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/sched/mm.h>
  81 #include <linux/perf_event.h>
  82 #include <linux/posix-timers.h>
  83 #include <linux/user-return-notifier.h>
  84 #include <linux/oom.h>
  85 #include <linux/khugepaged.h>
  86 #include <linux/signalfd.h>
  87 #include <linux/uprobes.h>
  88 #include <linux/aio.h>
  89 #include <linux/compiler.h>
  90 #include <linux/sysctl.h>
  91 #include <linux/kcov.h>
  92 #include <linux/livepatch.h>
  93 #include <linux/thread_info.h>
  94
  95 #include <asm/pgtable.h>
  96 #include <asm/pgalloc.h>
  97 #include <linux/uaccess.h>
  98 #include <asm/mmu_context.h>
  99 #include <asm/cacheflush.h>
 100 #include <asm/tlbflush.h>
 101
 102 #include <trace/events/sched.h>
 103
 104 #define CREATE_TRACE_POINTS
 105 #include <trace/events/task.h>
 106
 107 /*
 108  * Minimum number of threads to boot the kernel
 109  */
 110 #define MIN_THREADS 20
 111
 112 /*
 113  * Maximum number of threads
 114  */
 115 #define MAX_THREADS FUTEX_TID_MASK
 116
 117 /*
 118  * Protected counters by write_lock_irq(&tasklist_lock)
 119  */
 120 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 121 int nr_threads;                 /* The idle threads do not count.. */
 122
 123 int max_threads;                /* tunable limit on nr_threads */
 124
 125 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 126
 127 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 128
 129 #ifdef CONFIG_PROVE_RCU
 130 int lockdep_tasklist_lock_is_held(void)
 131 {
 132         return lockdep_is_held(&tasklist_lock);
 133 }
 134 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 135 #endif /* #ifdef CONFIG_PROVE_RCU */
 136
 137 int nr_processes(void)
 138 {
 139         int cpu;
 140         int total = 0;
 141
 142         for_each_possible_cpu(cpu)
 143                 total += per_cpu(process_counts, cpu);
 144
 145         return total;
 146 }
 147
 148 void __weak arch_release_task_struct(struct task_struct *tsk)
 149 {
 150 }
 151
 152 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 153 static struct kmem_cache *task_struct_cachep;
 154
 155 static inline struct task_struct *alloc_task_struct_node(int node)
 156 {
 157         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 158 }
 159
 160 static inline void free_task_struct(struct task_struct *tsk)
 161 {
 162         kmem_cache_free(task_struct_cachep, tsk);
 163 }
 164 #endif
 165
 166 void __weak arch_release_thread_stack(unsigned long *stack)
 167 {
 168 }
 169
 170 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 171
 172 /*
 173  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 174  * kmemcache based allocator.
 175  */
 176 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 177
 178 #ifdef CONFIG_VMAP_STACK
 179 /*
 180  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 181  * flush.  Try to minimize the number of calls by caching stacks.
 182  */
 183 #define NR_CACHED_STACKS 2
 184 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 185
 186 static int free_vm_stack_cache(unsigned int cpu)
 187 {
 188         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 189         int i;
 190
 191         for (i = 0; i < NR_CACHED_STACKS; i++) {
 192                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 193
 194                 if (!vm_stack)
 195                         continue;
 196
 197                 vfree(vm_stack->addr);
 198                 cached_vm_stacks[i] = NULL;
 199         }
 200
 201         return 0;
 202 }
 203 #endif
 204
 205 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 206 {
 207 #ifdef CONFIG_VMAP_STACK
 208         void *stack;
 209         int i;
 210
 211         for (i = 0; i < NR_CACHED_STACKS; i++) {
 212                 struct vm_struct *s;
 213
 214                 s = this_cpu_xchg(cached_stacks[i], NULL);
 215
 216                 if (!s)
 217                         continue;
 218
 219                 /* Clear stale pointers from reused stack. */
 220                 memset(s->addr, 0, THREAD_SIZE);
 221
 222                 tsk->stack_vm_area = s;
 223                 return s->addr;
 224         }
 225
 226         /*
 227          * Allocated stacks are cached and later reused by new threads,
 228          * so memcg accounting is performed manually on assigning/releasing
 229          * stacks to tasks. Drop __GFP_ACCOUNT.
 230          */
 231         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 232                                      VMALLOC_START, VMALLOC_END,
 233                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 234                                      PAGE_KERNEL,
 235                                      0, node, __builtin_return_address(0));
 236
 237         /*
 238          * We can't call find_vm_area() in interrupt context, and
 239          * free_thread_stack() can be called in interrupt context,
 240          * so cache the vm_struct.
 241          */
 242         if (stack)
 243                 tsk->stack_vm_area = find_vm_area(stack);
 244         return stack;
 245 #else
 246         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 247                                              THREAD_SIZE_ORDER);
 248
 249         return page ? page_address(page) : NULL;
 250 #endif
 251 }
 252
 253 static inline void free_thread_stack(struct task_struct *tsk)
 254 {
 255 #ifdef CONFIG_VMAP_STACK
 256         struct vm_struct *vm = task_stack_vm_area(tsk);
 257
 258         if (vm) {
 259                 int i;
 260
 261                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 262                         mod_memcg_page_state(vm->pages[i],
 263                                              MEMCG_KERNEL_STACK_KB,
 264                                              -(int)(PAGE_SIZE / 1024));
 265
 266                         memcg_kmem_uncharge(vm->pages[i], 0);
 267                 }
 268
 269                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 270                         if (this_cpu_cmpxchg(cached_stacks[i],
 271                                         NULL, tsk->stack_vm_area) != NULL)
 272                                 continue;
 273
 274                         return;
 275                 }
 276
 277                 vfree_atomic(tsk->stack);
 278                 return;
 279         }
 280 #endif
 281
 282         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 283 }
 284 # else
 285 static struct kmem_cache *thread_stack_cache;
 286
 287 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 288                                                   int node)
 289 {
 290         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 291 }
 292
 293 static void free_thread_stack(struct task_struct *tsk)
 294 {
 295         kmem_cache_free(thread_stack_cache, tsk->stack);
 296 }
 297
 298 void thread_stack_cache_init(void)
 299 {
 300         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 301                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 302                                         THREAD_SIZE, NULL);
 303         BUG_ON(thread_stack_cache == NULL);
 304 }
 305 # endif
 306 #endif
 307
 308 /* SLAB cache for signal_struct structures (tsk->signal) */
 309 static struct kmem_cache *signal_cachep;
 310
 311 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 312 struct kmem_cache *sighand_cachep;
 313
 314 /* SLAB cache for files_struct structures (tsk->files) */
 315 struct kmem_cache *files_cachep;
 316
 317 /* SLAB cache for fs_struct structures (tsk->fs) */
 318 struct kmem_cache *fs_cachep;
 319
 320 /* SLAB cache for vm_area_struct structures */
 321 static struct kmem_cache *vm_area_cachep;
 322
 323 /* SLAB cache for mm_struct structures (tsk->mm) */
 324 static struct kmem_cache *mm_cachep;
 325
 326 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 327 {
 328         struct vm_area_struct *vma;
 329
 330         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 331         if (vma)
 332                 vma_init(vma, mm);
 333         return vma;
 334 }
 335
 336 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 337 {
 338         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 339
 340         if (new) {
 341                 *new = *orig;
 342                 INIT_LIST_HEAD(&new->anon_vma_chain);
 343         }
 344         return new;
 345 }
 346
 347 void vm_area_free(struct vm_area_struct *vma)
 348 {
 349         kmem_cache_free(vm_area_cachep, vma);
 350 }
 351
 352 static void account_kernel_stack(struct task_struct *tsk, int account)
 353 {
 354         void *stack = task_stack_page(tsk);
 355         struct vm_struct *vm = task_stack_vm_area(tsk);
 356
 357         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 358
 359         if (vm) {
 360                 int i;
 361
 362                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 363
 364                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 365                         mod_zone_page_state(page_zone(vm->pages[i]),
 366                                             NR_KERNEL_STACK_KB,
 367                                             PAGE_SIZE / 1024 * account);
 368                 }
 369         } else {
 370                 /*
 371                  * All stack pages are in the same zone and belong to the
 372                  * same memcg.
 373                  */
 374                 struct page *first_page = virt_to_page(stack);
 375
 376                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 377                                     THREAD_SIZE / 1024 * account);
 378
 379                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 380                                      account * (THREAD_SIZE / 1024));
 381         }
 382 }
 383
 384 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 385 {
 386 #ifdef CONFIG_VMAP_STACK
 387         struct vm_struct *vm = task_stack_vm_area(tsk);
 388         int ret;
 389
 390         if (vm) {
 391                 int i;
 392
 393                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 394                         /*
 395                          * If memcg_kmem_charge() fails, page->mem_cgroup
 396                          * pointer is NULL, and both memcg_kmem_uncharge()
 397                          * and mod_memcg_page_state() in free_thread_stack()
 398                          * will ignore this page. So it's safe.
 399                          */
 400                         ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0);
 401                         if (ret)
 402                                 return ret;
 403
 404                         mod_memcg_page_state(vm->pages[i],
 405                                              MEMCG_KERNEL_STACK_KB,
 406                                              PAGE_SIZE / 1024);
 407                 }
 408         }
 409 #endif
 410         return 0;
 411 }
 412
 413 static void release_task_stack(struct task_struct *tsk)
 414 {
 415         if (WARN_ON(tsk->state != TASK_DEAD))
 416                 return;  /* Better to leak the stack than to free prematurely */
 417
 418         account_kernel_stack(tsk, -1);
 419         arch_release_thread_stack(tsk->stack);
 420         free_thread_stack(tsk);
 421         tsk->stack = NULL;
 422 #ifdef CONFIG_VMAP_STACK
 423         tsk->stack_vm_area = NULL;
 424 #endif
 425 }
 426
 427 #ifdef CONFIG_THREAD_INFO_IN_TASK
 428 void put_task_stack(struct task_struct *tsk)
 429 {
 430         if (atomic_dec_and_test(&tsk->stack_refcount))
 431                 release_task_stack(tsk);
 432 }
 433 #endif
 434
 435 void free_task(struct task_struct *tsk)
 436 {
 437 #ifndef CONFIG_THREAD_INFO_IN_TASK
 438         /*
 439          * The task is finally done with both the stack and thread_info,
 440          * so free both.
