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/sched/autogroup.h>
  16 #include <linux/sched/mm.h>
  17 #include <linux/sched/coredump.h>
  18 #include <linux/sched/user.h>
  19 #include <linux/sched/numa_balancing.h>
  20 #include <linux/sched/stat.h>
  21 #include <linux/sched/task.h>
  22 #include <linux/sched/task_stack.h>
  23 #include <linux/sched/cputime.h>
  24 #include <linux/rtmutex.h>
  25 #include <linux/init.h>
  26 #include <linux/unistd.h>
  27 #include <linux/module.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/completion.h>
  30 #include <linux/personality.h>
  31 #include <linux/mempolicy.h>
  32 #include <linux/sem.h>
  33 #include <linux/file.h>
  34 #include <linux/fdtable.h>
  35 #include <linux/iocontext.h>
  36 #include <linux/key.h>
  37 #include <linux/binfmts.h>
  38 #include <linux/mman.h>
  39 #include <linux/mmu_notifier.h>
  40 #include <linux/hmm.h>
  41 #include <linux/fs.h>
  42 #include <linux/mm.h>
  43 #include <linux/vmacache.h>
  44 #include <linux/nsproxy.h>
  45 #include <linux/capability.h>
  46 #include <linux/cpu.h>
  47 #include <linux/cgroup.h>
  48 #include <linux/security.h>
  49 #include <linux/hugetlb.h>
  50 #include <linux/seccomp.h>
  51 #include <linux/swap.h>
  52 #include <linux/syscalls.h>
  53 #include <linux/jiffies.h>
  54 #include <linux/futex.h>
  55 #include <linux/compat.h>
  56 #include <linux/kthread.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/rcupdate.h>
  59 #include <linux/ptrace.h>
  60 #include <linux/mount.h>
  61 #include <linux/audit.h>
  62 #include <linux/memcontrol.h>
  63 #include <linux/ftrace.h>
  64 #include <linux/proc_fs.h>
  65 #include <linux/profile.h>
  66 #include <linux/rmap.h>
  67 #include <linux/ksm.h>
  68 #include <linux/acct.h>
  69 #include <linux/userfaultfd_k.h>
  70 #include <linux/tsacct_kern.h>
  71 #include <linux/cn_proc.h>
  72 #include <linux/freezer.h>
  73 #include <linux/delayacct.h>
  74 #include <linux/taskstats_kern.h>
  75 #include <linux/random.h>
  76 #include <linux/tty.h>
  77 #include <linux/blkdev.h>
  78 #include <linux/fs_struct.h>
  79 #include <linux/magic.h>
  80 #include <linux/perf_event.h>
  81 #include <linux/posix-timers.h>
  82 #include <linux/user-return-notifier.h>
  83 #include <linux/oom.h>
  84 #include <linux/khugepaged.h>
  85 #include <linux/signalfd.h>
  86 #include <linux/uprobes.h>
  87 #include <linux/aio.h>
  88 #include <linux/compiler.h>
  89 #include <linux/sysctl.h>
  90 #include <linux/kcov.h>
  91 #include <linux/livepatch.h>
  92 #include <linux/thread_info.h>
  93
  94 #include <asm/pgtable.h>
  95 #include <asm/pgalloc.h>
  96 #include <linux/uaccess.h>
  97 #include <asm/mmu_context.h>
  98 #include <asm/cacheflush.h>
  99 #include <asm/tlbflush.h>
 100
 101 #include <trace/events/sched.h>
 102
 103 #define CREATE_TRACE_POINTS
 104 #include <trace/events/task.h>
 105
 106 /*
 107  * Minimum number of threads to boot the kernel
 108  */
 109 #define MIN_THREADS 20
 110
 111 /*
 112  * Maximum number of threads
 113  */
 114 #define MAX_THREADS FUTEX_TID_MASK
 115
 116 /*
 117  * Protected counters by write_lock_irq(&tasklist_lock)
 118  */
 119 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 120 int nr_threads;                 /* The idle threads do not count.. */
 121
 122 int max_threads;                /* tunable limit on nr_threads */
 123
 124 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 125
 126 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 127
 128 #ifdef CONFIG_PROVE_RCU
 129 int lockdep_tasklist_lock_is_held(void)
 130 {
 131         return lockdep_is_held(&tasklist_lock);
 132 }
 133 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 134 #endif /* #ifdef CONFIG_PROVE_RCU */
 135
 136 int nr_processes(void)
 137 {
 138         int cpu;
 139         int total = 0;
 140
 141         for_each_possible_cpu(cpu)
 142                 total += per_cpu(process_counts, cpu);
 143
 144         return total;
 145 }
 146
 147 void __weak arch_release_task_struct(struct task_struct *tsk)
 148 {
 149 }
 150
 151 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 152 static struct kmem_cache *task_struct_cachep;
 153
 154 static inline struct task_struct *alloc_task_struct_node(int node)
 155 {
 156         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 157 }
 158
 159 static inline void free_task_struct(struct task_struct *tsk)
 160 {
 161         kmem_cache_free(task_struct_cachep, tsk);
 162 }
 163 #endif
 164
 165 void __weak arch_release_thread_stack(unsigned long *stack)
 166 {
 167 }
 168
 169 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 170
 171 /*
 172  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 173  * kmemcache based allocator.
 174  */
 175 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 176
 177 #ifdef CONFIG_VMAP_STACK
 178 /*
 179  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 180  * flush.  Try to minimize the number of calls by caching stacks.
 181  */
 182 #define NR_CACHED_STACKS 2
 183 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 184
 185 static int free_vm_stack_cache(unsigned int cpu)
 186 {
 187         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 188         int i;
 189
 190         for (i = 0; i < NR_CACHED_STACKS; i++) {
 191                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 192
 193                 if (!vm_stack)
 194                         continue;
 195
 196                 vfree(vm_stack->addr);
 197                 cached_vm_stacks[i] = NULL;
 198         }
 199
 200         return 0;
 201 }
 202 #endif
 203
 204 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 205 {
 206 #ifdef CONFIG_VMAP_STACK
 207         void *stack;
 208         int i;
 209
 210         for (i = 0; i < NR_CACHED_STACKS; i++) {
 211                 struct vm_struct *s;
 212
 213                 s = this_cpu_xchg(cached_stacks[i], NULL);
 214
 215                 if (!s)
 216                         continue;
 217
 218 #ifdef CONFIG_DEBUG_KMEMLEAK
 219                 /* Clear stale pointers from reused stack. */
 220                 memset(s->addr, 0, THREAD_SIZE);
 221 #endif
 222                 tsk->stack_vm_area = s;
 223                 return s->addr;
 224         }
 225
 226         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 227                                      VMALLOC_START, VMALLOC_END,
 228                                      THREADINFO_GFP,
 229                                      PAGE_KERNEL,
 230                                      0, node, __builtin_return_address(0));
 231
 232         /*
 233          * We can't call find_vm_area() in interrupt context, and
 234          * free_thread_stack() can be called in interrupt context,
 235          * so cache the vm_struct.
 236          */
 237         if (stack)
 238                 tsk->stack_vm_area = find_vm_area(stack);
 239         return stack;
 240 #else
 241         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 242                                              THREAD_SIZE_ORDER);
 243
 244         return page ? page_address(page) : NULL;
 245 #endif
 246 }
 247
 248 static inline void free_thread_stack(struct task_struct *tsk)
 249 {
 250 #ifdef CONFIG_VMAP_STACK
 251         if (task_stack_vm_area(tsk)) {
 252                 int i;
 253
 254                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 255                         if (this_cpu_cmpxchg(cached_stacks[i],
 256                                         NULL, tsk->stack_vm_area) != NULL)
 257                                 continue;
 258
 259                         return;
 260                 }
 261
 262                 vfree_atomic(tsk->stack);
 263                 return;
 264         }
 265 #endif
 266
 267         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 268 }
 269 # else
 270 static struct kmem_cache *thread_stack_cache;
 271
 272 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 273                                                   int node)
 274 {
 275         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 276 }
 277
 278 static void free_thread_stack(struct task_struct *tsk)
 279 {
 280         kmem_cache_free(thread_stack_cache, tsk->stack);
 281 }
 282
 283 void thread_stack_cache_init(void)
 284 {
 285         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 286                                               THREAD_SIZE, 0, NULL);
 287         BUG_ON(thread_stack_cache == NULL);
 288 }
 289 # endif
 290 #endif
 291
 292 /* SLAB cache for signal_struct structures (tsk->signal) */
 293 static struct kmem_cache *signal_cachep;
 294
 295 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 296 struct kmem_cache *sighand_cachep;
 297
 298 /* SLAB cache for files_struct structures (tsk->files) */
 299 struct kmem_cache *files_cachep;
 300
 301 /* SLAB cache for fs_struct structures (tsk->fs) */
 302 struct kmem_cache *fs_cachep;
 303
 304 /* SLAB cache for vm_area_struct structures */
 305 struct kmem_cache *vm_area_cachep;
 306
 307 /* SLAB cache for mm_struct structures (tsk->mm) */
 308 static struct kmem_cache *mm_cachep;
 309
 310 static void account_kernel_stack(struct task_struct *tsk, int account)
 311 {
 312         void *stack = task_stack_page(tsk);
 313         struct vm_struct *vm = task_stack_vm_area(tsk);
 314
 315         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 316
 317         if (vm) {
 318                 int i;
 319
 320                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 321
 322                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 323                         mod_zone_page_state(page_zone(vm->pages[i]),
 324                                             NR_KERNEL_STACK_KB,
 325                                             PAGE_SIZE / 1024 * account);
 326                 }
 327
 328                 /* All stack pages belong to the same memcg. */
 329                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 330                                      account * (THREAD_SIZE / 1024));
 331         } else {
 332                 /*
 333                  * All stack pages are in the same zone and belong to the
 334                  * same memcg.
