kernel/fork.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/kernel/fork.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  *  'fork.c' contains the help-routines for the 'fork' system call
  10  * (see also entry.S and others).
  11  * Fork is rather simple, once you get the hang of it, but the memory
  12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13  */
  14
  15 #include <linux/anon_inodes.h>
  16 #include <linux/slab.h>
  17 #include <linux/sched/autogroup.h>
  18 #include <linux/sched/mm.h>
  19 #include <linux/sched/coredump.h>
  20 #include <linux/sched/user.h>
  21 #include <linux/sched/numa_balancing.h>
  22 #include <linux/sched/stat.h>
  23 #include <linux/sched/task.h>
  24 #include <linux/sched/task_stack.h>
  25 #include <linux/sched/cputime.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/rtmutex.h>
  28 #include <linux/init.h>
  29 #include <linux/unistd.h>
  30 #include <linux/module.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/completion.h>
  33 #include <linux/personality.h>
  34 #include <linux/mempolicy.h>
  35 #include <linux/sem.h>
  36 #include <linux/file.h>
  37 #include <linux/fdtable.h>
  38 #include <linux/iocontext.h>
  39 #include <linux/key.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/mman.h>
  42 #include <linux/mmu_notifier.h>
  43 #include <linux/hmm.h>
  44 #include <linux/fs.h>
  45 #include <linux/mm.h>
  46 #include <linux/vmacache.h>
  47 #include <linux/nsproxy.h>
  48 #include <linux/capability.h>
  49 #include <linux/cpu.h>
  50 #include <linux/cgroup.h>
  51 #include <linux/security.h>
  52 #include <linux/hugetlb.h>
  53 #include <linux/seccomp.h>
  54 #include <linux/swap.h>
  55 #include <linux/syscalls.h>
  56 #include <linux/jiffies.h>
  57 #include <linux/futex.h>
  58 #include <linux/compat.h>
  59 #include <linux/kthread.h>
  60 #include <linux/task_io_accounting_ops.h>
  61 #include <linux/rcupdate.h>
  62 #include <linux/ptrace.h>
  63 #include <linux/mount.h>
  64 #include <linux/audit.h>
  65 #include <linux/memcontrol.h>
  66 #include <linux/ftrace.h>
  67 #include <linux/proc_fs.h>
  68 #include <linux/profile.h>
  69 #include <linux/rmap.h>
  70 #include <linux/ksm.h>
  71 #include <linux/acct.h>
  72 #include <linux/userfaultfd_k.h>
  73 #include <linux/tsacct_kern.h>
  74 #include <linux/cn_proc.h>
  75 #include <linux/freezer.h>
  76 #include <linux/delayacct.h>
  77 #include <linux/taskstats_kern.h>
  78 #include <linux/random.h>
  79 #include <linux/tty.h>
  80 #include <linux/blkdev.h>
  81 #include <linux/fs_struct.h>
  82 #include <linux/magic.h>
  83 #include <linux/perf_event.h>
  84 #include <linux/posix-timers.h>
  85 #include <linux/user-return-notifier.h>
  86 #include <linux/oom.h>
  87 #include <linux/khugepaged.h>
  88 #include <linux/signalfd.h>
  89 #include <linux/uprobes.h>
  90 #include <linux/aio.h>
  91 #include <linux/compiler.h>
  92 #include <linux/sysctl.h>
  93 #include <linux/kcov.h>
  94 #include <linux/livepatch.h>
  95 #include <linux/thread_info.h>
  96 #include <linux/stackleak.h>
  97
  98 #include <asm/pgtable.h>
  99 #include <asm/pgalloc.h>
 100 #include <linux/uaccess.h>
 101 #include <asm/mmu_context.h>
 102 #include <asm/cacheflush.h>
 103 #include <asm/tlbflush.h>
 104
 105 #include <trace/events/sched.h>
 106
 107 #define CREATE_TRACE_POINTS
 108 #include <trace/events/task.h>
 109
 110 /*
 111  * Minimum number of threads to boot the kernel
 112  */
 113 #define MIN_THREADS 20
 114
 115 /*
 116  * Maximum number of threads
 117  */
 118 #define MAX_THREADS FUTEX_TID_MASK
 119
 120 /*
 121  * Protected counters by write_lock_irq(&tasklist_lock)
 122  */
 123 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 124 int nr_threads;                 /* The idle threads do not count.. */
 125
 126 int max_threads;                /* tunable limit on nr_threads */
 127
 128 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 129
 130 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 131
 132 #ifdef CONFIG_PROVE_RCU
 133 int lockdep_tasklist_lock_is_held(void)
 134 {
 135         return lockdep_is_held(&tasklist_lock);
 136 }
 137 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 138 #endif /* #ifdef CONFIG_PROVE_RCU */
 139
 140 int nr_processes(void)
 141 {
 142         int cpu;
 143         int total = 0;
 144
 145         for_each_possible_cpu(cpu)
 146                 total += per_cpu(process_counts, cpu);
 147
 148         return total;
 149 }
 150
 151 void __weak arch_release_task_struct(struct task_struct *tsk)
 152 {
 153 }
 154
 155 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 156 static struct kmem_cache *task_struct_cachep;
 157
 158 static inline struct task_struct *alloc_task_struct_node(int node)
 159 {
 160         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 161 }
 162
 163 static inline void free_task_struct(struct task_struct *tsk)
 164 {
 165         kmem_cache_free(task_struct_cachep, tsk);
 166 }
 167 #endif
 168
 169 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 170
 171 /*
 172  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 173  * kmemcache based allocator.
 174  */
 175 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 176
 177 #ifdef CONFIG_VMAP_STACK
 178 /*
 179  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 180  * flush.  Try to minimize the number of calls by caching stacks.
 181  */
 182 #define NR_CACHED_STACKS 2
 183 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 184
 185 static int free_vm_stack_cache(unsigned int cpu)
 186 {
 187         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 188         int i;
 189
 190         for (i = 0; i < NR_CACHED_STACKS; i++) {
 191                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 192
 193                 if (!vm_stack)
 194                         continue;
 195
 196                 vfree(vm_stack->addr);
 197                 cached_vm_stacks[i] = NULL;
 198         }
 199
 200         return 0;
 201 }
 202 #endif
 203
 204 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 205 {
 206 #ifdef CONFIG_VMAP_STACK
 207         void *stack;
 208         int i;
 209
 210         for (i = 0; i < NR_CACHED_STACKS; i++) {
 211                 struct vm_struct *s;
 212
 213                 s = this_cpu_xchg(cached_stacks[i], NULL);
 214
 215                 if (!s)
 216                         continue;
 217
 218                 /* Clear stale pointers from reused stack. */
 219                 memset(s->addr, 0, THREAD_SIZE);
 220
 221                 tsk->stack_vm_area = s;
 222                 tsk->stack = s->addr;
 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                 tsk->stack = stack;
 245         }
 246         return stack;
 247 #else
 248         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 249                                              THREAD_SIZE_ORDER);
 250
 251         return page ? page_address(page) : NULL;
 252 #endif
 253 }
 254
 255 static inline void free_thread_stack(struct task_struct *tsk)
 256 {
 257 #ifdef CONFIG_VMAP_STACK
 258         struct vm_struct *vm = task_stack_vm_area(tsk);
 259
 260         if (vm) {
 261                 int i;
 262
 263                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 264                         mod_memcg_page_state(vm->pages[i],
 265                                              MEMCG_KERNEL_STACK_KB,
 266                                              -(int)(PAGE_SIZE / 1024));
 267
 268                         memcg_kmem_uncharge(vm->pages[i], 0);
 269                 }
 270
 271                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 272                         if (this_cpu_cmpxchg(cached_stacks[i],
 273                                         NULL, tsk->stack_vm_area) != NULL)
 274                                 continue;
 275
 276                         return;
 277                 }
 278
 279                 vfree_atomic(tsk->stack);
 280                 return;
 281         }
 282 #endif
 283
 284         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 285 }
 286 # else
 287 static struct kmem_cache *thread_stack_cache;
 288
 289 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 290                                                   int node)
 291 {
 292         unsigned long *stack;
 293         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 294         tsk->stack = stack;
 295         return stack;
 296 }
 297
 298 static void free_thread_stack(struct task_struct *tsk)
 299 {
 300         kmem_cache_free(thread_stack_cache, tsk->stack);
 301 }
 302
 303 void thread_stack_cache_init(void)
 304 {
 305         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 306                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 307                                         THREAD_SIZE, NULL);
 308         BUG_ON(thread_stack_cache == NULL);
 309 }
 310 # endif
 311 #endif
 312
 313 /* SLAB cache for signal_struct structures (tsk->signal) */
 314 static struct kmem_cache *signal_cachep;
 315
 316 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 317 struct kmem_cache *sighand_cachep;
 318
 319 /* SLAB cache for files_struct structures (tsk->files) */
 320 struct kmem_cache *files_cachep;
 321
 322 /* SLAB cache for fs_struct structures (tsk->fs) */
 323 struct kmem_cache *fs_cachep;
 324
 325 /* SLAB cache for vm_area_struct structures */
 326 static struct kmem_cache *vm_area_cachep;
 327
 328 /* SLAB cache for mm_struct structures (tsk->mm) */
 329 static struct kmem_cache *mm_cachep;
 330
 331 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 332 {
 333         struct vm_area_struct *vma;
 334
 335         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 336         if (vma)
 337                 vma_init(vma, mm);
 338         return vma;
 339 }
 340
 341 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 342 {
 343         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 344
 345         if (new) {
 346                 *new = *orig;
 347                 INIT_LIST_HEAD(&new->anon_vma_chain);
 348         }
 349         return new;
 350 }
 351
 352 void vm_area_free(struct vm_area_struct *vma)
 353 {
 354         kmem_cache_free(vm_area_cachep, vma);
 355 }
 356
 357 static void account_kernel_stack(struct task_struct *tsk, int account)
 358 {
 359         void *stack = task_stack_page(tsk);
 360         struct vm_struct *vm = task_stack_vm_area(tsk);
 361
 362         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 363
 364         if (vm) {
 365                 int i;
 366
 367                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 368
 369                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 370                         mod_zone_page_state(page_zone(vm->pages[i]),
 371                                             NR_KERNEL_STACK_KB,
 372                                             PAGE_SIZE / 1024 * account);
 373                 }
 374         } else {
 375                 /*
 376                  * All stack pages are in the same zone and belong to the
 377                  * same memcg.
 378                  */
 379                 struct page *first_page = virt_to_page(stack);
 380
 381                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 382                                     THREAD_SIZE / 1024 * account);
 383
 384                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 385                                      account * (THREAD_SIZE / 1024));
 386         }
 387 }
 388
 389 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 390 {
 391 #ifdef CONFIG_VMAP_STACK
 392         struct vm_struct *vm = task_stack_vm_area(tsk);
 393         int ret;
 394
 395         if (vm) {
 396                 int i;
 397
 398                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 399                         /*
 400                          * If memcg_kmem_charge() fails, page->mem_cgroup
 401                          * pointer is NULL, and both memcg_kmem_uncharge()
 402                          * and mod_memcg_page_state() in free_thread_stack()
 403                          * will ignore this page. So it's safe.