 441          */
 442         release_task_stack(tsk);
 443 #else
 444         /*
 445          * If the task had a separate stack allocation, it should be gone
 446          * by now.
 447          */
 448         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 449 #endif
 450         rt_mutex_debug_task_free(tsk);
 451         ftrace_graph_exit_task(tsk);
 452         put_seccomp_filter(tsk);
 453         arch_release_task_struct(tsk);
 454         if (tsk->flags & PF_KTHREAD)
 455                 free_kthread_struct(tsk);
 456         free_task_struct(tsk);
 457 }
 458 EXPORT_SYMBOL(free_task);
 459
 460 #ifdef CONFIG_MMU
 461 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 462                                         struct mm_struct *oldmm)
 463 {
 464         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 465         struct rb_node **rb_link, *rb_parent;
 466         int retval;
 467         unsigned long charge;
 468         LIST_HEAD(uf);
 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                 /*
 510                  * Don't duplicate many vmas if we've been oom-killed (for
 511                  * example)
 512                  */
 513                 if (fatal_signal_pending(current)) {
 514                         retval = -EINTR;
 515                         goto out;
 516                 }
 517                 if (mpnt->vm_flags & VM_ACCOUNT) {
 518                         unsigned long len = vma_pages(mpnt);
 519
 520                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 521                                 goto fail_nomem;
 522                         charge = len;
 523                 }
 524                 tmp = vm_area_dup(mpnt);
 525                 if (!tmp)
 526                         goto fail_nomem;
 527                 retval = vma_dup_policy(mpnt, tmp);
 528                 if (retval)
 529                         goto fail_nomem_policy;
 530                 tmp->vm_mm = mm;
 531                 retval = dup_userfaultfd(tmp, &uf);
 532                 if (retval)
 533                         goto fail_nomem_anon_vma_fork;
 534                 if (tmp->vm_flags & VM_WIPEONFORK) {
 535                         /* VM_WIPEONFORK gets a clean slate in the child. */
 536                         tmp->anon_vma = NULL;
 537                         if (anon_vma_prepare(tmp))
 538                                 goto fail_nomem_anon_vma_fork;
 539                 } else if (anon_vma_fork(tmp, mpnt))
 540                         goto fail_nomem_anon_vma_fork;
 541                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 542                 tmp->vm_next = tmp->vm_prev = NULL;
 543                 file = tmp->vm_file;
 544                 if (file) {
 545                         struct inode *inode = file_inode(file);
 546                         struct address_space *mapping = file->f_mapping;
 547
 548                         get_file(file);
 549                         if (tmp->vm_flags & VM_DENYWRITE)
 550                                 atomic_dec(&inode->i_writecount);
 551                         i_mmap_lock_write(mapping);
 552                         if (tmp->vm_flags & VM_SHARED)
 553                                 atomic_inc(&mapping->i_mmap_writable);
 554                         flush_dcache_mmap_lock(mapping);
 555                         /* insert tmp into the share list, just after mpnt */
 556                         vma_interval_tree_insert_after(tmp, mpnt,
 557                                         &mapping->i_mmap);
 558                         flush_dcache_mmap_unlock(mapping);
 559                         i_mmap_unlock_write(mapping);
 560                 }
 561
 562                 /*
 563                  * Clear hugetlb-related page reserves for children. This only
 564                  * affects MAP_PRIVATE mappings. Faults generated by the child
 565                  * are not guaranteed to succeed, even if read-only
 566                  */
 567                 if (is_vm_hugetlb_page(tmp))
 568                         reset_vma_resv_huge_pages(tmp);
 569
 570                 /*
 571                  * Link in the new vma and copy the page table entries.
 572                  */
 573                 *pprev = tmp;
 574                 pprev = &tmp->vm_next;
 575                 tmp->vm_prev = prev;
 576                 prev = tmp;
 577
 578                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 579                 rb_link = &tmp->vm_rb.rb_right;
 580                 rb_parent = &tmp->vm_rb;
 581
 582                 mm->map_count++;
 583                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 584                         retval = copy_page_range(mm, oldmm, mpnt);
 585
 586                 if (tmp->vm_ops && tmp->vm_ops->open)
 587                         tmp->vm_ops->open(tmp);
 588
 589                 if (retval)
 590                         goto out;
 591         }
 592         /* a new mm has just been created */
 593         retval = arch_dup_mmap(oldmm, mm);
 594 out:
 595         up_write(&mm->mmap_sem);
 596         flush_tlb_mm(oldmm);
 597         up_write(&oldmm->mmap_sem);
 598         dup_userfaultfd_complete(&uf);
 599 fail_uprobe_end:
 600         uprobe_end_dup_mmap();
 601         return retval;
 602 fail_nomem_anon_vma_fork:
 603         mpol_put(vma_policy(tmp));
 604 fail_nomem_policy:
 605         vm_area_free(tmp);
 606 fail_nomem:
 607         retval = -ENOMEM;
 608         vm_unacct_memory(charge);
 609         goto out;
 610 }
 611
 612 static inline int mm_alloc_pgd(struct mm_struct *mm)
 613 {
 614         mm->pgd = pgd_alloc(mm);
 615         if (unlikely(!mm->pgd))
 616                 return -ENOMEM;
 617         return 0;
 618 }
 619
 620 static inline void mm_free_pgd(struct mm_struct *mm)
 621 {
 622         pgd_free(mm, mm->pgd);
 623 }
 624 #else
 625 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 626 {
 627         down_write(&oldmm->mmap_sem);
 628         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 629         up_write(&oldmm->mmap_sem);
 630         return 0;
 631 }
 632 #define mm_alloc_pgd(mm)        (0)
 633 #define mm_free_pgd(mm)
 634 #endif /* CONFIG_MMU */
 635
 636 static void check_mm(struct mm_struct *mm)
 637 {
 638         int i;
 639
 640         for (i = 0; i < NR_MM_COUNTERS; i++) {
 641                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 642
 643                 if (unlikely(x))
 644                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 645                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 646         }
 647
 648         if (mm_pgtables_bytes(mm))
 649                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 650                                 mm_pgtables_bytes(mm));
 651
 652 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 653         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 654 #endif
 655 }
 656
 657 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 658 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 659
 660 /*
 661  * Called when the last reference to the mm
 662  * is dropped: either by a lazy thread or by
 663  * mmput. Free the page directory and the mm.
 664  */
 665 void __mmdrop(struct mm_struct *mm)
 666 {
 667         BUG_ON(mm == &init_mm);
 668         WARN_ON_ONCE(mm == current->mm);
 669         WARN_ON_ONCE(mm == current->active_mm);
 670         mm_free_pgd(mm);
 671         destroy_context(mm);
 672         hmm_mm_destroy(mm);
 673         mmu_notifier_mm_destroy(mm);
 674         check_mm(mm);
 675         put_user_ns(mm->user_ns);
 676         free_mm(mm);
 677 }
 678 EXPORT_SYMBOL_GPL(__mmdrop);
 679
 680 static void mmdrop_async_fn(struct work_struct *work)
 681 {
 682         struct mm_struct *mm;
 683
 684         mm = container_of(work, struct mm_struct, async_put_work);
 685         __mmdrop(mm);
 686 }
 687
 688 static void mmdrop_async(struct mm_struct *mm)
 689 {
 690         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 691                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 692                 schedule_work(&mm->async_put_work);
 693         }
 694 }
 695
 696 static inline void free_signal_struct(struct signal_struct *sig)
 697 {
 698         taskstats_tgid_free(sig);
 699         sched_autogroup_exit(sig);
 700         /*
 701          * __mmdrop is not safe to call from softirq context on x86 due to
 702          * pgd_dtor so postpone it to the async context
 703          */
 704         if (sig->oom_mm)
 705                 mmdrop_async(sig->oom_mm);
 706         kmem_cache_free(signal_cachep, sig);
 707 }
 708
 709 static inline void put_signal_struct(struct signal_struct *sig)
 710 {
 711         if (atomic_dec_and_test(&sig->sigcnt))
 712                 free_signal_struct(sig);
 713 }
 714
 715 void __put_task_struct(struct task_struct *tsk)
 716 {
 717         WARN_ON(!tsk->exit_state);
 718         WARN_ON(atomic_read(&tsk->usage));
 719         WARN_ON(tsk == current);
 720
 721         cgroup_free(tsk);
 722         task_numa_free(tsk);
 723         security_task_free(tsk);
 724         exit_creds(tsk);
 725         delayacct_tsk_free(tsk);
 726         put_signal_struct(tsk->signal);
 727
 728         if (!profile_handoff_task(tsk))
 729                 free_task(tsk);
 730 }
 731 EXPORT_SYMBOL_GPL(__put_task_struct);
 732
 733 void __init __weak arch_task_cache_init(void) { }
 734
 735 /*
 736  * set_max_threads
 737  */
 738 static void set_max_threads(unsigned int max_threads_suggested)
 739 {
 740         u64 threads;
 741
 742         /*
 743          * The number of threads shall be limited such that the thread
 744          * structures may only consume a small part of the available memory.
 745          */
 746         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 747                 threads = MAX_THREADS;
 748         else
 749                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 750                                     (u64) THREAD_SIZE * 8UL);
 751
 752         if (threads > max_threads_suggested)
 753                 threads = max_threads_suggested;
 754
 755         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 756 }
 757
 758 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 759 /* Initialized by the architecture: */
 760 int arch_task_struct_size __read_mostly;
 761 #endif
 762
 763 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 764 {
 765         /* Fetch thread_struct whitelist for the architecture. */
 766         arch_thread_struct_whitelist(offset, size);
 767
 768         /*
 769          * Handle zero-sized whitelist or empty thread_struct, otherwise
 770          * adjust offset to position of thread_struct in task_struct.
