TRACE_IRQFLAGS
1531         p->irq_events = 0;
1532         p->hardirqs_enabled = 0;
1533         p->hardirq_enable_ip = 0;
1534         p->hardirq_enable_event = 0;
1535         p->hardirq_disable_ip = _THIS_IP_;
1536         p->hardirq_disable_event = 0;
1537         p->softirqs_enabled = 1;
1538         p->softirq_enable_ip = _THIS_IP_;
1539         p->softirq_enable_event = 0;
1540         p->softirq_disable_ip = 0;
1541         p->softirq_disable_event = 0;
1542         p->hardirq_context = 0;
1543         p->softirq_context = 0;
1544 #endif
1545
1546         p->pagefault_disabled = 0;
1547
1548 #ifdef CONFIG_LOCKDEP
1549         p->lockdep_depth = 0; /* no locks held yet */
1550         p->curr_chain_key = 0;
1551         p->lockdep_recursion = 0;
1552 #endif
1553
1554 #ifdef CONFIG_DEBUG_MUTEXES
1555         p->blocked_on = NULL; /* not blocked yet */
1556 #endif
1557 #ifdef CONFIG_BCACHE
1558         p->sequential_io        = 0;
1559         p->sequential_io_avg    = 0;
1560 #endif
1561
1562         /* Perform scheduler related setup. Assign this task to a CPU. */
1563         retval = sched_fork(clone_flags, p);
1564         if (retval)
1565                 goto bad_fork_cleanup_policy;
1566
1567         retval = perf_event_init_task(p);
1568         if (retval)
1569                 goto bad_fork_cleanup_policy;
1570         retval = audit_alloc(p);
1571         if (retval)
1572                 goto bad_fork_cleanup_perf;
1573         /* copy all the process information */
1574         shm_init_task(p);
1575         retval = copy_semundo(clone_flags, p);
1576         if (retval)
1577                 goto bad_fork_cleanup_audit;
1578         retval = copy_files(clone_flags, p);
1579         if (retval)
1580                 goto bad_fork_cleanup_semundo;
1581         retval = copy_fs(clone_flags, p);
1582         if (retval)
1583                 goto bad_fork_cleanup_files;
1584         retval = copy_sighand(clone_flags, p);
1585         if (retval)
1586                 goto bad_fork_cleanup_fs;
1587         retval = copy_signal(clone_flags, p);
1588         if (retval)
1589                 goto bad_fork_cleanup_sighand;
1590         retval = copy_mm(clone_flags, p);
1591         if (retval)
1592                 goto bad_fork_cleanup_signal;
1593         retval = copy_namespaces(clone_flags, p);
1594         if (retval)
1595                 goto bad_fork_cleanup_mm;
1596         retval = copy_io(clone_flags, p);
1597         if (retval)
1598                 goto bad_fork_cleanup_namespaces;
1599         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1600         if (retval)
1601                 goto bad_fork_cleanup_io;
1602
1603         if (pid != &init_struct_pid) {
1604                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1605                 if (IS_ERR(pid)) {
1606                         retval = PTR_ERR(pid);
1607                         goto bad_fork_cleanup_thread;
1608                 }
1609         }
1610
1611         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1612         /*
1613          * Clear TID on mm_release()?
1614          */
1615         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1616 #ifdef CONFIG_BLOCK
1617         p->plug = NULL;
1618 #endif
1619 #ifdef CONFIG_FUTEX
1620         p->robust_list = NULL;
1621 #ifdef CONFIG_COMPAT
1622         p->compat_robust_list = NULL;
1623 #endif
1624         INIT_LIST_HEAD(&p->pi_state_list);
1625         p->pi_state_cache = NULL;
1626 #endif
1627         /*
1628          * sigaltstack should be cleared when sharing the same VM
1629          */
1630         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1631                 sas_ss_reset(p);
1632
1633         /*
1634          * Syscall tracing and stepping should be turned off in the
1635          * child regardless of CLONE_PTRACE.