 335                  */
 336                 struct page *first_page = virt_to_page(stack);
 337
 338                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 339                                     THREAD_SIZE / 1024 * account);
 340
 341                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 342                                      account * (THREAD_SIZE / 1024));
 343         }
 344 }
 345
 346 static void release_task_stack(struct task_struct *tsk)
 347 {
 348         if (WARN_ON(tsk->state != TASK_DEAD))
 349                 return;  /* Better to leak the stack than to free prematurely */
 350
 351         account_kernel_stack(tsk, -1);
 352         arch_release_thread_stack(tsk->stack);
 353         free_thread_stack(tsk);
 354         tsk->stack = NULL;
 355 #ifdef CONFIG_VMAP_STACK
 356         tsk->stack_vm_area = NULL;
 357 #endif
 358 }
 359
 360 #ifdef CONFIG_THREAD_INFO_IN_TASK
 361 void put_task_stack(struct task_struct *tsk)
 362 {
 363         if (atomic_dec_and_test(&tsk->stack_refcount))
 364                 release_task_stack(tsk);
 365 }
 366 #endif
 367
 368 void free_task(struct task_struct *tsk)
 369 {
 370 #ifndef CONFIG_THREAD_INFO_IN_TASK
 371         /*
 372          * The task is finally done with both the stack and thread_info,
 373          * so free both.
 374          */
 375         release_task_stack(tsk);
 376 #else
 377         /*
 378          * If the task had a separate stack allocation, it should be gone
 379          * by now.
 380          */
 381         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 382 #endif
 383         rt_mutex_debug_task_free(tsk);
 384         ftrace_graph_exit_task(tsk);
 385         put_seccomp_filter(tsk);
 386         arch_release_task_struct(tsk);
 387         if (tsk->flags & PF_KTHREAD)
 388                 free_kthread_struct(tsk);
 389         free_task_struct(tsk);
 390 }
 391 EXPORT_SYMBOL(free_task);
 392
 393 static inline void free_signal_struct(struct signal_struct *sig)
 394 {
 395         taskstats_tgid_free(sig);
 396         sched_autogroup_exit(sig);
 397         /*
 398          * __mmdrop is not safe to call from softirq context on x86 due to
 399          * pgd_dtor so postpone it to the async context
 400          */
 401         if (sig->oom_mm)
 402                 mmdrop_async(sig->oom_mm);
 403         kmem_cache_free(signal_cachep, sig);
 404 }
 405
 406 static inline void put_signal_struct(struct signal_struct *sig)
 407 {
 408         if (atomic_dec_and_test(&sig->sigcnt))
 409                 free_signal_struct(sig);
 410 }
 411
 412 void __put_task_struct(struct task_struct *tsk)
 413 {
 414         WARN_ON(!tsk->exit_state);
 415         WARN_ON(atomic_read(&tsk->usage));
 416         WARN_ON(tsk == current);
 417
 418         cgroup_free(tsk);
 419         task_numa_free(tsk);
 420         security_task_free(tsk);
 421         exit_creds(tsk);
 422         delayacct_tsk_free(tsk);
 423         put_signal_struct(tsk->signal);
 424
 425         if (!profile_handoff_task(tsk))
 426                 free_task(tsk);
 427 }
 428 EXPORT_SYMBOL_GPL(__put_task_struct);
 429
 430 void __init __weak arch_task_cache_init(void) { }
 431
 432 /*
 433  * set_max_threads
 434  */
 435 static void set_max_threads(unsigned int max_threads_suggested)
 436 {
 437         u64 threads;
 438
 439         /*
 440          * The number of threads shall be limited such that the thread
 441          * structures may only consume a small part of the available memory.
 442          */
 443         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 444                 threads = MAX_THREADS;
 445         else
 446                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 447                                     (u64) THREAD_SIZE * 8UL);
 448
 449         if (threads > max_threads_suggested)
 450                 threads = max_threads_suggested;
 451
 452         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 453 }
 454
 455 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 456 /* Initialized by the architecture: */
 457 int arch_task_struct_size __read_mostly;
 458 #endif
 459
 460 void __init fork_init(void)
 461 {
 462         int i;
 463 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 464 #ifndef ARCH_MIN_TASKALIGN
 465 #define ARCH_MIN_TASKALIGN      0
 466 #endif
 467         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 468
 469         /* create a slab on which task_structs can be allocated */
 470         task_struct_cachep = kmem_cache_create("task_struct",
 471                         arch_task_struct_size, align,
 472                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 473 #endif
 474
 475         /* do the arch specific task caches init */
 476         arch_task_cache_init();
 477
 478         set_max_threads(MAX_THREADS);
 479
 480         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 481         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 482         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 483                 init_task.signal->rlim[RLIMIT_NPROC];
 484
 485         for (i = 0; i < UCOUNT_COUNTS; i++) {
 486                 init_user_ns.ucount_max[i] = max_threads/2;
 487         }
 488
 489 #ifdef CONFIG_VMAP_STACK
 490         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 491                           NULL, free_vm_stack_cache);
 492 #endif
 493
 494         lockdep_init_task(&init_task);
 495 }
 496
 497 int __weak arch_dup_task_struct(struct task_struct *dst,
 498                                                struct task_struct *src)
 499 {
 500         *dst = *src;
 501         return 0;
 502 }
 503
 504 void set_task_stack_end_magic(struct task_struct *tsk)
 505 {
 506         unsigned long *stackend;
 507
 508         stackend = end_of_stack(tsk);
 509         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 510 }
 511
 512 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 513 {
 514         struct task_struct *tsk;
 515         unsigned long *stack;
 516         struct vm_struct *stack_vm_area;
 517         int err;
 518
 519         if (node == NUMA_NO_NODE)
 520                 node = tsk_fork_get_node(orig);
 521         tsk = alloc_task_struct_node(node);
 522         if (!tsk)
 523                 return NULL;
 524
 525         stack = alloc_thread_stack_node(tsk, node);
 526         if (!stack)
 527                 goto free_tsk;
 528
 529         stack_vm_area = task_stack_vm_area(tsk);
 530
 531         err = arch_dup_task_struct(tsk, orig);
 532
 533         /*
 534          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 535          * sure they're properly initialized before using any stack-related
 536          * functions again.
 537          */
 538         tsk->stack = stack;
 539 #ifdef CONFIG_VMAP_STACK
 540         tsk->stack_vm_area = stack_vm_area;
 541 #endif
 542 #ifdef CONFIG_THREAD_INFO_IN_TASK
 543         atomic_set(&tsk->stack_refcount, 1);
 544 #endif
 545
 546         if (err)
 547                 goto free_stack;
 548
 549 #ifdef CONFIG_SECCOMP
 550         /*
 551          * We must handle setting up seccomp filters once we're under
 552          * the sighand lock in case orig has changed between now and
 553          * then. Until then, filter must be NULL to avoid messing up
 554          * the usage counts on the error path calling free_task.
 555          */
 556         tsk->seccomp.filter = NULL;
 557 #endif
 558
 559         setup_thread_stack(tsk, orig);
 560         clear_user_return_notifier(tsk);
 561         clear_tsk_need_resched(tsk);
 562         set_task_stack_end_magic(tsk);
 563
 564 #ifdef CONFIG_CC_STACKPROTECTOR
 565         tsk->stack_canary = get_random_canary();
 566 #endif
 567
 568         /*
 569          * One for us, one for whoever does the "release_task()" (usually
 570          * parent)
 571          */
 572         atomic_set(&tsk->usage, 2);
 573 #ifdef CONFIG_BLK_DEV_IO_TRACE
 574         tsk->btrace_seq = 0;
 575 #endif
 576         tsk->splice_pipe = NULL;
 577         tsk->task_frag.page = NULL;
 578         tsk->wake_q.next = NULL;
 579
 580         account_kernel_stack(tsk, 1);
 581
 582         kcov_task_init(tsk);
 583
 584 #ifdef CONFIG_FAULT_INJECTION
 585         tsk->fail_nth = 0;
 586 #endif
 587
 588         return tsk;
 589
 590 free_stack:
 591         free_thread_stack(tsk);
 592 free_tsk:
 593         free_task_struct(tsk);
 594         return NULL;
 595 }
 596
 597 #ifdef CONFIG_MMU
 598 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 599                                         struct mm_struct *oldmm)
 600 {
 601         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 602         struct rb_node **rb_link, *rb_parent;
 603         int retval;
 604         unsigned long charge;
 605         LIST_HEAD(uf);
 606
 607         uprobe_start_dup_mmap();
 608         if (down_write_killable(&oldmm->mmap_sem)) {
 609                 retval = -EINTR;
 610                 goto fail_uprobe_end;
 611         }
 612         flush_cache_dup_mm(oldmm);
 613         uprobe_dup_mmap(oldmm, mm);
 614         /*
 615          * Not linked in yet - no deadlock potential:
 616          */
 617         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 618
 619         /* No ordering required: file already has been exposed. */
 620         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 621
 622         mm->total_vm = oldmm->total_vm;
 623         mm->data_vm = oldmm->data_vm;
 624         mm->exec_vm = oldmm->exec_vm;
 625         mm->stack_vm = oldmm->stack_vm;
 626
 627         rb_link = &mm->mm_rb.rb_node;
 628         rb_parent = NULL;
 629         pprev = &mm->mmap;
 630         retval = ksm_fork(mm, oldmm);
 631         if (retval)
 632                 goto out;
 633         retval = khugepaged_fork(mm, oldmm);
 634         if (retval)
 635                 goto out;
 636
 637         prev = NULL;
 638         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 639                 struct file *file;
 640
 641                 if (mpnt->vm_flags & VM_DONTCOPY) {
 642                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 643                         continue;
 644                 }
 645                 charge = 0;
 646                 if (mpnt->vm_flags & VM_ACCOUNT) {
 647                         unsigned long len = vma_pages(mpnt);
 648
 649                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 650                                 goto fail_nomem;
 651                         charge = len;
 652                 }
 653                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 654                 if (!tmp)
 655                         goto fail_nomem;
 656                 *tmp = *mpnt;
 657                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 658                 retval = vma_dup_policy(mpnt, tmp);
 659                 if (retval)
 660                         goto fail_nomem_policy;
 661                 tmp->vm_mm = mm;
 662                 retval = dup_userfaultfd(tmp, &uf);
 663                 if (retval)
 664                         goto fail_nomem_anon_vma_fork;
 665                 if (tmp->vm_flags & VM_WIPEONFORK) {
 666                         /* VM_WIPEONFORK gets a clean slate in the child. */
 667                         tmp->anon_vma = NULL;
 668                         if (anon_vma_prepare(tmp))
 669                                 goto fail_nomem_anon_vma_fork;
 670                 } else if (anon_vma_fork(tmp, mpnt))
 671                         goto fail_nomem_anon_vma_fork;
 672                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 673                 tmp->vm_next = tmp->vm_prev = NULL;
 674                 file = tmp->vm_file;
 675                 if (file) {
 676                         struct inode *inode = file_inode(file);
 677                         struct address_space *mapping = file->f_mapping;
 678
 679                         get_file(file);
 680                         if (tmp->vm_flags & VM_DENYWRITE)
 681                                 atomic_dec(&inode->i_writecount);
 682                         i_mmap_lock_write(mapping);
 683                         if (tmp->vm_flags & VM_SHARED)
 684                                 atomic_inc(&mapping->i_mmap_writable);
 685                         flush_dcache_mmap_lock(mapping);
 686                         /* insert tmp into the share list, just after mpnt */
 687                         vma_interval_tree_insert_after(tmp, mpnt,
 688                                         &mapping->i_mmap);
 689                         flush_dcache_mmap_unlock(mapping);
 690                         i_mmap_unlock_write(mapping);
 691                 }
 692
 693                 /*
 694                  * Clear hugetlb-related page reserves for children. This only
 695                  * affects MAP_PRIVATE mappings. Faults generated by the child
 696                  * are not guaranteed to succeed, even if read-only
 697                  */
 698                 if (is_vm_hugetlb_page(tmp))
 699                         reset_vma_resv_huge_pages(tmp);
 700
 701                 /*
 702                  * Link in the new vma and copy the page table entries.
