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/fs.h>
  44 #include <linux/mm.h>
  45 #include <linux/vmacache.h>
  46 #include <linux/nsproxy.h>
  47 #include <linux/capability.h>
  48 #include <linux/cpu.h>
  49 #include <linux/cgroup.h>
  50 #include <linux/security.h>
  51 #include <linux/hugetlb.h>
  52 #include <linux/seccomp.h>
  53 #include <linux/swap.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/jiffies.h>
  56 #include <linux/futex.h>
  57 #include <linux/compat.h>
  58 #include <linux/kthread.h>
  59 #include <linux/task_io_accounting_ops.h>
  60 #include <linux/rcupdate.h>
  61 #include <linux/ptrace.h>
  62 #include <linux/mount.h>
  63 #include <linux/audit.h>
  64 #include <linux/memcontrol.h>
  65 #include <linux/ftrace.h>
  66 #include <linux/proc_fs.h>
  67 #include <linux/profile.h>
  68 #include <linux/rmap.h>
  69 #include <linux/ksm.h>
  70 #include <linux/acct.h>
  71 #include <linux/userfaultfd_k.h>
  72 #include <linux/tsacct_kern.h>
  73 #include <linux/cn_proc.h>
  74 #include <linux/freezer.h>
  75 #include <linux/delayacct.h>
  76 #include <linux/taskstats_kern.h>
  77 #include <linux/random.h>
  78 #include <linux/tty.h>
  79 #include <linux/blkdev.h>
  80 #include <linux/fs_struct.h>
  81 #include <linux/magic.h>
  82 #include <linux/perf_event.h>
  83 #include <linux/posix-timers.h>
  84 #include <linux/user-return-notifier.h>
  85 #include <linux/oom.h>
  86 #include <linux/khugepaged.h>
  87 #include <linux/signalfd.h>
  88 #include <linux/uprobes.h>
  89 #include <linux/aio.h>
  90 #include <linux/compiler.h>
  91 #include <linux/sysctl.h>
  92 #include <linux/kcov.h>
  93 #include <linux/livepatch.h>
  94 #include <linux/thread_info.h>
  95 #include <linux/stackleak.h>
  96 #include <linux/kasan.h>
  97 #include <linux/scs.h>
  98
  99 #include <asm/pgtable.h>
 100 #include <asm/pgalloc.h>
 101 #include <linux/uaccess.h>
 102 #include <asm/mmu_context.h>
 103 #include <asm/cacheflush.h>
 104 #include <asm/tlbflush.h>
 105
 106 #include <trace/events/sched.h>
 107
 108 #define CREATE_TRACE_POINTS
 109 #include <trace/events/task.h>
 110
 111 /*
 112  * Minimum number of threads to boot the kernel
 113  */
 114 #define MIN_THREADS 20
 115
 116 /*
 117  * Maximum number of threads
 118  */
 119 #define MAX_THREADS FUTEX_TID_MASK
 120
 121 /*
 122  * Protected counters by write_lock_irq(&tasklist_lock)
 123  */
 124 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 125 int nr_threads;                 /* The idle threads do not count.. */
 126
 127 static int max_threads;         /* tunable limit on nr_threads */
 128
 129 #define NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 130
 131 static const char * const resident_page_types[] = {
 132         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 133         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 134         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 135         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 136 };
 137
 138 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 139
 140 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 141
 142 #ifdef CONFIG_PROVE_RCU
 143 int lockdep_tasklist_lock_is_held(void)
 144 {
 145         return lockdep_is_held(&tasklist_lock);
 146 }
 147 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 148 #endif /* #ifdef CONFIG_PROVE_RCU */
 149
 150 int nr_processes(void)
 151 {
 152         int cpu;
 153         int total = 0;
 154
 155         for_each_possible_cpu(cpu)
 156                 total += per_cpu(process_counts, cpu);
 157
 158         return total;
 159 }
 160
 161 void __weak arch_release_task_struct(struct task_struct *tsk)
 162 {
 163 }
 164
 165 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 166 static struct kmem_cache *task_struct_cachep;
 167
 168 static inline struct task_struct *alloc_task_struct_node(int node)
 169 {
 170         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 171 }
 172
 173 static inline void free_task_struct(struct task_struct *tsk)
 174 {
 175         kmem_cache_free(task_struct_cachep, tsk);
 176 }
 177 #endif
 178
 179 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 180
 181 /*
 182  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 183  * kmemcache based allocator.
 184  */
 185 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 186
 187 #ifdef CONFIG_VMAP_STACK
 188 /*
 189  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 190  * flush.  Try to minimize the number of calls by caching stacks.
 191  */
 192 #define NR_CACHED_STACKS 2
 193 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 194
 195 static int free_vm_stack_cache(unsigned int cpu)
 196 {
 197         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 198         int i;
 199
 200         for (i = 0; i < NR_CACHED_STACKS; i++) {
 201                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 202
 203                 if (!vm_stack)
 204                         continue;
 205
 206                 vfree(vm_stack->addr);
 207                 cached_vm_stacks[i] = NULL;
 208         }
 209
 210         return 0;
 211 }
 212 #endif
 213
 214 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 215 {
 216 #ifdef CONFIG_VMAP_STACK
 217         void *stack;
 218         int i;
 219
 220         for (i = 0; i < NR_CACHED_STACKS; i++) {
 221                 struct vm_struct *s;
 222
 223                 s = this_cpu_xchg(cached_stacks[i], NULL);
 224
 225                 if (!s)
 226                         continue;
 227
 228                 /* Clear the KASAN shadow of the stack. */
 229                 kasan_unpoison_shadow(s->addr, THREAD_SIZE);
 230
 231                 /* Clear stale pointers from reused stack. */
 232                 memset(s->addr, 0, THREAD_SIZE);
 233
 234                 tsk->stack_vm_area = s;
 235                 tsk->stack = s->addr;
 236                 return s->addr;
 237         }
 238
 239         /*
 240          * Allocated stacks are cached and later reused by new threads,
 241          * so memcg accounting is performed manually on assigning/releasing
 242          * stacks to tasks. Drop __GFP_ACCOUNT.
 243          */
 244         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 245                                      VMALLOC_START, VMALLOC_END,
 246                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 247                                      PAGE_KERNEL,
 248                                      0, node, __builtin_return_address(0));
 249
 250         /*
 251          * We can't call find_vm_area() in interrupt context, and
 252          * free_thread_stack() can be called in interrupt context,
 253          * so cache the vm_struct.
 254          */
 255         if (stack) {
 256                 tsk->stack_vm_area = find_vm_area(stack);
 257                 tsk->stack = stack;
 258         }
 259         return stack;
 260 #else
 261         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 262                                              THREAD_SIZE_ORDER);
 263
 264         if (likely(page)) {
 265                 tsk->stack = page_address(page);
 266                 return tsk->stack;
 267         }
 268         return NULL;
 269 #endif
 270 }
 271
 272 static inline void free_thread_stack(struct task_struct *tsk)
 273 {
 274 #ifdef CONFIG_VMAP_STACK
 275         struct vm_struct *vm = task_stack_vm_area(tsk);
 276
 277         if (vm) {
 278                 int i;
 279
 280                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 281                         mod_memcg_page_state(vm->pages[i],
 282                                              MEMCG_KERNEL_STACK_KB,
 283                                              -(int)(PAGE_SIZE / 1024));
 284
 285                         memcg_kmem_uncharge_page(vm->pages[i], 0);
 286                 }
 287
 288                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 289                         if (this_cpu_cmpxchg(cached_stacks[i],
 290                                         NULL, tsk->stack_vm_area) != NULL)
 291                                 continue;
 292
 293                         return;
 294                 }
 295
 296                 vfree_atomic(tsk->stack);
 297                 return;
 298         }
 299 #endif
 300
 301         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 302 }
 303 # else
 304 static struct kmem_cache *thread_stack_cache;
 305
 306 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 307                                                   int node)
 308 {
 309         unsigned long *stack;
 310         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 311         tsk->stack = stack;
 312         return stack;
 313 }
 314
 315 static void free_thread_stack(struct task_struct *tsk)
 316 {
 317         kmem_cache_free(thread_stack_cache, tsk->stack);
 318 }
 319
 320 void thread_stack_cache_init(void)
 321 {
 322         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 323                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 324                                         THREAD_SIZE, NULL);
 325         BUG_ON(thread_stack_cache == NULL);
 326 }
 327 # endif
 328 #endif
 329
 330 /* SLAB cache for signal_struct structures (tsk->signal) */
 331 static struct kmem_cache *signal_cachep;
 332
 333 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 334 struct kmem_cache *sighand_cachep;
 335
 336 /* SLAB cache for files_struct structures (tsk->files) */
 337 struct kmem_cache *files_cachep;
 338
 339 /* SLAB cache for fs_struct structures (tsk->fs) */
 340 struct kmem_cache *fs_cachep;
 341
 342 /* SLAB cache for vm_area_struct structures */
 343 static struct kmem_cache *vm_area_cachep;
 344
 345 /* SLAB cache for mm_struct structures (tsk->mm) */
 346 static struct kmem_cache *mm_cachep;
 347
 348 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 349 {
 350         struct vm_area_struct *vma;
 351
 352         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 353         if (vma)
 354                 vma_init(vma, mm);
 355         return vma;
 356 }
 357
 358 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 359 {
 360         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 361
 362         if (new) {
 363                 *new = *orig;
 364                 INIT_LIST_HEAD(&new->anon_vma_chain);
 365                 new->vm_next = new->vm_prev = NULL;
 366         }
 367         return new;
 368 }
 369
 370 void vm_area_free(struct vm_area_struct *vma)
 371 {
 372         kmem_cache_free(vm_area_cachep, vma);
 373 }
 374
 375 static void account_kernel_stack(struct task_struct *tsk, int account)
 376 {
 377         void *stack = task_stack_page(tsk);
 378         struct vm_struct *vm = task_stack_vm_area(tsk);
 379
 380         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 381
 382         if (vm) {
 383                 int i;
 384
 385                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 386
 387                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 388                         mod_zone_page_state(page_zone(vm->pages[i]),
 389                                             NR_KERNEL_STACK_KB,
 390                                             PAGE_SIZE / 1024 * account);
 391                 }
 392         } else {
 393                 /*
 394                  * All stack pages are in the same zone and belong to the
 395                  * same memcg.