1636          */
1637         user_disable_single_step(p);
1638         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1639 #ifdef TIF_SYSCALL_EMU
1640         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1641 #endif
1642         clear_all_latency_tracing(p);
1643
1644         /* ok, now we should be set up.. */
1645         p->pid = pid_nr(pid);
1646         if (clone_flags & CLONE_THREAD) {
1647                 p->exit_signal = -1;
1648                 p->group_leader = current->group_leader;
1649                 p->tgid = current->tgid;
1650         } else {
1651                 if (clone_flags & CLONE_PARENT)
1652                         p->exit_signal = current->group_leader->exit_signal;
1653                 else
1654                         p->exit_signal = (clone_flags & CSIGNAL);
1655                 p->group_leader = p;
1656                 p->tgid = p->pid;
1657         }
1658
1659         p->nr_dirtied = 0;
1660         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1661         p->dirty_paused_when = 0;
1662
1663         p->pdeath_signal = 0;
1664         INIT_LIST_HEAD(&p->thread_group);
1665         p->task_works = NULL;
1666
1667         threadgroup_change_begin(current);
1668         /*
1669          * Ensure that the cgroup subsystem policies allow the new process to be
1670          * forked. It should be noted the the new process's css_set can be changed
1671          * between here and cgroup_post_fork() if an organisation operation is in
1672          * progress.
1673          */
1674         retval = cgroup_can_fork(p);
1675         if (retval)
1676                 goto bad_fork_free_pid;
1677
1678         /*
1679          * Make it visible to the rest of the system, but dont wake it up yet.
1680          * Need tasklist lock for parent etc handling!
1681          */
1682         write_lock_irq(&tasklist_lock);
1683
1684         /* CLONE_PARENT re-uses the old parent */
1685         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1686                 p->real_parent = current->real_parent;
1687                 p->parent_exec_id = current->parent_exec_id;
1688         } else {
1689                 p->real_parent = current;
1690                 p->parent_exec_id = current->self_exec_id;
1691         }
1692
1693         spin_lock(&current->sighand->siglock);
1694
1695         /*
1696          * Copy seccomp details explicitly here, in case they were changed
1697          * before holding sighand lock.
1698          */
1699         copy_seccomp(p);
1700
1701         /*
1702          * Process group and session signals need to be delivered to just the
1703          * parent before the fork or both the parent and the child after the
1704          * fork. Restart if a signal comes in before we add the new process to
1705          * it's process group.
1706          * A fatal signal pending means that current will exit, so the new
1707          * thread can't slip out of an OOM kill (or normal SIGKILL).
1708         */
1709         recalc_sigpending();
1710         if (signal_pending(current)) {
1711                 spin_unlock(&current->sighand->siglock);
1712                 write_unlock_irq(&tasklist_lock);
1713                 retval = -ERESTARTNOINTR;
1714                 goto bad_fork_cancel_cgroup;
1715         }
1716
1717         if (likely(p->pid)) {
1718                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1719
1720                 init_task_pid(p, PIDTYPE_PID, pid);
1721                 if (thread_group_leader(p)) {
1722                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1723                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1724
1725                         if (is_child_reaper(pid)) {
1726                                 ns_of_pid(pid)->child_reaper = p;
1727                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1728                         }
1729
1730                         p->signal->leader_pid = pid;
1731                         p->signal->tty = tty_kref_get(current->signal->tty);
1732                         list_add_tail(&p->sibling, &p->real_parent->children);
1733                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1734                         attach_pid(p, PIDTYPE_PGID);
1735                         attach_pid(p, PIDTYPE_SID);
1736                         __this_cpu_inc(process_counts);
1737                 } else {
1738                         current->signal->nr_threads++;
1739                         atomic_inc(&current->signal->live);
1740                         atomic_inc(&current->signal->sigcnt);
1741                         list_add_tail_rcu(&p->thread_group,