 703                  */
 704                 *pprev = tmp;
 705                 pprev = &tmp->vm_next;
 706                 tmp->vm_prev = prev;
 707                 prev = tmp;
 708
 709                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 710                 rb_link = &tmp->vm_rb.rb_right;
 711                 rb_parent = &tmp->vm_rb;
 712
 713                 mm->map_count++;
 714                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 715                         retval = copy_page_range(mm, oldmm, mpnt);
 716
 717                 if (tmp->vm_ops && tmp->vm_ops->open)
 718                         tmp->vm_ops->open(tmp);
 719
 720                 if (retval)
 721                         goto out;
 722         }
 723         /* a new mm has just been created */
 724         arch_dup_mmap(oldmm, mm);
 725         retval = 0;
 726 out:
 727         up_write(&mm->mmap_sem);
 728         flush_tlb_mm(oldmm);
 729         up_write(&oldmm->mmap_sem);
 730         dup_userfaultfd_complete(&uf);
 731 fail_uprobe_end:
 732         uprobe_end_dup_mmap();
 733         return retval;
 734 fail_nomem_anon_vma_fork:
 735         mpol_put(vma_policy(tmp));
 736 fail_nomem_policy:
 737         kmem_cache_free(vm_area_cachep, tmp);
 738 fail_nomem:
 739         retval = -ENOMEM;
 740         vm_unacct_memory(charge);
 741         goto out;
 742 }
 743
 744 static inline int mm_alloc_pgd(struct mm_struct *mm)
 745 {
 746         mm->pgd = pgd_alloc(mm);
 747         if (unlikely(!mm->pgd))
 748                 return -ENOMEM;
 749         return 0;
 750 }
 751
 752 static inline void mm_free_pgd(struct mm_struct *mm)
 753 {
 754         pgd_free(mm, mm->pgd);
 755 }
 756 #else
 757 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 758 {
 759         down_write(&oldmm->mmap_sem);
 760         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 761         up_write(&oldmm->mmap_sem);
 762         return 0;
 763 }
 764 #define mm_alloc_pgd(mm)        (0)
 765 #define mm_free_pgd(mm)
 766 #endif /* CONFIG_MMU */
 767
 768 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 769
 770 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 771 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 772
 773 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 774
 775 static int __init coredump_filter_setup(char *s)
 776 {
 777         default_dump_filter =
 778                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 779                 MMF_DUMP_FILTER_MASK;
 780         return 1;
 781 }
 782
 783 __setup("coredump_filter=", coredump_filter_setup);
 784
 785 #include <linux/init_task.h>
 786
 787 static void mm_init_aio(struct mm_struct *mm)
 788 {
 789 #ifdef CONFIG_AIO
 790         spin_lock_init(&mm->ioctx_lock);
 791         mm->ioctx_table = NULL;
 792 #endif
 793 }
 794
 795 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 796 {
 797 #ifdef CONFIG_MEMCG
 798         mm->owner = p;
 799 #endif
 800 }
 801
 802 static void mm_init_uprobes_state(struct mm_struct *mm)
 803 {
 804 #ifdef CONFIG_UPROBES
 805         mm->uprobes_state.xol_area = NULL;
 806 #endif
 807 }
 808
 809 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 810         struct user_namespace *user_ns)
 811 {
 812         mm->mmap = NULL;
 813         mm->mm_rb = RB_ROOT;
 814         mm->vmacache_seqnum = 0;
 815         atomic_set(&mm->mm_users, 1);
 816         atomic_set(&mm->mm_count, 1);
 817         init_rwsem(&mm->mmap_sem);
 818         INIT_LIST_HEAD(&mm->mmlist);
 819         mm->core_state = NULL;
 820         atomic_long_set(&mm->nr_ptes, 0);
 821         mm_nr_pmds_init(mm);
 822         mm->map_count = 0;
 823         mm->locked_vm = 0;
 824         mm->pinned_vm = 0;
 825         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 826         spin_lock_init(&mm->page_table_lock);
 827         mm_init_cpumask(mm);
 828         mm_init_aio(mm);
 829         mm_init_owner(mm, p);
 830         RCU_INIT_POINTER(mm->exe_file, NULL);
 831         mmu_notifier_mm_init(mm);
 832         hmm_mm_init(mm);
 833         init_tlb_flush_pending(mm);
 834 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 835         mm->pmd_huge_pte = NULL;
 836 #endif
 837         mm_init_uprobes_state(mm);
 838
 839         if (current->mm) {
 840                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 841                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 842         } else {
 843                 mm->flags = default_dump_filter;
 844                 mm->def_flags = 0;
 845         }
 846
 847         if (mm_alloc_pgd(mm))
 848                 goto fail_nopgd;
 849
 850         if (init_new_context(p, mm))
 851                 goto fail_nocontext;
 852
 853         mm->user_ns = get_user_ns(user_ns);
 854         return mm;
 855
 856 fail_nocontext:
 857         mm_free_pgd(mm);
 858 fail_nopgd:
 859         free_mm(mm);
 860         return NULL;
 861 }
 862
 863 static void check_mm(struct mm_struct *mm)
 864 {
 865         int i;
 866
 867         for (i = 0; i < NR_MM_COUNTERS; i++) {
 868                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 869
 870                 if (unlikely(x))
 871                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 872                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 873         }
 874
 875         if (atomic_long_read(&mm->nr_ptes))
 876                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 877                                 atomic_long_read(&mm->nr_ptes));
 878         if (mm_nr_pmds(mm))
 879                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 880                                 mm_nr_pmds(mm));
 881
 882 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 883         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 884 #endif
 885 }
 886
 887 /*
 888  * Allocate and initialize an mm_struct.
 889  */
 890 struct mm_struct *mm_alloc(void)
 891 {
 892         struct mm_struct *mm;
 893
 894         mm = allocate_mm();
 895         if (!mm)
 896                 return NULL;
 897
 898         memset(mm, 0, sizeof(*mm));
 899         return mm_init(mm, current, current_user_ns());
 900 }
 901
 902 /*
 903  * Called when the last reference to the mm
 904  * is dropped: either by a lazy thread or by
 905  * mmput. Free the page directory and the mm.
 906  */
 907 void __mmdrop(struct mm_struct *mm)
 908 {
 909         BUG_ON(mm == &init_mm);
 910         mm_free_pgd(mm);
 911         destroy_context(mm);
 912         hmm_mm_destroy(mm);
 913         mmu_notifier_mm_destroy(mm);
 914         check_mm(mm);
 915         put_user_ns(mm->user_ns);
 916         free_mm(mm);
 917 }
 918 EXPORT_SYMBOL_GPL(__mmdrop);
 919
 920 static inline void __mmput(struct mm_struct *mm)
 921 {
 922         VM_BUG_ON(atomic_read(&mm->mm_users));
 923
 924         uprobe_clear_state(mm);
 925         exit_aio(mm);
 926         ksm_exit(mm);
 927         khugepaged_exit(mm); /* must run before exit_mmap */
 928         exit_mmap(mm);
 929         mm_put_huge_zero_page(mm);
 930         set_mm_exe_file(mm, NULL);
 931         if (!list_empty(&mm->mmlist)) {
 932                 spin_lock(&mmlist_lock);
 933                 list_del(&mm->mmlist);
 934                 spin_unlock(&mmlist_lock);
 935         }
 936         if (mm->binfmt)
 937                 module_put(mm->binfmt->module);
 938         mmdrop(mm);
 939 }
 940
 941 /*
 942  * Decrement the use count and release all resources for an mm.