 396                  */
 397                 struct page *first_page = virt_to_page(stack);
 398
 399                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 400                                     THREAD_SIZE / 1024 * account);
 401
 402                 mod_memcg_obj_state(stack, MEMCG_KERNEL_STACK_KB,
 403                                     account * (THREAD_SIZE / 1024));
 404         }
 405 }
 406
 407 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 408 {
 409 #ifdef CONFIG_VMAP_STACK
 410         struct vm_struct *vm = task_stack_vm_area(tsk);
 411         int ret;
 412
 413         if (vm) {
 414                 int i;
 415
 416                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 417                         /*
 418                          * If memcg_kmem_charge_page() fails, page->mem_cgroup
 419                          * pointer is NULL, and both memcg_kmem_uncharge_page()
 420                          * and mod_memcg_page_state() in free_thread_stack()
 421                          * will ignore this page. So it's safe.
 422                          */
 423                         ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
 424                                                      0);
 425                         if (ret)
 426                                 return ret;
 427
 428                         mod_memcg_page_state(vm->pages[i],
 429                                              MEMCG_KERNEL_STACK_KB,
 430                                              PAGE_SIZE / 1024);
 431                 }
 432         }
 433 #endif
 434         return 0;
 435 }
 436
 437 static void release_task_stack(struct task_struct *tsk)
 438 {
 439         if (WARN_ON(tsk->state != TASK_DEAD))
 440                 return;  /* Better to leak the stack than to free prematurely */
 441
 442         account_kernel_stack(tsk, -1);
 443         free_thread_stack(tsk);
 444         tsk->stack = NULL;
 445 #ifdef CONFIG_VMAP_STACK
 446         tsk->stack_vm_area = NULL;
 447 #endif
 448 }
 449
 450 #ifdef CONFIG_THREAD_INFO_IN_TASK
 451 void put_task_stack(struct task_struct *tsk)
 452 {
 453         if (refcount_dec_and_test(&tsk->stack_refcount))
 454                 release_task_stack(tsk);
 455 }
 456 #endif
 457
 458 void free_task(struct task_struct *tsk)
 459 {
 460         scs_release(tsk);
 461
 462 #ifndef CONFIG_THREAD_INFO_IN_TASK
 463         /*
 464          * The task is finally done with both the stack and thread_info,
 465          * so free both.
 466          */
 467         release_task_stack(tsk);
 468 #else
 469         /*
 470          * If the task had a separate stack allocation, it should be gone
 471          * by now.
 472          */
 473         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 474 #endif
 475         rt_mutex_debug_task_free(tsk);
 476         ftrace_graph_exit_task(tsk);
 477         put_seccomp_filter(tsk);
 478         arch_release_task_struct(tsk);
 479         if (tsk->flags & PF_KTHREAD)
 480                 free_kthread_struct(tsk);
 481         free_task_struct(tsk);
 482 }
 483 EXPORT_SYMBOL(free_task);
 484
 485 #ifdef CONFIG_MMU
 486 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 487                                         struct mm_struct *oldmm)
 488 {
 489         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 490         struct rb_node **rb_link, *rb_parent;
 491         int retval;
 492         unsigned long charge;
 493         LIST_HEAD(uf);
 494
 495         uprobe_start_dup_mmap();
 496         if (down_write_killable(&oldmm->mmap_sem)) {
 497                 retval = -EINTR;
 498                 goto fail_uprobe_end;
 499         }
 500         flush_cache_dup_mm(oldmm);
 501         uprobe_dup_mmap(oldmm, mm);
 502         /*
 503          * Not linked in yet - no deadlock potential:
 504          */
 505         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 506
 507         /* No ordering required: file already has been exposed. */
 508         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 509
 510         mm->total_vm = oldmm->total_vm;
 511         mm->data_vm = oldmm->data_vm;
 512         mm->exec_vm = oldmm->exec_vm;
 513         mm->stack_vm = oldmm->stack_vm;
 514
 515         rb_link = &mm->mm_rb.rb_node;
 516         rb_parent = NULL;
 517         pprev = &mm->mmap;
 518         retval = ksm_fork(mm, oldmm);
 519         if (retval)
 520                 goto out;
 521         retval = khugepaged_fork(mm, oldmm);
 522         if (retval)
 523                 goto out;
 524
 525         prev = NULL;
 526         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 527                 struct file *file;
 528
 529                 if (mpnt->vm_flags & VM_DONTCOPY) {
 530                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 531                         continue;
 532                 }
 533                 charge = 0;
 534                 /*
 535                  * Don't duplicate many vmas if we've been oom-killed (for
 536                  * example)
 537                  */
 538                 if (fatal_signal_pending(current)) {
 539                         retval = -EINTR;
 540                         goto out;
 541                 }
 542                 if (mpnt->vm_flags & VM_ACCOUNT) {
 543                         unsigned long len = vma_pages(mpnt);
 544
 545                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 546                                 goto fail_nomem;
 547                         charge = len;
 548                 }
 549                 tmp = vm_area_dup(mpnt);
 550                 if (!tmp)
 551                         goto fail_nomem;
 552                 retval = vma_dup_policy(mpnt, tmp);
 553                 if (retval)
 554                         goto fail_nomem_policy;
 555                 tmp->vm_mm = mm;
 556                 retval = dup_userfaultfd(tmp, &uf);
 557                 if (retval)
 558                         goto fail_nomem_anon_vma_fork;
 559                 if (tmp->vm_flags & VM_WIPEONFORK) {
 560                         /*
 561                          * VM_WIPEONFORK gets a clean slate in the child.
 562                          * Don't prepare anon_vma until fault since we don't
 563                          * copy page for current vma.
 564                          */
 565                         tmp->anon_vma = NULL;
 566                 } else if (anon_vma_fork(tmp, mpnt))
 567                         goto fail_nomem_anon_vma_fork;
 568                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 569                 file = tmp->vm_file;
 570                 if (file) {
 571                         struct inode *inode = file_inode(file);
 572                         struct address_space *mapping = file->f_mapping;
 573
 574                         get_file(file);
 575                         if (tmp->vm_flags & VM_DENYWRITE)
 576                                 atomic_dec(&inode->i_writecount);
 577                         i_mmap_lock_write(mapping);
 578                         if (tmp->vm_flags & VM_SHARED)
 579                                 atomic_inc(&mapping->i_mmap_writable);
 580                         flush_dcache_mmap_lock(mapping);
 581                         /* insert tmp into the share list, just after mpnt */
 582                         vma_interval_tree_insert_after(tmp, mpnt,
 583                                         &mapping->i_mmap);
 584                         flush_dcache_mmap_unlock(mapping);
 585                         i_mmap_unlock_write(mapping);
 586                 }
 587
 588                 /*
 589                  * Clear hugetlb-related page reserves for children. This only
 590                  * affects MAP_PRIVATE mappings. Faults generated by the child
 591                  * are not guaranteed to succeed, even if read-only
 592                  */
 593                 if (is_vm_hugetlb_page(tmp))
 594                         reset_vma_resv_huge_pages(tmp);
 595
 596                 /*
 597                  * Link in the new vma and copy the page table entries.
 598                  */
 599                 *pprev = tmp;
 600                 pprev = &tmp->vm_next;
 601                 tmp->vm_prev = prev;
 602                 prev = tmp;
 603
 604                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 605                 rb_link = &tmp->vm_rb.rb_right;
 606                 rb_parent = &tmp->vm_rb;
 607
 608                 mm->map_count++;
 609                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 610                         retval = copy_page_range(mm, oldmm, mpnt);
 611
 612                 if (tmp->vm_ops && tmp->vm_ops->open)
 613                         tmp->vm_ops->open(tmp);
 614
 615                 if (retval)
 616                         goto out;
 617         }
 618         /* a new mm has just been created */
 619         retval = arch_dup_mmap(oldmm, mm);
 620 out:
 621         up_write(&mm->mmap_sem);
 622         flush_tlb_mm(oldmm);
 623         up_write(&oldmm->mmap_sem);
 624         dup_userfaultfd_complete(&uf);
 625 fail_uprobe_end:
 626         uprobe_end_dup_mmap();
 627         return retval;
 628 fail_nomem_anon_vma_fork:
 629         mpol_put(vma_policy(tmp));
 630 fail_nomem_policy:
 631         vm_area_free(tmp);
 632 fail_nomem:
 633         retval = -ENOMEM;
 634         vm_unacct_memory(charge);
 635         goto out;
 636 }
 637
 638 static inline int mm_alloc_pgd(struct mm_struct *mm)
 639 {
 640         mm->pgd = pgd_alloc(mm);
 641         if (unlikely(!mm->pgd))
 642                 return -ENOMEM;
 643         return 0;
 644 }
 645
 646 static inline void mm_free_pgd(struct mm_struct *mm)
 647 {
 648         pgd_free(mm, mm->pgd);
 649 }
 650 #else
 651 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 652 {
 653         down_write(&oldmm->mmap_sem);
 654         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 655         up_write(&oldmm->mmap_sem);
 656         return 0;
 657 }
 658 #define mm_alloc_pgd(mm)        (0)
 659 #define mm_free_pgd(mm)
 660 #endif /* CONFIG_MMU */
 661
 662 static void check_mm(struct mm_struct *mm)
 663 {
 664         int i;
 665
 666         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 667                          "Please make sure 'struct resident_page_types[]' is updated as well");
 668
 669         for (i = 0; i < NR_MM_COUNTERS; i++) {
 670                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 671
 672                 if (unlikely(x))
 673                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 674                                  mm, resident_page_types[i], x);
 675         }
 676
 677         if (mm_pgtables_bytes(mm))
 678                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 679                                 mm_pgtables_bytes(mm));
 680
 681 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 682         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 683 #endif
 684 }
 685
 686 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 687 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 688
 689 /*
 690  * Called when the last reference to the mm
 691  * is dropped: either by a lazy thread or by
 692  * mmput. Free the page directory and the mm.