1742                                           &p->group_leader->thread_group);
1743                         list_add_tail_rcu(&p->thread_node,
1744                                           &p->signal->thread_head);
1745                 }
1746                 attach_pid(p, PIDTYPE_PID);
1747                 nr_threads++;
1748         }
1749
1750         total_forks++;
1751         spin_unlock(&current->sighand->siglock);
1752         syscall_tracepoint_update(p);
1753         write_unlock_irq(&tasklist_lock);
1754
1755         proc_fork_connector(p);
1756         cgroup_post_fork(p);
1757         threadgroup_change_end(current);
1758         perf_event_fork(p);
1759
1760         trace_task_newtask(p, clone_flags);
1761         uprobe_copy_process(p, clone_flags);
1762
1763         return p;
1764
1765 bad_fork_cancel_cgroup:
1766         cgroup_cancel_fork(p);
1767 bad_fork_free_pid:
1768         threadgroup_change_end(current);
1769         if (pid != &init_struct_pid)
1770                 free_pid(pid);
1771 bad_fork_cleanup_thread:
1772         exit_thread(p);
1773 bad_fork_cleanup_io:
1774         if (p->io_context)
1775                 exit_io_context(p);
1776 bad_fork_cleanup_namespaces:
1777         exit_task_namespaces(p);
1778 bad_fork_cleanup_mm:
1779         if (p->mm)
1780                 mmput(p->mm);
1781 bad_fork_cleanup_signal:
1782         if (!(clone_flags & CLONE_THREAD))
1783                 free_signal_struct(p->signal);
1784 bad_fork_cleanup_sighand:
1785         __cleanup_sighand(p->sighand);
1786 bad_fork_cleanup_fs:
1787         exit_fs(p); /* blocking */
1788 bad_fork_cleanup_files:
1789         exit_files(p); /* blocking */
1790 bad_fork_cleanup_semundo:
1791         exit_sem(p);
1792 bad_fork_cleanup_audit:
1793         audit_free(p);
1794 bad_fork_cleanup_perf:
1795         perf_event_free_task(p);
1796 bad_fork_cleanup_policy:
1797 #ifdef CONFIG_NUMA
1798         mpol_put(p->mempolicy);
1799 bad_fork_cleanup_threadgroup_lock:
1800 #endif
1801         delayacct_tsk_free(p);
1802 bad_fork_cleanup_count:
1803         atomic_dec(&p->cred->user->processes);
1804         exit_creds(p);
1805 bad_fork_free:
1806         put_task_stack(p);
1807         free_task(p);
1808 fork_out:
1809         return ERR_PTR(retval);
1810 }
1811
1812 static inline void init_idle_pids(struct pid_link *links)
1813 {
1814         enum pid_type type;
1815
1816         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1817                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1818                 links[type].pid = &init_struct_pid;
1819         }
1820 }
1821
1822 struct task_struct *fork_idle(int cpu)
1823 {
1824         struct task_struct *task;
1825         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1826                             cpu_to_node(cpu));
1827         if (!IS_ERR(task)) {
1828                 init_idle_pids(task->pids);
1829                 init_idle(task, cpu);
1830         }
1831
1832         return task;
1833 }
1834
1835 /*
1836  *  Ok, this is the main fork-routine.
1837  *
1838  * It copies the process, and if successful kick-starts
1839  * it and waits for it to finish using the VM if required.
1840  */
1841 long _do_fork(unsigned long clone_flags,
1842               unsigned long stack_start,
1843               unsigned long stack_size,
1844               int __user *parent_tidptr,
1845               int __user *child_tidptr,
1846               unsigned long tls)
1847 {
1848         struct task_struct *p;
1849         int trace = 0;
1850         long nr;
1851
1852         /*
1853          * Determine whether and which event to report to ptracer.  When
1854          * called from kernel_thread or CLONE_UNTRACED is explicitly
1855          * requested, no event is reported; otherwise, report if the event
1856          * for the type of forking is enabled.
1857          */
1858         if (!(clone_flags & CLONE_UNTRACED)) {
1859                 if (clone_flags & CLONE_VFORK)
1860                         trace = PTRACE_EVENT_VFORK;
1861                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1862                         trace = PTRACE_EVENT_CLONE;
1863                 else
1864                         trace = PTRACE_EVENT_FORK;
1865
1866                 if (likely(!ptrace_event_enabled(current, trace)))
1867                         trace = 0;
1868         }
1869
1870         p = copy_process(clone_flags, stack_start, stack_size,
1871                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1872         /*
1873          * Do this prior waking up the new thread - the thread pointer
1874          * might get invalid after that point, if the thread exits quickly.