 943  */
 944 void mmput(struct mm_struct *mm)
 945 {
 946         might_sleep();
 947
 948         if (atomic_dec_and_test(&mm->mm_users))
 949                 __mmput(mm);
 950 }
 951 EXPORT_SYMBOL_GPL(mmput);
 952
 953 #ifdef CONFIG_MMU
 954 static void mmput_async_fn(struct work_struct *work)
 955 {
 956         struct mm_struct *mm = container_of(work, struct mm_struct,
 957                                             async_put_work);
 958
 959         __mmput(mm);
 960 }
 961
 962 void mmput_async(struct mm_struct *mm)
 963 {
 964         if (atomic_dec_and_test(&mm->mm_users)) {
 965                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 966                 schedule_work(&mm->async_put_work);
 967         }
 968 }
 969 #endif
 970
 971 /**
 972  * set_mm_exe_file - change a reference to the mm's executable file
 973  *
 974  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 975  *
 976  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 977  * invocations: in mmput() nobody alive left, in execve task is single
 978  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 979  * mm->exe_file, but does so without using set_mm_exe_file() in order
 980  * to do avoid the need for any locks.
 981  */
 982 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 983 {
 984         struct file *old_exe_file;
 985
 986         /*
 987          * It is safe to dereference the exe_file without RCU as
 988          * this function is only called if nobody else can access
 989          * this mm -- see comment above for justification.
 990          */
 991         old_exe_file = rcu_dereference_raw(mm->exe_file);
 992
 993         if (new_exe_file)
 994                 get_file(new_exe_file);
 995         rcu_assign_pointer(mm->exe_file, new_exe_file);
 996         if (old_exe_file)
 997                 fput(old_exe_file);
 998 }
 999
1000 /**
1001  * get_mm_exe_file - acquire a reference to the mm's executable file
1002  *
1003  * Returns %NULL if mm has no associated executable file.
1004  * User must release file via fput().
1005  */
1006 struct file *get_mm_exe_file(struct mm_struct *mm)
1007 {
1008         struct file *exe_file;
1009
1010         rcu_read_lock();
1011         exe_file = rcu_dereference(mm->exe_file);
1012         if (exe_file && !get_file_rcu(exe_file))
1013                 exe_file = NULL;
1014         rcu_read_unlock();
1015         return exe_file;
1016 }
1017 EXPORT_SYMBOL(get_mm_exe_file);
1018
1019 /**
1020  * get_task_exe_file - acquire a reference to the task's executable file
1021  *
1022  * Returns %NULL if task's mm (if any) has no associated executable file or
1023  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1024  * User must release file via fput().
1025  */
1026 struct file *get_task_exe_file(struct task_struct *task)
1027 {
1028         struct file *exe_file = NULL;
1029         struct mm_struct *mm;
1030
1031         task_lock(task);
1032         mm = task->mm;
1033         if (mm) {
1034                 if (!(task->flags & PF_KTHREAD))
1035                         exe_file = get_mm_exe_file(mm);
1036         }
1037         task_unlock(task);
1038         return exe_file;
1039 }
1040 EXPORT_SYMBOL(get_task_exe_file);
1041
1042 /**
1043  * get_task_mm - acquire a reference to the task's mm
1044  *
1045  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1046  * this kernel workthread has transiently adopted a user mm with use_mm,
1047  * to do its AIO) is not set and if so returns a reference to it, after
1048  * bumping up the use count.  User must release the mm via mmput()
1049  * after use.  Typically used by /proc and ptrace.
1050  */
1051 struct mm_struct *get_task_mm(struct task_struct *task)
1052 {
1053         struct mm_struct *mm;
1054
1055         task_lock(task);
1056         mm = task->mm;
1057         if (mm) {
1058                 if (task->flags & PF_KTHREAD)
1059                         mm = NULL;
1060                 else
1061                         mmget(mm);
1062         }
1063         task_unlock(task);
1064         return mm;
1065 }
1066 EXPORT_SYMBOL_GPL(get_task_mm);
1067
1068 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1069 {
1070         struct mm_struct *mm;
1071         int err;
1072
1073         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1074         if (err)
1075                 return ERR_PTR(err);
1076
1077         mm = get_task_mm(task);
1078         if (mm && mm != current->mm &&
1079                         !ptrace_may_access(task, mode)) {
1080                 mmput(mm);
1081                 mm = ERR_PTR(-EACCES);
1082         }
1083         mutex_unlock(&task->signal->cred_guard_mutex);
1084
1085         return mm;
1086 }
1087
1088 static void complete_vfork_done(struct task_struct *tsk)
1089 {
1090         struct completion *vfork;
1091
1092         task_lock(tsk);
1093         vfork = tsk->vfork_done;
1094         if (likely(vfork)) {
1095                 tsk->vfork_done = NULL;
1096                 complete(vfork);
1097         }
1098         task_unlock(tsk);
1099 }
1100
1101 static int wait_for_vfork_done(struct task_struct *child,
1102                                 struct completion *vfork)
1103 {
1104         int killed;
1105
1106         freezer_do_not_count();
1107         killed = wait_for_completion_killable(vfork);
1108         freezer_count();
1109
1110         if (killed) {
1111                 task_lock(child);
1112                 child->vfork_done = NULL;
1113                 task_unlock(child);
1114         }
1115
1116         put_task_struct(child);
1117         return killed;
1118 }
1119
1120 /* Please note the differences between mmput and mm_release.
1121  * mmput is called whenever we stop holding onto a mm_struct,
1122  * error success whatever.
1123  *
1124  * mm_release is called after a mm_struct has been removed
1125  * from the current process.
1126  *
1127  * This difference is important for error handling, when we
1128  * only half set up a mm_struct for a new process and need to restore
1129  * the old one.  Because we mmput the new mm_struct before
1130  * restoring the old one. . .
1131  * Eric Biederman 10 January 1998
1132  */
1133 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1134 {
1135         /* Get rid of any futexes when releasing the mm */
1136 #ifdef CONFIG_FUTEX
1137         if (unlikely(tsk->robust_list)) {
1138                 exit_robust_list(tsk);
1139                 tsk->robust_list = NULL;
1140         }
1141 #ifdef CONFIG_COMPAT
1142         if (unlikely(tsk->compat_robust_list)) {
1143                 compat_exit_robust_list(tsk);
1144                 tsk->compat_robust_list = NULL;
1145         }
1146 #endif
1147         if (unlikely(!list_empty(&tsk->pi_state_list)))
1148                 exit_pi_state_list(tsk);
1149 #endif
1150
1151         uprobe_free_utask(tsk);
1152
1153         /* Get rid of any cached register state */
1154         deactivate_mm(tsk, mm);
1155
1156         /*
1157          * Signal userspace if we're not exiting with a core dump
1158          * because we want to leave the value intact for debugging
1159          * purposes.
1160          */
1161         if (tsk->clear_child_tid) {
1162                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1163                     atomic_read(&mm->mm_users) > 1) {
1164                         /*
1165                          * We don't check the error code - if userspace has
1166                          * not set up a proper pointer then tough luck.
1167                          */
1168                         put_user(0, tsk->clear_child_tid);
1169                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1170                                         1, NULL, NULL, 0);
1171                 }
1172                 tsk->clear_child_tid = NULL;
1173         }
1174
1175         /*
1176          * All done, finally we can wake up parent and return this mm to him.
1177          * Also kthread_stop() uses this completion for synchronization.
1178          */
1179         if (tsk->vfork_done)
1180                 complete_vfork_done(tsk);
1181 }
1182
1183 /*
1184  * Allocate a new mm structure and copy contents from the
1185  * mm structure of the passed in task structure.
1186  */
1187 static struct mm_struct *dup_mm(struct task_struct *tsk)
1188 {
1189         struct mm_struct *mm, *oldmm = current->mm;
1190         int err;
1191
1192         mm = allocate_mm();
1193         if (!mm)
1194                 goto fail_nomem;
1195
1196         memcpy(mm, oldmm, sizeof(*mm));
1197
1198         if (!mm_init(mm, tsk, mm->user_ns))
1199                 goto fail_nomem;
1200
1201         err = dup_mmap(mm, oldmm);
1202         if (err)
1203                 goto free_pt;
1204
1205         mm->hiwater_rss = get_mm_rss(mm);
1206         mm->hiwater_vm = mm->total_vm;
1207
1208         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1209                 goto free_pt;
1210
1211         return mm;
1212
1213 free_pt:
1214         /* don't put binfmt in mmput, we haven't got module yet */
1215         mm->binfmt = NULL;
1216         mmput(mm);
1217
1218 fail_nomem:
1219         return NULL;
1220 }
1221
1222 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1223 {
1224         struct mm_struct *mm, *oldmm;
1225         int retval;
1226
1227         tsk->min_flt = tsk->maj_flt = 0;
1228         tsk->nvcsw = tsk->nivcsw = 0;
1229 #ifdef CONFIG_DETECT_HUNG_TASK
1230         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1231 #endif
1232
1233         tsk->mm = NULL;
1234         tsk->active_mm = NULL;
1235
1236         /*
1237          * Are we cloning a kernel thread?
1238          *
1239          * We need to steal a active VM for that..