 404                          */
 405                         ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0);
 406                         if (ret)
 407                                 return ret;
 408
 409                         mod_memcg_page_state(vm->pages[i],
 410                                              MEMCG_KERNEL_STACK_KB,
 411                                              PAGE_SIZE / 1024);
 412                 }
 413         }
 414 #endif
 415         return 0;
 416 }
 417
 418 static void release_task_stack(struct task_struct *tsk)
 419 {
 420         if (WARN_ON(tsk->state != TASK_DEAD))
 421                 return;  /* Better to leak the stack than to free prematurely */
 422
 423         account_kernel_stack(tsk, -1);
 424         free_thread_stack(tsk);
 425         tsk->stack = NULL;
 426 #ifdef CONFIG_VMAP_STACK
 427         tsk->stack_vm_area = NULL;
 428 #endif
 429 }
 430
 431 #ifdef CONFIG_THREAD_INFO_IN_TASK
 432 void put_task_stack(struct task_struct *tsk)
 433 {
 434         if (refcount_dec_and_test(&tsk->stack_refcount))
 435                 release_task_stack(tsk);
 436 }
 437 #endif
 438
 439 void free_task(struct task_struct *tsk)
 440 {
 441 #ifndef CONFIG_THREAD_INFO_IN_TASK
 442         /*
 443          * The task is finally done with both the stack and thread_info,
 444          * so free both.
 445          */
 446         release_task_stack(tsk);
 447 #else
 448         /*
 449          * If the task had a separate stack allocation, it should be gone
 450          * by now.
 451          */
 452         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 453 #endif
 454         rt_mutex_debug_task_free(tsk);
 455         ftrace_graph_exit_task(tsk);
 456         put_seccomp_filter(tsk);
 457         arch_release_task_struct(tsk);
 458         if (tsk->flags & PF_KTHREAD)
 459                 free_kthread_struct(tsk);
 460         free_task_struct(tsk);
 461 }
 462 EXPORT_SYMBOL(free_task);
 463
 464 #ifdef CONFIG_MMU
 465 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 466                                         struct mm_struct *oldmm)
 467 {
 468         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 469         struct rb_node **rb_link, *rb_parent;
 470         int retval;
 471         unsigned long charge;
 472         LIST_HEAD(uf);
 473
 474         uprobe_start_dup_mmap();
 475         if (down_write_killable(&oldmm->mmap_sem)) {
 476                 retval = -EINTR;
 477                 goto fail_uprobe_end;
 478         }
 479         flush_cache_dup_mm(oldmm);
 480         uprobe_dup_mmap(oldmm, mm);
 481         /*
 482          * Not linked in yet - no deadlock potential:
 483          */
 484         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 485
 486         /* No ordering required: file already has been exposed. */
 487         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 488
 489         mm->total_vm = oldmm->total_vm;
 490         mm->data_vm = oldmm->data_vm;
 491         mm->exec_vm = oldmm->exec_vm;
 492         mm->stack_vm = oldmm->stack_vm;
 493
 494         rb_link = &mm->mm_rb.rb_node;
 495         rb_parent = NULL;
 496         pprev = &mm->mmap;
 497         retval = ksm_fork(mm, oldmm);
 498         if (retval)
 499                 goto out;
 500         retval = khugepaged_fork(mm, oldmm);
 501         if (retval)
 502                 goto out;
 503
 504         prev = NULL;
 505         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 506                 struct file *file;
 507
 508                 if (mpnt->vm_flags & VM_DONTCOPY) {
 509                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 510                         continue;
 511                 }
 512                 charge = 0;
 513                 /*
 514                  * Don't duplicate many vmas if we've been oom-killed (for
 515                  * example)
 516                  */
 517                 if (fatal_signal_pending(current)) {
 518                         retval = -EINTR;
 519                         goto out;
 520                 }
 521                 if (mpnt->vm_flags & VM_ACCOUNT) {
 522                         unsigned long len = vma_pages(mpnt);
 523
 524                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 525                                 goto fail_nomem;
 526                         charge = len;
 527                 }
 528                 tmp = vm_area_dup(mpnt);
 529                 if (!tmp)
 530                         goto fail_nomem;
 531                 retval = vma_dup_policy(mpnt, tmp);
 532                 if (retval)
 533                         goto fail_nomem_policy;
 534                 tmp->vm_mm = mm;
 535                 retval = dup_userfaultfd(tmp, &uf);
 536                 if (retval)
 537                         goto fail_nomem_anon_vma_fork;
 538                 if (tmp->vm_flags & VM_WIPEONFORK) {
 539                         /* VM_WIPEONFORK gets a clean slate in the child. */
 540                         tmp->anon_vma = NULL;
 541                         if (anon_vma_prepare(tmp))
 542                                 goto fail_nomem_anon_vma_fork;
 543                 } else if (anon_vma_fork(tmp, mpnt))
 544                         goto fail_nomem_anon_vma_fork;
 545                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 546                 tmp->vm_next = tmp->vm_prev = NULL;
 547                 file = tmp->vm_file;
 548                 if (file) {
 549                         struct inode *inode = file_inode(file);
 550                         struct address_space *mapping = file->f_mapping;
 551
 552                         get_file(file);
 553                         if (tmp->vm_flags & VM_DENYWRITE)
 554                                 atomic_dec(&inode->i_writecount);
 555                         i_mmap_lock_write(mapping);
 556                         if (tmp->vm_flags & VM_SHARED)
 557                                 atomic_inc(&mapping->i_mmap_writable);
 558                         flush_dcache_mmap_lock(mapping);
 559                         /* insert tmp into the share list, just after mpnt */
 560                         vma_interval_tree_insert_after(tmp, mpnt,
 561                                         &mapping->i_mmap);
 562                         flush_dcache_mmap_unlock(mapping);
 563                         i_mmap_unlock_write(mapping);
 564                 }
 565
 566                 /*
 567                  * Clear hugetlb-related page reserves for children. This only
 568                  * affects MAP_PRIVATE mappings. Faults generated by the child
 569                  * are not guaranteed to succeed, even if read-only
 570                  */
 571                 if (is_vm_hugetlb_page(tmp))
 572                         reset_vma_resv_huge_pages(tmp);
 573
 574                 /*
 575                  * Link in the new vma and copy the page table entries.
 576                  */
 577                 *pprev = tmp;
 578                 pprev = &tmp->vm_next;
 579                 tmp->vm_prev = prev;
 580                 prev = tmp;
 581
 582                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 583                 rb_link = &tmp->vm_rb.rb_right;
 584                 rb_parent = &tmp->vm_rb;
 585
 586                 mm->map_count++;
 587                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 588                         retval = copy_page_range(mm, oldmm, mpnt);
 589
 590                 if (tmp->vm_ops && tmp->vm_ops->open)
 591                         tmp->vm_ops->open(tmp);
 592
 593                 if (retval)
 594                         goto out;
 595         }
 596         /* a new mm has just been created */
 597         retval = arch_dup_mmap(oldmm, mm);
 598 out:
 599         up_write(&mm->mmap_sem);
 600         flush_tlb_mm(oldmm);
 601         up_write(&oldmm->mmap_sem);
 602         dup_userfaultfd_complete(&uf);
 603 fail_uprobe_end:
 604         uprobe_end_dup_mmap();
 605         return retval;
 606 fail_nomem_anon_vma_fork:
 607         mpol_put(vma_policy(tmp));
 608 fail_nomem_policy:
 609         vm_area_free(tmp);
 610 fail_nomem:
 611         retval = -ENOMEM;
 612         vm_unacct_memory(charge);
 613         goto out;
 614 }
 615
 616 static inline int mm_alloc_pgd(struct mm_struct *mm)
 617 {
 618         mm->pgd = pgd_alloc(mm);
 619         if (unlikely(!mm->pgd))
 620                 return -ENOMEM;
 621         return 0;
 622 }
 623
 624 static inline void mm_free_pgd(struct mm_struct *mm)
 625 {
 626         pgd_free(mm, mm->pgd);
 627 }
 628 #else
 629 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 630 {
 631         down_write(&oldmm->mmap_sem);
 632         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 633         up_write(&oldmm->mmap_sem);
 634         return 0;
 635 }
 636 #define mm_alloc_pgd(mm)        (0)
 637 #define mm_free_pgd(mm)
 638 #endif /* CONFIG_MMU */
 639
 640 static void check_mm(struct mm_struct *mm)
 641 {
 642         int i;
 643
 644         for (i = 0; i < NR_MM_COUNTERS; i++) {
 645                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 646
 647                 if (unlikely(x))
 648                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 649                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 650         }
 651
 652         if (mm_pgtables_bytes(mm))
 653                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 654                                 mm_pgtables_bytes(mm));
 655
 656 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 657         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 658 #endif
 659 }
 660
 661 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 662 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 663
 664 /*
 665  * Called when the last reference to the mm
 666  * is dropped: either by a lazy thread or by
 667  * mmput. Free the page directory and the mm.
 668  */
 669 void __mmdrop(struct mm_struct *mm)
 670 {
 671         BUG_ON(mm == &init_mm);
 672         WARN_ON_ONCE(mm == current->mm);
 673         WARN_ON_ONCE(mm == current->active_mm);
 674         mm_free_pgd(mm);
 675         destroy_context(mm);
 676         hmm_mm_destroy(mm);
 677         mmu_notifier_mm_destroy(mm);
 678         check_mm(mm);
 679         put_user_ns(mm->user_ns);
 680         free_mm(mm);
 681 }
 682 EXPORT_SYMBOL_GPL(__mmdrop);
 683
 684 static void mmdrop_async_fn(struct work_struct *work)
 685 {
 686         struct mm_struct *mm;
 687
 688         mm = container_of(work, struct mm_struct, async_put_work);
 689         __mmdrop(mm);
 690 }
 691
 692 static void mmdrop_async(struct mm_struct *mm)
 693 {
 694         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 695                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 696                 schedule_work(&mm->async_put_work);
 697         }
 698 }
 699
 700 static inline void free_signal_struct(struct signal_struct *sig)
 701 {
 702         taskstats_tgid_free(sig);
 703         sched_autogroup_exit(sig);
 704         /*
 705          * __mmdrop is not safe to call from softirq context on x86 due to
 706          * pgd_dtor so postpone it to the async context
 707          */
 708         if (sig->oom_mm)
 709                 mmdrop_async(sig->oom_mm);
 710         kmem_cache_free(signal_cachep, sig);
 711 }
 712
 713 static inline void put_signal_struct(struct signal_struct *sig)
 714 {
 715         if (refcount_dec_and_test(&sig->sigcnt))
 716                 free_signal_struct(sig);
 717 }
 718
 719 void __put_task_struct(struct task_struct *tsk)
 720 {
 721         WARN_ON(!tsk->exit_state);
 722         WARN_ON(refcount_read(&tsk->usage));
 723         WARN_ON(tsk == current);
 724
 725         cgroup_free(tsk);
 726         task_numa_free(tsk);
 727         security_task_free(tsk);
 728         exit_creds(tsk);
 729         delayacct_tsk_free(tsk);
 730         put_signal_struct(tsk->signal);
 731
 732         if (!profile_handoff_task(tsk))
 733                 free_task(tsk);
 734 }
 735 EXPORT_SYMBOL_GPL(__put_task_struct);
 736
 737 void __init __weak arch_task_cache_init(void) { }
 738
 739 /*
 740  * set_max_threads
 741  */
 742 static void set_max_threads(unsigned int max_threads_suggested)
 743 {
 744         u64 threads;
 745         unsigned long nr_pages = totalram_pages();
 746
 747         /*
 748          * The number of threads shall be limited such that the thread
 749          * structures may only consume a small part of the available memory.