 693  */
 694 void __mmdrop(struct mm_struct *mm)
 695 {
 696         BUG_ON(mm == &init_mm);
 697         WARN_ON_ONCE(mm == current->mm);
 698         WARN_ON_ONCE(mm == current->active_mm);
 699         mm_free_pgd(mm);
 700         destroy_context(mm);
 701         mmu_notifier_subscriptions_destroy(mm);
 702         check_mm(mm);
 703         put_user_ns(mm->user_ns);
 704         free_mm(mm);
 705 }
 706 EXPORT_SYMBOL_GPL(__mmdrop);
 707
 708 static void mmdrop_async_fn(struct work_struct *work)
 709 {
 710         struct mm_struct *mm;
 711
 712         mm = container_of(work, struct mm_struct, async_put_work);
 713         __mmdrop(mm);
 714 }
 715
 716 static void mmdrop_async(struct mm_struct *mm)
 717 {
 718         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 719                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 720                 schedule_work(&mm->async_put_work);
 721         }
 722 }
 723
 724 static inline void free_signal_struct(struct signal_struct *sig)
 725 {
 726         taskstats_tgid_free(sig);
 727         sched_autogroup_exit(sig);
 728         /*
 729          * __mmdrop is not safe to call from softirq context on x86 due to
 730          * pgd_dtor so postpone it to the async context
 731          */
 732         if (sig->oom_mm)
 733                 mmdrop_async(sig->oom_mm);
 734         kmem_cache_free(signal_cachep, sig);
 735 }
 736
 737 static inline void put_signal_struct(struct signal_struct *sig)
 738 {
 739         if (refcount_dec_and_test(&sig->sigcnt))
 740                 free_signal_struct(sig);
 741 }
 742
 743 void __put_task_struct(struct task_struct *tsk)
 744 {
 745         WARN_ON(!tsk->exit_state);
 746         WARN_ON(refcount_read(&tsk->usage));
 747         WARN_ON(tsk == current);
 748
 749         cgroup_free(tsk);
 750         task_numa_free(tsk, true);
 751         security_task_free(tsk);
 752         exit_creds(tsk);
 753         delayacct_tsk_free(tsk);
 754         put_signal_struct(tsk->signal);
 755
 756         if (!profile_handoff_task(tsk))
 757                 free_task(tsk);
 758 }
 759 EXPORT_SYMBOL_GPL(__put_task_struct);
 760
 761 void __init __weak arch_task_cache_init(void) { }
 762
 763 /*
 764  * set_max_threads
 765  */
 766 static void set_max_threads(unsigned int max_threads_suggested)
 767 {
 768         u64 threads;
 769         unsigned long nr_pages = totalram_pages();
 770
 771         /*
 772          * The number of threads shall be limited such that the thread
 773          * structures may only consume a small part of the available memory.
 774          */
 775         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 776                 threads = MAX_THREADS;
 777         else
 778                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 779                                     (u64) THREAD_SIZE * 8UL);
 780
 781         if (threads > max_threads_suggested)
 782                 threads = max_threads_suggested;
 783
 784         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 785 }
 786
 787 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 788 /* Initialized by the architecture: */
 789 int arch_task_struct_size __read_mostly;
 790 #endif
 791
 792 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 793 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 794 {
 795         /* Fetch thread_struct whitelist for the architecture. */
 796         arch_thread_struct_whitelist(offset, size);
 797
 798         /*
 799          * Handle zero-sized whitelist or empty thread_struct, otherwise
 800          * adjust offset to position of thread_struct in task_struct.
 801          */
 802         if (unlikely(*size == 0))
 803                 *offset = 0;
 804         else
 805                 *offset += offsetof(struct task_struct, thread);
 806 }
 807 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 808
 809 void __init fork_init(void)
 810 {
 811         int i;
 812 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 813 #ifndef ARCH_MIN_TASKALIGN
 814 #define ARCH_MIN_TASKALIGN      0
 815 #endif
 816         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 817         unsigned long useroffset, usersize;
 818
 819         /* create a slab on which task_structs can be allocated */
 820         task_struct_whitelist(&useroffset, &usersize);
 821         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 822                         arch_task_struct_size, align,
 823                         SLAB_PANIC|SLAB_ACCOUNT,
 824                         useroffset, usersize, NULL);
 825 #endif
 826
 827         /* do the arch specific task caches init */
 828         arch_task_cache_init();
 829
 830         set_max_threads(MAX_THREADS);
 831
 832         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 833         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 834         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 835                 init_task.signal->rlim[RLIMIT_NPROC];
 836
 837         for (i = 0; i < UCOUNT_COUNTS; i++) {
 838                 init_user_ns.ucount_max[i] = max_threads/2;
 839         }
 840
 841 #ifdef CONFIG_VMAP_STACK
 842         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 843                           NULL, free_vm_stack_cache);
 844 #endif
 845
 846         scs_init();
 847
 848         lockdep_init_task(&init_task);
 849         uprobes_init();
 850 }
 851
 852 int __weak arch_dup_task_struct(struct task_struct *dst,
 853                                                struct task_struct *src)
 854 {
 855         *dst = *src;
 856         return 0;
 857 }
 858
 859 void set_task_stack_end_magic(struct task_struct *tsk)
 860 {
 861         unsigned long *stackend;
 862
 863         stackend = end_of_stack(tsk);
 864         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 865 }
 866
 867 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 868 {
 869         struct task_struct *tsk;
 870         unsigned long *stack;
 871         struct vm_struct *stack_vm_area __maybe_unused;
 872         int err;
 873
 874         if (node == NUMA_NO_NODE)
 875                 node = tsk_fork_get_node(orig);
 876         tsk = alloc_task_struct_node(node);
 877         if (!tsk)
 878                 return NULL;
 879
 880         stack = alloc_thread_stack_node(tsk, node);
 881         if (!stack)
 882                 goto free_tsk;
 883
 884         if (memcg_charge_kernel_stack(tsk))
 885                 goto free_stack;
 886
 887         stack_vm_area = task_stack_vm_area(tsk);
 888
 889         err = arch_dup_task_struct(tsk, orig);
 890
 891         /*
 892          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 893          * sure they're properly initialized before using any stack-related
 894          * functions again.
 895          */
 896         tsk->stack = stack;
 897 #ifdef CONFIG_VMAP_STACK
 898         tsk->stack_vm_area = stack_vm_area;
 899 #endif
 900 #ifdef CONFIG_THREAD_INFO_IN_TASK
 901         refcount_set(&tsk->stack_refcount, 1);
 902 #endif
 903
 904         if (err)
 905                 goto free_stack;
 906
 907         err = scs_prepare(tsk, node);
 908         if (err)
 909                 goto free_stack;
 910
 911 #ifdef CONFIG_SECCOMP
 912         /*
 913          * We must handle setting up seccomp filters once we're under
 914          * the sighand lock in case orig has changed between now and
 915          * then. Until then, filter must be NULL to avoid messing up
 916          * the usage counts on the error path calling free_task.
 917          */
 918         tsk->seccomp.filter = NULL;
 919 #endif
 920
 921         setup_thread_stack(tsk, orig);
 922         clear_user_return_notifier(tsk);
 923         clear_tsk_need_resched(tsk);
 924         set_task_stack_end_magic(tsk);
 925
 926 #ifdef CONFIG_STACKPROTECTOR
 927         tsk->stack_canary = get_random_canary();
 928 #endif
 929         if (orig->cpus_ptr == &orig->cpus_mask)
 930                 tsk->cpus_ptr = &tsk->cpus_mask;
 931
 932         /*
 933          * One for the user space visible state that goes away when reaped.
 934          * One for the scheduler.