1240          */
1241         oldmm = current->mm;
1242         if (!oldmm)
1243                 return 0;
1244
1245         /* initialize the new vmacache entries */
1246         vmacache_flush(tsk);
1247
1248         if (clone_flags & CLONE_VM) {
1249                 mmget(oldmm);
1250                 mm = oldmm;
1251                 goto good_mm;
1252         }
1253
1254         retval = -ENOMEM;
1255         mm = dup_mm(tsk);
1256         if (!mm)
1257                 goto fail_nomem;
1258
1259 good_mm:
1260         tsk->mm = mm;
1261         tsk->active_mm = mm;
1262         return 0;
1263
1264 fail_nomem:
1265         return retval;
1266 }
1267
1268 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1269 {
1270         struct fs_struct *fs = current->fs;
1271         if (clone_flags & CLONE_FS) {
1272                 /* tsk->fs is already what we want */
1273                 spin_lock(&fs->lock);
1274                 if (fs->in_exec) {
1275                         spin_unlock(&fs->lock);
1276                         return -EAGAIN;
1277                 }
1278                 fs->users++;
1279                 spin_unlock(&fs->lock);
1280                 return 0;
1281         }
1282         tsk->fs = copy_fs_struct(fs);
1283         if (!tsk->fs)
1284                 return -ENOMEM;
1285         return 0;
1286 }
1287
1288 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1289 {
1290         struct files_struct *oldf, *newf;
1291         int error = 0;
1292
1293         /*
1294          * A background process may not have any files ...
1295          */
1296         oldf = current->files;
1297         if (!oldf)
1298                 goto out;
1299
1300         if (clone_flags & CLONE_FILES) {
1301                 atomic_inc(&oldf->count);
1302                 goto out;
1303         }
1304
1305         newf = dup_fd(oldf, &error);
1306         if (!newf)
1307                 goto out;
1308
1309         tsk->files = newf;
1310         error = 0;
1311 out:
1312         return error;
1313 }
1314
1315 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1316 {
1317 #ifdef CONFIG_BLOCK
1318         struct io_context *ioc = current->io_context;
1319         struct io_context *new_ioc;
1320
1321         if (!ioc)
1322                 return 0;
1323         /*
1324          * Share io context with parent, if CLONE_IO is set
1325          */
1326         if (clone_flags & CLONE_IO) {
1327                 ioc_task_link(ioc);
1328                 tsk->io_context = ioc;
1329         } else if (ioprio_valid(ioc->ioprio)) {
1330                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1331                 if (unlikely(!new_ioc))
1332                         return -ENOMEM;
1333
1334                 new_ioc->ioprio = ioc->ioprio;
1335                 put_io_context(new_ioc);
1336         }
1337 #endif
1338         return 0;
1339 }
1340
1341 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1342 {
1343         struct sighand_struct *sig;
1344
1345         if (clone_flags & CLONE_SIGHAND) {
1346                 atomic_inc(&current->sighand->count);
1347                 return 0;
1348         }
1349         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1350         rcu_assign_pointer(tsk->sighand, sig);
1351         if (!sig)
1352                 return -ENOMEM;
1353
1354         atomic_set(&sig->count, 1);
1355         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1356         return 0;
1357 }
1358
1359 void __cleanup_sighand(struct sighand_struct *sighand)
1360 {
1361         if (atomic_dec_and_test(&sighand->count)) {
1362                 signalfd_cleanup(sighand);
1363                 /*
1364                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1365                  * without an RCU grace period, see __lock_task_sighand().
1366                  */
1367                 kmem_cache_free(sighand_cachep, sighand);
1368         }
1369 }
1370
1371 #ifdef CONFIG_POSIX_TIMERS
1372 /*
1373  * Initialize POSIX timer handling for a thread group.
1374  */
1375 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1376 {
1377         unsigned long cpu_limit;
1378
1379         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1380         if (cpu_limit != RLIM_INFINITY) {
1381                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1382                 sig->cputimer.running = true;
1383         }
1384
1385         /* The timer lists. */
1386         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1387         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1388         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1389 }
1390 #else
1391 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1392 #endif
1393
1394 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1395 {
1396         struct signal_struct *sig;
1397
1398         if (clone_flags & CLONE_THREAD)
1399                 return 0;
1400
1401         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1402         tsk->signal = sig;
1403         if (!sig)
1404                 return -ENOMEM;
1405
1406         sig->nr_threads = 1;
1407         atomic_set(&sig->live, 1);
1408         atomic_set(&sig->sigcnt, 1);
1409
1410         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1411         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1412         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1413
1414         init_waitqueue_head(&sig->wait_chldexit);
1415         sig->curr_target = tsk;
1416         init_sigpending(&sig->shared_pending);
1417         seqlock_init(&sig->stats_lock);
1418         prev_cputime_init(&sig->prev_cputime);
1419
1420 #ifdef CONFIG_POSIX_TIMERS
1421         INIT_LIST_HEAD(&sig->posix_timers);
1422         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1423         sig->real_timer.function = it_real_fn;
1424 #endif
1425
1426         task_lock(current->group_leader);
1427         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1428         task_unlock(current->group_leader);
1429
1430         posix_cpu_timers_init_group(sig);
1431
1432         tty_audit_fork(sig);
1433         sched_autogroup_fork(sig);
1434
1435         sig->oom_score_adj = current->signal->oom_score_adj;
1436         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1437
1438         mutex_init(&sig->cred_guard_mutex);
1439
1440         return 0;
1441 }
1442
1443 static void copy_seccomp(struct task_struct *p)
1444 {
1445 #ifdef CONFIG_SECCOMP
1446         /*
1447          * Must be called with sighand->lock held, which is common to
1448          * all threads in the group. Holding cred_guard_mutex is not
1449          * needed because this new task is not yet running and cannot
1450          * be racing exec.
1451          */
1452         assert_spin_locked(&current->sighand->siglock);
1453
1454         /* Ref-count the new filter user, and assign it. */
1455         get_seccomp_filter(current);
1456         p->seccomp = current->seccomp;
1457
1458         /*
1459          * Explicitly enable no_new_privs here in case it got set
1460          * between the task_struct being duplicated and holding the
1461          * sighand lock. The seccomp state and nnp must be in sync.
1462          */
1463         if (task_no_new_privs(current))
1464                 task_set_no_new_privs(p);
1465
1466         /*
1467          * If the parent gained a seccomp mode after copying thread
1468          * flags and between before we held the sighand lock, we have
1469          * to manually enable the seccomp thread flag here.
1470          */
1471         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1472                 set_tsk_thread_flag(p, TIF_SECCOMP);
1473 #endif
1474 }
1475
1476 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1477 {
1478         current->clear_child_tid = tidptr;
1479
1480         return task_pid_vnr(current);
1481 }
1482
1483 static void rt_mutex_init_task(struct task_struct *p)
1484 {
1485         raw_spin_lock_init(&p->pi_lock);
1486 #ifdef CONFIG_RT_MUTEXES
1487         p->pi_waiters = RB_ROOT_CACHED;
1488         p->pi_top_task = NULL;
1489         p->pi_blocked_on = NULL;
1490 #endif
1491 }
1492
1493 #ifdef CONFIG_POSIX_TIMERS
1494 /*
1495  * Initialize POSIX timer handling for a single task.
1496  */
1497 static void posix_cpu_timers_init(struct task_struct *tsk)
1498 {
1499         tsk->cputime_expires.prof_exp = 0;
1500         tsk->cputime_expires.virt_exp = 0;
1501         tsk->cputime_expires.sched_exp = 0;
1502         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1503         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1504         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1505 }
1506 #else
1507 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1508 #endif
1509
1510 static inline void
1511 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1512 {
1513          task->pids[type].pid = pid;
1514 }
1515
1516 static inline void rcu_copy_process(struct task_struct *p)
1517 {
1518 #ifdef CONFIG_PREEMPT_RCU
1519         p->rcu_read_lock_nesting = 0;
1520         p->rcu_read_unlock_special.s = 0;
1521         p->rcu_blocked_node = NULL;
1522         INIT_LIST_HEAD(&p->rcu_node_entry);
1523 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1524 #ifdef CONFIG_TASKS_RCU
1525         p->rcu_tasks_holdout = false;
1526         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1527         p->rcu_tasks_idle_cpu = -1;
1528 #endif /* #ifdef CONFIG_TASKS_RCU */
1529 }
1530
1531 /*
1532  * This creates a new process as a copy of the old one,
1533  * but does not actually start it yet.
1534  *
1535  * It copies the registers, and all the appropriate
1536  * parts of the process environment (as per the clone
1537  * flags). The actual kick-off is left to the caller.
1538  */
1539 static __latent_entropy struct task_struct *copy_process(
1540                                         unsigned long clone_flags,
1541                                         unsigned long stack_start,
1542                                         unsigned long stack_size,
1543                                         int __user *child_tidptr,
1544                                         struct pid *pid,
1545                                         int trace,
1546                                         unsigned long tls,
1547                                         int node)
1548 {
1549         int retval;
1550         struct task_struct *p;
1551
1552         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1553                 return ERR_PTR(-EINVAL);
1554
1555         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1556                 return ERR_PTR(-EINVAL);
1557
1558         /*
1559          * Thread groups must share signals as well, and detached threads
1560          * can only be started up within the thread group.
1561          */
1562         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1563                 return ERR_PTR(-EINVAL);
1564
1565         /*
1566          * Shared signal handlers imply shared VM. By way of the above,
1567          * thread groups also imply shared VM. Blocking this case allows
1568          * for various simplifications in other code.
1569          */
1570         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1571                 return ERR_PTR(-EINVAL);
1572
1573         /*
1574          * Siblings of global init remain as zombies on exit since they are
1575          * not reaped by their parent (swapper). To solve this and to avoid
1576          * multi-rooted process trees, prevent global and container-inits
1577          * from creating siblings.
1578          */
1579         if ((clone_flags & CLONE_PARENT) &&
1580                                 current->signal->flags & SIGNAL_UNKILLABLE)
1581                 return ERR_PTR(-EINVAL);
1582
1583         /*
1584          * If the new process will be in a different pid or user namespace
1585          * do not allow it to share a thread group with the forking task.