 750          */
 751         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 752                 threads = MAX_THREADS;
 753         else
 754                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 755                                     (u64) THREAD_SIZE * 8UL);
 756
 757         if (threads > max_threads_suggested)
 758                 threads = max_threads_suggested;
 759
 760         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 761 }
 762
 763 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 764 /* Initialized by the architecture: */
 765 int arch_task_struct_size __read_mostly;
 766 #endif
 767
 768 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 769 {
 770         /* Fetch thread_struct whitelist for the architecture. */
 771         arch_thread_struct_whitelist(offset, size);
 772
 773         /*
 774          * Handle zero-sized whitelist or empty thread_struct, otherwise
 775          * adjust offset to position of thread_struct in task_struct.
 776          */
 777         if (unlikely(*size == 0))
 778                 *offset = 0;
 779         else
 780                 *offset += offsetof(struct task_struct, thread);
 781 }
 782
 783 void __init fork_init(void)
 784 {
 785         int i;
 786 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 787 #ifndef ARCH_MIN_TASKALIGN
 788 #define ARCH_MIN_TASKALIGN      0
 789 #endif
 790         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 791         unsigned long useroffset, usersize;
 792
 793         /* create a slab on which task_structs can be allocated */
 794         task_struct_whitelist(&useroffset, &usersize);
 795         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 796                         arch_task_struct_size, align,
 797                         SLAB_PANIC|SLAB_ACCOUNT,
 798                         useroffset, usersize, NULL);
 799 #endif
 800
 801         /* do the arch specific task caches init */
 802         arch_task_cache_init();
 803
 804         set_max_threads(MAX_THREADS);
 805
 806         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 807         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 808         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 809                 init_task.signal->rlim[RLIMIT_NPROC];
 810
 811         for (i = 0; i < UCOUNT_COUNTS; i++) {
 812                 init_user_ns.ucount_max[i] = max_threads/2;
 813         }
 814
 815 #ifdef CONFIG_VMAP_STACK
 816         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 817                           NULL, free_vm_stack_cache);
 818 #endif
 819
 820         lockdep_init_task(&init_task);
 821         uprobes_init();
 822 }
 823
 824 int __weak arch_dup_task_struct(struct task_struct *dst,
 825                                                struct task_struct *src)
 826 {
 827         *dst = *src;
 828         return 0;
 829 }
 830
 831 void set_task_stack_end_magic(struct task_struct *tsk)
 832 {
 833         unsigned long *stackend;
 834
 835         stackend = end_of_stack(tsk);
 836         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 837 }
 838
 839 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 840 {
 841         struct task_struct *tsk;
 842         unsigned long *stack;
 843         struct vm_struct *stack_vm_area __maybe_unused;
 844         int err;
 845
 846         if (node == NUMA_NO_NODE)
 847                 node = tsk_fork_get_node(orig);
 848         tsk = alloc_task_struct_node(node);
 849         if (!tsk)
 850                 return NULL;
 851
 852         stack = alloc_thread_stack_node(tsk, node);
 853         if (!stack)
 854                 goto free_tsk;
 855
 856         if (memcg_charge_kernel_stack(tsk))
 857                 goto free_stack;
 858
 859         stack_vm_area = task_stack_vm_area(tsk);
 860
 861         err = arch_dup_task_struct(tsk, orig);
 862
 863         /*
 864          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 865          * sure they're properly initialized before using any stack-related
 866          * functions again.
 867          */
 868         tsk->stack = stack;
 869 #ifdef CONFIG_VMAP_STACK
 870         tsk->stack_vm_area = stack_vm_area;
 871 #endif
 872 #ifdef CONFIG_THREAD_INFO_IN_TASK
 873         refcount_set(&tsk->stack_refcount, 1);
 874 #endif
 875
 876         if (err)
 877                 goto free_stack;
 878
 879 #ifdef CONFIG_SECCOMP
 880         /*
 881          * We must handle setting up seccomp filters once we're under
 882          * the sighand lock in case orig has changed between now and
 883          * then. Until then, filter must be NULL to avoid messing up
 884          * the usage counts on the error path calling free_task.
 885          */
 886         tsk->seccomp.filter = NULL;
 887 #endif
 888
 889         setup_thread_stack(tsk, orig);
 890         clear_user_return_notifier(tsk);
 891         clear_tsk_need_resched(tsk);
 892         set_task_stack_end_magic(tsk);
 893
 894 #ifdef CONFIG_STACKPROTECTOR
 895         tsk->stack_canary = get_random_canary();
 896 #endif
 897
 898         /*
 899          * One for us, one for whoever does the "release_task()" (usually
 900          * parent)
 901          */
 902         refcount_set(&tsk->usage, 2);
 903 #ifdef CONFIG_BLK_DEV_IO_TRACE
 904         tsk->btrace_seq = 0;
 905 #endif
 906         tsk->splice_pipe = NULL;
 907         tsk->task_frag.page = NULL;
 908         tsk->wake_q.next = NULL;
 909
 910         account_kernel_stack(tsk, 1);
 911
 912         kcov_task_init(tsk);
 913
 914 #ifdef CONFIG_FAULT_INJECTION
 915         tsk->fail_nth = 0;
 916 #endif
 917
 918 #ifdef CONFIG_BLK_CGROUP
 919         tsk->throttle_queue = NULL;
 920         tsk->use_memdelay = 0;
 921 #endif
 922
 923 #ifdef CONFIG_MEMCG
 924         tsk->active_memcg = NULL;
 925 #endif
 926         return tsk;
 927
 928 free_stack:
 929         free_thread_stack(tsk);
 930 free_tsk:
 931         free_task_struct(tsk);
 932         return NULL;
 933 }
 934
 935 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 936
 937 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 938
 939 static int __init coredump_filter_setup(char *s)
 940 {
 941         default_dump_filter =
 942                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 943                 MMF_DUMP_FILTER_MASK;
 944         return 1;
 945 }
 946
 947 __setup("coredump_filter=", coredump_filter_setup);
 948
 949 #include <linux/init_task.h>
 950
 951 static void mm_init_aio(struct mm_struct *mm)
 952 {
 953 #ifdef CONFIG_AIO
 954         spin_lock_init(&mm->ioctx_lock);
 955         mm->ioctx_table = NULL;
 956 #endif
 957 }
 958
 959 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 960                                            struct task_struct *p)
 961 {
 962 #ifdef CONFIG_MEMCG
 963         if (mm->owner == p)
 964                 WRITE_ONCE(mm->owner, NULL);
 965 #endif
 966 }
 967
 968 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 969 {
 970 #ifdef CONFIG_MEMCG
 971         mm->owner = p;
 972 #endif
 973 }
 974
 975 static void mm_init_uprobes_state(struct mm_struct *mm)
 976 {
 977 #ifdef CONFIG_UPROBES
 978         mm->uprobes_state.xol_area = NULL;
 979 #endif
 980 }
 981
 982 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 983         struct user_namespace *user_ns)
 984 {
 985         mm->mmap = NULL;
 986         mm->mm_rb = RB_ROOT;
 987         mm->vmacache_seqnum = 0;
 988         atomic_set(&mm->mm_users, 1);
 989         atomic_set(&mm->mm_count, 1);
 990         init_rwsem(&mm->mmap_sem);
 991         INIT_LIST_HEAD(&mm->mmlist);
 992         mm->core_state = NULL;
 993         mm_pgtables_bytes_init(mm);
 994         mm->map_count = 0;
 995         mm->locked_vm = 0;
 996         atomic64_set(&mm->pinned_vm, 0);
 997         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 998         spin_lock_init(&mm->page_table_lock);
 999         spin_lock_init(&mm->arg_lock);
1000         mm_init_cpumask(mm);
1001         mm_init_aio(mm);
1002         mm_init_owner(mm, p);
1003         RCU_INIT_POINTER(mm->exe_file, NULL);
1004         mmu_notifier_mm_init(mm);
1005         hmm_mm_init(mm);
1006         init_tlb_flush_pending(mm);
1007 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1008         mm->pmd_huge_pte = NULL;
1009 #endif
1010         mm_init_uprobes_state(mm);
1011
1012         if (current->mm) {
1013                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1014                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1015         } else {
1016                 mm->flags = default_dump_filter;
1017                 mm->def_flags = 0;
1018         }
1019
1020         if (mm_alloc_pgd(mm))
1021                 goto fail_nopgd;
1022
1023         if (init_new_context(p, mm))
1024                 goto fail_nocontext;
1025
1026         mm->user_ns = get_user_ns(user_ns);
1027         return mm;
1028
1029 fail_nocontext:
1030         mm_free_pgd(mm);
1031 fail_nopgd:
1032         free_mm(mm);
1033         return NULL;
1034 }
1035
1036 /*
1037  * Allocate and initialize an mm_struct.
1038  */
1039 struct mm_struct *mm_alloc(void)
1040 {
1041         struct mm_struct *mm;
1042
1043         mm = allocate_mm();
1044         if (!mm)
1045                 return NULL;
1046
1047         memset(mm, 0, sizeof(*mm));
1048         return mm_init(mm, current, current_user_ns());
1049 }
1050
1051 static inline void __mmput(struct mm_struct *mm)
1052 {
1053         VM_BUG_ON(atomic_read(&mm->mm_users));
1054
1055         uprobe_clear_state(mm);
1056         exit_aio(mm);
1057         ksm_exit(mm);
1058         khugepaged_exit(mm); /* must run before exit_mmap */
1059         exit_mmap(mm);
1060         mm_put_huge_zero_page(mm);
1061         set_mm_exe_file(mm, NULL);
1062         if (!list_empty(&mm->mmlist)) {
1063                 spin_lock(&mmlist_lock);
1064                 list_del(&mm->mmlist);
1065                 spin_unlock(&mmlist_lock);
1066         }
1067         if (mm->binfmt)
1068                 module_put(mm->binfmt->module);
1069         mmdrop(mm);
1070 }
1071
1072 /*
1073  * Decrement the use count and release all resources for an mm.
