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