1875          */
1876         if (!IS_ERR(p)) {
1877                 struct completion vfork;
1878                 struct pid *pid;
1879
1880                 trace_sched_process_fork(current, p);
1881
1882                 pid = get_task_pid(p, PIDTYPE_PID);
1883                 nr = pid_vnr(pid);
1884
1885                 if (clone_flags & CLONE_PARENT_SETTID)
1886                         put_user(nr, parent_tidptr);
1887
1888                 if (clone_flags & CLONE_VFORK) {
1889                         p->vfork_done = &vfork;
1890                         init_completion(&vfork);
1891                         get_task_struct(p);
1892                 }
1893
1894                 wake_up_new_task(p);
1895
1896                 /* forking complete and child started to run, tell ptracer */
1897                 if (unlikely(trace))
1898                         ptrace_event_pid(trace, pid);
1899
1900                 if (clone_flags & CLONE_VFORK) {
1901                         if (!wait_for_vfork_done(p, &vfork))
1902                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1903                 }
1904
1905                 put_pid(pid);
1906         } else {
1907                 nr = PTR_ERR(p);
1908         }
1909         return nr;
1910 }
1911
1912 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1913 /* For compatibility with architectures that call do_fork directly rather than
1914  * using the syscall entry points below. */
1915 long do_fork(unsigned long clone_flags,
1916               unsigned long stack_start,
1917               unsigned long stack_size,
1918               int __user *parent_tidptr,
1919               int __user *child_tidptr)
1920 {
1921         return _do_fork(clone_flags, stack_start, stack_size,
1922                         parent_tidptr, child_tidptr, 0);
1923 }
1924 #endif
1925
1926 /*
1927  * Create a kernel thread.
1928  */
1929 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1930 {
1931         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1932                 (unsigned long)arg, NULL, NULL, 0);
1933 }
1934
1935 #ifdef __ARCH_WANT_SYS_FORK
1936 SYSCALL_DEFINE0(fork)
1937 {
1938 #ifdef CONFIG_MMU
1939         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1940 #else
1941         /* can not support in nommu mode */
1942         return -EINVAL;
1943 #endif
1944 }
1945 #endif
1946
1947 #ifdef __ARCH_WANT_SYS_VFORK
1948 SYSCALL_DEFINE0(vfork)
1949 {
1950         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1951                         0, NULL, NULL, 0);
1952 }
1953 #endif
1954
1955 #ifdef __ARCH_WANT_SYS_CLONE
1956 #ifdef CONFIG_CLONE_BACKWARDS
1957 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1958                  int __user *, parent_tidptr,
1959                  unsigned long, tls,
1960                  int __user *, child_tidptr)
1961 #elif defined(CONFIG_CLONE_BACKWARDS2)
1962 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1963                  int __user *, parent_tidptr,
1964                  int __user *, child_tidptr,
1965                  unsigned long, tls)
1966 #elif defined(CONFIG_CLONE_BACKWARDS3)
1967 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1968                 int, stack_size,
1969                 int __user *, parent_tidptr,
1970                 int __user *, child_tidptr,
1971                 unsigned long, tls)
1972 #else
1973 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1974                  int __user *, parent_tidptr,
1975                  int __user *, child_tidptr,
1976                  unsigned long, tls)
1977 #endif
1978 {
1979         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1980 }
1981 #endif
1982
1983 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1984 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1985 #endif
1986
1987 static void sighand_ctor(void *data)
1988 {
1989         struct sighand_struct *sighand = data;
1990
1991         spin_lock_init(&sighand->siglock);
1992         init_waitqueue_head(&sighand->signalfd_wqh);
1993 }
1994
1995 void __init proc_caches_init(void)
1996 {
1997         sighand_cachep = kmem_cache_create("sighand_cache",
1998                         sizeof(struct sighand_struct), 0,
1999                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2000                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2001         signal_cachep = kmem_cache_create("signal_cache",
2002                         sizeof(struct signal_struct), 0,
2003                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2004                         NULL);
2005         files_cachep = kmem_cache_create("files_cache",
2006                         sizeof(struct files_struct), 0,
2007                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2008                         NULL);
2009         fs_cachep = kmem_cache_create("fs_cache",
2010                         sizeof(struct fs_struct), 0,
2011                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2012                         NULL);
2013         /*
2014          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2015          * whole struct cpumask for the OFFSTACK case. We could change
2016          * this to *only* allocate as much of it as required by the
2017          * maximum number of CPU's we can ever have.  The cpumask_allocation
2018          * is at the end of the structure, exactly for that reason.