1074  */
1075 void mmput(struct mm_struct *mm)
1076 {
1077         might_sleep();
1078
1079         if (atomic_dec_and_test(&mm->mm_users))
1080                 __mmput(mm);
1081 }
1082 EXPORT_SYMBOL_GPL(mmput);
1083
1084 #ifdef CONFIG_MMU
1085 static void mmput_async_fn(struct work_struct *work)
1086 {
1087         struct mm_struct *mm = container_of(work, struct mm_struct,
1088                                             async_put_work);
1089
1090         __mmput(mm);
1091 }
1092
1093 void mmput_async(struct mm_struct *mm)
1094 {
1095         if (atomic_dec_and_test(&mm->mm_users)) {
1096                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1097                 schedule_work(&mm->async_put_work);
1098         }
1099 }
1100 #endif
1101
1102 /**
1103  * set_mm_exe_file - change a reference to the mm's executable file
1104  *
1105  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1106  *
1107  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1108  * invocations: in mmput() nobody alive left, in execve task is single
1109  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1110  * mm->exe_file, but does so without using set_mm_exe_file() in order
1111  * to do avoid the need for any locks.
1112  */
1113 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1114 {
1115         struct file *old_exe_file;
1116
1117         /*
1118          * It is safe to dereference the exe_file without RCU as
1119          * this function is only called if nobody else can access
1120          * this mm -- see comment above for justification.
1121          */
1122         old_exe_file = rcu_dereference_raw(mm->exe_file);
1123
1124         if (new_exe_file)
1125                 get_file(new_exe_file);
1126         rcu_assign_pointer(mm->exe_file, new_exe_file);
1127         if (old_exe_file)
1128                 fput(old_exe_file);
1129 }
1130
1131 /**
1132  * get_mm_exe_file - acquire a reference to the mm's executable file
1133  *
1134  * Returns %NULL if mm has no associated executable file.
1135  * User must release file via fput().
1136  */
1137 struct file *get_mm_exe_file(struct mm_struct *mm)
1138 {
1139         struct file *exe_file;
1140
1141         rcu_read_lock();
1142         exe_file = rcu_dereference(mm->exe_file);
1143         if (exe_file && !get_file_rcu(exe_file))
1144                 exe_file = NULL;
1145         rcu_read_unlock();
1146         return exe_file;
1147 }
1148 EXPORT_SYMBOL(get_mm_exe_file);
1149
1150 /**
1151  * get_task_exe_file - acquire a reference to the task's executable file
1152  *
1153  * Returns %NULL if task's mm (if any) has no associated executable file or
1154  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1155  * User must release file via fput().
1156  */
1157 struct file *get_task_exe_file(struct task_struct *task)
1158 {
1159         struct file *exe_file = NULL;
1160         struct mm_struct *mm;
1161
1162         task_lock(task);
1163         mm = task->mm;
1164         if (mm) {
1165                 if (!(task->flags & PF_KTHREAD))
1166                         exe_file = get_mm_exe_file(mm);
1167         }
1168         task_unlock(task);
1169         return exe_file;
1170 }
1171 EXPORT_SYMBOL(get_task_exe_file);
1172
1173 /**
1174  * get_task_mm - acquire a reference to the task's mm
1175  *
1176  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1177  * this kernel workthread has transiently adopted a user mm with use_mm,
1178  * to do its AIO) is not set and if so returns a reference to it, after
1179  * bumping up the use count.  User must release the mm via mmput()
1180  * after use.  Typically used by /proc and ptrace.
1181  */
1182 struct mm_struct *get_task_mm(struct task_struct *task)
1183 {
1184         struct mm_struct *mm;
1185
1186         task_lock(task);
1187         mm = task->mm;
1188         if (mm) {
1189                 if (task->flags & PF_KTHREAD)
1190                         mm = NULL;
1191                 else
1192                         mmget(mm);
1193         }
1194         task_unlock(task);
1195         return mm;
1196 }
1197 EXPORT_SYMBOL_GPL(get_task_mm);
1198
1199 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1200 {
1201         struct mm_struct *mm;
1202         int err;
1203
1204         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1205         if (err)
1206                 return ERR_PTR(err);
1207
1208         mm = get_task_mm(task);
1209         if (mm && mm != current->mm &&
1210                         !ptrace_may_access(task, mode)) {
1211                 mmput(mm);
1212                 mm = ERR_PTR(-EACCES);
1213         }
1214         mutex_unlock(&task->signal->cred_guard_mutex);
1215
1216         return mm;
1217 }
1218
1219 static void complete_vfork_done(struct task_struct *tsk)
1220 {
1221         struct completion *vfork;
1222
1223         task_lock(tsk);
1224         vfork = tsk->vfork_done;
1225         if (likely(vfork)) {
1226                 tsk->vfork_done = NULL;
1227                 complete(vfork);
1228         }
1229         task_unlock(tsk);
1230 }
1231
1232 static int wait_for_vfork_done(struct task_struct *child,
1233                                 struct completion *vfork)
1234 {
1235         int killed;
1236
1237         freezer_do_not_count();
1238         cgroup_enter_frozen();
1239         killed = wait_for_completion_killable(vfork);
1240         cgroup_leave_frozen(false);
1241         freezer_count();
1242
1243         if (killed) {
1244                 task_lock(child);
1245                 child->vfork_done = NULL;
1246                 task_unlock(child);
1247         }
1248
1249         put_task_struct(child);
1250         return killed;
1251 }
1252
1253 /* Please note the differences between mmput and mm_release.
1254  * mmput is called whenever we stop holding onto a mm_struct,
1255  * error success whatever.
1256  *
1257  * mm_release is called after a mm_struct has been removed
1258  * from the current process.
1259  *
1260  * This difference is important for error handling, when we
1261  * only half set up a mm_struct for a new process and need to restore
1262  * the old one.  Because we mmput the new mm_struct before
1263  * restoring the old one. . .
1264  * Eric Biederman 10 January 1998
1265  */
1266 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1267 {
1268         /* Get rid of any futexes when releasing the mm */
1269 #ifdef CONFIG_FUTEX
1270         if (unlikely(tsk->robust_list)) {
1271                 exit_robust_list(tsk);
1272                 tsk->robust_list = NULL;
1273         }
1274 #ifdef CONFIG_COMPAT
1275         if (unlikely(tsk->compat_robust_list)) {
1276                 compat_exit_robust_list(tsk);
1277                 tsk->compat_robust_list = NULL;
1278         }
1279 #endif
1280         if (unlikely(!list_empty(&tsk->pi_state_list)))
1281                 exit_pi_state_list(tsk);
1282 #endif
1283
1284         uprobe_free_utask(tsk);
1285
1286         /* Get rid of any cached register state */
1287         deactivate_mm(tsk, mm);
1288
1289         /*
1290          * Signal userspace if we're not exiting with a core dump
1291          * because we want to leave the value intact for debugging
1292          * purposes.
1293          */
1294         if (tsk->clear_child_tid) {
1295                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1296                     atomic_read(&mm->mm_users) > 1) {
1297                         /*
1298                          * We don't check the error code - if userspace has
1299                          * not set up a proper pointer then tough luck.
1300                          */
1301                         put_user(0, tsk->clear_child_tid);
1302                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1303                                         1, NULL, NULL, 0, 0);
1304                 }
1305                 tsk->clear_child_tid = NULL;
1306         }
1307
1308         /*
1309          * All done, finally we can wake up parent and return this mm to him.
1310          * Also kthread_stop() uses this completion for synchronization.
1311          */
1312         if (tsk->vfork_done)
1313                 complete_vfork_done(tsk);
1314 }
1315
1316 /**
1317  * dup_mm() - duplicates an existing mm structure
1318  * @tsk: the task_struct with which the new mm will be associated.
1319  * @oldmm: the mm to duplicate.
1320  *
1321  * Allocates a new mm structure and duplicates the provided @oldmm structure
1322  * content into it.
1323  *
1324  * Return: the duplicated mm or NULL on failure.
1325  */
1326 static struct mm_struct *dup_mm(struct task_struct *tsk,
1327                                 struct mm_struct *oldmm)
1328 {
1329         struct mm_struct *mm;
1330         int err;
1331
1332         mm = allocate_mm();
1333         if (!mm)
1334                 goto fail_nomem;
1335
1336         memcpy(mm, oldmm, sizeof(*mm));
1337
1338         if (!mm_init(mm, tsk, mm->user_ns))
1339                 goto fail_nomem;
1340
1341         err = dup_mmap(mm, oldmm);
1342         if (err)
1343                 goto free_pt;
1344
1345         mm->hiwater_rss = get_mm_rss(mm);
1346         mm->hiwater_vm = mm->total_vm;
1347
1348         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1349                 goto free_pt;
1350
1351         return mm;
1352
1353 free_pt:
1354         /* don't put binfmt in mmput, we haven't got module yet */
1355         mm->binfmt = NULL;
1356         mm_init_owner(mm, NULL);
1357         mmput(mm);
1358
1359 fail_nomem:
1360         return NULL;
1361 }
1362
1363 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1364 {
1365         struct mm_struct *mm, *oldmm;
1366         int retval;
1367
1368         tsk->min_flt = tsk->maj_flt = 0;
1369         tsk->nvcsw = tsk->nivcsw = 0;
1370 #ifdef CONFIG_DETECT_HUNG_TASK
1371         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1372         tsk->last_switch_time = 0;
1373 #endif
1374
1375         tsk->mm = NULL;
1376         tsk->active_mm = NULL;
1377
1378         /*
1379          * Are we cloning a kernel thread?
1380          *
1381          * We need to steal a active VM for that..
1382          */
1383         oldmm = current->mm;
1384         if (!oldmm)
1385                 return 0;
1386
1387         /* initialize the new vmacache entries */
1388         vmacache_flush(tsk);
1389
1390         if (clone_flags & CLONE_VM) {
1391                 mmget(oldmm);
1392                 mm = oldmm;
1393                 goto good_mm;
1394         }
1395
1396         retval = -ENOMEM;
1397         mm = dup_mm(tsk, current->mm);
1398         if (!mm)
1399                 goto fail_nomem;
1400
1401 good_mm:
1402         tsk->mm = mm;
1403         tsk->active_mm = mm;
1404         return 0;
1405
1406 fail_nomem:
1407         return retval;
1408 }
1409
1410 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1411 {
1412         struct fs_struct *fs = current->fs;
1413         if (clone_flags & CLONE_FS) {
1414                 /* tsk->fs is already what we want */
1415                 spin_lock(&fs->lock);
1416                 if (fs->in_exec) {
1417                         spin_unlock(&fs->lock);
1418                         return -EAGAIN;
1419                 }
1420                 fs->users++;
1421                 spin_unlock(&fs->lock);
1422                 return 0;
1423         }
1424         tsk->fs = copy_fs_struct(fs);
1425         if (!tsk->fs)
1426                 return -ENOMEM;
1427         return 0;
1428 }
1429
1430 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1431 {
1432         struct files_struct *oldf, *newf;
1433         int error = 0;
1434
1435         /*
1436          * A background process may not have any files ...