 771          */
 772         if (unlikely(*size == 0))
 773                 *offset = 0;
 774         else
 775                 *offset += offsetof(struct task_struct, thread);
 776 }
 777
 778 void __init fork_init(void)
 779 {
 780         int i;
 781 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 782 #ifndef ARCH_MIN_TASKALIGN
 783 #define ARCH_MIN_TASKALIGN      0
 784 #endif
 785         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 786         unsigned long useroffset, usersize;
 787
 788         /* create a slab on which task_structs can be allocated */
 789         task_struct_whitelist(&useroffset, &usersize);
 790         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 791                         arch_task_struct_size, align,
 792                         SLAB_PANIC|SLAB_ACCOUNT,
 793                         useroffset, usersize, NULL);
 794 #endif
 795
 796         /* do the arch specific task caches init */
 797         arch_task_cache_init();
 798
 799         set_max_threads(MAX_THREADS);
 800
 801         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 802         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 803         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 804                 init_task.signal->rlim[RLIMIT_NPROC];
 805
 806         for (i = 0; i < UCOUNT_COUNTS; i++) {
 807                 init_user_ns.ucount_max[i] = max_threads/2;
 808         }
 809
 810 #ifdef CONFIG_VMAP_STACK
 811         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 812                           NULL, free_vm_stack_cache);
 813 #endif
 814
 815         lockdep_init_task(&init_task);
 816 }
 817
 818 int __weak arch_dup_task_struct(struct task_struct *dst,
 819                                                struct task_struct *src)
 820 {
 821         *dst = *src;
 822         return 0;
 823 }
 824
 825 void set_task_stack_end_magic(struct task_struct *tsk)
 826 {
 827         unsigned long *stackend;
 828
 829         stackend = end_of_stack(tsk);
 830         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 831 }
 832
 833 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 834 {
 835         struct task_struct *tsk;
 836         unsigned long *stack;
 837         struct vm_struct *stack_vm_area;
 838         int err;
 839
 840         if (node == NUMA_NO_NODE)
 841                 node = tsk_fork_get_node(orig);
 842         tsk = alloc_task_struct_node(node);
 843         if (!tsk)
 844                 return NULL;
 845
 846         stack = alloc_thread_stack_node(tsk, node);
 847         if (!stack)
 848                 goto free_tsk;
 849
 850         if (memcg_charge_kernel_stack(tsk))
 851                 goto free_stack;
 852
 853         stack_vm_area = task_stack_vm_area(tsk);
 854
 855         err = arch_dup_task_struct(tsk, orig);
 856
 857         /*
 858          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 859          * sure they're properly initialized before using any stack-related
 860          * functions again.
 861          */
 862         tsk->stack = stack;
 863 #ifdef CONFIG_VMAP_STACK
 864         tsk->stack_vm_area = stack_vm_area;
 865 #endif
 866 #ifdef CONFIG_THREAD_INFO_IN_TASK
 867         atomic_set(&tsk->stack_refcount, 1);
 868 #endif
 869
 870         if (err)
 871                 goto free_stack;
 872
 873 #ifdef CONFIG_SECCOMP
 874         /*
 875          * We must handle setting up seccomp filters once we're under
 876          * the sighand lock in case orig has changed between now and
 877          * then. Until then, filter must be NULL to avoid messing up
 878          * the usage counts on the error path calling free_task.
 879          */
 880         tsk->seccomp.filter = NULL;
 881 #endif
 882
 883         setup_thread_stack(tsk, orig);
 884         clear_user_return_notifier(tsk);
 885         clear_tsk_need_resched(tsk);
 886         set_task_stack_end_magic(tsk);
 887
 888 #ifdef CONFIG_STACKPROTECTOR
 889         tsk->stack_canary = get_random_canary();
 890 #endif
 891
 892         /*
 893          * One for us, one for whoever does the "release_task()" (usually
 894          * parent)
 895          */
 896         atomic_set(&tsk->usage, 2);
 897 #ifdef CONFIG_BLK_DEV_IO_TRACE
 898         tsk->btrace_seq = 0;
 899 #endif
 900         tsk->splice_pipe = NULL;
 901         tsk->task_frag.page = NULL;
 902         tsk->wake_q.next = NULL;
 903
 904         account_kernel_stack(tsk, 1);
 905
 906         kcov_task_init(tsk);
 907
 908 #ifdef CONFIG_FAULT_INJECTION
 909         tsk->fail_nth = 0;
 910 #endif
 911
 912 #ifdef CONFIG_BLK_CGROUP
 913         tsk->throttle_queue = NULL;
 914         tsk->use_memdelay = 0;
 915 #endif
 916
 917 #ifdef CONFIG_MEMCG
 918         tsk->active_memcg = NULL;
 919 #endif
 920         return tsk;
 921
 922 free_stack:
 923         free_thread_stack(tsk);
 924 free_tsk:
 925         free_task_struct(tsk);
 926         return NULL;
 927 }
 928
 929 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 930
 931 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 932
 933 static int __init coredump_filter_setup(char *s)
 934 {
 935         default_dump_filter =
 936                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 937                 MMF_DUMP_FILTER_MASK;
 938         return 1;
 939 }
 940
 941 __setup("coredump_filter=", coredump_filter_setup);
 942
 943 #include <linux/init_task.h>
 944
 945 static void mm_init_aio(struct mm_struct *mm)
 946 {
 947 #ifdef CONFIG_AIO
 948         spin_lock_init(&mm->ioctx_lock);
 949         mm->ioctx_table = NULL;
 950 #endif
 951 }
 952
 953 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 954 {
 955 #ifdef CONFIG_MEMCG
 956         mm->owner = p;
 957 #endif
 958 }
 959
 960 static void mm_init_uprobes_state(struct mm_struct *mm)
 961 {
 962 #ifdef CONFIG_UPROBES
 963         mm->uprobes_state.xol_area = NULL;
 964 #endif
 965 }
 966
 967 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 968         struct user_namespace *user_ns)
 969 {
 970         mm->mmap = NULL;
 971         mm->mm_rb = RB_ROOT;
 972         mm->vmacache_seqnum = 0;
 973         atomic_set(&mm->mm_users, 1);
 974         atomic_set(&mm->mm_count, 1);
 975         init_rwsem(&mm->mmap_sem);
 976         INIT_LIST_HEAD(&mm->mmlist);
 977         mm->core_state = NULL;
 978         mm_pgtables_bytes_init(mm);
 979         mm->map_count = 0;
 980         mm->locked_vm = 0;
 981         mm->pinned_vm = 0;
 982         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 983         spin_lock_init(&mm->page_table_lock);
 984         spin_lock_init(&mm->arg_lock);
 985         mm_init_cpumask(mm);
 986         mm_init_aio(mm);
 987         mm_init_owner(mm, p);
 988         RCU_INIT_POINTER(mm->exe_file, NULL);
 989         mmu_notifier_mm_init(mm);
 990         hmm_mm_init(mm);
 991         init_tlb_flush_pending(mm);
 992 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 993         mm->pmd_huge_pte = NULL;
 994 #endif
 995         mm_init_uprobes_state(mm);
 996
 997         if (current->mm) {
 998                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 999                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1000         } else {
1001                 mm->flags = default_dump_filter;
1002                 mm->def_flags = 0;
1003         }
1004
1005         if (mm_alloc_pgd(mm))
1006                 goto fail_nopgd;
1007
1008         if (init_new_context(p, mm))
1009                 goto fail_nocontext;
1010
1011         mm->user_ns = get_user_ns(user_ns);
1012         return mm;
1013
1014 fail_nocontext:
1015         mm_free_pgd(mm);
1016 fail_nopgd:
1017         free_mm(mm);
1018         return NULL;
1019 }
1020
1021 /*
1022  * Allocate and initialize an mm_struct.
1023  */
1024 struct mm_struct *mm_alloc(void)
1025 {
1026         struct mm_struct *mm;
1027
1028         mm = allocate_mm();
1029         if (!mm)
1030                 return NULL;
1031
1032         memset(mm, 0, sizeof(*mm));
1033         return mm_init(mm, current, current_user_ns());
1034 }
1035
1036 static inline void __mmput(struct mm_struct *mm)
1037 {
1038         VM_BUG_ON(atomic_read(&mm->mm_users));
1039
1040         uprobe_clear_state(mm);
1041         exit_aio(mm);
1042         ksm_exit(mm);
1043         khugepaged_exit(mm); /* must run before exit_mmap */
1044         exit_mmap(mm);
1045         mm_put_huge_zero_page(mm);
1046         set_mm_exe_file(mm, NULL);
1047         if (!list_empty(&mm->mmlist)) {
1048                 spin_lock(&mmlist_lock);
1049                 list_del(&mm->mmlist);
1050                 spin_unlock(&mmlist_lock);
1051         }
1052         if (mm->binfmt)
1053                 module_put(mm->binfmt->module);
1054         mmdrop(mm);
1055 }
1056
1057 /*
1058  * Decrement the use count and release all resources for an mm.
1059  */
1060 void mmput(struct mm_struct *mm)
1061 {
1062         might_sleep();
1063
1064         if (atomic_dec_and_test(&mm->mm_users))
1065                 __mmput(mm);
1066 }
1067 EXPORT_SYMBOL_GPL(mmput);
1068
1069 #ifdef CONFIG_MMU
1070 static void mmput_async_fn(struct work_struct *work)
1071 {
1072         struct mm_struct *mm = container_of(work, struct mm_struct,
1073                                             async_put_work);
1074
1075         __mmput(mm);
1076 }
1077
1078 void mmput_async(struct mm_struct *mm)
1079 {
1080         if (atomic_dec_and_test(&mm->mm_users)) {
1081                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1082                 schedule_work(&mm->async_put_work);
1083         }
1084 }
1085 #endif
1086
1087 /**
1088  * set_mm_exe_file - change a reference to the mm's executable file
1089  *
1090  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1091  *
1092  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1093  * invocations: in mmput() nobody alive left, in execve task is single
1094  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1095  * mm->exe_file, but does so without using set_mm_exe_file() in order
1096  * to do avoid the need for any locks.