2019          */
2020         mm_cachep = kmem_cache_create("mm_struct",
2021                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2022                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2023                         NULL);
2024         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2025         mmap_init();
2026         nsproxy_cache_init();
2027 }
2028
2029 /*
2030  * Check constraints on flags passed to the unshare system call.
2031  */
2032 static int check_unshare_flags(unsigned long unshare_flags)
2033 {
2034         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2035                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2036                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2037                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2038                 return -EINVAL;
2039         /*
2040          * Not implemented, but pretend it works if there is nothing
2041          * to unshare.  Note that unsharing the address space or the
2042          * signal handlers also need to unshare the signal queues (aka
2043          * CLONE_THREAD).
2044          */
2045         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2046                 if (!thread_group_empty(current))
2047                         return -EINVAL;
2048         }
2049         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2050                 if (atomic_read(&current->sighand->count) > 1)
2051                         return -EINVAL;
2052         }
2053         if (unshare_flags & CLONE_VM) {
2054                 if (!current_is_single_threaded())
2055                         return -EINVAL;
2056         }
2057
2058         return 0;
2059 }
2060
2061 /*
2062  * Unshare the filesystem structure if it is being shared
2063  */
2064 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2065 {
2066         struct fs_struct *fs = current->fs;
2067
2068         if (!(unshare_flags & CLONE_FS) || !fs)
2069                 return 0;
2070
2071         /* don't need lock here; in the worst case we'll do useless copy */
2072         if (fs->users == 1)
2073                 return 0;
2074
2075         *new_fsp = copy_fs_struct(fs);
2076         if (!*new_fsp)
2077                 return -ENOMEM;
2078
2079         return 0;
2080 }
2081
2082 /*
2083  * Unshare file descriptor table if it is being shared
2084  */
2085 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2086 {
2087         struct files_struct *fd = current->files;
2088         int error = 0;
2089
2090         if ((unshare_flags & CLONE_FILES) &&
2091             (fd && atomic_read(&fd->count) > 1)) {
2092                 *new_fdp = dup_fd(fd, &error);
2093                 if (!*new_fdp)
2094                         return error;
2095         }
2096
2097         return 0;
2098 }
2099
2100 /*
2101  * unshare allows a process to 'unshare' part of the process
2102  * context which was originally shared using clone.  copy_*
2103  * functions used by do_fork() cannot be used here directly
2104  * because they modify an inactive task_struct that is being
2105  * constructed. Here we are modifying the current, active,
2106  * task_struct.
2107  */
2108 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2109 {
2110         struct fs_struct *fs, *new_fs = NULL;
2111         struct files_struct *fd, *new_fd = NULL;
2112         struct cred *new_cred = NULL;
2113         struct nsproxy *new_nsproxy = NULL;
2114         int do_sysvsem = 0;
2115         int err;
2116
2117         /*
2118          * If unsharing a user namespace must also unshare the thread group
2119          * and unshare the filesystem root and working directories.
2120          */
2121         if (unshare_flags & CLONE_NEWUSER)
2122                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2123         /*
2124          * If unsharing vm, must also unshare signal handlers.
2125          */
2126         if (unshare_flags & CLONE_VM)
2127                 unshare_flags |= CLONE_SIGHAND;
2128         /*
2129          * If unsharing a signal handlers, must also unshare the signal queues.
2130          */
2131         if (unshare_flags & CLONE_SIGHAND)
2132                 unshare_flags |= CLONE_THREAD;
2133         /*
2134          * If unsharing namespace, must also unshare filesystem information.