1437          */
1438         oldf = current->files;
1439         if (!oldf)
1440                 goto out;
1441
1442         if (clone_flags & CLONE_FILES) {
1443                 atomic_inc(&oldf->count);
1444                 goto out;
1445         }
1446
1447         newf = dup_fd(oldf, &error);
1448         if (!newf)
1449                 goto out;
1450
1451         tsk->files = newf;
1452         error = 0;
1453 out:
1454         return error;
1455 }
1456
1457 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1458 {
1459 #ifdef CONFIG_BLOCK
1460         struct io_context *ioc = current->io_context;
1461         struct io_context *new_ioc;
1462
1463         if (!ioc)
1464                 return 0;
1465         /*
1466          * Share io context with parent, if CLONE_IO is set
1467          */
1468         if (clone_flags & CLONE_IO) {
1469                 ioc_task_link(ioc);
1470                 tsk->io_context = ioc;
1471         } else if (ioprio_valid(ioc->ioprio)) {
1472                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1473                 if (unlikely(!new_ioc))
1474                         return -ENOMEM;
1475
1476                 new_ioc->ioprio = ioc->ioprio;
1477                 put_io_context(new_ioc);
1478         }
1479 #endif
1480         return 0;
1481 }
1482
1483 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1484 {
1485         struct sighand_struct *sig;
1486
1487         if (clone_flags & CLONE_SIGHAND) {
1488                 refcount_inc(&current->sighand->count);
1489                 return 0;
1490         }
1491         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1492         rcu_assign_pointer(tsk->sighand, sig);
1493         if (!sig)
1494                 return -ENOMEM;
1495
1496         refcount_set(&sig->count, 1);
1497         spin_lock_irq(&current->sighand->siglock);
1498         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1499         spin_unlock_irq(&current->sighand->siglock);
1500         return 0;
1501 }
1502
1503 void __cleanup_sighand(struct sighand_struct *sighand)
1504 {
1505         if (refcount_dec_and_test(&sighand->count)) {
1506                 signalfd_cleanup(sighand);
1507                 /*
1508                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1509                  * without an RCU grace period, see __lock_task_sighand().
1510                  */
1511                 kmem_cache_free(sighand_cachep, sighand);
1512         }
1513 }
1514
1515 #ifdef CONFIG_POSIX_TIMERS
1516 /*
1517  * Initialize POSIX timer handling for a thread group.
1518  */
1519 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1520 {
1521         unsigned long cpu_limit;
1522
1523         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1524         if (cpu_limit != RLIM_INFINITY) {
1525                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1526                 sig->cputimer.running = true;
1527         }
1528
1529         /* The timer lists. */
1530         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1531         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1532         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1533 }
1534 #else
1535 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1536 #endif
1537
1538 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1539 {
1540         struct signal_struct *sig;
1541
1542         if (clone_flags & CLONE_THREAD)
1543                 return 0;
1544
1545         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1546         tsk->signal = sig;
1547         if (!sig)
1548                 return -ENOMEM;
1549
1550         sig->nr_threads = 1;
1551         atomic_set(&sig->live, 1);
1552         refcount_set(&sig->sigcnt, 1);
1553
1554         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1555         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1556         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1557
1558         init_waitqueue_head(&sig->wait_chldexit);
1559         sig->curr_target = tsk;
1560         init_sigpending(&sig->shared_pending);
1561         INIT_HLIST_HEAD(&sig->multiprocess);
1562         seqlock_init(&sig->stats_lock);
1563         prev_cputime_init(&sig->prev_cputime);
1564
1565 #ifdef CONFIG_POSIX_TIMERS
1566         INIT_LIST_HEAD(&sig->posix_timers);
1567         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1568         sig->real_timer.function = it_real_fn;
1569 #endif
1570
1571         task_lock(current->group_leader);
1572         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1573         task_unlock(current->group_leader);
1574
1575         posix_cpu_timers_init_group(sig);
1576
1577         tty_audit_fork(sig);
1578         sched_autogroup_fork(sig);
1579
1580         sig->oom_score_adj = current->signal->oom_score_adj;
1581         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1582
1583         mutex_init(&sig->cred_guard_mutex);
1584
1585         return 0;
1586 }
1587
1588 static void copy_seccomp(struct task_struct *p)
1589 {
1590 #ifdef CONFIG_SECCOMP
1591         /*
1592          * Must be called with sighand->lock held, which is common to
1593          * all threads in the group. Holding cred_guard_mutex is not
1594          * needed because this new task is not yet running and cannot
1595          * be racing exec.
1596          */
1597         assert_spin_locked(&current->sighand->siglock);
1598
1599         /* Ref-count the new filter user, and assign it. */
1600         get_seccomp_filter(current);
1601         p->seccomp = current->seccomp;
1602
1603         /*
1604          * Explicitly enable no_new_privs here in case it got set
1605          * between the task_struct being duplicated and holding the
1606          * sighand lock. The seccomp state and nnp must be in sync.
1607          */
1608         if (task_no_new_privs(current))
1609                 task_set_no_new_privs(p);
1610
1611         /*
1612          * If the parent gained a seccomp mode after copying thread
1613          * flags and between before we held the sighand lock, we have
1614          * to manually enable the seccomp thread flag here.
1615          */
1616         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1617                 set_tsk_thread_flag(p, TIF_SECCOMP);
1618 #endif
1619 }
1620
1621 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1622 {
1623         current->clear_child_tid = tidptr;
1624
1625         return task_pid_vnr(current);
1626 }
1627
1628 static void rt_mutex_init_task(struct task_struct *p)
1629 {
1630         raw_spin_lock_init(&p->pi_lock);
1631 #ifdef CONFIG_RT_MUTEXES
1632         p->pi_waiters = RB_ROOT_CACHED;
1633         p->pi_top_task = NULL;
1634         p->pi_blocked_on = NULL;
1635 #endif
1636 }
1637
1638 #ifdef CONFIG_POSIX_TIMERS
1639 /*
1640  * Initialize POSIX timer handling for a single task.
1641  */
1642 static void posix_cpu_timers_init(struct task_struct *tsk)
1643 {
1644         tsk->cputime_expires.prof_exp = 0;
1645         tsk->cputime_expires.virt_exp = 0;
1646         tsk->cputime_expires.sched_exp = 0;
1647         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1648         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1649         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1650 }
1651 #else
1652 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1653 #endif
1654
1655 static inline void init_task_pid_links(struct task_struct *task)
1656 {
1657         enum pid_type type;
1658
1659         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1660                 INIT_HLIST_NODE(&task->pid_links[type]);
1661         }
1662 }
1663
1664 static inline void
1665 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1666 {
1667         if (type == PIDTYPE_PID)
1668                 task->thread_pid = pid;
1669         else
1670                 task->signal->pids[type] = pid;
1671 }
1672
1673 static inline void rcu_copy_process(struct task_struct *p)
1674 {
1675 #ifdef CONFIG_PREEMPT_RCU
1676         p->rcu_read_lock_nesting = 0;
1677         p->rcu_read_unlock_special.s = 0;
1678         p->rcu_blocked_node = NULL;
1679         INIT_LIST_HEAD(&p->rcu_node_entry);
1680 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1681 #ifdef CONFIG_TASKS_RCU
1682         p->rcu_tasks_holdout = false;
1683         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1684         p->rcu_tasks_idle_cpu = -1;
1685 #endif /* #ifdef CONFIG_TASKS_RCU */
1686 }
1687
1688 static int pidfd_release(struct inode *inode, struct file *file)
1689 {
1690         struct pid *pid = file->private_data;
1691
1692         file->private_data = NULL;
1693         put_pid(pid);
1694         return 0;
1695 }
1696
1697 #ifdef CONFIG_PROC_FS
1698 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1699 {
1700         struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1701         struct pid *pid = f->private_data;
1702
1703         seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1704         seq_putc(m, '\n');
1705 }
1706 #endif
1707
1708 const struct file_operations pidfd_fops = {
1709         .release = pidfd_release,
1710 #ifdef CONFIG_PROC_FS
1711         .show_fdinfo = pidfd_show_fdinfo,
1712 #endif
1713 };
1714
1715 /**
1716  * pidfd_create() - Create a new pid file descriptor.
1717  *
1718  * @pid:  struct pid that the pidfd will reference
1719  *
1720  * This creates a new pid file descriptor with the O_CLOEXEC flag set.
1721  *
1722  * Note, that this function can only be called after the fd table has
1723  * been unshared to avoid leaking the pidfd to the new process.
1724  *
1725  * Return: On success, a cloexec pidfd is returned.
1726  *         On error, a negative errno number will be returned.
1727  */
1728 static int pidfd_create(struct pid *pid)
1729 {
1730         int fd;
1731
1732         fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid),
1733                               O_RDWR | O_CLOEXEC);
1734         if (fd < 0)
1735                 put_pid(pid);
1736
1737         return fd;
1738 }
1739
1740 static void __delayed_free_task(struct rcu_head *rhp)
1741 {
1742         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1743
1744         free_task(tsk);
1745 }
1746
1747 static __always_inline void delayed_free_task(struct task_struct *tsk)
1748 {
1749         if (IS_ENABLED(CONFIG_MEMCG))
1750                 call_rcu(&tsk->rcu, __delayed_free_task);
1751         else
1752                 free_task(tsk);
1753 }
1754
1755 /*
1756  * This creates a new process as a copy of the old one,
1757  * but does not actually start it yet.
1758  *
1759  * It copies the registers, and all the appropriate
1760  * parts of the process environment (as per the clone
1761  * flags). The actual kick-off is left to the caller.
1762  */
1763 static __latent_entropy struct task_struct *copy_process(
1764                                         unsigned long clone_flags,
1765                                         unsigned long stack_start,
1766                                         unsigned long stack_size,
1767                                         int __user *parent_tidptr,
1768                                         int __user *child_tidptr,
1769                                         struct pid *pid,
1770                                         int trace,
1771                                         unsigned long tls,
1772                                         int node)
1773 {
1774         int pidfd = -1, retval;
1775         struct task_struct *p;
1776         struct multiprocess_signals delayed;
1777
1778         /*
1779          * Don't allow sharing the root directory with processes in a different
1780          * namespace
1781          */
1782         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1783                 return ERR_PTR(-EINVAL);
1784
1785         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1786                 return ERR_PTR(-EINVAL);
1787
1788         /*
1789          * Thread groups must share signals as well, and detached threads
1790          * can only be started up within the thread group.
1791          */
1792         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1793                 return ERR_PTR(-EINVAL);
1794
1795         /*
1796          * Shared signal handlers imply shared VM. By way of the above,
1797          * thread groups also imply shared VM. Blocking this case allows
1798          * for various simplifications in other code.
1799          */
1800         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1801                 return ERR_PTR(-EINVAL);
1802
1803         /*
1804          * Siblings of global init remain as zombies on exit since they are
1805          * not reaped by their parent (swapper). To solve this and to avoid
1806          * multi-rooted process trees, prevent global and container-inits
1807          * from creating siblings.
1808          */
1809         if ((clone_flags & CLONE_PARENT) &&
1810                                 current->signal->flags & SIGNAL_UNKILLABLE)
1811                 return ERR_PTR(-EINVAL);
1812
1813         /*
1814          * If the new process will be in a different pid or user namespace
1815          * do not allow it to share a thread group with the forking task.