1097  */
1098 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1099 {
1100         struct file *old_exe_file;
1101
1102         /*
1103          * It is safe to dereference the exe_file without RCU as
1104          * this function is only called if nobody else can access
1105          * this mm -- see comment above for justification.
1106          */
1107         old_exe_file = rcu_dereference_raw(mm->exe_file);
1108
1109         if (new_exe_file)
1110                 get_file(new_exe_file);
1111         rcu_assign_pointer(mm->exe_file, new_exe_file);
1112         if (old_exe_file)
1113                 fput(old_exe_file);
1114 }
1115
1116 /**
1117  * get_mm_exe_file - acquire a reference to the mm's executable file
1118  *
1119  * Returns %NULL if mm has no associated executable file.
1120  * User must release file via fput().
1121  */
1122 struct file *get_mm_exe_file(struct mm_struct *mm)
1123 {
1124         struct file *exe_file;
1125
1126         rcu_read_lock();
1127         exe_file = rcu_dereference(mm->exe_file);
1128         if (exe_file && !get_file_rcu(exe_file))
1129                 exe_file = NULL;
1130         rcu_read_unlock();
1131         return exe_file;
1132 }
1133 EXPORT_SYMBOL(get_mm_exe_file);
1134
1135 /**
1136  * get_task_exe_file - acquire a reference to the task's executable file
1137  *
1138  * Returns %NULL if task's mm (if any) has no associated executable file or
1139  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1140  * User must release file via fput().
1141  */
1142 struct file *get_task_exe_file(struct task_struct *task)
1143 {
1144         struct file *exe_file = NULL;
1145         struct mm_struct *mm;
1146
1147         task_lock(task);
1148         mm = task->mm;
1149         if (mm) {
1150                 if (!(task->flags & PF_KTHREAD))
1151                         exe_file = get_mm_exe_file(mm);
1152         }
1153         task_unlock(task);
1154         return exe_file;
1155 }
1156 EXPORT_SYMBOL(get_task_exe_file);
1157
1158 /**
1159  * get_task_mm - acquire a reference to the task's mm
1160  *
1161  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1162  * this kernel workthread has transiently adopted a user mm with use_mm,
1163  * to do its AIO) is not set and if so returns a reference to it, after
1164  * bumping up the use count.  User must release the mm via mmput()
1165  * after use.  Typically used by /proc and ptrace.
1166  */
1167 struct mm_struct *get_task_mm(struct task_struct *task)
1168 {
1169         struct mm_struct *mm;
1170
1171         task_lock(task);
1172         mm = task->mm;
1173         if (mm) {
1174                 if (task->flags & PF_KTHREAD)
1175                         mm = NULL;
1176                 else
1177                         mmget(mm);
1178         }
1179         task_unlock(task);
1180         return mm;
1181 }
1182 EXPORT_SYMBOL_GPL(get_task_mm);
1183
1184 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1185 {
1186         struct mm_struct *mm;
1187         int err;
1188
1189         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1190         if (err)
1191                 return ERR_PTR(err);
1192
1193         mm = get_task_mm(task);
1194         if (mm && mm != current->mm &&
1195                         !ptrace_may_access(task, mode)) {
1196                 mmput(mm);
1197                 mm = ERR_PTR(-EACCES);
1198         }
1199         mutex_unlock(&task->signal->cred_guard_mutex);
1200
1201         return mm;
1202 }
1203
1204 static void complete_vfork_done(struct task_struct *tsk)
1205 {
1206         struct completion *vfork;
1207
1208         task_lock(tsk);
1209         vfork = tsk->vfork_done;
1210         if (likely(vfork)) {
1211                 tsk->vfork_done = NULL;
1212                 complete(vfork);
1213         }
1214         task_unlock(tsk);
1215 }
1216
1217 static int wait_for_vfork_done(struct task_struct *child,
1218                                 struct completion *vfork)
1219 {
1220         int killed;
1221
1222         freezer_do_not_count();
1223         killed = wait_for_completion_killable(vfork);
1224         freezer_count();
1225
1226         if (killed) {
1227                 task_lock(child);
1228                 child->vfork_done = NULL;
1229                 task_unlock(child);
1230         }
1231
1232         put_task_struct(child);
1233         return killed;
1234 }
1235
1236 /* Please note the differences between mmput and mm_release.
1237  * mmput is called whenever we stop holding onto a mm_struct,
1238  * error success whatever.
1239  *
1240  * mm_release is called after a mm_struct has been removed
1241  * from the current process.
1242  *
1243  * This difference is important for error handling, when we
1244  * only half set up a mm_struct for a new process and need to restore
1245  * the old one.  Because we mmput the new mm_struct before
1246  * restoring the old one. . .
1247  * Eric Biederman 10 January 1998
1248  */
1249 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1250 {
1251         /* Get rid of any futexes when releasing the mm */
1252 #ifdef CONFIG_FUTEX
1253         if (unlikely(tsk->robust_list)) {
1254                 exit_robust_list(tsk);
1255                 tsk->robust_list = NULL;
1256         }
1257 #ifdef CONFIG_COMPAT
1258         if (unlikely(tsk->compat_robust_list)) {
1259                 compat_exit_robust_list(tsk);
1260                 tsk->compat_robust_list = NULL;
1261         }
1262 #endif
1263         if (unlikely(!list_empty(&tsk->pi_state_list)))
1264                 exit_pi_state_list(tsk);
1265 #endif
1266
1267         uprobe_free_utask(tsk);
1268
1269         /* Get rid of any cached register state */
1270         deactivate_mm(tsk, mm);
1271
1272         /*
1273          * Signal userspace if we're not exiting with a core dump
1274          * because we want to leave the value intact for debugging
1275          * purposes.
1276          */
1277         if (tsk->clear_child_tid) {
1278                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1279                     atomic_read(&mm->mm_users) > 1) {
1280                         /*
1281                          * We don't check the error code - if userspace has
1282                          * not set up a proper pointer then tough luck.
1283                          */
1284                         put_user(0, tsk->clear_child_tid);
1285                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1286                                         1, NULL, NULL, 0, 0);
1287                 }
1288                 tsk->clear_child_tid = NULL;
1289         }
1290
1291         /*
1292          * All done, finally we can wake up parent and return this mm to him.
1293          * Also kthread_stop() uses this completion for synchronization.
1294          */
1295         if (tsk->vfork_done)
1296                 complete_vfork_done(tsk);
1297 }
1298
1299 /*
1300  * Allocate a new mm structure and copy contents from the
1301  * mm structure of the passed in task structure.
1302  */
1303 static struct mm_struct *dup_mm(struct task_struct *tsk)
1304 {
1305         struct mm_struct *mm, *oldmm = current->mm;
1306         int err;
1307
1308         mm = allocate_mm();
1309         if (!mm)
1310                 goto fail_nomem;
1311
1312         memcpy(mm, oldmm, sizeof(*mm));
1313
1314         if (!mm_init(mm, tsk, mm->user_ns))
1315                 goto fail_nomem;
1316
1317         err = dup_mmap(mm, oldmm);
1318         if (err)
1319                 goto free_pt;
1320
1321         mm->hiwater_rss = get_mm_rss(mm);
1322         mm->hiwater_vm = mm->total_vm;
1323
1324         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1325                 goto free_pt;
1326
1327         return mm;
1328
1329 free_pt:
1330         /* don't put binfmt in mmput, we haven't got module yet */
1331         mm->binfmt = NULL;
1332         mmput(mm);
1333
1334 fail_nomem:
1335         return NULL;
1336 }
1337
1338 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1339 {
1340         struct mm_struct *mm, *oldmm;
1341         int retval;
1342
1343         tsk->min_flt = tsk->maj_flt = 0;
1344         tsk->nvcsw = tsk->nivcsw = 0;
1345 #ifdef CONFIG_DETECT_HUNG_TASK
1346         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1347         tsk->last_switch_time = 0;
1348 #endif
1349
1350         tsk->mm = NULL;
1351         tsk->active_mm = NULL;
1352
1353         /*
1354          * Are we cloning a kernel thread?
1355          *
1356          * We need to steal a active VM for that..
1357          */
1358         oldmm = current->mm;
1359         if (!oldmm)
1360                 return 0;
1361
1362         /* initialize the new vmacache entries */
1363         vmacache_flush(tsk);
1364
1365         if (clone_flags & CLONE_VM) {
1366                 mmget(oldmm);
1367                 mm = oldmm;
1368                 goto good_mm;
1369         }
1370
1371         retval = -ENOMEM;
1372         mm = dup_mm(tsk);
1373         if (!mm)
1374                 goto fail_nomem;
1375
1376 good_mm:
1377         tsk->mm = mm;
1378         tsk->active_mm = mm;
1379         return 0;
1380
1381 fail_nomem:
1382         return retval;
1383 }
1384
1385 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1386 {
1387         struct fs_struct *fs = current->fs;
1388         if (clone_flags & CLONE_FS) {
1389                 /* tsk->fs is already what we want */
1390                 spin_lock(&fs->lock);
1391                 if (fs->in_exec) {
1392                         spin_unlock(&fs->lock);
1393                         return -EAGAIN;
1394                 }
1395                 fs->users++;
1396                 spin_unlock(&fs->lock);
1397                 return 0;
1398         }
1399         tsk->fs = copy_fs_struct(fs);
1400         if (!tsk->fs)
1401                 return -ENOMEM;
1402         return 0;
1403 }
1404
1405 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1406 {
1407         struct files_struct *oldf, *newf;
1408         int error = 0;
1409
1410         /*
1411          * A background process may not have any files ...