 935          */
 936         refcount_set(&tsk->rcu_users, 2);
 937         /* One for the rcu users */
 938         refcount_set(&tsk->usage, 1);
 939 #ifdef CONFIG_BLK_DEV_IO_TRACE
 940         tsk->btrace_seq = 0;
 941 #endif
 942         tsk->splice_pipe = NULL;
 943         tsk->task_frag.page = NULL;
 944         tsk->wake_q.next = NULL;
 945
 946         account_kernel_stack(tsk, 1);
 947
 948         kcov_task_init(tsk);
 949
 950 #ifdef CONFIG_FAULT_INJECTION
 951         tsk->fail_nth = 0;
 952 #endif
 953
 954 #ifdef CONFIG_BLK_CGROUP
 955         tsk->throttle_queue = NULL;
 956         tsk->use_memdelay = 0;
 957 #endif
 958
 959 #ifdef CONFIG_MEMCG
 960         tsk->active_memcg = NULL;
 961 #endif
 962         return tsk;
 963
 964 free_stack:
 965         free_thread_stack(tsk);
 966 free_tsk:
 967         free_task_struct(tsk);
 968         return NULL;
 969 }
 970
 971 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 972
 973 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 974
 975 static int __init coredump_filter_setup(char *s)
 976 {
 977         default_dump_filter =
 978                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 979                 MMF_DUMP_FILTER_MASK;
 980         return 1;
 981 }
 982
 983 __setup("coredump_filter=", coredump_filter_setup);
 984
 985 #include <linux/init_task.h>
 986
 987 static void mm_init_aio(struct mm_struct *mm)
 988 {
 989 #ifdef CONFIG_AIO
 990         spin_lock_init(&mm->ioctx_lock);
 991         mm->ioctx_table = NULL;
 992 #endif
 993 }
 994
 995 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 996                                            struct task_struct *p)
 997 {
 998 #ifdef CONFIG_MEMCG
 999         if (mm->owner == p)
1000                 WRITE_ONCE(mm->owner, NULL);
1001 #endif
1002 }
1003
1004 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1005 {
1006 #ifdef CONFIG_MEMCG
1007         mm->owner = p;
1008 #endif
1009 }
1010
1011 static void mm_init_uprobes_state(struct mm_struct *mm)
1012 {
1013 #ifdef CONFIG_UPROBES
1014         mm->uprobes_state.xol_area = NULL;
1015 #endif
1016 }
1017
1018 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1019         struct user_namespace *user_ns)
1020 {
1021         mm->mmap = NULL;
1022         mm->mm_rb = RB_ROOT;
1023         mm->vmacache_seqnum = 0;
1024         atomic_set(&mm->mm_users, 1);
1025         atomic_set(&mm->mm_count, 1);
1026         init_rwsem(&mm->mmap_sem);
1027         INIT_LIST_HEAD(&mm->mmlist);
1028         mm->core_state = NULL;
1029         mm_pgtables_bytes_init(mm);
1030         mm->map_count = 0;
1031         mm->locked_vm = 0;
1032         atomic64_set(&mm->pinned_vm, 0);
1033         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1034         spin_lock_init(&mm->page_table_lock);
1035         spin_lock_init(&mm->arg_lock);
1036         mm_init_cpumask(mm);
1037         mm_init_aio(mm);
1038         mm_init_owner(mm, p);
1039         RCU_INIT_POINTER(mm->exe_file, NULL);
1040         mmu_notifier_subscriptions_init(mm);
1041         init_tlb_flush_pending(mm);
1042 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1043         mm->pmd_huge_pte = NULL;
1044 #endif
1045         mm_init_uprobes_state(mm);
1046
1047         if (current->mm) {
1048                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1049                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1050         } else {
1051                 mm->flags = default_dump_filter;
1052                 mm->def_flags = 0;
1053         }
1054
1055         if (mm_alloc_pgd(mm))
1056                 goto fail_nopgd;
1057
1058         if (init_new_context(p, mm))
1059                 goto fail_nocontext;
1060
1061         mm->user_ns = get_user_ns(user_ns);
1062         return mm;
1063
1064 fail_nocontext:
1065         mm_free_pgd(mm);
1066 fail_nopgd:
1067         free_mm(mm);
1068         return NULL;
1069 }
1070
1071 /*
1072  * Allocate and initialize an mm_struct.
1073  */
1074 struct mm_struct *mm_alloc(void)
1075 {
1076         struct mm_struct *mm;
1077
1078         mm = allocate_mm();
1079         if (!mm)
1080                 return NULL;
1081
1082         memset(mm, 0, sizeof(*mm));
1083         return mm_init(mm, current, current_user_ns());
1084 }
1085
1086 static inline void __mmput(struct mm_struct *mm)
1087 {
1088         VM_BUG_ON(atomic_read(&mm->mm_users));
1089
1090         uprobe_clear_state(mm);
1091         exit_aio(mm);
1092         ksm_exit(mm);
1093         khugepaged_exit(mm); /* must run before exit_mmap */
1094         exit_mmap(mm);
1095         mm_put_huge_zero_page(mm);
1096         set_mm_exe_file(mm, NULL);
1097         if (!list_empty(&mm->mmlist)) {
1098                 spin_lock(&mmlist_lock);
1099                 list_del(&mm->mmlist);
1100                 spin_unlock(&mmlist_lock);
1101         }
1102         if (mm->binfmt)
1103                 module_put(mm->binfmt->module);
1104         mmdrop(mm);
1105 }
1106
1107 /*
1108  * Decrement the use count and release all resources for an mm.
1109  */
1110 void mmput(struct mm_struct *mm)
1111 {
1112         might_sleep();
1113
1114         if (atomic_dec_and_test(&mm->mm_users))
1115                 __mmput(mm);
1116 }
1117 EXPORT_SYMBOL_GPL(mmput);
1118
1119 #ifdef CONFIG_MMU
1120 static void mmput_async_fn(struct work_struct *work)
1121 {
1122         struct mm_struct *mm = container_of(work, struct mm_struct,
1123                                             async_put_work);
1124
1125         __mmput(mm);
1126 }
1127
1128 void mmput_async(struct mm_struct *mm)
1129 {
1130         if (atomic_dec_and_test(&mm->mm_users)) {
1131                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1132                 schedule_work(&mm->async_put_work);
1133         }
1134 }
1135 #endif
1136
1137 /**
1138  * set_mm_exe_file - change a reference to the mm's executable file
1139  *
1140  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1141  *
1142  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1143  * invocations: in mmput() nobody alive left, in execve task is single
1144  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1145  * mm->exe_file, but does so without using set_mm_exe_file() in order
1146  * to do avoid the need for any locks.
1147  */
1148 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1149 {
1150         struct file *old_exe_file;
1151
1152         /*
1153          * It is safe to dereference the exe_file without RCU as
1154          * this function is only called if nobody else can access
1155          * this mm -- see comment above for justification.
1156          */
1157         old_exe_file = rcu_dereference_raw(mm->exe_file);
1158
1159         if (new_exe_file)
1160                 get_file(new_exe_file);
1161         rcu_assign_pointer(mm->exe_file, new_exe_file);
1162         if (old_exe_file)
1163                 fput(old_exe_file);
1164 }
1165
1166 /**
1167  * get_mm_exe_file - acquire a reference to the mm's executable file
1168  *
1169  * Returns %NULL if mm has no associated executable file.
1170  * User must release file via fput().
1171  */
1172 struct file *get_mm_exe_file(struct mm_struct *mm)
1173 {
1174         struct file *exe_file;
1175
1176         rcu_read_lock();
1177         exe_file = rcu_dereference(mm->exe_file);
1178         if (exe_file && !get_file_rcu(exe_file))
1179                 exe_file = NULL;
1180         rcu_read_unlock();
1181         return exe_file;
1182 }
1183 EXPORT_SYMBOL(get_mm_exe_file);
1184
1185 /**
1186  * get_task_exe_file - acquire a reference to the task's executable file
1187  *
1188  * Returns %NULL if task's mm (if any) has no associated executable file or
1189  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1190  * User must release file via fput().
1191  */
1192 struct file *get_task_exe_file(struct task_struct *task)
1193 {
1194         struct file *exe_file = NULL;
1195         struct mm_struct *mm;
1196
1197         task_lock(task);
1198         mm = task->mm;
1199         if (mm) {
1200                 if (!(task->flags & PF_KTHREAD))
1201                         exe_file = get_mm_exe_file(mm);
1202         }
1203         task_unlock(task);
1204         return exe_file;
1205 }
1206 EXPORT_SYMBOL(get_task_exe_file);
1207
1208 /**
1209  * get_task_mm - acquire a reference to the task's mm
1210  *
1211  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1212  * this kernel workthread has transiently adopted a user mm with use_mm,
1213  * to do its AIO) is not set and if so returns a reference to it, after
1214  * bumping up the use count.  User must release the mm via mmput()
1215  * after use.  Typically used by /proc and ptrace.
1216  */
1217 struct mm_struct *get_task_mm(struct task_struct *task)
1218 {
1219         struct mm_struct *mm;
1220
1221         task_lock(task);
1222         mm = task->mm;
1223         if (mm) {
1224                 if (task->flags & PF_KTHREAD)
1225                         mm = NULL;
1226                 else
1227                         mmget(mm);
1228         }
1229         task_unlock(task);
1230         return mm;
1231 }
1232 EXPORT_SYMBOL_GPL(get_task_mm);
1233
1234 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1235 {
1236         struct mm_struct *mm;
1237         int err;
1238
1239         err =  mutex_lock_killable(&task->signal->exec_update_mutex);
1240         if (err)
1241                 return ERR_PTR(err);
1242
1243         mm = get_task_mm(task);
1244         if (mm && mm != current->mm &&
1245                         !ptrace_may_access(task, mode)) {
1246                 mmput(mm);
1247                 mm = ERR_PTR(-EACCES);
1248         }
1249         mutex_unlock(&task->signal->exec_update_mutex);
1250
1251         return mm;
1252 }
1253
1254 static void complete_vfork_done(struct task_struct *tsk)
1255 {
1256         struct completion *vfork;
1257
1258         task_lock(tsk);
1259         vfork = tsk->vfork_done;
1260         if (likely(vfork)) {
1261                 tsk->vfork_done = NULL;
1262                 complete(vfork);
1263         }
1264         task_unlock(tsk);
1265 }
1266
1267 static int wait_for_vfork_done(struct task_struct *child,
1268                                 struct completion *vfork)
1269 {
1270         int killed;
1271
1272         freezer_do_not_count();
1273         cgroup_enter_frozen();
1274         killed = wait_for_completion_killable(vfork);
1275         cgroup_leave_frozen(false);
1276         freezer_count();
1277
1278         if (killed) {
1279                 task_lock(child);
1280                 child->vfork_done = NULL;
1281                 task_unlock(child);
1282         }
1283
1284         put_task_struct(child);
1285         return killed;
1286 }
1287
1288 /* Please note the differences between mmput and mm_release.
1289  * mmput is called whenever we stop holding onto a mm_struct,
1290  * error success whatever.
1291  *
1292  * mm_release is called after a mm_struct has been removed
1293  * from the current process.
1294  *
1295  * This difference is important for error handling, when we
1296  * only half set up a mm_struct for a new process and need to restore
1297  * the old one.  Because we mmput the new mm_struct before
1298  * restoring the old one. . .
1299  * Eric Biederman 10 January 1998
1300  */
1301 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1302 {
1303         uprobe_free_utask(tsk);
1304
1305         /* Get rid of any cached register state */
1306         deactivate_mm(tsk, mm);
1307
1308         /*
1309          * Signal userspace if we're not exiting with a core dump
1310          * because we want to leave the value intact for debugging
1311          * purposes.