1816          */
1817         if (clone_flags & CLONE_THREAD) {
1818                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1819                     (task_active_pid_ns(current) !=
1820                                 current->nsproxy->pid_ns_for_children))
1821                         return ERR_PTR(-EINVAL);
1822         }
1823
1824         if (clone_flags & CLONE_PIDFD) {
1825                 int reserved;
1826
1827                 /*
1828                  * - CLONE_PARENT_SETTID is useless for pidfds and also
1829                  *   parent_tidptr is used to return pidfds.
1830                  * - CLONE_DETACHED is blocked so that we can potentially
1831                  *   reuse it later for CLONE_PIDFD.
1832                  * - CLONE_THREAD is blocked until someone really needs it.
1833                  */
1834                 if (clone_flags &
1835                     (CLONE_DETACHED | CLONE_PARENT_SETTID | CLONE_THREAD))
1836                         return ERR_PTR(-EINVAL);
1837
1838                 /*
1839                  * Verify that parent_tidptr is sane so we can potentially
1840                  * reuse it later.
1841                  */
1842                 if (get_user(reserved, parent_tidptr))
1843                         return ERR_PTR(-EFAULT);
1844
1845                 if (reserved != 0)
1846                         return ERR_PTR(-EINVAL);
1847         }
1848
1849         /*
1850          * Force any signals received before this point to be delivered
1851          * before the fork happens.  Collect up signals sent to multiple
1852          * processes that happen during the fork and delay them so that
1853          * they appear to happen after the fork.
1854          */
1855         sigemptyset(&delayed.signal);
1856         INIT_HLIST_NODE(&delayed.node);
1857
1858         spin_lock_irq(&current->sighand->siglock);
1859         if (!(clone_flags & CLONE_THREAD))
1860                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1861         recalc_sigpending();
1862         spin_unlock_irq(&current->sighand->siglock);
1863         retval = -ERESTARTNOINTR;
1864         if (signal_pending(current))
1865                 goto fork_out;
1866
1867         retval = -ENOMEM;
1868         p = dup_task_struct(current, node);
1869         if (!p)
1870                 goto fork_out;
1871
1872         /*
1873          * This _must_ happen before we call free_task(), i.e. before we jump
1874          * to any of the bad_fork_* labels. This is to avoid freeing
1875          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1876          * kernel threads (PF_KTHREAD).
1877          */
1878         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1879         /*
1880          * Clear TID on mm_release()?
1881          */
1882         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1883
1884         ftrace_graph_init_task(p);
1885
1886         rt_mutex_init_task(p);
1887
1888 #ifdef CONFIG_PROVE_LOCKING
1889         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1890         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1891 #endif
1892         retval = -EAGAIN;
1893         if (atomic_read(&p->real_cred->user->processes) >=
1894                         task_rlimit(p, RLIMIT_NPROC)) {
1895                 if (p->real_cred->user != INIT_USER &&
1896                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1897                         goto bad_fork_free;
1898         }
1899         current->flags &= ~PF_NPROC_EXCEEDED;
1900
1901         retval = copy_creds(p, clone_flags);
1902         if (retval < 0)
1903                 goto bad_fork_free;
1904
1905         /*
1906          * If multiple threads are within copy_process(), then this check
1907          * triggers too late. This doesn't hurt, the check is only there
1908          * to stop root fork bombs.
1909          */
1910         retval = -EAGAIN;
1911         if (nr_threads >= max_threads)
1912                 goto bad_fork_cleanup_count;
1913
1914         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1915         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1916         p->flags |= PF_FORKNOEXEC;
1917         INIT_LIST_HEAD(&p->children);
1918         INIT_LIST_HEAD(&p->sibling);
1919         rcu_copy_process(p);
1920         p->vfork_done = NULL;
1921         spin_lock_init(&p->alloc_lock);
1922
1923         init_sigpending(&p->pending);
1924
1925         p->utime = p->stime = p->gtime = 0;
1926 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1927         p->utimescaled = p->stimescaled = 0;
1928 #endif
1929         prev_cputime_init(&p->prev_cputime);
1930
1931 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1932         seqcount_init(&p->vtime.seqcount);
1933         p->vtime.starttime = 0;
1934         p->vtime.state = VTIME_INACTIVE;
1935 #endif
1936
1937 #if defined(SPLIT_RSS_COUNTING)
1938         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1939 #endif
1940
1941         p->default_timer_slack_ns = current->timer_slack_ns;
1942
1943 #ifdef CONFIG_PSI
1944         p->psi_flags = 0;
1945 #endif
1946
1947         task_io_accounting_init(&p->ioac);
1948         acct_clear_integrals(p);
1949
1950         posix_cpu_timers_init(p);
1951
1952         p->io_context = NULL;
1953         audit_set_context(p, NULL);
1954         cgroup_fork(p);
1955 #ifdef CONFIG_NUMA
1956         p->mempolicy = mpol_dup(p->mempolicy);
1957         if (IS_ERR(p->mempolicy)) {
1958                 retval = PTR_ERR(p->mempolicy);
1959                 p->mempolicy = NULL;
1960                 goto bad_fork_cleanup_threadgroup_lock;
1961         }
1962 #endif
1963 #ifdef CONFIG_CPUSETS
1964         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1965         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1966         seqcount_init(&p->mems_allowed_seq);
1967 #endif
1968 #ifdef CONFIG_TRACE_IRQFLAGS
1969         p->irq_events = 0;
1970         p->hardirqs_enabled = 0;
1971         p->hardirq_enable_ip = 0;
1972         p->hardirq_enable_event = 0;
1973         p->hardirq_disable_ip = _THIS_IP_;
1974         p->hardirq_disable_event = 0;
1975         p->softirqs_enabled = 1;
1976         p->softirq_enable_ip = _THIS_IP_;
1977         p->softirq_enable_event = 0;
1978         p->softirq_disable_ip = 0;
1979         p->softirq_disable_event = 0;
1980         p->hardirq_context = 0;
1981         p->softirq_context = 0;
1982 #endif
1983
1984         p->pagefault_disabled = 0;
1985
1986 #ifdef CONFIG_LOCKDEP
1987         p->lockdep_depth = 0; /* no locks held yet */
1988         p->curr_chain_key = 0;
1989         p->lockdep_recursion = 0;
1990         lockdep_init_task(p);
1991 #endif
1992
1993 #ifdef CONFIG_DEBUG_MUTEXES
1994         p->blocked_on = NULL; /* not blocked yet */
1995 #endif
1996 #ifdef CONFIG_BCACHE
1997         p->sequential_io        = 0;
1998         p->sequential_io_avg    = 0;
1999 #endif
2000
2001         /* Perform scheduler related setup. Assign this task to a CPU. */
2002         retval = sched_fork(clone_flags, p);
2003         if (retval)
2004                 goto bad_fork_cleanup_policy;
2005
2006         retval = perf_event_init_task(p);
2007         if (retval)
2008                 goto bad_fork_cleanup_policy;
2009         retval = audit_alloc(p);
2010         if (retval)
2011                 goto bad_fork_cleanup_perf;
2012         /* copy all the process information */
2013         shm_init_task(p);
2014         retval = security_task_alloc(p, clone_flags);
2015         if (retval)
2016                 goto bad_fork_cleanup_audit;
2017         retval = copy_semundo(clone_flags, p);
2018         if (retval)
2019                 goto bad_fork_cleanup_security;
2020         retval = copy_files(clone_flags, p);
2021         if (retval)
2022                 goto bad_fork_cleanup_semundo;
2023         retval = copy_fs(clone_flags, p);
2024         if (retval)
2025                 goto bad_fork_cleanup_files;
2026         retval = copy_sighand(clone_flags, p);
2027         if (retval)
2028                 goto bad_fork_cleanup_fs;
2029         retval = copy_signal(clone_flags, p);
2030         if (retval)
2031                 goto bad_fork_cleanup_sighand;
2032         retval = copy_mm(clone_flags, p);
2033         if (retval)
2034                 goto bad_fork_cleanup_signal;
2035         retval = copy_namespaces(clone_flags, p);
2036         if (retval)
2037                 goto bad_fork_cleanup_mm;
2038         retval = copy_io(clone_flags, p);
2039         if (retval)
2040                 goto bad_fork_cleanup_namespaces;
2041         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
2042         if (retval)
2043                 goto bad_fork_cleanup_io;
2044
2045         stackleak_task_init(p);
2046
2047         if (pid != &init_struct_pid) {
2048                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
2049                 if (IS_ERR(pid)) {
2050                         retval = PTR_ERR(pid);
2051                         goto bad_fork_cleanup_thread;
2052                 }
2053         }
2054
2055         /*
2056          * This has to happen after we've potentially unshared the file
2057          * descriptor table (so that the pidfd doesn't leak into the child
2058          * if the fd table isn't shared).
2059          */
2060         if (clone_flags & CLONE_PIDFD) {
2061                 retval = pidfd_create(pid);
2062                 if (retval < 0)
2063                         goto bad_fork_free_pid;
2064
2065                 pidfd = retval;
2066                 retval = put_user(pidfd, parent_tidptr);
2067                 if (retval)
2068                         goto bad_fork_put_pidfd;
2069         }
2070
2071 #ifdef CONFIG_BLOCK
2072         p->plug = NULL;
2073 #endif
2074 #ifdef CONFIG_FUTEX
2075         p->robust_list = NULL;
2076 #ifdef CONFIG_COMPAT
2077         p->compat_robust_list = NULL;
2078 #endif
2079         INIT_LIST_HEAD(&p->pi_state_list);
2080         p->pi_state_cache = NULL;
2081 #endif
2082         /*
2083          * sigaltstack should be cleared when sharing the same VM
2084          */
2085         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2086                 sas_ss_reset(p);
2087
2088         /*
2089          * Syscall tracing and stepping should be turned off in the
2090          * child regardless of CLONE_PTRACE.
2091          */
2092         user_disable_single_step(p);
2093         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2094 #ifdef TIF_SYSCALL_EMU
2095         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2096 #endif
2097         clear_tsk_latency_tracing(p);
2098
2099         /* ok, now we should be set up.. */
2100         p->pid = pid_nr(pid);
2101         if (clone_flags & CLONE_THREAD) {
2102                 p->exit_signal = -1;
2103                 p->group_leader = current->group_leader;
2104                 p->tgid = current->tgid;
2105         } else {
2106                 if (clone_flags & CLONE_PARENT)
2107                         p->exit_signal = current->group_leader->exit_signal;
2108                 else
2109                         p->exit_signal = (clone_flags & CSIGNAL);
2110                 p->group_leader = p;
2111                 p->tgid = p->pid;
2112         }
2113
2114         p->nr_dirtied = 0;
2115         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2116         p->dirty_paused_when = 0;
2117
2118         p->pdeath_signal = 0;
2119         INIT_LIST_HEAD(&p->thread_group);
2120         p->task_works = NULL;
2121
2122         cgroup_threadgroup_change_begin(current);
2123         /*
2124          * Ensure that the cgroup subsystem policies allow the new process to be
2125          * forked. It should be noted the the new process's css_set can be changed
2126          * between here and cgroup_post_fork() if an organisation operation is in
2127          * progress.