1412          */
1413         oldf = current->files;
1414         if (!oldf)
1415                 goto out;
1416
1417         if (clone_flags & CLONE_FILES) {
1418                 atomic_inc(&oldf->count);
1419                 goto out;
1420         }
1421
1422         newf = dup_fd(oldf, &error);
1423         if (!newf)
1424                 goto out;
1425
1426         tsk->files = newf;
1427         error = 0;
1428 out:
1429         return error;
1430 }
1431
1432 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1433 {
1434 #ifdef CONFIG_BLOCK
1435         struct io_context *ioc = current->io_context;
1436         struct io_context *new_ioc;
1437
1438         if (!ioc)
1439                 return 0;
1440         /*
1441          * Share io context with parent, if CLONE_IO is set
1442          */
1443         if (clone_flags & CLONE_IO) {
1444                 ioc_task_link(ioc);
1445                 tsk->io_context = ioc;
1446         } else if (ioprio_valid(ioc->ioprio)) {
1447                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1448                 if (unlikely(!new_ioc))
1449                         return -ENOMEM;
1450
1451                 new_ioc->ioprio = ioc->ioprio;
1452                 put_io_context(new_ioc);
1453         }
1454 #endif
1455         return 0;
1456 }
1457
1458 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1459 {
1460         struct sighand_struct *sig;
1461
1462         if (clone_flags & CLONE_SIGHAND) {
1463                 atomic_inc(&current->sighand->count);
1464                 return 0;
1465         }
1466         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1467         rcu_assign_pointer(tsk->sighand, sig);
1468         if (!sig)
1469                 return -ENOMEM;
1470
1471         atomic_set(&sig->count, 1);
1472         spin_lock_irq(&current->sighand->siglock);
1473         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1474         spin_unlock_irq(&current->sighand->siglock);
1475         return 0;
1476 }
1477
1478 void __cleanup_sighand(struct sighand_struct *sighand)
1479 {
1480         if (atomic_dec_and_test(&sighand->count)) {
1481                 signalfd_cleanup(sighand);
1482                 /*
1483                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1484                  * without an RCU grace period, see __lock_task_sighand().
1485                  */
1486                 kmem_cache_free(sighand_cachep, sighand);
1487         }
1488 }
1489
1490 #ifdef CONFIG_POSIX_TIMERS
1491 /*
1492  * Initialize POSIX timer handling for a thread group.
1493  */
1494 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1495 {
1496         unsigned long cpu_limit;
1497
1498         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1499         if (cpu_limit != RLIM_INFINITY) {
1500                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1501                 sig->cputimer.running = true;
1502         }
1503
1504         /* The timer lists. */
1505         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1506         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1507         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1508 }
1509 #else
1510 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1511 #endif
1512
1513 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1514 {
1515         struct signal_struct *sig;
1516
1517         if (clone_flags & CLONE_THREAD)
1518                 return 0;
1519
1520         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1521         tsk->signal = sig;
1522         if (!sig)
1523                 return -ENOMEM;
1524
1525         sig->nr_threads = 1;
1526         atomic_set(&sig->live, 1);
1527         atomic_set(&sig->sigcnt, 1);
1528
1529         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1530         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1531         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1532
1533         init_waitqueue_head(&sig->wait_chldexit);
1534         sig->curr_target = tsk;
1535         init_sigpending(&sig->shared_pending);
1536         INIT_HLIST_HEAD(&sig->multiprocess);
1537         seqlock_init(&sig->stats_lock);
1538         prev_cputime_init(&sig->prev_cputime);
1539
1540 #ifdef CONFIG_POSIX_TIMERS
1541         INIT_LIST_HEAD(&sig->posix_timers);
1542         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1543         sig->real_timer.function = it_real_fn;
1544 #endif
1545
1546         task_lock(current->group_leader);
1547         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1548         task_unlock(current->group_leader);
1549
1550         posix_cpu_timers_init_group(sig);
1551
1552         tty_audit_fork(sig);
1553         sched_autogroup_fork(sig);
1554
1555         sig->oom_score_adj = current->signal->oom_score_adj;
1556         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1557
1558         mutex_init(&sig->cred_guard_mutex);
1559
1560         return 0;
1561 }
1562
1563 static void copy_seccomp(struct task_struct *p)
1564 {
1565 #ifdef CONFIG_SECCOMP
1566         /*
1567          * Must be called with sighand->lock held, which is common to
1568          * all threads in the group. Holding cred_guard_mutex is not
1569          * needed because this new task is not yet running and cannot
1570          * be racing exec.
1571          */
1572         assert_spin_locked(&current->sighand->siglock);
1573
1574         /* Ref-count the new filter user, and assign it. */
1575         get_seccomp_filter(current);
1576         p->seccomp = current->seccomp;
1577
1578         /*
1579          * Explicitly enable no_new_privs here in case it got set
1580          * between the task_struct being duplicated and holding the
1581          * sighand lock. The seccomp state and nnp must be in sync.
1582          */
1583         if (task_no_new_privs(current))
1584                 task_set_no_new_privs(p);
1585
1586         /*
1587          * If the parent gained a seccomp mode after copying thread
1588          * flags and between before we held the sighand lock, we have
1589          * to manually enable the seccomp thread flag here.
1590          */
1591         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1592                 set_tsk_thread_flag(p, TIF_SECCOMP);
1593 #endif
1594 }
1595
1596 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1597 {
1598         current->clear_child_tid = tidptr;
1599
1600         return task_pid_vnr(current);
1601 }
1602
1603 static void rt_mutex_init_task(struct task_struct *p)
1604 {
1605         raw_spin_lock_init(&p->pi_lock);
1606 #ifdef CONFIG_RT_MUTEXES
1607         p->pi_waiters = RB_ROOT_CACHED;
1608         p->pi_top_task = NULL;
1609         p->pi_blocked_on = NULL;
1610 #endif
1611 }
1612
1613 #ifdef CONFIG_POSIX_TIMERS
1614 /*
1615  * Initialize POSIX timer handling for a single task.
1616  */
1617 static void posix_cpu_timers_init(struct task_struct *tsk)
1618 {
1619         tsk->cputime_expires.prof_exp = 0;
1620         tsk->cputime_expires.virt_exp = 0;
1621         tsk->cputime_expires.sched_exp = 0;
1622         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1623         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1624         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1625 }
1626 #else
1627 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1628 #endif
1629
1630 static inline void init_task_pid_links(struct task_struct *task)
1631 {
1632         enum pid_type type;
1633
1634         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1635                 INIT_HLIST_NODE(&task->pid_links[type]);
1636         }
1637 }
1638
1639 static inline void
1640 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1641 {
1642         if (type == PIDTYPE_PID)
1643                 task->thread_pid = pid;
1644         else
1645                 task->signal->pids[type] = pid;
1646 }
1647
1648 static inline void rcu_copy_process(struct task_struct *p)
1649 {
1650 #ifdef CONFIG_PREEMPT_RCU
1651         p->rcu_read_lock_nesting = 0;
1652         p->rcu_read_unlock_special.s = 0;
1653         p->rcu_blocked_node = NULL;
1654         INIT_LIST_HEAD(&p->rcu_node_entry);
1655 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1656 #ifdef CONFIG_TASKS_RCU
1657         p->rcu_tasks_holdout = false;
1658         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1659         p->rcu_tasks_idle_cpu = -1;
1660 #endif /* #ifdef CONFIG_TASKS_RCU */
1661 }
1662
1663 /*
1664  * This creates a new process as a copy of the old one,
1665  * but does not actually start it yet.
1666  *
1667  * It copies the registers, and all the appropriate
1668  * parts of the process environment (as per the clone
1669  * flags). The actual kick-off is left to the caller.
1670  */
1671 static __latent_entropy struct task_struct *copy_process(
1672                                         unsigned long clone_flags,
1673                                         unsigned long stack_start,
1674                                         unsigned long stack_size,
1675                                         int __user *child_tidptr,
1676                                         struct pid *pid,
1677                                         int trace,
1678                                         unsigned long tls,
1679                                         int node)
1680 {
1681         int retval;
1682         struct task_struct *p;
1683         struct multiprocess_signals delayed;
1684
1685         /*
1686          * Don't allow sharing the root directory with processes in a different
1687          * namespace
1688          */
1689         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1690                 return ERR_PTR(-EINVAL);
1691
1692         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1693                 return ERR_PTR(-EINVAL);
1694
1695         /*
1696          * Thread groups must share signals as well, and detached threads
1697          * can only be started up within the thread group.
1698          */
1699         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1700                 return ERR_PTR(-EINVAL);
1701
1702         /*
1703          * Shared signal handlers imply shared VM. By way of the above,
1704          * thread groups also imply shared VM. Blocking this case allows
1705          * for various simplifications in other code.
1706          */
1707         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1708                 return ERR_PTR(-EINVAL);
1709
1710         /*
1711          * Siblings of global init remain as zombies on exit since they are
1712          * not reaped by their parent (swapper). To solve this and to avoid
1713          * multi-rooted process trees, prevent global and container-inits
1714          * from creating siblings.
1715          */
1716         if ((clone_flags & CLONE_PARENT) &&
1717                                 current->signal->flags & SIGNAL_UNKILLABLE)
1718                 return ERR_PTR(-EINVAL);
1719
1720         /*
1721          * If the new process will be in a different pid or user namespace
1722          * do not allow it to share a thread group with the forking task.
1723          */
1724         if (clone_flags & CLONE_THREAD) {
1725                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1726                     (task_active_pid_ns(current) !=
1727                                 current->nsproxy->pid_ns_for_children))
1728                         return ERR_PTR(-EINVAL);
1729         }
1730
1731         /*
1732          * Force any signals received before this point to be delivered
1733          * before the fork happens.  Collect up signals sent to multiple
1734          * processes that happen during the fork and delay them so that
1735          * they appear to happen after the fork.