1586          */
1587         if (clone_flags & CLONE_THREAD) {
1588                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1589                     (task_active_pid_ns(current) !=
1590                                 current->nsproxy->pid_ns_for_children))
1591                         return ERR_PTR(-EINVAL);
1592         }
1593
1594         retval = -ENOMEM;
1595         p = dup_task_struct(current, node);
1596         if (!p)
1597                 goto fork_out;
1598
1599         /*
1600          * This _must_ happen before we call free_task(), i.e. before we jump
1601          * to any of the bad_fork_* labels. This is to avoid freeing
1602          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1603          * kernel threads (PF_KTHREAD).
1604          */
1605         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1606         /*
1607          * Clear TID on mm_release()?
1608          */
1609         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1610
1611         ftrace_graph_init_task(p);
1612
1613         rt_mutex_init_task(p);
1614
1615 #ifdef CONFIG_PROVE_LOCKING
1616         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1617         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1618 #endif
1619         retval = -EAGAIN;
1620         if (atomic_read(&p->real_cred->user->processes) >=
1621                         task_rlimit(p, RLIMIT_NPROC)) {
1622                 if (p->real_cred->user != INIT_USER &&
1623                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1624                         goto bad_fork_free;
1625         }
1626         current->flags &= ~PF_NPROC_EXCEEDED;
1627
1628         retval = copy_creds(p, clone_flags);
1629         if (retval < 0)
1630                 goto bad_fork_free;
1631
1632         /*
1633          * If multiple threads are within copy_process(), then this check
1634          * triggers too late. This doesn't hurt, the check is only there
1635          * to stop root fork bombs.
1636          */
1637         retval = -EAGAIN;
1638         if (nr_threads >= max_threads)
1639                 goto bad_fork_cleanup_count;
1640
1641         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1642         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1643         p->flags |= PF_FORKNOEXEC;
1644         INIT_LIST_HEAD(&p->children);
1645         INIT_LIST_HEAD(&p->sibling);
1646         rcu_copy_process(p);
1647         p->vfork_done = NULL;
1648         spin_lock_init(&p->alloc_lock);
1649
1650         init_sigpending(&p->pending);
1651
1652         p->utime = p->stime = p->gtime = 0;
1653 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1654         p->utimescaled = p->stimescaled = 0;
1655 #endif
1656         prev_cputime_init(&p->prev_cputime);
1657
1658 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1659         seqcount_init(&p->vtime.seqcount);
1660         p->vtime.starttime = 0;
1661         p->vtime.state = VTIME_INACTIVE;
1662 #endif
1663
1664 #if defined(SPLIT_RSS_COUNTING)
1665         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1666 #endif
1667
1668         p->default_timer_slack_ns = current->timer_slack_ns;
1669
1670         task_io_accounting_init(&p->ioac);
1671         acct_clear_integrals(p);
1672
1673         posix_cpu_timers_init(p);
1674
1675         p->start_time = ktime_get_ns();
1676         p->real_start_time = ktime_get_boot_ns();
1677         p->io_context = NULL;
1678         p->audit_context = NULL;
1679         cgroup_fork(p);
1680 #ifdef CONFIG_NUMA
1681         p->mempolicy = mpol_dup(p->mempolicy);
1682         if (IS_ERR(p->mempolicy)) {
1683                 retval = PTR_ERR(p->mempolicy);
1684                 p->mempolicy = NULL;
1685                 goto bad_fork_cleanup_threadgroup_lock;
1686         }
1687 #endif
1688 #ifdef CONFIG_CPUSETS
1689         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1690         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1691         seqcount_init(&p->mems_allowed_seq);
1692 #endif
1693 #ifdef CONFIG_TRACE_IRQFLAGS
1694         p->irq_events = 0;
1695         p->hardirqs_enabled = 0;
1696         p->hardirq_enable_ip = 0;
1697         p->hardirq_enable_event = 0;
1698         p->hardirq_disable_ip = _THIS_IP_;
1699         p->hardirq_disable_event = 0;
1700         p->softirqs_enabled = 1;
1701         p->softirq_enable_ip = _THIS_IP_;
1702         p->softirq_enable_event = 0;
1703         p->softirq_disable_ip = 0;
1704         p->softirq_disable_event = 0;
1705         p->hardirq_context = 0;
1706         p->softirq_context = 0;
1707 #endif
1708
1709         p->pagefault_disabled = 0;
1710
1711 #ifdef CONFIG_LOCKDEP
1712         p->lockdep_depth = 0; /* no locks held yet */
1713         p->curr_chain_key = 0;
1714         p->lockdep_recursion = 0;
1715         lockdep_init_task(p);
1716 #endif
1717
1718 #ifdef CONFIG_DEBUG_MUTEXES
1719         p->blocked_on = NULL; /* not blocked yet */
1720 #endif
1721 #ifdef CONFIG_BCACHE
1722         p->sequential_io        = 0;
1723         p->sequential_io_avg    = 0;
1724 #endif
1725
1726         /* Perform scheduler related setup. Assign this task to a CPU. */
1727         retval = sched_fork(clone_flags, p);
1728         if (retval)
1729                 goto bad_fork_cleanup_policy;
1730
1731         retval = perf_event_init_task(p);
1732         if (retval)
1733                 goto bad_fork_cleanup_policy;
1734         retval = audit_alloc(p);
1735         if (retval)
1736                 goto bad_fork_cleanup_perf;
1737         /* copy all the process information */
1738         shm_init_task(p);
1739         retval = security_task_alloc(p, clone_flags);
1740         if (retval)
1741                 goto bad_fork_cleanup_audit;
1742         retval = copy_semundo(clone_flags, p);
1743         if (retval)
1744                 goto bad_fork_cleanup_security;
1745         retval = copy_files(clone_flags, p);
1746         if (retval)
1747                 goto bad_fork_cleanup_semundo;
1748         retval = copy_fs(clone_flags, p);
1749         if (retval)
1750                 goto bad_fork_cleanup_files;
1751         retval = copy_sighand(clone_flags, p);
1752         if (retval)
1753                 goto bad_fork_cleanup_fs;
1754         retval = copy_signal(clone_flags, p);
1755         if (retval)
1756                 goto bad_fork_cleanup_sighand;
1757         retval = copy_mm(clone_flags, p);
1758         if (retval)
1759                 goto bad_fork_cleanup_signal;
1760         retval = copy_namespaces(clone_flags, p);
1761         if (retval)
1762                 goto bad_fork_cleanup_mm;
1763         retval = copy_io(clone_flags, p);
1764         if (retval)
1765                 goto bad_fork_cleanup_namespaces;
1766         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1767         if (retval)
1768                 goto bad_fork_cleanup_io;
1769
1770         if (pid != &init_struct_pid) {
1771                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1772                 if (IS_ERR(pid)) {
1773                         retval = PTR_ERR(pid);
1774                         goto bad_fork_cleanup_thread;
1775                 }
1776         }
1777
1778 #ifdef CONFIG_BLOCK
1779         p->plug = NULL;
1780 #endif
1781 #ifdef CONFIG_FUTEX
1782         p->robust_list = NULL;
1783 #ifdef CONFIG_COMPAT
1784         p->compat_robust_list = NULL;
1785 #endif
1786         INIT_LIST_HEAD(&p->pi_state_list);
1787         p->pi_state_cache = NULL;
1788 #endif
1789         /*
1790          * sigaltstack should be cleared when sharing the same VM
1791          */
1792         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1793                 sas_ss_reset(p);
1794
1795         /*
1796          * Syscall tracing and stepping should be turned off in the
1797          * child regardless of CLONE_PTRACE.
1798          */
1799         user_disable_single_step(p);
1800         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1801 #ifdef TIF_SYSCALL_EMU
1802         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1803 #endif
1804         clear_all_latency_tracing(p);
1805
1806         /* ok, now we should be set up.. */
1807         p->pid = pid_nr(pid);
1808         if (clone_flags & CLONE_THREAD) {
1809                 p->exit_signal = -1;
1810                 p->group_leader = current->group_leader;
1811                 p->tgid = current->tgid;
1812         } else {
1813                 if (clone_flags & CLONE_PARENT)
1814                         p->exit_signal = current->group_leader->exit_signal;
1815                 else
1816                         p->exit_signal = (clone_flags & CSIGNAL);
1817                 p->group_leader = p;
1818                 p->tgid = p->pid;
1819         }
1820
1821         p->nr_dirtied = 0;
1822         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1823         p->dirty_paused_when = 0;
1824
1825         p->pdeath_signal = 0;
1826         INIT_LIST_HEAD(&p->thread_group);
1827         p->task_works = NULL;
1828
1829         cgroup_threadgroup_change_begin(current);
1830         /*
1831          * Ensure that the cgroup subsystem policies allow the new process to be
1832          * forked. It should be noted the the new process's css_set can be changed
1833          * between here and cgroup_post_fork() if an organisation operation is in
1834          * progress.
1835          */
1836         retval = cgroup_can_fork(p);
1837         if (retval)
1838                 goto bad_fork_free_pid;
1839
1840         /*
1841          * Make it visible to the rest of the system, but dont wake it up yet.
1842          * Need tasklist lock for parent etc handling!
1843          */
1844         write_lock_irq(&tasklist_lock);
1845
1846         /* CLONE_PARENT re-uses the old parent */
1847         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1848                 p->real_parent = current->real_parent;
1849                 p->parent_exec_id = current->parent_exec_id;
1850         } else {
1851                 p->real_parent = current;
1852                 p->parent_exec_id = current->self_exec_id;
1853         }
1854
1855         klp_copy_process(p);
1856
1857         spin_lock(&current->sighand->siglock);
1858
1859         /*
1860          * Copy seccomp details explicitly here, in case they were changed
1861          * before holding sighand lock.