2128          */
2129         retval = cgroup_can_fork(p);
2130         if (retval)
2131                 goto bad_fork_cgroup_threadgroup_change_end;
2132
2133         /*
2134          * From this point on we must avoid any synchronous user-space
2135          * communication until we take the tasklist-lock. In particular, we do
2136          * not want user-space to be able to predict the process start-time by
2137          * stalling fork(2) after we recorded the start_time but before it is
2138          * visible to the system.
2139          */
2140
2141         p->start_time = ktime_get_ns();
2142         p->real_start_time = ktime_get_boot_ns();
2143
2144         /*
2145          * Make it visible to the rest of the system, but dont wake it up yet.
2146          * Need tasklist lock for parent etc handling!
2147          */
2148         write_lock_irq(&tasklist_lock);
2149
2150         /* CLONE_PARENT re-uses the old parent */
2151         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2152                 p->real_parent = current->real_parent;
2153                 p->parent_exec_id = current->parent_exec_id;
2154         } else {
2155                 p->real_parent = current;
2156                 p->parent_exec_id = current->self_exec_id;
2157         }
2158
2159         klp_copy_process(p);
2160
2161         spin_lock(&current->sighand->siglock);
2162
2163         /*
2164          * Copy seccomp details explicitly here, in case they were changed
2165          * before holding sighand lock.
2166          */
2167         copy_seccomp(p);
2168
2169         rseq_fork(p, clone_flags);
2170
2171         /* Don't start children in a dying pid namespace */
2172         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2173                 retval = -ENOMEM;
2174                 goto bad_fork_cancel_cgroup;
2175         }
2176
2177         /* Let kill terminate clone/fork in the middle */
2178         if (fatal_signal_pending(current)) {
2179                 retval = -EINTR;
2180                 goto bad_fork_cancel_cgroup;
2181         }
2182
2183
2184         init_task_pid_links(p);
2185         if (likely(p->pid)) {
2186                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2187
2188                 init_task_pid(p, PIDTYPE_PID, pid);
2189                 if (thread_group_leader(p)) {
2190                         init_task_pid(p, PIDTYPE_TGID, pid);
2191                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2192                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2193
2194                         if (is_child_reaper(pid)) {
2195                                 ns_of_pid(pid)->child_reaper = p;
2196                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2197                         }
2198                         p->signal->shared_pending.signal = delayed.signal;
2199                         p->signal->tty = tty_kref_get(current->signal->tty);
2200                         /*
2201                          * Inherit has_child_subreaper flag under the same
2202                          * tasklist_lock with adding child to the process tree
2203                          * for propagate_has_child_subreaper optimization.
2204                          */
2205                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2206                                                          p->real_parent->signal->is_child_subreaper;
2207                         list_add_tail(&p->sibling, &p->real_parent->children);
2208                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2209                         attach_pid(p, PIDTYPE_TGID);
2210                         attach_pid(p, PIDTYPE_PGID);
2211                         attach_pid(p, PIDTYPE_SID);
2212                         __this_cpu_inc(process_counts);
2213                 } else {
2214                         current->signal->nr_threads++;
2215                         atomic_inc(&current->signal->live);
2216                         refcount_inc(&current->signal->sigcnt);
2217                         task_join_group_stop(p);
2218                         list_add_tail_rcu(&p->thread_group,
2219                                           &p->group_leader->thread_group);
2220                         list_add_tail_rcu(&p->thread_node,
2221                                           &p->signal->thread_head);
2222                 }
2223                 attach_pid(p, PIDTYPE_PID);
2224                 nr_threads++;
2225         }
2226         total_forks++;
2227         hlist_del_init(&delayed.node);
2228         spin_unlock(&current->sighand->siglock);
2229         syscall_tracepoint_update(p);
2230         write_unlock_irq(&tasklist_lock);
2231
2232         proc_fork_connector(p);
2233         cgroup_post_fork(p);
2234         cgroup_threadgroup_change_end(current);
2235         perf_event_fork(p);
2236
2237         trace_task_newtask(p, clone_flags);
2238         uprobe_copy_process(p, clone_flags);
2239
2240         return p;
2241
2242 bad_fork_cancel_cgroup:
2243         spin_unlock(&current->sighand->siglock);
2244         write_unlock_irq(&tasklist_lock);
2245         cgroup_cancel_fork(p);
2246 bad_fork_cgroup_threadgroup_change_end:
2247         cgroup_threadgroup_change_end(current);
2248 bad_fork_put_pidfd:
2249         if (clone_flags & CLONE_PIDFD)
2250                 ksys_close(pidfd);
2251 bad_fork_free_pid:
2252         if (pid != &init_struct_pid)
2253                 free_pid(pid);
2254 bad_fork_cleanup_thread:
2255         exit_thread(p);
2256 bad_fork_cleanup_io:
2257         if (p->io_context)
2258                 exit_io_context(p);
2259 bad_fork_cleanup_namespaces:
2260         exit_task_namespaces(p);
2261 bad_fork_cleanup_mm:
2262         if (p->mm) {
2263                 mm_clear_owner(p->mm, p);
2264                 mmput(p->mm);
2265         }
2266 bad_fork_cleanup_signal:
2267         if (!(clone_flags & CLONE_THREAD))
2268                 free_signal_struct(p->signal);
2269 bad_fork_cleanup_sighand:
2270         __cleanup_sighand(p->sighand);
2271 bad_fork_cleanup_fs:
2272         exit_fs(p); /* blocking */
2273 bad_fork_cleanup_files:
2274         exit_files(p); /* blocking */
2275 bad_fork_cleanup_semundo:
2276         exit_sem(p);
2277 bad_fork_cleanup_security:
2278         security_task_free(p);
2279 bad_fork_cleanup_audit:
2280         audit_free(p);
2281 bad_fork_cleanup_perf:
2282         perf_event_free_task(p);
2283 bad_fork_cleanup_policy:
2284         lockdep_free_task(p);
2285 #ifdef CONFIG_NUMA
2286         mpol_put(p->mempolicy);
2287 bad_fork_cleanup_threadgroup_lock:
2288 #endif
2289         delayacct_tsk_free(p);
2290 bad_fork_cleanup_count:
2291         atomic_dec(&p->cred->user->processes);
2292         exit_creds(p);
2293 bad_fork_free:
2294         p->state = TASK_DEAD;
2295         put_task_stack(p);
2296         delayed_free_task(p);
2297 fork_out:
2298         spin_lock_irq(&current->sighand->siglock);
2299         hlist_del_init(&delayed.node);
2300         spin_unlock_irq(&current->sighand->siglock);
2301         return ERR_PTR(retval);
2302 }
2303
2304 static inline void init_idle_pids(struct task_struct *idle)
2305 {
2306         enum pid_type type;
2307
2308         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2309                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2310                 init_task_pid(idle, type, &init_struct_pid);
2311         }
2312 }
2313
2314 struct task_struct *fork_idle(int cpu)
2315 {
2316         struct task_struct *task;
2317         task = copy_process(CLONE_VM, 0, 0, NULL, NULL, &init_struct_pid, 0, 0,
2318                             cpu_to_node(cpu));
2319         if (!IS_ERR(task)) {
2320                 init_idle_pids(task);
2321                 init_idle(task, cpu);
2322         }
2323
2324         return task;
2325 }
2326
2327 struct mm_struct *copy_init_mm(void)
2328 {
2329         return dup_mm(NULL, &init_mm);
2330 }
2331
2332 /*
2333  *  Ok, this is the main fork-routine.
2334  *
2335  * It copies the process, and if successful kick-starts
2336  * it and waits for it to finish using the VM if required.
2337  */
2338 long _do_fork(unsigned long clone_flags,
2339               unsigned long stack_start,
2340               unsigned long stack_size,
2341               int __user *parent_tidptr,
2342               int __user *child_tidptr,
2343               unsigned long tls)
2344 {
2345         struct completion vfork;
2346         struct pid *pid;
2347         struct task_struct *p;
2348         int trace = 0;
2349         long nr;
2350
2351         /*
2352          * Determine whether and which event to report to ptracer.  When
2353          * called from kernel_thread or CLONE_UNTRACED is explicitly
2354          * requested, no event is reported; otherwise, report if the event
2355          * for the type of forking is enabled.
2356          */
2357         if (!(clone_flags & CLONE_UNTRACED)) {
2358                 if (clone_flags & CLONE_VFORK)
2359                         trace = PTRACE_EVENT_VFORK;
2360                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2361                         trace = PTRACE_EVENT_CLONE;
2362                 else
2363                         trace = PTRACE_EVENT_FORK;
2364
2365                 if (likely(!ptrace_event_enabled(current, trace)))
2366                         trace = 0;
2367         }
2368
2369         p = copy_process(clone_flags, stack_start, stack_size, parent_tidptr,
2370                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2371         add_latent_entropy();
2372
2373         if (IS_ERR(p))
2374                 return PTR_ERR(p);
2375
2376         /*
2377          * Do this prior waking up the new thread - the thread pointer
2378          * might get invalid after that point, if the thread exits quickly.
2379          */
2380         trace_sched_process_fork(current, p);
2381
2382         pid = get_task_pid(p, PIDTYPE_PID);
2383         nr = pid_vnr(pid);
2384
2385         if (clone_flags & CLONE_PARENT_SETTID)
2386                 put_user(nr, parent_tidptr);
2387
2388         if (clone_flags & CLONE_VFORK) {
2389                 p->vfork_done = &vfork;
2390                 init_completion(&vfork);
2391                 get_task_struct(p);
2392         }
2393
2394         wake_up_new_task(p);
2395
2396         /* forking complete and child started to run, tell ptracer */
2397         if (unlikely(trace))
2398                 ptrace_event_pid(trace, pid);
2399
2400         if (clone_flags & CLONE_VFORK) {
2401                 if (!wait_for_vfork_done(p, &vfork))
2402                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2403         }
2404
2405         put_pid(pid);
2406         return nr;
2407 }
2408
2409 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2410 /* For compatibility with architectures that call do_fork directly rather than
2411  * using the syscall entry points below. */
2412 long do_fork(unsigned long clone_flags,
2413               unsigned long stack_start,
2414               unsigned long stack_size,
2415               int __user *parent_tidptr,
2416               int __user *child_tidptr)
2417 {
2418         return _do_fork(clone_flags, stack_start, stack_size,
2419                         parent_tidptr, child_tidptr, 0);
2420 }
2421 #endif
2422
2423 /*
2424  * Create a kernel thread.