1736          */
1737         sigemptyset(&delayed.signal);
1738         INIT_HLIST_NODE(&delayed.node);
1739
1740         spin_lock_irq(&current->sighand->siglock);
1741         if (!(clone_flags & CLONE_THREAD))
1742                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1743         recalc_sigpending();
1744         spin_unlock_irq(&current->sighand->siglock);
1745         retval = -ERESTARTNOINTR;
1746         if (signal_pending(current))
1747                 goto fork_out;
1748
1749         retval = -ENOMEM;
1750         p = dup_task_struct(current, node);
1751         if (!p)
1752                 goto fork_out;
1753
1754         /*
1755          * This _must_ happen before we call free_task(), i.e. before we jump
1756          * to any of the bad_fork_* labels. This is to avoid freeing
1757          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1758          * kernel threads (PF_KTHREAD).
1759          */
1760         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1761         /*
1762          * Clear TID on mm_release()?
1763          */
1764         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1765
1766         ftrace_graph_init_task(p);
1767
1768         rt_mutex_init_task(p);
1769
1770 #ifdef CONFIG_PROVE_LOCKING
1771         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1772         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1773 #endif
1774         retval = -EAGAIN;
1775         if (atomic_read(&p->real_cred->user->processes) >=
1776                         task_rlimit(p, RLIMIT_NPROC)) {
1777                 if (p->real_cred->user != INIT_USER &&
1778                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1779                         goto bad_fork_free;
1780         }
1781         current->flags &= ~PF_NPROC_EXCEEDED;
1782
1783         retval = copy_creds(p, clone_flags);
1784         if (retval < 0)
1785                 goto bad_fork_free;
1786
1787         /*
1788          * If multiple threads are within copy_process(), then this check
1789          * triggers too late. This doesn't hurt, the check is only there
1790          * to stop root fork bombs.
1791          */
1792         retval = -EAGAIN;
1793         if (nr_threads >= max_threads)
1794                 goto bad_fork_cleanup_count;
1795
1796         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1797         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1798         p->flags |= PF_FORKNOEXEC;
1799         INIT_LIST_HEAD(&p->children);
1800         INIT_LIST_HEAD(&p->sibling);
1801         rcu_copy_process(p);
1802         p->vfork_done = NULL;
1803         spin_lock_init(&p->alloc_lock);
1804
1805         init_sigpending(&p->pending);
1806
1807         p->utime = p->stime = p->gtime = 0;
1808 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1809         p->utimescaled = p->stimescaled = 0;
1810 #endif
1811         prev_cputime_init(&p->prev_cputime);
1812
1813 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1814         seqcount_init(&p->vtime.seqcount);
1815         p->vtime.starttime = 0;
1816         p->vtime.state = VTIME_INACTIVE;
1817 #endif
1818
1819 #if defined(SPLIT_RSS_COUNTING)
1820         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1821 #endif
1822
1823         p->default_timer_slack_ns = current->timer_slack_ns;
1824
1825         task_io_accounting_init(&p->ioac);
1826         acct_clear_integrals(p);
1827
1828         posix_cpu_timers_init(p);
1829
1830         p->start_time = ktime_get_ns();
1831         p->real_start_time = ktime_get_boot_ns();
1832         p->io_context = NULL;
1833         audit_set_context(p, NULL);
1834         cgroup_fork(p);
1835 #ifdef CONFIG_NUMA
1836         p->mempolicy = mpol_dup(p->mempolicy);
1837         if (IS_ERR(p->mempolicy)) {
1838                 retval = PTR_ERR(p->mempolicy);
1839                 p->mempolicy = NULL;
1840                 goto bad_fork_cleanup_threadgroup_lock;
1841         }
1842 #endif
1843 #ifdef CONFIG_CPUSETS
1844         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1845         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1846         seqcount_init(&p->mems_allowed_seq);
1847 #endif
1848 #ifdef CONFIG_TRACE_IRQFLAGS
1849         p->irq_events = 0;
1850         p->hardirqs_enabled = 0;
1851         p->hardirq_enable_ip = 0;
1852         p->hardirq_enable_event = 0;
1853         p->hardirq_disable_ip = _THIS_IP_;
1854         p->hardirq_disable_event = 0;
1855         p->softirqs_enabled = 1;
1856         p->softirq_enable_ip = _THIS_IP_;
1857         p->softirq_enable_event = 0;
1858         p->softirq_disable_ip = 0;
1859         p->softirq_disable_event = 0;
1860         p->hardirq_context = 0;
1861         p->softirq_context = 0;
1862 #endif
1863
1864         p->pagefault_disabled = 0;
1865
1866 #ifdef CONFIG_LOCKDEP
1867         p->lockdep_depth = 0; /* no locks held yet */
1868         p->curr_chain_key = 0;
1869         p->lockdep_recursion = 0;
1870         lockdep_init_task(p);
1871 #endif
1872
1873 #ifdef CONFIG_DEBUG_MUTEXES
1874         p->blocked_on = NULL; /* not blocked yet */
1875 #endif
1876 #ifdef CONFIG_BCACHE
1877         p->sequential_io        = 0;
1878         p->sequential_io_avg    = 0;
1879 #endif
1880
1881         /* Perform scheduler related setup. Assign this task to a CPU. */
1882         retval = sched_fork(clone_flags, p);
1883         if (retval)
1884                 goto bad_fork_cleanup_policy;
1885
1886         retval = perf_event_init_task(p);
1887         if (retval)
1888                 goto bad_fork_cleanup_policy;
1889         retval = audit_alloc(p);
1890         if (retval)
1891                 goto bad_fork_cleanup_perf;
1892         /* copy all the process information */
1893         shm_init_task(p);
1894         retval = security_task_alloc(p, clone_flags);
1895         if (retval)
1896                 goto bad_fork_cleanup_audit;
1897         retval = copy_semundo(clone_flags, p);
1898         if (retval)
1899                 goto bad_fork_cleanup_security;
1900         retval = copy_files(clone_flags, p);
1901         if (retval)
1902                 goto bad_fork_cleanup_semundo;
1903         retval = copy_fs(clone_flags, p);
1904         if (retval)
1905                 goto bad_fork_cleanup_files;
1906         retval = copy_sighand(clone_flags, p);
1907         if (retval)
1908                 goto bad_fork_cleanup_fs;
1909         retval = copy_signal(clone_flags, p);
1910         if (retval)
1911                 goto bad_fork_cleanup_sighand;
1912         retval = copy_mm(clone_flags, p);
1913         if (retval)
1914                 goto bad_fork_cleanup_signal;
1915         retval = copy_namespaces(clone_flags, p);
1916         if (retval)
1917                 goto bad_fork_cleanup_mm;
1918         retval = copy_io(clone_flags, p);
1919         if (retval)
1920                 goto bad_fork_cleanup_namespaces;
1921         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1922         if (retval)
1923                 goto bad_fork_cleanup_io;
1924
1925         if (pid != &init_struct_pid) {
1926                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1927                 if (IS_ERR(pid)) {
1928                         retval = PTR_ERR(pid);
1929                         goto bad_fork_cleanup_thread;
1930                 }
1931         }
1932
1933 #ifdef CONFIG_BLOCK
1934         p->plug = NULL;
1935 #endif
1936 #ifdef CONFIG_FUTEX
1937         p->robust_list = NULL;
1938 #ifdef CONFIG_COMPAT
1939         p->compat_robust_list = NULL;
1940 #endif
1941         INIT_LIST_HEAD(&p->pi_state_list);
1942         p->pi_state_cache = NULL;
1943 #endif
1944         /*
1945          * sigaltstack should be cleared when sharing the same VM
1946          */
1947         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1948                 sas_ss_reset(p);
1949
1950         /*
1951          * Syscall tracing and stepping should be turned off in the
1952          * child regardless of CLONE_PTRACE.
1953          */
1954         user_disable_single_step(p);
1955         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1956 #ifdef TIF_SYSCALL_EMU
1957         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1958 #endif
1959         clear_all_latency_tracing(p);
1960
1961         /* ok, now we should be set up.. */
1962         p->pid = pid_nr(pid);
1963         if (clone_flags & CLONE_THREAD) {
1964                 p->exit_signal = -1;
1965                 p->group_leader = current->group_leader;
1966                 p->tgid = current->tgid;
1967         } else {
1968                 if (clone_flags & CLONE_PARENT)
1969                         p->exit_signal = current->group_leader->exit_signal;
1970                 else
1971                         p->exit_signal = (clone_flags & CSIGNAL);
1972                 p->group_leader = p;
1973                 p->tgid = p->pid;
1974         }
1975
1976         p->nr_dirtied = 0;
1977         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1978         p->dirty_paused_when = 0;
1979
1980         p->pdeath_signal = 0;
1981         INIT_LIST_HEAD(&p->thread_group);
1982         p->task_works = NULL;
1983
1984         cgroup_threadgroup_change_begin(current);
1985         /*
1986          * Ensure that the cgroup subsystem policies allow the new process to be
1987          * forked. It should be noted the the new process's css_set can be changed
1988          * between here and cgroup_post_fork() if an organisation operation is in
1989          * progress.
1990          */
1991         retval = cgroup_can_fork(p);
1992         if (retval)
1993                 goto bad_fork_free_pid;
1994
1995         /*
1996          * Make it visible to the rest of the system, but dont wake it up yet.
1997          * Need tasklist lock for parent etc handling!
1998          */
1999         write_lock_irq(&tasklist_lock);
2000
2001         /* CLONE_PARENT re-uses the old parent */
2002         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2003                 p->real_parent = current->real_parent;
2004                 p->parent_exec_id = current->parent_exec_id;
2005         } else {
2006                 p->real_parent = current;
2007                 p->parent_exec_id = current->self_exec_id;
2008         }
2009
2010         klp_copy_process(p);
2011
2012         spin_lock(&current->sighand->siglock);
2013
2014         /*
2015          * Copy seccomp details explicitly here, in case they were changed
2016          * before holding sighand lock.