1312          */
1313         if (tsk->clear_child_tid) {
1314                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1315                     atomic_read(&mm->mm_users) > 1) {
1316                         /*
1317                          * We don't check the error code - if userspace has
1318                          * not set up a proper pointer then tough luck.
1319                          */
1320                         put_user(0, tsk->clear_child_tid);
1321                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1322                                         1, NULL, NULL, 0, 0);
1323                 }
1324                 tsk->clear_child_tid = NULL;
1325         }
1326
1327         /*
1328          * All done, finally we can wake up parent and return this mm to him.
1329          * Also kthread_stop() uses this completion for synchronization.
1330          */
1331         if (tsk->vfork_done)
1332                 complete_vfork_done(tsk);
1333 }
1334
1335 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1336 {
1337         futex_exit_release(tsk);
1338         mm_release(tsk, mm);
1339 }
1340
1341 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1342 {
1343         futex_exec_release(tsk);
1344         mm_release(tsk, mm);
1345 }
1346
1347 /**
1348  * dup_mm() - duplicates an existing mm structure
1349  * @tsk: the task_struct with which the new mm will be associated.
1350  * @oldmm: the mm to duplicate.
1351  *
1352  * Allocates a new mm structure and duplicates the provided @oldmm structure
1353  * content into it.
1354  *
1355  * Return: the duplicated mm or NULL on failure.
1356  */
1357 static struct mm_struct *dup_mm(struct task_struct *tsk,
1358                                 struct mm_struct *oldmm)
1359 {
1360         struct mm_struct *mm;
1361         int err;
1362
1363         mm = allocate_mm();
1364         if (!mm)
1365                 goto fail_nomem;
1366
1367         memcpy(mm, oldmm, sizeof(*mm));
1368
1369         if (!mm_init(mm, tsk, mm->user_ns))
1370                 goto fail_nomem;
1371
1372         err = dup_mmap(mm, oldmm);
1373         if (err)
1374                 goto free_pt;
1375
1376         mm->hiwater_rss = get_mm_rss(mm);
1377         mm->hiwater_vm = mm->total_vm;
1378
1379         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1380                 goto free_pt;
1381
1382         return mm;
1383
1384 free_pt:
1385         /* don't put binfmt in mmput, we haven't got module yet */
1386         mm->binfmt = NULL;
1387         mm_init_owner(mm, NULL);
1388         mmput(mm);
1389
1390 fail_nomem:
1391         return NULL;
1392 }
1393
1394 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1395 {
1396         struct mm_struct *mm, *oldmm;
1397         int retval;
1398
1399         tsk->min_flt = tsk->maj_flt = 0;
1400         tsk->nvcsw = tsk->nivcsw = 0;
1401 #ifdef CONFIG_DETECT_HUNG_TASK
1402         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1403         tsk->last_switch_time = 0;
1404 #endif
1405
1406         tsk->mm = NULL;
1407         tsk->active_mm = NULL;
1408
1409         /*
1410          * Are we cloning a kernel thread?
1411          *
1412          * We need to steal a active VM for that..
1413          */
1414         oldmm = current->mm;
1415         if (!oldmm)
1416                 return 0;
1417
1418         /* initialize the new vmacache entries */
1419         vmacache_flush(tsk);
1420
1421         if (clone_flags & CLONE_VM) {
1422                 mmget(oldmm);
1423                 mm = oldmm;
1424                 goto good_mm;
1425         }
1426
1427         retval = -ENOMEM;
1428         mm = dup_mm(tsk, current->mm);
1429         if (!mm)
1430                 goto fail_nomem;
1431
1432 good_mm:
1433         tsk->mm = mm;
1434         tsk->active_mm = mm;
1435         return 0;
1436
1437 fail_nomem:
1438         return retval;
1439 }
1440
1441 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1442 {
1443         struct fs_struct *fs = current->fs;
1444         if (clone_flags & CLONE_FS) {
1445                 /* tsk->fs is already what we want */
1446                 spin_lock(&fs->lock);
1447                 if (fs->in_exec) {
1448                         spin_unlock(&fs->lock);
1449                         return -EAGAIN;
1450                 }
1451                 fs->users++;
1452                 spin_unlock(&fs->lock);
1453                 return 0;
1454         }
1455         tsk->fs = copy_fs_struct(fs);
1456         if (!tsk->fs)
1457                 return -ENOMEM;
1458         return 0;
1459 }
1460
1461 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1462 {
1463         struct files_struct *oldf, *newf;
1464         int error = 0;
1465
1466         /*
1467          * A background process may not have any files ...
1468          */
1469         oldf = current->files;
1470         if (!oldf)
1471                 goto out;
1472
1473         if (clone_flags & CLONE_FILES) {
1474                 atomic_inc(&oldf->count);
1475                 goto out;
1476         }
1477
1478         newf = dup_fd(oldf, &error);
1479         if (!newf)
1480                 goto out;
1481
1482         tsk->files = newf;
1483         error = 0;
1484 out:
1485         return error;
1486 }
1487
1488 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1489 {
1490 #ifdef CONFIG_BLOCK
1491         struct io_context *ioc = current->io_context;
1492         struct io_context *new_ioc;
1493
1494         if (!ioc)
1495                 return 0;
1496         /*
1497          * Share io context with parent, if CLONE_IO is set
1498          */
1499         if (clone_flags & CLONE_IO) {
1500                 ioc_task_link(ioc);
1501                 tsk->io_context = ioc;
1502         } else if (ioprio_valid(ioc->ioprio)) {
1503                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1504                 if (unlikely(!new_ioc))
1505                         return -ENOMEM;
1506
1507                 new_ioc->ioprio = ioc->ioprio;
1508                 put_io_context(new_ioc);
1509         }
1510 #endif
1511         return 0;
1512 }
1513
1514 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1515 {
1516         struct sighand_struct *sig;
1517
1518         if (clone_flags & CLONE_SIGHAND) {
1519                 refcount_inc(&current->sighand->count);
1520                 return 0;
1521         }
1522         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1523         RCU_INIT_POINTER(tsk->sighand, sig);
1524         if (!sig)
1525                 return -ENOMEM;
1526
1527         refcount_set(&sig->count, 1);
1528         spin_lock_irq(&current->sighand->siglock);
1529         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1530         spin_unlock_irq(&current->sighand->siglock);
1531
1532         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1533         if (clone_flags & CLONE_CLEAR_SIGHAND)
1534                 flush_signal_handlers(tsk, 0);
1535
1536         return 0;
1537 }
1538
1539 void __cleanup_sighand(struct sighand_struct *sighand)
1540 {
1541         if (refcount_dec_and_test(&sighand->count)) {
1542                 signalfd_cleanup(sighand);
1543                 /*
1544                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1545                  * without an RCU grace period, see __lock_task_sighand().
1546                  */
1547                 kmem_cache_free(sighand_cachep, sighand);
1548         }
1549 }
1550
1551 /*
1552  * Initialize POSIX timer handling for a thread group.
1553  */
1554 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1555 {
1556         struct posix_cputimers *pct = &sig->posix_cputimers;
1557         unsigned long cpu_limit;
1558
1559         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1560         posix_cputimers_group_init(pct, cpu_limit);
1561 }
1562
1563 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1564 {
1565         struct signal_struct *sig;
1566
1567         if (clone_flags & CLONE_THREAD)
1568                 return 0;
1569
1570         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1571         tsk->signal = sig;
1572         if (!sig)
1573                 return -ENOMEM;
1574
1575         sig->nr_threads = 1;
1576         atomic_set(&sig->live, 1);
1577         refcount_set(&sig->sigcnt, 1);
1578
1579         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1580         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1581         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1582
1583         init_waitqueue_head(&sig->wait_chldexit);
1584         sig->curr_target = tsk;
1585         init_sigpending(&sig->shared_pending);
1586         INIT_HLIST_HEAD(&sig->multiprocess);
1587         seqlock_init(&sig->stats_lock);
1588         prev_cputime_init(&sig->prev_cputime);
1589
1590 #ifdef CONFIG_POSIX_TIMERS
1591         INIT_LIST_HEAD(&sig->posix_timers);
1592         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1593         sig->real_timer.function = it_real_fn;
1594 #endif
1595
1596         task_lock(current->group_leader);
1597         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1598         task_unlock(current->group_leader);
1599
1600         posix_cpu_timers_init_group(sig);
1601
1602         tty_audit_fork(sig);
1603         sched_autogroup_fork(sig);
1604
1605         sig->oom_score_adj = current->signal->oom_score_adj;
1606         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1607
1608         mutex_init(&sig->cred_guard_mutex);
1609         mutex_init(&sig->exec_update_mutex);
1610
1611         return 0;
1612 }
1613
1614 static void copy_seccomp(struct task_struct *p)
1615 {
1616 #ifdef CONFIG_SECCOMP
1617         /*
1618          * Must be called with sighand->lock held, which is common to
1619          * all threads in the group. Holding cred_guard_mutex is not
1620          * needed because this new task is not yet running and cannot
1621          * be racing exec.
1622          */
1623         assert_spin_locked(&current->sighand->siglock);
1624
1625         /* Ref-count the new filter user, and assign it. */
1626         get_seccomp_filter(current);
1627         p->seccomp = current->seccomp;
1628
1629         /*
1630          * Explicitly enable no_new_privs here in case it got set
1631          * between the task_struct being duplicated and holding the
1632          * sighand lock. The seccomp state and nnp must be in sync.
1633          */
1634         if (task_no_new_privs(current))
1635                 task_set_no_new_privs(p);
1636
1637         /*
1638          * If the parent gained a seccomp mode after copying thread
1639          * flags and between before we held the sighand lock, we have
1640          * to manually enable the seccomp thread flag here.