1862          */
1863         copy_seccomp(p);
1864
1865         /*
1866          * Process group and session signals need to be delivered to just the
1867          * parent before the fork or both the parent and the child after the
1868          * fork. Restart if a signal comes in before we add the new process to
1869          * it's process group.
1870          * A fatal signal pending means that current will exit, so the new
1871          * thread can't slip out of an OOM kill (or normal SIGKILL).
1872         */
1873         recalc_sigpending();
1874         if (signal_pending(current)) {
1875                 retval = -ERESTARTNOINTR;
1876                 goto bad_fork_cancel_cgroup;
1877         }
1878         if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1879                 retval = -ENOMEM;
1880                 goto bad_fork_cancel_cgroup;
1881         }
1882
1883         if (likely(p->pid)) {
1884                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1885
1886                 init_task_pid(p, PIDTYPE_PID, pid);
1887                 if (thread_group_leader(p)) {
1888                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1889                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1890
1891                         if (is_child_reaper(pid)) {
1892                                 ns_of_pid(pid)->child_reaper = p;
1893                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1894                         }
1895
1896                         p->signal->leader_pid = pid;
1897                         p->signal->tty = tty_kref_get(current->signal->tty);
1898                         /*
1899                          * Inherit has_child_subreaper flag under the same
1900                          * tasklist_lock with adding child to the process tree
1901                          * for propagate_has_child_subreaper optimization.
1902                          */
1903                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1904                                                          p->real_parent->signal->is_child_subreaper;
1905                         list_add_tail(&p->sibling, &p->real_parent->children);
1906                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1907                         attach_pid(p, PIDTYPE_PGID);
1908                         attach_pid(p, PIDTYPE_SID);
1909                         __this_cpu_inc(process_counts);
1910                 } else {
1911                         current->signal->nr_threads++;
1912                         atomic_inc(&current->signal->live);
1913                         atomic_inc(&current->signal->sigcnt);
1914                         list_add_tail_rcu(&p->thread_group,
1915                                           &p->group_leader->thread_group);
1916                         list_add_tail_rcu(&p->thread_node,
1917                                           &p->signal->thread_head);
1918                 }
1919                 attach_pid(p, PIDTYPE_PID);
1920                 nr_threads++;
1921         }
1922
1923         total_forks++;
1924         spin_unlock(&current->sighand->siglock);
1925         syscall_tracepoint_update(p);
1926         write_unlock_irq(&tasklist_lock);
1927
1928         proc_fork_connector(p);
1929         cgroup_post_fork(p);
1930         cgroup_threadgroup_change_end(current);
1931         perf_event_fork(p);
1932
1933         trace_task_newtask(p, clone_flags);
1934         uprobe_copy_process(p, clone_flags);
1935
1936         return p;
1937
1938 bad_fork_cancel_cgroup:
1939         spin_unlock(&current->sighand->siglock);
1940         write_unlock_irq(&tasklist_lock);
1941         cgroup_cancel_fork(p);
1942 bad_fork_free_pid:
1943         cgroup_threadgroup_change_end(current);
1944         if (pid != &init_struct_pid)
1945                 free_pid(pid);
1946 bad_fork_cleanup_thread:
1947         exit_thread(p);
1948 bad_fork_cleanup_io:
1949         if (p->io_context)
1950                 exit_io_context(p);
1951 bad_fork_cleanup_namespaces:
1952         exit_task_namespaces(p);
1953 bad_fork_cleanup_mm:
1954         if (p->mm)
1955                 mmput(p->mm);
1956 bad_fork_cleanup_signal:
1957         if (!(clone_flags & CLONE_THREAD))
1958                 free_signal_struct(p->signal);
1959 bad_fork_cleanup_sighand:
1960         __cleanup_sighand(p->sighand);
1961 bad_fork_cleanup_fs:
1962         exit_fs(p); /* blocking */
1963 bad_fork_cleanup_files:
1964         exit_files(p); /* blocking */
1965 bad_fork_cleanup_semundo:
1966         exit_sem(p);
1967 bad_fork_cleanup_security:
1968         security_task_free(p);
1969 bad_fork_cleanup_audit:
1970         audit_free(p);
1971 bad_fork_cleanup_perf:
1972         perf_event_free_task(p);
1973 bad_fork_cleanup_policy:
1974         lockdep_free_task(p);
1975 #ifdef CONFIG_NUMA
1976         mpol_put(p->mempolicy);
1977 bad_fork_cleanup_threadgroup_lock:
1978 #endif
1979         delayacct_tsk_free(p);
1980 bad_fork_cleanup_count:
1981         atomic_dec(&p->cred->user->processes);
1982         exit_creds(p);
1983 bad_fork_free:
1984         p->state = TASK_DEAD;
1985         put_task_stack(p);
1986         free_task(p);
1987 fork_out:
1988         return ERR_PTR(retval);
1989 }
1990
1991 static inline void init_idle_pids(struct pid_link *links)
1992 {
1993         enum pid_type type;
1994
1995         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1996                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1997                 links[type].pid = &init_struct_pid;
1998         }
1999 }
2000
2001 struct task_struct *fork_idle(int cpu)
2002 {
2003         struct task_struct *task;
2004         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2005                             cpu_to_node(cpu));
2006         if (!IS_ERR(task)) {
2007                 init_idle_pids(task->pids);
2008                 init_idle(task, cpu);
2009         }
2010
2011         return task;
2012 }
2013
2014 /*
2015  *  Ok, this is the main fork-routine.
2016  *
2017  * It copies the process, and if successful kick-starts
2018  * it and waits for it to finish using the VM if required.
2019  */
2020 long _do_fork(unsigned long clone_flags,
2021               unsigned long stack_start,
2022               unsigned long stack_size,
2023               int __user *parent_tidptr,
2024               int __user *child_tidptr,
2025               unsigned long tls)
2026 {
2027         struct task_struct *p;
2028         int trace = 0;
2029         long nr;
2030
2031         /*
2032          * Determine whether and which event to report to ptracer.  When
2033          * called from kernel_thread or CLONE_UNTRACED is explicitly
2034          * requested, no event is reported; otherwise, report if the event
2035          * for the type of forking is enabled.
2036          */
2037         if (!(clone_flags & CLONE_UNTRACED)) {
2038                 if (clone_flags & CLONE_VFORK)
2039                         trace = PTRACE_EVENT_VFORK;
2040                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2041                         trace = PTRACE_EVENT_CLONE;
2042                 else
2043                         trace = PTRACE_EVENT_FORK;
2044
2045                 if (likely(!ptrace_event_enabled(current, trace)))
2046                         trace = 0;
2047         }
2048
2049         p = copy_process(clone_flags, stack_start, stack_size,
2050                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2051         add_latent_entropy();
2052         /*
2053          * Do this prior waking up the new thread - the thread pointer
2054          * might get invalid after that point, if the thread exits quickly.
2055          */
2056         if (!IS_ERR(p)) {
2057                 struct completion vfork;
2058                 struct pid *pid;
2059
2060                 trace_sched_process_fork(current, p);
2061
2062                 pid = get_task_pid(p, PIDTYPE_PID);
2063                 nr = pid_vnr(pid);
2064
2065                 if (clone_flags & CLONE_PARENT_SETTID)
2066                         put_user(nr, parent_tidptr);
2067
2068                 if (clone_flags & CLONE_VFORK) {
2069                         p->vfork_done = &vfork;
2070                         init_completion(&vfork);
2071                         get_task_struct(p);
2072                 }
2073
2074                 wake_up_new_task(p);
2075
2076                 /* forking complete and child started to run, tell ptracer */
2077                 if (unlikely(trace))
2078                         ptrace_event_pid(trace, pid);
2079
2080                 if (clone_flags & CLONE_VFORK) {
2081                         if (!wait_for_vfork_done(p, &vfork))
2082                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2083                 }
2084
2085                 put_pid(pid);
2086         } else {
2087                 nr = PTR_ERR(p);
2088         }
2089         return nr;
2090 }
2091
2092 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2093 /* For compatibility with architectures that call do_fork directly rather than
2094  * using the syscall entry points below. */
2095 long do_fork(unsigned long clone_flags,
2096               unsigned long stack_start,
2097               unsigned long stack_size,
2098               int __user *parent_tidptr,
2099               int __user *child_tidptr)
2100 {
2101         return _do_fork(clone_flags, stack_start, stack_size,
2102                         parent_tidptr, child_tidptr, 0);
2103 }
2104 #endif
2105
2106 /*
2107  * Create a kernel thread.