2135          */
2136         if (unshare_flags & CLONE_NEWNS)
2137                 unshare_flags |= CLONE_FS;
2138
2139         err = check_unshare_flags(unshare_flags);
2140         if (err)
2141                 goto bad_unshare_out;
2142         /*
2143          * CLONE_NEWIPC must also detach from the undolist: after switching
2144          * to a new ipc namespace, the semaphore arrays from the old
2145          * namespace are unreachable.
2146          */
2147         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2148                 do_sysvsem = 1;
2149         err = unshare_fs(unshare_flags, &new_fs);
2150         if (err)
2151                 goto bad_unshare_out;
2152         err = unshare_fd(unshare_flags, &new_fd);
2153         if (err)
2154                 goto bad_unshare_cleanup_fs;
2155         err = unshare_userns(unshare_flags, &new_cred);
2156         if (err)
2157                 goto bad_unshare_cleanup_fd;
2158         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2159                                          new_cred, new_fs);
2160         if (err)
2161                 goto bad_unshare_cleanup_cred;
2162
2163         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2164                 if (do_sysvsem) {
2165                         /*
2166                          * CLONE_SYSVSEM is equivalent to sys_exit().
2167                          */
2168                         exit_sem(current);
2169                 }
2170                 if (unshare_flags & CLONE_NEWIPC) {
2171                         /* Orphan segments in old ns (see sem above). */
2172                         exit_shm(current);
2173                         shm_init_task(current);
2174                 }
2175
2176                 if (new_nsproxy)
2177                         switch_task_namespaces(current, new_nsproxy);
2178
2179                 task_lock(current);
2180
2181                 if (new_fs) {
2182                         fs = current->fs;
2183                         spin_lock(&fs->lock);
2184                         current->fs = new_fs;
2185                         if (--fs->users)
2186                                 new_fs = NULL;
2187                         else
2188                                 new_fs = fs;
2189                         spin_unlock(&fs->lock);
2190                 }
2191
2192                 if (new_fd) {
2193                         fd = current->files;
2194                         current->files = new_fd;
2195                         new_fd = fd;
2196                 }
2197
2198                 task_unlock(current);
2199
2200                 if (new_cred) {
2201                         /* Install the new user namespace */
2202                         commit_creds(new_cred);
2203                         new_cred = NULL;
2204                 }
2205         }
2206
2207 bad_unshare_cleanup_cred:
2208         if (new_cred)
2209                 put_cred(new_cred);
2210 bad_unshare_cleanup_fd:
2211         if (new_fd)
2212                 put_files_struct(new_fd);
2213
2214 bad_unshare_cleanup_fs:
2215         if (new_fs)
2216                 free_fs_struct(new_fs);
2217
2218 bad_unshare_out:
2219         return err;
2220 }
2221
2222 /*
2223  *      Helper to unshare the files of the current task.
2224  *      We don't want to expose copy_files internals to
2225  *      the exec layer of the kernel.
2226  */
2227
2228 int unshare_files(struct files_struct **displaced)
2229 {
2230         struct task_struct *task = current;
2231         struct files_struct *copy = NULL;
2232         int error;
2233
2234         error = unshare_fd(CLONE_FILES, &copy);
2235         if (error || !copy) {
2236                 *displaced = NULL;
2237                 return error;
2238         }
2239         *displaced = task->files;
2240         task_lock(task);
2241         task->files = copy;
2242         task_unlock(task);
2243         return 0;
2244 }
2245
2246 int sysctl_max_threads(struct ctl_table *table, int write,
2247                        void __user *buffer, size_t *lenp, loff_t *ppos)
2248 {
2249         struct ctl_table t;
2250         int ret;
2251         int threads = max_threads;
2252         int min = MIN_THREADS;
2253         int max = MAX_THREADS;
2254
2255         t = *table;
2256         t.data = &threads;
2257         t.extra1 = &min;
2258         t.extra2 = &max;
2259
2260         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2261         if (ret || !write)
2262                 return ret;
2263
2264         set_max_threads(threads);
2265
2266         return 0;
2267 }