1641          */
1642         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1643                 set_tsk_thread_flag(p, TIF_SECCOMP);
1644 #endif
1645 }
1646
1647 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1648 {
1649         current->clear_child_tid = tidptr;
1650
1651         return task_pid_vnr(current);
1652 }
1653
1654 static void rt_mutex_init_task(struct task_struct *p)
1655 {
1656         raw_spin_lock_init(&p->pi_lock);
1657 #ifdef CONFIG_RT_MUTEXES
1658         p->pi_waiters = RB_ROOT_CACHED;
1659         p->pi_top_task = NULL;
1660         p->pi_blocked_on = NULL;
1661 #endif
1662 }
1663
1664 static inline void init_task_pid_links(struct task_struct *task)
1665 {
1666         enum pid_type type;
1667
1668         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1669                 INIT_HLIST_NODE(&task->pid_links[type]);
1670         }
1671 }
1672
1673 static inline void
1674 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1675 {
1676         if (type == PIDTYPE_PID)
1677                 task->thread_pid = pid;
1678         else
1679                 task->signal->pids[type] = pid;
1680 }
1681
1682 static inline void rcu_copy_process(struct task_struct *p)
1683 {
1684 #ifdef CONFIG_PREEMPT_RCU
1685         p->rcu_read_lock_nesting = 0;
1686         p->rcu_read_unlock_special.s = 0;
1687         p->rcu_blocked_node = NULL;
1688         INIT_LIST_HEAD(&p->rcu_node_entry);
1689 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1690 #ifdef CONFIG_TASKS_RCU
1691         p->rcu_tasks_holdout = false;
1692         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1693         p->rcu_tasks_idle_cpu = -1;
1694 #endif /* #ifdef CONFIG_TASKS_RCU */
1695 #ifdef CONFIG_TASKS_TRACE_RCU
1696         p->trc_reader_nesting = 0;
1697         p->trc_reader_special.s = 0;
1698         INIT_LIST_HEAD(&p->trc_holdout_list);
1699 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1700 }
1701
1702 struct pid *pidfd_pid(const struct file *file)
1703 {
1704         if (file->f_op == &pidfd_fops)
1705                 return file->private_data;
1706
1707         return ERR_PTR(-EBADF);
1708 }
1709
1710 static int pidfd_release(struct inode *inode, struct file *file)
1711 {
1712         struct pid *pid = file->private_data;
1713
1714         file->private_data = NULL;
1715         put_pid(pid);
1716         return 0;
1717 }
1718
1719 #ifdef CONFIG_PROC_FS
1720 /**
1721  * pidfd_show_fdinfo - print information about a pidfd
1722  * @m: proc fdinfo file
1723  * @f: file referencing a pidfd
1724  *
1725  * Pid:
1726  * This function will print the pid that a given pidfd refers to in the
1727  * pid namespace of the procfs instance.
1728  * If the pid namespace of the process is not a descendant of the pid
1729  * namespace of the procfs instance 0 will be shown as its pid. This is
1730  * similar to calling getppid() on a process whose parent is outside of
1731  * its pid namespace.
1732  *
1733  * NSpid:
1734  * If pid namespaces are supported then this function will also print
1735  * the pid of a given pidfd refers to for all descendant pid namespaces
1736  * starting from the current pid namespace of the instance, i.e. the
1737  * Pid field and the first entry in the NSpid field will be identical.
1738  * If the pid namespace of the process is not a descendant of the pid
1739  * namespace of the procfs instance 0 will be shown as its first NSpid
1740  * entry and no others will be shown.
1741  * Note that this differs from the Pid and NSpid fields in
1742  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1743  * the  pid namespace of the procfs instance. The difference becomes
1744  * obvious when sending around a pidfd between pid namespaces from a
1745  * different branch of the tree, i.e. where no ancestoral relation is
1746  * present between the pid namespaces:
1747  * - create two new pid namespaces ns1 and ns2 in the initial pid
1748  *   namespace (also take care to create new mount namespaces in the
1749  *   new pid namespace and mount procfs)
1750  * - create a process with a pidfd in ns1
1751  * - send pidfd from ns1 to ns2
1752  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1753  *   have exactly one entry, which is 0
1754  */
1755 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1756 {
1757         struct pid *pid = f->private_data;
1758         struct pid_namespace *ns;
1759         pid_t nr = -1;
1760
1761         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1762                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1763                 nr = pid_nr_ns(pid, ns);
1764         }
1765
1766         seq_put_decimal_ll(m, "Pid:\t", nr);
1767
1768 #ifdef CONFIG_PID_NS
1769         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1770         if (nr > 0) {
1771                 int i;
1772
1773                 /* If nr is non-zero it means that 'pid' is valid and that
1774                  * ns, i.e. the pid namespace associated with the procfs
1775                  * instance, is in the pid namespace hierarchy of pid.
1776                  * Start at one below the already printed level.
1777                  */
1778                 for (i = ns->level + 1; i <= pid->level; i++)
1779                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1780         }
1781 #endif
1782         seq_putc(m, '\n');
1783 }
1784 #endif
1785
1786 /*
1787  * Poll support for process exit notification.
1788  */
1789 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1790 {
1791         struct task_struct *task;
1792         struct pid *pid = file->private_data;
1793         __poll_t poll_flags = 0;
1794
1795         poll_wait(file, &pid->wait_pidfd, pts);
1796
1797         rcu_read_lock();
1798         task = pid_task(pid, PIDTYPE_PID);
1799         /*
1800          * Inform pollers only when the whole thread group exits.
1801          * If the thread group leader exits before all other threads in the
1802          * group, then poll(2) should block, similar to the wait(2) family.
1803          */
1804         if (!task || (task->exit_state && thread_group_empty(task)))
1805                 poll_flags = EPOLLIN | EPOLLRDNORM;
1806         rcu_read_unlock();
1807
1808         return poll_flags;
1809 }
1810
1811 const struct file_operations pidfd_fops = {
1812         .release = pidfd_release,
1813         .poll = pidfd_poll,
1814 #ifdef CONFIG_PROC_FS
1815         .show_fdinfo = pidfd_show_fdinfo,
1816 #endif
1817 };
1818
1819 static void __delayed_free_task(struct rcu_head *rhp)
1820 {
1821         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1822
1823         free_task(tsk);
1824 }
1825
1826 static __always_inline void delayed_free_task(struct task_struct *tsk)
1827 {
1828         if (IS_ENABLED(CONFIG_MEMCG))
1829                 call_rcu(&tsk->rcu, __delayed_free_task);
1830         else
1831                 free_task(tsk);
1832 }
1833
1834 /*
1835  * This creates a new process as a copy of the old one,
1836  * but does not actually start it yet.
1837  *
1838  * It copies the registers, and all the appropriate
1839  * parts of the process environment (as per the clone
1840  * flags). The actual kick-off is left to the caller.
1841  */
1842 static __latent_entropy struct task_struct *copy_process(
1843                                         struct pid *pid,
1844                                         int trace,
1845                                         int node,
1846                                         struct kernel_clone_args *args)
1847 {
1848         int pidfd = -1, retval;
1849         struct task_struct *p;
1850         struct multiprocess_signals delayed;
1851         struct file *pidfile = NULL;
1852         u64 clone_flags = args->flags;
1853         struct nsproxy *nsp = current->nsproxy;
1854
1855         /*
1856          * Don't allow sharing the root directory with processes in a different
1857          * namespace
1858          */
1859         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1860                 return ERR_PTR(-EINVAL);
1861
1862         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1863                 return ERR_PTR(-EINVAL);
1864
1865         /*
1866          * Thread groups must share signals as well, and detached threads
1867          * can only be started up within the thread group.
1868          */
1869         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1870                 return ERR_PTR(-EINVAL);
1871
1872         /*
1873          * Shared signal handlers imply shared VM. By way of the above,
1874          * thread groups also imply shared VM. Blocking this case allows
1875          * for various simplifications in other code.
1876          */
1877         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1878                 return ERR_PTR(-EINVAL);
1879
1880         /*
1881          * Siblings of global init remain as zombies on exit since they are
1882          * not reaped by their parent (swapper). To solve this and to avoid
1883          * multi-rooted process trees, prevent global and container-inits
1884          * from creating siblings.
1885          */
1886         if ((clone_flags & CLONE_PARENT) &&
1887                                 current->signal->flags & SIGNAL_UNKILLABLE)
1888                 return ERR_PTR(-EINVAL);
1889
1890         /*
1891          * If the new process will be in a different pid or user namespace
1892          * do not allow it to share a thread group with the forking task.
1893          */
1894         if (clone_flags & CLONE_THREAD) {
1895                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1896                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1897                         return ERR_PTR(-EINVAL);
1898         }
1899
1900         /*
1901          * If the new process will be in a different time namespace
1902          * do not allow it to share VM or a thread group with the forking task.
1903          */
1904         if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1905                 if (nsp->time_ns != nsp->time_ns_for_children)
1906                         return ERR_PTR(-EINVAL);
1907         }
1908
1909         if (clone_flags & CLONE_PIDFD) {
1910                 /*
1911                  * - CLONE_DETACHED is blocked so that we can potentially
1912                  *   reuse it later for CLONE_PIDFD.
1913                  * - CLONE_THREAD is blocked until someone really needs it.
1914                  */
1915                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1916                         return ERR_PTR(-EINVAL);
1917         }
1918
1919         /*
1920          * Force any signals received before this point to be delivered
1921          * before the fork happens.  Collect up signals sent to multiple
1922          * processes that happen during the fork and delay them so that
1923          * they appear to happen after the fork.
1924          */
1925         sigemptyset(&delayed.signal);
1926         INIT_HLIST_NODE(&delayed.node);
1927
1928         spin_lock_irq(&current->sighand->siglock);
1929         if (!(clone_flags & CLONE_THREAD))
1930                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1931         recalc_sigpending();
1932         spin_unlock_irq(&current->sighand->siglock);
1933         retval = -ERESTARTNOINTR;
1934         if (signal_pending(current))
1935                 goto fork_out;
1936
1937         retval = -ENOMEM;
1938         p = dup_task_struct(current, node);
1939         if (!p)
1940                 goto fork_out;
1941
1942         /*
1943          * This _must_ happen before we call free_task(), i.e. before we jump
1944          * to any of the bad_fork_* labels. This is to avoid freeing
1945          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1946          * kernel threads (PF_KTHREAD).