2425  */
2426 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2427 {
2428         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2429                 (unsigned long)arg, NULL, NULL, 0);
2430 }
2431
2432 #ifdef __ARCH_WANT_SYS_FORK
2433 SYSCALL_DEFINE0(fork)
2434 {
2435 #ifdef CONFIG_MMU
2436         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2437 #else
2438         /* can not support in nommu mode */
2439         return -EINVAL;
2440 #endif
2441 }
2442 #endif
2443
2444 #ifdef __ARCH_WANT_SYS_VFORK
2445 SYSCALL_DEFINE0(vfork)
2446 {
2447         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2448                         0, NULL, NULL, 0);
2449 }
2450 #endif
2451
2452 #ifdef __ARCH_WANT_SYS_CLONE
2453 #ifdef CONFIG_CLONE_BACKWARDS
2454 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2455                  int __user *, parent_tidptr,
2456                  unsigned long, tls,
2457                  int __user *, child_tidptr)
2458 #elif defined(CONFIG_CLONE_BACKWARDS2)
2459 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2460                  int __user *, parent_tidptr,
2461                  int __user *, child_tidptr,
2462                  unsigned long, tls)
2463 #elif defined(CONFIG_CLONE_BACKWARDS3)
2464 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2465                 int, stack_size,
2466                 int __user *, parent_tidptr,
2467                 int __user *, child_tidptr,
2468                 unsigned long, tls)
2469 #else
2470 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2471                  int __user *, parent_tidptr,
2472                  int __user *, child_tidptr,
2473                  unsigned long, tls)
2474 #endif
2475 {
2476         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2477 }
2478 #endif
2479
2480 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2481 {
2482         struct task_struct *leader, *parent, *child;
2483         int res;
2484
2485         read_lock(&tasklist_lock);
2486         leader = top = top->group_leader;
2487 down:
2488         for_each_thread(leader, parent) {
2489                 list_for_each_entry(child, &parent->children, sibling) {
2490                         res = visitor(child, data);
2491                         if (res) {
2492                                 if (res < 0)
2493                                         goto out;
2494                                 leader = child;
2495                                 goto down;
2496                         }
2497 up:
2498                         ;
2499                 }
2500         }
2501
2502         if (leader != top) {
2503                 child = leader;
2504                 parent = child->real_parent;
2505                 leader = parent->group_leader;
2506                 goto up;
2507         }
2508 out:
2509         read_unlock(&tasklist_lock);
2510 }
2511
2512 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2513 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2514 #endif
2515
2516 static void sighand_ctor(void *data)
2517 {
2518         struct sighand_struct *sighand = data;
2519
2520         spin_lock_init(&sighand->siglock);
2521         init_waitqueue_head(&sighand->signalfd_wqh);
2522 }
2523
2524 void __init proc_caches_init(void)
2525 {
2526         unsigned int mm_size;
2527
2528         sighand_cachep = kmem_cache_create("sighand_cache",
2529                         sizeof(struct sighand_struct), 0,
2530                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2531                         SLAB_ACCOUNT, sighand_ctor);
2532         signal_cachep = kmem_cache_create("signal_cache",
2533                         sizeof(struct signal_struct), 0,
2534                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2535                         NULL);
2536         files_cachep = kmem_cache_create("files_cache",
2537                         sizeof(struct files_struct), 0,
2538                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2539                         NULL);
2540         fs_cachep = kmem_cache_create("fs_cache",
2541                         sizeof(struct fs_struct), 0,
2542                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2543                         NULL);
2544
2545         /*
2546          * The mm_cpumask is located at the end of mm_struct, and is
2547          * dynamically sized based on the maximum CPU number this system
2548          * can have, taking hotplug into account (nr_cpu_ids).
2549          */
2550         mm_size = sizeof(struct mm_struct) + cpumask_size();
2551
2552         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2553                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2554                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2555                         offsetof(struct mm_struct, saved_auxv),
2556                         sizeof_field(struct mm_struct, saved_auxv),
2557                         NULL);
2558         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2559         mmap_init();
2560         nsproxy_cache_init();
2561 }
2562
2563 /*
2564  * Check constraints on flags passed to the unshare system call.
2565  */
2566 static int check_unshare_flags(unsigned long unshare_flags)
2567 {
2568         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2569                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2570                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2571                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2572                 return -EINVAL;
2573         /*
2574          * Not implemented, but pretend it works if there is nothing
2575          * to unshare.  Note that unsharing the address space or the
2576          * signal handlers also need to unshare the signal queues (aka
2577          * CLONE_THREAD).
2578          */
2579         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2580                 if (!thread_group_empty(current))
2581                         return -EINVAL;
2582         }
2583         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2584                 if (refcount_read(&current->sighand->count) > 1)
2585                         return -EINVAL;
2586         }
2587         if (unshare_flags & CLONE_VM) {
2588                 if (!current_is_single_threaded())
2589                         return -EINVAL;
2590         }
2591
2592         return 0;
2593 }
2594
2595 /*
2596  * Unshare the filesystem structure if it is being shared
2597  */
2598 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2599 {
2600         struct fs_struct *fs = current->fs;
2601
2602         if (!(unshare_flags & CLONE_FS) || !fs)
2603                 return 0;
2604
2605         /* don't need lock here; in the worst case we'll do useless copy */
2606         if (fs->users == 1)
2607                 return 0;
2608
2609         *new_fsp = copy_fs_struct(fs);
2610         if (!*new_fsp)
2611                 return -ENOMEM;
2612
2613         return 0;
2614 }
2615
2616 /*
2617  * Unshare file descriptor table if it is being shared
2618  */
2619 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2620 {
2621         struct files_struct *fd = current->files;
2622         int error = 0;
2623
2624         if ((unshare_flags & CLONE_FILES) &&
2625             (fd && atomic_read(&fd->count) > 1)) {
2626                 *new_fdp = dup_fd(fd, &error);
2627                 if (!*new_fdp)
2628                         return error;
2629         }
2630
2631         return 0;
2632 }
2633
2634 /*
2635  * unshare allows a process to 'unshare' part of the process
2636  * context which was originally shared using clone.  copy_*
2637  * functions used by do_fork() cannot be used here directly
2638  * because they modify an inactive task_struct that is being
2639  * constructed. Here we are modifying the current, active,
2640  * task_struct.
2641  */
2642 int ksys_unshare(unsigned long unshare_flags)
2643 {
2644         struct fs_struct *fs, *new_fs = NULL;
2645         struct files_struct *fd, *new_fd = NULL;
2646         struct cred *new_cred = NULL;
2647         struct nsproxy *new_nsproxy = NULL;
2648         int do_sysvsem = 0;
2649         int err;
2650
2651         /*
2652          * If unsharing a user namespace must also unshare the thread group
2653          * and unshare the filesystem root and working directories.
2654          */
2655         if (unshare_flags & CLONE_NEWUSER)
2656                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2657         /*
2658          * If unsharing vm, must also unshare signal handlers.
2659          */
2660         if (unshare_flags & CLONE_VM)
2661                 unshare_flags |= CLONE_SIGHAND;
2662         /*
2663          * If unsharing a signal handlers, must also unshare the signal queues.
2664          */
2665         if (unshare_flags & CLONE_SIGHAND)
2666                 unshare_flags |= CLONE_THREAD;
2667         /*
2668          * If unsharing namespace, must also unshare filesystem information.
2669          */
2670         if (unshare_flags & CLONE_NEWNS)
2671                 unshare_flags |= CLONE_FS;
2672
2673         err = check_unshare_flags(unshare_flags);
2674         if (err)
2675                 goto bad_unshare_out;
2676         /*
2677          * CLONE_NEWIPC must also detach from the undolist: after switching
2678          * to a new ipc namespace, the semaphore arrays from the old
2679          * namespace are unreachable.
2680          */
2681         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2682                 do_sysvsem = 1;
2683         err = unshare_fs(unshare_flags, &new_fs);
2684         if (err)
2685                 goto bad_unshare_out;
2686         err = unshare_fd(unshare_flags, &new_fd);
2687         if (err)
2688                 goto bad_unshare_cleanup_fs;
2689         err = unshare_userns(unshare_flags, &new_cred);
2690         if (err)
2691                 goto bad_unshare_cleanup_fd;
2692         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2693                                          new_cred, new_fs);
2694         if (err)
2695                 goto bad_unshare_cleanup_cred;
2696
2697         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2698                 if (do_sysvsem) {
2699                         /*
2700                          * CLONE_SYSVSEM is equivalent to sys_exit().
2701                          */
2702                         exit_sem(current);
2703                 }
2704                 if (unshare_flags & CLONE_NEWIPC) {
2705                         /* Orphan segments in old ns (see sem above). */
2706                         exit_shm(current);
2707                         shm_init_task(current);
2708                 }
2709
2710                 if (new_nsproxy)
2711                         switch_task_namespaces(current, new_nsproxy);
2712
2713                 task_lock(current);
2714
2715                 if (new_fs) {
2716                         fs = current->fs;
2717                         spin_lock(&fs->lock);
2718                         current->fs = new_fs;
2719                         if (--fs->users)
2720                                 new_fs = NULL;
2721                         else
2722                                 new_fs = fs;
2723                         spin_unlock(&fs->lock);
2724                 }
2725
2726                 if (new_fd) {
2727                         fd = current->files;
2728                         current->files = new_fd;
2729                         new_fd = fd;
2730                 }
2731
2732                 task_unlock(current);
2733
2734                 if (new_cred) {
2735                         /* Install the new user namespace */
2736                         commit_creds(new_cred);
2737                         new_cred = NULL;
2738                 }
2739         }
2740
2741         perf_event_namespaces(current);
2742
2743 bad_unshare_cleanup_cred:
2744         if (new_cred)
2745                 put_cred(new_cred);
2746 bad_unshare_cleanup_fd:
2747         if (new_fd)
2748                 put_files_struct(new_fd);
2749
2750 bad_unshare_cleanup_fs:
2751         if (new_fs)
2752                 free_fs_struct(new_fs);
2753
2754 bad_unshare_out:
2755         return err;
2756 }
2757
2758 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2759 {
2760         return ksys_unshare(unshare_flags);
2761 }
2762
2763 /*
2764  *      Helper to unshare the files of the current task.
2765  *      We don't want to expose copy_files internals to
2766  *      the exec layer of the kernel.
2767  */
2768
2769 int unshare_files(struct files_struct **displaced)
2770 {
2771         struct task_struct *task = current;
2772         struct files_struct *copy = NULL;
2773         int error;
2774
2775         error = unshare_fd(CLONE_FILES, &copy);
2776         if (error || !copy) {
2777                 *displaced = NULL;
2778                 return error;
2779         }
2780         *displaced = task->files;
2781         task_lock(task);
2782         task->files = copy;
2783         task_unlock(task);
2784         return 0;
2785 }
2786
2787 int sysctl_max_threads(struct ctl_table *table, int write,
2788                        void __user *buffer, size_t *lenp, loff_t *ppos)
2789 {
2790         struct ctl_table t;
2791         int ret;
2792         int threads = max_threads;
2793         int min = MIN_THREADS;
2794         int max = MAX_THREADS;
2795
2796         t = *table;
2797         t.data = &threads;
2798         t.extra1 = &min;
2799         t.extra2 = &max;
2800
2801         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2802         if (ret || !write)
2803                 return ret;
2804
2805         set_max_threads(threads);
2806
2807         return 0;
2808 }