2108  */
2109 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2110 {
2111         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2112                 (unsigned long)arg, NULL, NULL, 0);
2113 }
2114
2115 #ifdef __ARCH_WANT_SYS_FORK
2116 SYSCALL_DEFINE0(fork)
2117 {
2118 #ifdef CONFIG_MMU
2119         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2120 #else
2121         /* can not support in nommu mode */
2122         return -EINVAL;
2123 #endif
2124 }
2125 #endif
2126
2127 #ifdef __ARCH_WANT_SYS_VFORK
2128 SYSCALL_DEFINE0(vfork)
2129 {
2130         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2131                         0, NULL, NULL, 0);
2132 }
2133 #endif
2134
2135 #ifdef __ARCH_WANT_SYS_CLONE
2136 #ifdef CONFIG_CLONE_BACKWARDS
2137 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2138                  int __user *, parent_tidptr,
2139                  unsigned long, tls,
2140                  int __user *, child_tidptr)
2141 #elif defined(CONFIG_CLONE_BACKWARDS2)
2142 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2143                  int __user *, parent_tidptr,
2144                  int __user *, child_tidptr,
2145                  unsigned long, tls)
2146 #elif defined(CONFIG_CLONE_BACKWARDS3)
2147 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2148                 int, stack_size,
2149                 int __user *, parent_tidptr,
2150                 int __user *, child_tidptr,
2151                 unsigned long, tls)
2152 #else
2153 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2154                  int __user *, parent_tidptr,
2155                  int __user *, child_tidptr,
2156                  unsigned long, tls)
2157 #endif
2158 {
2159         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2160 }
2161 #endif
2162
2163 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2164 {
2165         struct task_struct *leader, *parent, *child;
2166         int res;
2167
2168         read_lock(&tasklist_lock);
2169         leader = top = top->group_leader;
2170 down:
2171         for_each_thread(leader, parent) {
2172                 list_for_each_entry(child, &parent->children, sibling) {
2173                         res = visitor(child, data);
2174                         if (res) {
2175                                 if (res < 0)
2176                                         goto out;
2177                                 leader = child;
2178                                 goto down;
2179                         }
2180 up:
2181                         ;
2182                 }
2183         }
2184
2185         if (leader != top) {
2186                 child = leader;
2187                 parent = child->real_parent;
2188                 leader = parent->group_leader;
2189                 goto up;
2190         }
2191 out:
2192         read_unlock(&tasklist_lock);
2193 }
2194
2195 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2196 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2197 #endif
2198
2199 static void sighand_ctor(void *data)
2200 {
2201         struct sighand_struct *sighand = data;
2202
2203         spin_lock_init(&sighand->siglock);
2204         init_waitqueue_head(&sighand->signalfd_wqh);
2205 }
2206
2207 void __init proc_caches_init(void)
2208 {
2209         sighand_cachep = kmem_cache_create("sighand_cache",
2210                         sizeof(struct sighand_struct), 0,
2211                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2212                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2213         signal_cachep = kmem_cache_create("signal_cache",
2214                         sizeof(struct signal_struct), 0,
2215                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2216                         NULL);
2217         files_cachep = kmem_cache_create("files_cache",
2218                         sizeof(struct files_struct), 0,
2219                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2220                         NULL);
2221         fs_cachep = kmem_cache_create("fs_cache",
2222                         sizeof(struct fs_struct), 0,
2223                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2224                         NULL);
2225         /*
2226          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2227          * whole struct cpumask for the OFFSTACK case. We could change
2228          * this to *only* allocate as much of it as required by the
2229          * maximum number of CPU's we can ever have.  The cpumask_allocation
2230          * is at the end of the structure, exactly for that reason.
2231          */
2232         mm_cachep = kmem_cache_create("mm_struct",
2233                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2234                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2235                         NULL);
2236         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2237         mmap_init();
2238         nsproxy_cache_init();
2239 }
2240
2241 /*
2242  * Check constraints on flags passed to the unshare system call.
2243  */
2244 static int check_unshare_flags(unsigned long unshare_flags)
2245 {
2246         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2247                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2248                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2249                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2250                 return -EINVAL;
2251         /*
2252          * Not implemented, but pretend it works if there is nothing
2253          * to unshare.  Note that unsharing the address space or the
2254          * signal handlers also need to unshare the signal queues (aka
2255          * CLONE_THREAD).
2256          */
2257         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2258                 if (!thread_group_empty(current))
2259                         return -EINVAL;
2260         }
2261         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2262                 if (atomic_read(&current->sighand->count) > 1)
2263                         return -EINVAL;
2264         }
2265         if (unshare_flags & CLONE_VM) {
2266                 if (!current_is_single_threaded())
2267                         return -EINVAL;
2268         }
2269
2270         return 0;
2271 }
2272
2273 /*
2274  * Unshare the filesystem structure if it is being shared
2275  */
2276 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2277 {
2278         struct fs_struct *fs = current->fs;
2279
2280         if (!(unshare_flags & CLONE_FS) || !fs)
2281                 return 0;
2282
2283         /* don't need lock here; in the worst case we'll do useless copy */
2284         if (fs->users == 1)
2285                 return 0;
2286
2287         *new_fsp = copy_fs_struct(fs);
2288         if (!*new_fsp)
2289                 return -ENOMEM;
2290
2291         return 0;
2292 }
2293
2294 /*
2295  * Unshare file descriptor table if it is being shared
2296  */
2297 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2298 {
2299         struct files_struct *fd = current->files;
2300         int error = 0;
2301
2302         if ((unshare_flags & CLONE_FILES) &&
2303             (fd && atomic_read(&fd->count) > 1)) {
2304                 *new_fdp = dup_fd(fd, &error);
2305                 if (!*new_fdp)
2306                         return error;
2307         }
2308
2309         return 0;
2310 }
2311
2312 /*
2313  * unshare allows a process to 'unshare' part of the process
2314  * context which was originally shared using clone.  copy_*
2315  * functions used by do_fork() cannot be used here directly
2316  * because they modify an inactive task_struct that is being
2317  * constructed. Here we are modifying the current, active,
2318  * task_struct.
2319  */
2320 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2321 {
2322         struct fs_struct *fs, *new_fs = NULL;
2323         struct files_struct *fd, *new_fd = NULL;
2324         struct cred *new_cred = NULL;
2325         struct nsproxy *new_nsproxy = NULL;
2326         int do_sysvsem = 0;
2327         int err;
2328
2329         /*
2330          * If unsharing a user namespace must also unshare the thread group
2331          * and unshare the filesystem root and working directories.
2332          */
2333         if (unshare_flags & CLONE_NEWUSER)
2334                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2335         /*
2336          * If unsharing vm, must also unshare signal handlers.
2337          */
2338         if (unshare_flags & CLONE_VM)
2339                 unshare_flags |= CLONE_SIGHAND;
2340         /*
2341          * If unsharing a signal handlers, must also unshare the signal queues.
2342          */
2343         if (unshare_flags & CLONE_SIGHAND)
2344                 unshare_flags |= CLONE_THREAD;
2345         /*
2346          * If unsharing namespace, must also unshare filesystem information.
2347          */
2348         if (unshare_flags & CLONE_NEWNS)
2349                 unshare_flags |= CLONE_FS;
2350
2351         err = check_unshare_flags(unshare_flags);
2352         if (err)
2353                 goto bad_unshare_out;
2354         /*
2355          * CLONE_NEWIPC must also detach from the undolist: after switching
2356          * to a new ipc namespace, the semaphore arrays from the old
2357          * namespace are unreachable.
2358          */
2359         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2360                 do_sysvsem = 1;
2361         err = unshare_fs(unshare_flags, &new_fs);
2362         if (err)
2363                 goto bad_unshare_out;
2364         err = unshare_fd(unshare_flags, &new_fd);
2365         if (err)
2366                 goto bad_unshare_cleanup_fs;
2367         err = unshare_userns(unshare_flags, &new_cred);
2368         if (err)
2369                 goto bad_unshare_cleanup_fd;
2370         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2371                                          new_cred, new_fs);
2372         if (err)
2373                 goto bad_unshare_cleanup_cred;
2374
2375         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2376                 if (do_sysvsem) {
2377                         /*
2378                          * CLONE_SYSVSEM is equivalent to sys_exit().
2379                          */
2380                         exit_sem(current);
2381                 }
2382                 if (unshare_flags & CLONE_NEWIPC) {
2383                         /* Orphan segments in old ns (see sem above). */
2384                         exit_shm(current);
2385                         shm_init_task(current);
2386                 }
2387
2388                 if (new_nsproxy)
2389                         switch_task_namespaces(current, new_nsproxy);
2390
2391                 task_lock(current);
2392
2393                 if (new_fs) {
2394                         fs = current->fs;
2395                         spin_lock(&fs->lock);
2396                         current->fs = new_fs;
2397                         if (--fs->users)
2398                                 new_fs = NULL;
2399                         else
2400                                 new_fs = fs;
2401                         spin_unlock(&fs->lock);
2402                 }
2403
2404                 if (new_fd) {
2405                         fd = current->files;
2406                         current->files = new_fd;
2407                         new_fd = fd;
2408                 }
2409
2410                 task_unlock(current);
2411
2412                 if (new_cred) {
2413                         /* Install the new user namespace */
2414                         commit_creds(new_cred);
2415                         new_cred = NULL;
2416                 }
2417         }
2418
2419         perf_event_namespaces(current);
2420
2421 bad_unshare_cleanup_cred:
2422         if (new_cred)
2423                 put_cred(new_cred);
2424 bad_unshare_cleanup_fd:
2425         if (new_fd)
2426                 put_files_struct(new_fd);
2427
2428 bad_unshare_cleanup_fs:
2429         if (new_fs)
2430                 free_fs_struct(new_fs);
2431
2432 bad_unshare_out:
2433         return err;
2434 }
2435
2436 /*
2437  *      Helper to unshare the files of the current task.
2438  *      We don't want to expose copy_files internals to
2439  *      the exec layer of the kernel.
2440  */
2441
2442 int unshare_files(struct files_struct **displaced)
2443 {
2444         struct task_struct *task = current;
2445         struct files_struct *copy = NULL;
2446         int error;
2447
2448         error = unshare_fd(CLONE_FILES, &copy);
2449         if (error || !copy) {
2450                 *displaced = NULL;
2451                 return error;
2452         }
2453         *displaced = task->files;
2454         task_lock(task);
2455         task->files = copy;
2456         task_unlock(task);
2457         return 0;
2458 }
2459
2460 int sysctl_max_threads(struct ctl_table *table, int write,
2461                        void __user *buffer, size_t *lenp, loff_t *ppos)
2462 {
2463         struct ctl_table t;
2464         int ret;
2465         int threads = max_threads;
2466         int min = MIN_THREADS;
2467         int max = MAX_THREADS;
2468
2469         t = *table;
2470         t.data = &threads;
2471         t.extra1 = &min;
2472         t.extra2 = &max;
2473
2474         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2475         if (ret || !write)
2476                 return ret;
2477
2478         set_max_threads(threads);
2479
2480         return 0;
2481 }