1947          */
1948         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1949         /*
1950          * Clear TID on mm_release()?
1951          */
1952         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1953
1954         ftrace_graph_init_task(p);
1955
1956         rt_mutex_init_task(p);
1957
1958 #ifdef CONFIG_PROVE_LOCKING
1959         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1960         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1961 #endif
1962         retval = -EAGAIN;
1963         if (atomic_read(&p->real_cred->user->processes) >=
1964                         task_rlimit(p, RLIMIT_NPROC)) {
1965                 if (p->real_cred->user != INIT_USER &&
1966                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1967                         goto bad_fork_free;
1968         }
1969         current->flags &= ~PF_NPROC_EXCEEDED;
1970
1971         retval = copy_creds(p, clone_flags);
1972         if (retval < 0)
1973                 goto bad_fork_free;
1974
1975         /*
1976          * If multiple threads are within copy_process(), then this check
1977          * triggers too late. This doesn't hurt, the check is only there
1978          * to stop root fork bombs.
1979          */
1980         retval = -EAGAIN;
1981         if (nr_threads >= max_threads)
1982                 goto bad_fork_cleanup_count;
1983
1984         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1985         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1986         p->flags |= PF_FORKNOEXEC;
1987         INIT_LIST_HEAD(&p->children);
1988         INIT_LIST_HEAD(&p->sibling);
1989         rcu_copy_process(p);
1990         p->vfork_done = NULL;
1991         spin_lock_init(&p->alloc_lock);
1992
1993         init_sigpending(&p->pending);
1994
1995         p->utime = p->stime = p->gtime = 0;
1996 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1997         p->utimescaled = p->stimescaled = 0;
1998 #endif
1999         prev_cputime_init(&p->prev_cputime);
2000
2001 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2002         seqcount_init(&p->vtime.seqcount);
2003         p->vtime.starttime = 0;
2004         p->vtime.state = VTIME_INACTIVE;
2005 #endif
2006
2007 #if defined(SPLIT_RSS_COUNTING)
2008         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2009 #endif
2010
2011         p->default_timer_slack_ns = current->timer_slack_ns;
2012
2013 #ifdef CONFIG_PSI
2014         p->psi_flags = 0;
2015 #endif
2016
2017         task_io_accounting_init(&p->ioac);
2018         acct_clear_integrals(p);
2019
2020         posix_cputimers_init(&p->posix_cputimers);
2021
2022         p->io_context = NULL;
2023         audit_set_context(p, NULL);
2024         cgroup_fork(p);
2025 #ifdef CONFIG_NUMA
2026         p->mempolicy = mpol_dup(p->mempolicy);
2027         if (IS_ERR(p->mempolicy)) {
2028                 retval = PTR_ERR(p->mempolicy);
2029                 p->mempolicy = NULL;
2030                 goto bad_fork_cleanup_threadgroup_lock;
2031         }
2032 #endif
2033 #ifdef CONFIG_CPUSETS
2034         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2035         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2036         seqcount_init(&p->mems_allowed_seq);
2037 #endif
2038 #ifdef CONFIG_TRACE_IRQFLAGS
2039         p->irq_events = 0;
2040         p->hardirqs_enabled = 0;
2041         p->hardirq_enable_ip = 0;
2042         p->hardirq_enable_event = 0;
2043         p->hardirq_disable_ip = _THIS_IP_;
2044         p->hardirq_disable_event = 0;
2045         p->softirqs_enabled = 1;
2046         p->softirq_enable_ip = _THIS_IP_;
2047         p->softirq_enable_event = 0;
2048         p->softirq_disable_ip = 0;
2049         p->softirq_disable_event = 0;
2050         p->hardirq_context = 0;
2051         p->softirq_context = 0;
2052 #endif
2053
2054         p->pagefault_disabled = 0;
2055
2056 #ifdef CONFIG_LOCKDEP
2057         lockdep_init_task(p);
2058 #endif
2059
2060 #ifdef CONFIG_DEBUG_MUTEXES
2061         p->blocked_on = NULL; /* not blocked yet */
2062 #endif
2063 #ifdef CONFIG_BCACHE
2064         p->sequential_io        = 0;
2065         p->sequential_io_avg    = 0;
2066 #endif
2067
2068         /* Perform scheduler related setup. Assign this task to a CPU. */
2069         retval = sched_fork(clone_flags, p);
2070         if (retval)
2071                 goto bad_fork_cleanup_policy;
2072
2073         retval = perf_event_init_task(p);
2074         if (retval)
2075                 goto bad_fork_cleanup_policy;
2076         retval = audit_alloc(p);
2077         if (retval)
2078                 goto bad_fork_cleanup_perf;
2079         /* copy all the process information */
2080         shm_init_task(p);
2081         retval = security_task_alloc(p, clone_flags);
2082         if (retval)
2083                 goto bad_fork_cleanup_audit;
2084         retval = copy_semundo(clone_flags, p);
2085         if (retval)
2086                 goto bad_fork_cleanup_security;
2087         retval = copy_files(clone_flags, p);
2088         if (retval)
2089                 goto bad_fork_cleanup_semundo;
2090         retval = copy_fs(clone_flags, p);
2091         if (retval)
2092                 goto bad_fork_cleanup_files;
2093         retval = copy_sighand(clone_flags, p);
2094         if (retval)
2095                 goto bad_fork_cleanup_fs;
2096         retval = copy_signal(clone_flags, p);
2097         if (retval)
2098                 goto bad_fork_cleanup_sighand;
2099         retval = copy_mm(clone_flags, p);
2100         if (retval)
2101                 goto bad_fork_cleanup_signal;
2102         retval = copy_namespaces(clone_flags, p);
2103         if (retval)
2104                 goto bad_fork_cleanup_mm;
2105         retval = copy_io(clone_flags, p);
2106         if (retval)
2107                 goto bad_fork_cleanup_namespaces;
2108         retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p,
2109                                  args->tls);
2110         if (retval)
2111                 goto bad_fork_cleanup_io;
2112
2113         stackleak_task_init(p);
2114
2115         if (pid != &init_struct_pid) {
2116                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2117                                 args->set_tid_size);
2118                 if (IS_ERR(pid)) {
2119                         retval = PTR_ERR(pid);
2120                         goto bad_fork_cleanup_thread;
2121                 }
2122         }
2123
2124         /*
2125          * This has to happen after we've potentially unshared the file
2126          * descriptor table (so that the pidfd doesn't leak into the child
2127          * if the fd table isn't shared).
2128          */
2129         if (clone_flags & CLONE_PIDFD) {
2130                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2131                 if (retval < 0)
2132                         goto bad_fork_free_pid;
2133
2134                 pidfd = retval;
2135
2136                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2137                                               O_RDWR | O_CLOEXEC);
2138                 if (IS_ERR(pidfile)) {
2139                         put_unused_fd(pidfd);
2140                         retval = PTR_ERR(pidfile);
2141                         goto bad_fork_free_pid;
2142                 }
2143                 get_pid(pid);   /* held by pidfile now */
2144
2145                 retval = put_user(pidfd, args->pidfd);
2146                 if (retval)
2147                         goto bad_fork_put_pidfd;
2148         }
2149
2150 #ifdef CONFIG_BLOCK
2151         p->plug = NULL;
2152 #endif
2153         futex_init_task(p);
2154
2155         /*
2156          * sigaltstack should be cleared when sharing the same VM
2157          */
2158         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2159                 sas_ss_reset(p);
2160
2161         /*
2162          * Syscall tracing and stepping should be turned off in the
2163          * child regardless of CLONE_PTRACE.
2164          */
2165         user_disable_single_step(p);
2166         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2167 #ifdef TIF_SYSCALL_EMU
2168         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2169 #endif
2170         clear_tsk_latency_tracing(p);
2171
2172         /* ok, now we should be set up.. */
2173         p->pid = pid_nr(pid);
2174         if (clone_flags & CLONE_THREAD) {
2175                 p->exit_signal = -1;
2176                 p->group_leader = current->group_leader;
2177                 p->tgid = current->tgid;
2178         } else {
2179                 if (clone_flags & CLONE_PARENT)
2180                         p->exit_signal = current->group_leader->exit_signal;
2181                 else
2182                         p->exit_signal = args->exit_signal;
2183                 p->group_leader = p;
2184                 p->tgid = p->pid;
2185         }
2186
2187         p->nr_dirtied = 0;
2188         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2189         p->dirty_paused_when = 0;
2190
2191         p->pdeath_signal = 0;
2192         INIT_LIST_HEAD(&p->thread_group);
2193         p->task_works = NULL;
2194
2195         /*
2196          * Ensure that the cgroup subsystem policies allow the new process to be
2197          * forked. It should be noted the the new process's css_set can be changed
2198          * between here and cgroup_post_fork() if an organisation operation is in
2199          * progress.
2200          */
2201         retval = cgroup_can_fork(p, args);
2202         if (retval)
2203                 goto bad_fork_put_pidfd;
2204
2205         /*
2206          * From this point on we must avoid any synchronous user-space
2207          * communication until we take the tasklist-lock. In particular, we do
2208          * not want user-space to be able to predict the process start-time by
2209          * stalling fork(2) after we recorded the start_time but before it is
2210          * visible to the system.
2211          */
2212
2213         p->start_time = ktime_get_ns();
2214         p->start_boottime = ktime_get_boottime_ns();
2215
2216         /*
2217          * Make it visible to the rest of the system, but dont wake it up yet.
2218          * Need tasklist lock for parent etc handling!