2017          */
2018         copy_seccomp(p);
2019
2020         rseq_fork(p, clone_flags);
2021
2022         /* Don't start children in a dying pid namespace */
2023         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2024                 retval = -ENOMEM;
2025                 goto bad_fork_cancel_cgroup;
2026         }
2027
2028         /* Let kill terminate clone/fork in the middle */
2029         if (fatal_signal_pending(current)) {
2030                 retval = -EINTR;
2031                 goto bad_fork_cancel_cgroup;
2032         }
2033
2034
2035         init_task_pid_links(p);
2036         if (likely(p->pid)) {
2037                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2038
2039                 init_task_pid(p, PIDTYPE_PID, pid);
2040                 if (thread_group_leader(p)) {
2041                         init_task_pid(p, PIDTYPE_TGID, pid);
2042                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2043                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2044
2045                         if (is_child_reaper(pid)) {
2046                                 ns_of_pid(pid)->child_reaper = p;
2047                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2048                         }
2049                         p->signal->shared_pending.signal = delayed.signal;
2050                         p->signal->tty = tty_kref_get(current->signal->tty);
2051                         /*
2052                          * Inherit has_child_subreaper flag under the same
2053                          * tasklist_lock with adding child to the process tree
2054                          * for propagate_has_child_subreaper optimization.
2055                          */
2056                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2057                                                          p->real_parent->signal->is_child_subreaper;
2058                         list_add_tail(&p->sibling, &p->real_parent->children);
2059                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2060                         attach_pid(p, PIDTYPE_TGID);
2061                         attach_pid(p, PIDTYPE_PGID);
2062                         attach_pid(p, PIDTYPE_SID);
2063                         __this_cpu_inc(process_counts);
2064                 } else {
2065                         current->signal->nr_threads++;
2066                         atomic_inc(&current->signal->live);
2067                         atomic_inc(&current->signal->sigcnt);
2068                         task_join_group_stop(p);
2069                         list_add_tail_rcu(&p->thread_group,
2070                                           &p->group_leader->thread_group);
2071                         list_add_tail_rcu(&p->thread_node,
2072                                           &p->signal->thread_head);
2073                 }
2074                 attach_pid(p, PIDTYPE_PID);
2075                 nr_threads++;
2076         }
2077         total_forks++;
2078         hlist_del_init(&delayed.node);
2079         spin_unlock(&current->sighand->siglock);
2080         syscall_tracepoint_update(p);
2081         write_unlock_irq(&tasklist_lock);
2082
2083         proc_fork_connector(p);
2084         cgroup_post_fork(p);
2085         cgroup_threadgroup_change_end(current);
2086         perf_event_fork(p);
2087
2088         trace_task_newtask(p, clone_flags);
2089         uprobe_copy_process(p, clone_flags);
2090
2091         return p;
2092
2093 bad_fork_cancel_cgroup:
2094         spin_unlock(&current->sighand->siglock);
2095         write_unlock_irq(&tasklist_lock);
2096         cgroup_cancel_fork(p);
2097 bad_fork_free_pid:
2098         cgroup_threadgroup_change_end(current);
2099         if (pid != &init_struct_pid)
2100                 free_pid(pid);
2101 bad_fork_cleanup_thread:
2102         exit_thread(p);
2103 bad_fork_cleanup_io:
2104         if (p->io_context)
2105                 exit_io_context(p);
2106 bad_fork_cleanup_namespaces:
2107         exit_task_namespaces(p);
2108 bad_fork_cleanup_mm:
2109         if (p->mm)
2110                 mmput(p->mm);
2111 bad_fork_cleanup_signal:
2112         if (!(clone_flags & CLONE_THREAD))
2113                 free_signal_struct(p->signal);
2114 bad_fork_cleanup_sighand:
2115         __cleanup_sighand(p->sighand);
2116 bad_fork_cleanup_fs:
2117         exit_fs(p); /* blocking */
2118 bad_fork_cleanup_files:
2119         exit_files(p); /* blocking */
2120 bad_fork_cleanup_semundo:
2121         exit_sem(p);
2122 bad_fork_cleanup_security:
2123         security_task_free(p);
2124 bad_fork_cleanup_audit:
2125         audit_free(p);
2126 bad_fork_cleanup_perf:
2127         perf_event_free_task(p);
2128 bad_fork_cleanup_policy:
2129         lockdep_free_task(p);
2130 #ifdef CONFIG_NUMA
2131         mpol_put(p->mempolicy);
2132 bad_fork_cleanup_threadgroup_lock:
2133 #endif
2134         delayacct_tsk_free(p);
2135 bad_fork_cleanup_count:
2136         atomic_dec(&p->cred->user->processes);
2137         exit_creds(p);
2138 bad_fork_free:
2139         p->state = TASK_DEAD;
2140         put_task_stack(p);
2141         free_task(p);
2142 fork_out:
2143         spin_lock_irq(&current->sighand->siglock);
2144         hlist_del_init(&delayed.node);
2145         spin_unlock_irq(&current->sighand->siglock);
2146         return ERR_PTR(retval);
2147 }
2148
2149 static inline void init_idle_pids(struct task_struct *idle)
2150 {
2151         enum pid_type type;
2152
2153         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2154                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2155                 init_task_pid(idle, type, &init_struct_pid);
2156         }
2157 }
2158
2159 struct task_struct *fork_idle(int cpu)
2160 {
2161         struct task_struct *task;
2162         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2163                             cpu_to_node(cpu));
2164         if (!IS_ERR(task)) {
2165                 init_idle_pids(task);
2166                 init_idle(task, cpu);
2167         }
2168
2169         return task;
2170 }
2171
2172 /*
2173  *  Ok, this is the main fork-routine.
2174  *
2175  * It copies the process, and if successful kick-starts
2176  * it and waits for it to finish using the VM if required.
2177  */
2178 long _do_fork(unsigned long clone_flags,
2179               unsigned long stack_start,
2180               unsigned long stack_size,
2181               int __user *parent_tidptr,
2182               int __user *child_tidptr,
2183               unsigned long tls)
2184 {
2185         struct completion vfork;
2186         struct pid *pid;
2187         struct task_struct *p;
2188         int trace = 0;
2189         long nr;
2190
2191         /*
2192          * Determine whether and which event to report to ptracer.  When
2193          * called from kernel_thread or CLONE_UNTRACED is explicitly
2194          * requested, no event is reported; otherwise, report if the event
2195          * for the type of forking is enabled.
2196          */
2197         if (!(clone_flags & CLONE_UNTRACED)) {
2198                 if (clone_flags & CLONE_VFORK)
2199                         trace = PTRACE_EVENT_VFORK;
2200                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2201                         trace = PTRACE_EVENT_CLONE;
2202                 else
2203                         trace = PTRACE_EVENT_FORK;
2204
2205                 if (likely(!ptrace_event_enabled(current, trace)))
2206                         trace = 0;
2207         }
2208
2209         p = copy_process(clone_flags, stack_start, stack_size,
2210                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2211         add_latent_entropy();
2212
2213         if (IS_ERR(p))
2214                 return PTR_ERR(p);
2215
2216         /*
2217          * Do this prior waking up the new thread - the thread pointer
2218          * might get invalid after that point, if the thread exits quickly.
2219          */
2220         trace_sched_process_fork(current, p);
2221
2222         pid = get_task_pid(p, PIDTYPE_PID);
2223         nr = pid_vnr(pid);
2224
2225         if (clone_flags & CLONE_PARENT_SETTID)
2226                 put_user(nr, parent_tidptr);
2227
2228         if (clone_flags & CLONE_VFORK) {
2229                 p->vfork_done = &vfork;
2230                 init_completion(&vfork);
2231                 get_task_struct(p);
2232         }
2233
2234         wake_up_new_task(p);
2235
2236         /* forking complete and child started to run, tell ptracer */
2237         if (unlikely(trace))
2238                 ptrace_event_pid(trace, pid);
2239
2240         if (clone_flags & CLONE_VFORK) {
2241                 if (!wait_for_vfork_done(p, &vfork))
2242                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2243         }
2244
2245         put_pid(pid);
2246         return nr;
2247 }
2248
2249 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2250 /* For compatibility with architectures that call do_fork directly rather than
2251  * using the syscall entry points below. */
2252 long do_fork(unsigned long clone_flags,
2253               unsigned long stack_start,
2254               unsigned long stack_size,
2255               int __user *parent_tidptr,
2256               int __user *child_tidptr)
2257 {
2258         return _do_fork(clone_flags, stack_start, stack_size,
2259                         parent_tidptr, child_tidptr, 0);
2260 }
2261 #endif
2262
2263 /*
2264  * Create a kernel thread.
2265  */
2266 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2267 {
2268         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2269                 (unsigned long)arg, NULL, NULL, 0);
2270 }
2271
2272 #ifdef __ARCH_WANT_SYS_FORK
2273 SYSCALL_DEFINE0(fork)
2274 {
2275 #ifdef CONFIG_MMU
2276         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2277 #else
2278         /* can not support in nommu mode */
2279         return -EINVAL;
2280 #endif
2281 }
2282 #endif
2283
2284 #ifdef __ARCH_WANT_SYS_VFORK
2285 SYSCALL_DEFINE0(vfork)
2286 {
2287         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2288                         0, NULL, NULL, 0);
2289 }
2290 #endif
2291
2292 #ifdef __ARCH_WANT_SYS_CLONE
2293 #ifdef CONFIG_CLONE_BACKWARDS
2294 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2295                  int __user *, parent_tidptr,
2296                  unsigned long, tls,
2297                  int __user *, child_tidptr)
2298 #elif defined(CONFIG_CLONE_BACKWARDS2)
2299 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2300                  int __user *, parent_tidptr,
2301                  int __user *, child_tidptr,
2302                  unsigned long, tls)
2303 #elif defined(CONFIG_CLONE_BACKWARDS3)
2304 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2305                 int, stack_size,
2306                 int __user *, parent_tidptr,
2307                 int __user *, child_tidptr,
2308                 unsigned long, tls)
2309 #else
2310 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2311                  int __user *, parent_tidptr,
2312                  int __user *, child_tidptr,
2313                  unsigned long, tls)
2314 #endif
2315 {
2316         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2317 }
2318 #endif
2319
2320 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2321 {
2322         struct task_struct *leader, *parent, *child;
2323         int res;
2324
2325         read_lock(&tasklist_lock);
2326         leader = top = top->group_leader;
2327 down:
2328         for_each_thread(leader, parent) {
2329                 list_for_each_entry(child, &parent->children, sibling) {
2330                         res = visitor(child, data);
2331                         if (res) {
2332                                 if (res < 0)
2333                                         goto out;
2334                                 leader = child;
2335                                 goto down;
2336                         }
2337 up:
2338                         ;
2339                 }
2340         }
2341
2342         if (leader != top) {
2343                 child = leader;
2344                 parent = child->real_parent;
2345                 leader = parent->group_leader;
2346                 goto up;
2347         }
2348 out:
2349         read_unlock(&tasklist_lock);
2350 }
2351
2352 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2353 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2354 #endif
2355
2356 static void sighand_ctor(void *data)
2357 {
2358         struct sighand_struct *sighand = data;
2359
2360         spin_lock_init(&sighand->siglock);
2361         init_waitqueue_head(&sighand->signalfd_wqh);
2362 }
2363
2364 void __init proc_caches_init(void)
2365 {
2366         unsigned int mm_size;
2367
2368         sighand_cachep = kmem_cache_create("sighand_cache",
2369                         sizeof(struct sighand_struct), 0,
2370                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2371                         SLAB_ACCOUNT, sighand_ctor);
2372         signal_cachep = kmem_cache_create("signal_cache",
2373                         sizeof(struct signal_struct), 0,
2374                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2375                         NULL);
2376         files_cachep = kmem_cache_create("files_cache",
2377                         sizeof(struct files_struct), 0,
2378                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2379                         NULL);
2380         fs_cachep = kmem_cache_create("fs_cache",
2381                         sizeof(struct fs_struct), 0,
2382                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2383                         NULL);
2384
2385         /*
2386          * The mm_cpumask is located at the end of mm_struct, and is
2387          * dynamically sized based on the maximum CPU number this system
2388          * can have, taking hotplug into account (nr_cpu_ids).
2389          */
2390         mm_size = sizeof(struct mm_struct) + cpumask_size();
2391
2392         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2393                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2394                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2395                         offsetof(struct mm_struct, saved_auxv),
2396                         sizeof_field(struct mm_struct, saved_auxv),
2397                         NULL);
2398         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2399         mmap_init();
2400         nsproxy_cache_init();
2401 }
2402
2403 /*
2404  * Check constraints on flags passed to the unshare system call.
2405  */
2406 static int check_unshare_flags(unsigned long unshare_flags)
2407 {
2408         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2409                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2410                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2411                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2412                 return -EINVAL;
2413         /*
2414          * Not implemented, but pretend it works if there is nothing
2415          * to unshare.  Note that unsharing the address space or the
2416          * signal handlers also need to unshare the signal queues (aka
2417          * CLONE_THREAD).
2418          */
2419         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2420                 if (!thread_group_empty(current))
2421                         return -EINVAL;
2422         }
2423         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2424                 if (atomic_read(&current->sighand->count) > 1)
2425                         return -EINVAL;
2426         }
2427         if (unshare_flags & CLONE_VM) {
2428                 if (!current_is_single_threaded())
2429                         return -EINVAL;
2430         }
2431
2432         return 0;
2433 }
2434
2435 /*
2436  * Unshare the filesystem structure if it is being shared
2437  */
2438 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2439 {
2440         struct fs_struct *fs = current->fs;
2441
2442         if (!(unshare_flags & CLONE_FS) || !fs)
2443                 return 0;
2444
2445         /* don't need lock here; in the worst case we'll do useless copy */
2446         if (fs->users == 1)
2447                 return 0;
2448
2449         *new_fsp = copy_fs_struct(fs);
2450         if (!*new_fsp)
2451                 return -ENOMEM;
2452
2453         return 0;
2454 }
2455
2456 /*
2457  * Unshare file descriptor table if it is being shared
2458  */
2459 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2460 {
2461         struct files_struct *fd = current->files;
2462         int error = 0;
2463
2464         if ((unshare_flags & CLONE_FILES) &&
2465             (fd && atomic_read(&fd->count) > 1)) {
2466                 *new_fdp = dup_fd(fd, &error);
2467                 if (!*new_fdp)
2468                         return error;
2469         }
2470
2471         return 0;
2472 }
2473
2474 /*
2475  * unshare allows a process to 'unshare' part of the process
2476  * context which was originally shared using clone.  copy_*
2477  * functions used by do_fork() cannot be used here directly
2478  * because they modify an inactive task_struct that is being
2479  * constructed. Here we are modifying the current, active,
2480  * task_struct.
2481  */
2482 int ksys_unshare(unsigned long unshare_flags)
2483 {
2484         struct fs_struct *fs, *new_fs = NULL;
2485         struct files_struct *fd, *new_fd = NULL;
2486         struct cred *new_cred = NULL;
2487         struct nsproxy *new_nsproxy = NULL;
2488         int do_sysvsem = 0;
2489         int err;
2490
2491         /*
2492          * If unsharing a user namespace must also unshare the thread group
2493          * and unshare the filesystem root and working directories.
2494          */
2495         if (unshare_flags & CLONE_NEWUSER)
2496                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2497         /*
2498          * If unsharing vm, must also unshare signal handlers.
2499          */
2500         if (unshare_flags & CLONE_VM)
2501                 unshare_flags |= CLONE_SIGHAND;
2502         /*
2503          * If unsharing a signal handlers, must also unshare the signal queues.
2504          */
2505         if (unshare_flags & CLONE_SIGHAND)
2506                 unshare_flags |= CLONE_THREAD;
2507         /*
2508          * If unsharing namespace, must also unshare filesystem information.
2509          */
2510         if (unshare_flags & CLONE_NEWNS)
2511                 unshare_flags |= CLONE_FS;
2512
2513         err = check_unshare_flags(unshare_flags);
2514         if (err)
2515                 goto bad_unshare_out;
2516         /*
2517          * CLONE_NEWIPC must also detach from the undolist: after switching
2518          * to a new ipc namespace, the semaphore arrays from the old
2519          * namespace are unreachable.
2520          */
2521         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2522                 do_sysvsem = 1;
2523         err = unshare_fs(unshare_flags, &new_fs);
2524         if (err)
2525                 goto bad_unshare_out;
2526         err = unshare_fd(unshare_flags, &new_fd);
2527         if (err)
2528                 goto bad_unshare_cleanup_fs;
2529         err = unshare_userns(unshare_flags, &new_cred);
2530         if (err)
2531                 goto bad_unshare_cleanup_fd;
2532         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2533                                          new_cred, new_fs);
2534         if (err)
2535                 goto bad_unshare_cleanup_cred;
2536
2537         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2538                 if (do_sysvsem) {
2539                         /*
2540                          * CLONE_SYSVSEM is equivalent to sys_exit().
2541                          */
2542                         exit_sem(current);
2543                 }
2544                 if (unshare_flags & CLONE_NEWIPC) {
2545                         /* Orphan segments in old ns (see sem above). */
2546                         exit_shm(current);
2547                         shm_init_task(current);
2548                 }
2549
2550                 if (new_nsproxy)
2551                         switch_task_namespaces(current, new_nsproxy);
2552
2553                 task_lock(current);
2554
2555                 if (new_fs) {
2556                         fs = current->fs;
2557                         spin_lock(&fs->lock);
2558                         current->fs = new_fs;
2559                         if (--fs->users)
2560                                 new_fs = NULL;
2561                         else
2562                                 new_fs = fs;
2563                         spin_unlock(&fs->lock);
2564                 }
2565
2566                 if (new_fd) {
2567                         fd = current->files;
2568                         current->files = new_fd;
2569                         new_fd = fd;
2570                 }
2571
2572                 task_unlock(current);
2573
2574                 if (new_cred) {
2575                         /* Install the new user namespace */
2576                         commit_creds(new_cred);
2577                         new_cred = NULL;
2578                 }
2579         }
2580
2581         perf_event_namespaces(current);
2582
2583 bad_unshare_cleanup_cred:
2584         if (new_cred)
2585                 put_cred(new_cred);
2586 bad_unshare_cleanup_fd:
2587         if (new_fd)
2588                 put_files_struct(new_fd);
2589
2590 bad_unshare_cleanup_fs:
2591         if (new_fs)
2592                 free_fs_struct(new_fs);
2593
2594 bad_unshare_out:
2595         return err;
2596 }
2597
2598 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2599 {
2600         return ksys_unshare(unshare_flags);
2601 }
2602
2603 /*
2604  *      Helper to unshare the files of the current task.
2605  *      We don't want to expose copy_files internals to
2606  *      the exec layer of the kernel.
2607  */
2608
2609 int unshare_files(struct files_struct **displaced)
2610 {
2611         struct task_struct *task = current;
2612         struct files_struct *copy = NULL;
2613         int error;
2614
2615         error = unshare_fd(CLONE_FILES, &copy);
2616         if (error || !copy) {
2617                 *displaced = NULL;
2618                 return error;
2619         }
2620         *displaced = task->files;
2621         task_lock(task);
2622         task->files = copy;
2623         task_unlock(task);
2624         return 0;
2625 }
2626
2627 int sysctl_max_threads(struct ctl_table *table, int write,
2628                        void __user *buffer, size_t *lenp, loff_t *ppos)
2629 {
2630         struct ctl_table t;
2631         int ret;
2632         int threads = max_threads;
2633         int min = MIN_THREADS;
2634         int max = MAX_THREADS;
2635
2636         t = *table;
2637         t.data = &threads;
2638         t.extra1 = &min;
2639         t.extra2 = &max;
2640
2641         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2642         if (ret || !write)
2643                 return ret;
2644
2645         set_max_threads(threads);
2646
2647         return 0;
2648 }