2219          */
2220         write_lock_irq(&tasklist_lock);
2221
2222         /* CLONE_PARENT re-uses the old parent */
2223         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2224                 p->real_parent = current->real_parent;
2225                 p->parent_exec_id = current->parent_exec_id;
2226         } else {
2227                 p->real_parent = current;
2228                 p->parent_exec_id = current->self_exec_id;
2229         }
2230
2231         klp_copy_process(p);
2232
2233         spin_lock(&current->sighand->siglock);
2234
2235         /*
2236          * Copy seccomp details explicitly here, in case they were changed
2237          * before holding sighand lock.
2238          */
2239         copy_seccomp(p);
2240
2241         rseq_fork(p, clone_flags);
2242
2243         /* Don't start children in a dying pid namespace */
2244         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2245                 retval = -ENOMEM;
2246                 goto bad_fork_cancel_cgroup;
2247         }
2248
2249         /* Let kill terminate clone/fork in the middle */
2250         if (fatal_signal_pending(current)) {
2251                 retval = -EINTR;
2252                 goto bad_fork_cancel_cgroup;
2253         }
2254
2255         /* past the last point of failure */
2256         if (pidfile)
2257                 fd_install(pidfd, pidfile);
2258
2259         init_task_pid_links(p);
2260         if (likely(p->pid)) {
2261                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2262
2263                 init_task_pid(p, PIDTYPE_PID, pid);
2264                 if (thread_group_leader(p)) {
2265                         init_task_pid(p, PIDTYPE_TGID, pid);
2266                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2267                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2268
2269                         if (is_child_reaper(pid)) {
2270                                 ns_of_pid(pid)->child_reaper = p;
2271                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2272                         }
2273                         p->signal->shared_pending.signal = delayed.signal;
2274                         p->signal->tty = tty_kref_get(current->signal->tty);
2275                         /*
2276                          * Inherit has_child_subreaper flag under the same
2277                          * tasklist_lock with adding child to the process tree
2278                          * for propagate_has_child_subreaper optimization.
2279                          */
2280                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2281                                                          p->real_parent->signal->is_child_subreaper;
2282                         list_add_tail(&p->sibling, &p->real_parent->children);
2283                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2284                         attach_pid(p, PIDTYPE_TGID);
2285                         attach_pid(p, PIDTYPE_PGID);
2286                         attach_pid(p, PIDTYPE_SID);
2287                         __this_cpu_inc(process_counts);
2288                 } else {
2289                         current->signal->nr_threads++;
2290                         atomic_inc(&current->signal->live);
2291                         refcount_inc(&current->signal->sigcnt);
2292                         task_join_group_stop(p);
2293                         list_add_tail_rcu(&p->thread_group,
2294                                           &p->group_leader->thread_group);
2295                         list_add_tail_rcu(&p->thread_node,
2296                                           &p->signal->thread_head);
2297                 }
2298                 attach_pid(p, PIDTYPE_PID);
2299                 nr_threads++;
2300         }
2301         total_forks++;
2302         hlist_del_init(&delayed.node);
2303         spin_unlock(&current->sighand->siglock);
2304         syscall_tracepoint_update(p);
2305         write_unlock_irq(&tasklist_lock);
2306
2307         proc_fork_connector(p);
2308         cgroup_post_fork(p, args);
2309         perf_event_fork(p);
2310
2311         trace_task_newtask(p, clone_flags);
2312         uprobe_copy_process(p, clone_flags);
2313
2314         return p;
2315
2316 bad_fork_cancel_cgroup:
2317         spin_unlock(&current->sighand->siglock);
2318         write_unlock_irq(&tasklist_lock);
2319         cgroup_cancel_fork(p, args);
2320 bad_fork_put_pidfd:
2321         if (clone_flags & CLONE_PIDFD) {
2322                 fput(pidfile);
2323                 put_unused_fd(pidfd);
2324         }
2325 bad_fork_free_pid:
2326         if (pid != &init_struct_pid)
2327                 free_pid(pid);
2328 bad_fork_cleanup_thread:
2329         exit_thread(p);
2330 bad_fork_cleanup_io:
2331         if (p->io_context)
2332                 exit_io_context(p);
2333 bad_fork_cleanup_namespaces:
2334         exit_task_namespaces(p);
2335 bad_fork_cleanup_mm:
2336         if (p->mm) {
2337                 mm_clear_owner(p->mm, p);
2338                 mmput(p->mm);
2339         }
2340 bad_fork_cleanup_signal:
2341         if (!(clone_flags & CLONE_THREAD))
2342                 free_signal_struct(p->signal);
2343 bad_fork_cleanup_sighand:
2344         __cleanup_sighand(p->sighand);
2345 bad_fork_cleanup_fs:
2346         exit_fs(p); /* blocking */
2347 bad_fork_cleanup_files:
2348         exit_files(p); /* blocking */
2349 bad_fork_cleanup_semundo:
2350         exit_sem(p);
2351 bad_fork_cleanup_security:
2352         security_task_free(p);
2353 bad_fork_cleanup_audit:
2354         audit_free(p);
2355 bad_fork_cleanup_perf:
2356         perf_event_free_task(p);
2357 bad_fork_cleanup_policy:
2358         lockdep_free_task(p);
2359 #ifdef CONFIG_NUMA
2360         mpol_put(p->mempolicy);
2361 bad_fork_cleanup_threadgroup_lock:
2362 #endif
2363         delayacct_tsk_free(p);
2364 bad_fork_cleanup_count:
2365         atomic_dec(&p->cred->user->processes);
2366         exit_creds(p);
2367 bad_fork_free:
2368         p->state = TASK_DEAD;
2369         put_task_stack(p);
2370         delayed_free_task(p);
2371 fork_out:
2372         spin_lock_irq(&current->sighand->siglock);
2373         hlist_del_init(&delayed.node);
2374         spin_unlock_irq(&current->sighand->siglock);
2375         return ERR_PTR(retval);
2376 }
2377
2378 static inline void init_idle_pids(struct task_struct *idle)
2379 {
2380         enum pid_type type;
2381
2382         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2383                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2384                 init_task_pid(idle, type, &init_struct_pid);
2385         }
2386 }
2387
2388 struct task_struct *fork_idle(int cpu)
2389 {
2390         struct task_struct *task;
2391         struct kernel_clone_args args = {
2392                 .flags = CLONE_VM,
2393         };
2394
2395         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2396         if (!IS_ERR(task)) {
2397                 init_idle_pids(task);
2398                 init_idle(task, cpu);
2399         }
2400
2401         return task;
2402 }
2403
2404 struct mm_struct *copy_init_mm(void)
2405 {
2406         return dup_mm(NULL, &init_mm);
2407 }
2408
2409 /*
2410  *  Ok, this is the main fork-routine.
2411  *
2412  * It copies the process, and if successful kick-starts
2413  * it and waits for it to finish using the VM if required.
2414  *
2415  * args->exit_signal is expected to be checked for sanity by the caller.
2416  */
2417 long _do_fork(struct kernel_clone_args *args)
2418 {
2419         u64 clone_flags = args->flags;
2420         struct completion vfork;
2421         struct pid *pid;
2422         struct task_struct *p;
2423         int trace = 0;
2424         long nr;
2425
2426         /*
2427          * Determine whether and which event to report to ptracer.  When
2428          * called from kernel_thread or CLONE_UNTRACED is explicitly
2429          * requested, no event is reported; otherwise, report if the event
2430          * for the type of forking is enabled.
2431          */
2432         if (!(clone_flags & CLONE_UNTRACED)) {
2433                 if (clone_flags & CLONE_VFORK)
2434                         trace = PTRACE_EVENT_VFORK;
2435                 else if (args->exit_signal != SIGCHLD)
2436                         trace = PTRACE_EVENT_CLONE;
2437                 else
2438                         trace = PTRACE_EVENT_FORK;
2439
2440                 if (likely(!ptrace_event_enabled(current, trace)))
2441                         trace = 0;
2442         }
2443
2444         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2445         add_latent_entropy();
2446
2447         if (IS_ERR(p))
2448                 return PTR_ERR(p);
2449
2450         /*
2451          * Do this prior waking up the new thread - the thread pointer
2452          * might get invalid after that point, if the thread exits quickly.
2453          */
2454         trace_sched_process_fork(current, p);
2455
2456         pid = get_task_pid(p, PIDTYPE_PID);
2457         nr = pid_vnr(pid);
2458
2459         if (clone_flags & CLONE_PARENT_SETTID)
2460                 put_user(nr, args->parent_tid);
2461
2462         if (clone_flags & CLONE_VFORK) {
2463                 p->vfork_done = &vfork;
2464                 init_completion(&vfork);
2465                 get_task_struct(p);
2466         }
2467
2468         wake_up_new_task(p);
2469
2470         /* forking complete and child started to run, tell ptracer */
2471         if (unlikely(trace))
2472                 ptrace_event_pid(trace, pid);
2473
2474         if (clone_flags & CLONE_VFORK) {
2475                 if (!wait_for_vfork_done(p, &vfork))
2476                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2477         }
2478
2479         put_pid(pid);
2480         return nr;
2481 }
2482
2483 bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
2484 {
2485         /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2486         if ((kargs->flags & CLONE_PIDFD) &&
2487             (kargs->flags & CLONE_PARENT_SETTID))
2488                 return false;
2489
2490         return true;
2491 }
2492
2493 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2494 /* For compatibility with architectures that call do_fork directly rather than
2495  * using the syscall entry points below. */
2496 long do_fork(unsigned long clone_flags,
2497               unsigned long stack_start,
2498               unsigned long stack_size,
2499               int __user *parent_tidptr,
2500               int __user *child_tidptr)
2501 {
2502         struct kernel_clone_args args = {
2503                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2504                 .pidfd          = parent_tidptr,
2505                 .child_tid      = child_tidptr,
2506                 .parent_tid     = parent_tidptr,
2507                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2508                 .stack          = stack_start,
2509                 .stack_size     = stack_size,
2510         };
2511
2512         if (!legacy_clone_args_valid(&args))
2513                 return -EINVAL;
2514
2515         return _do_fork(&args);
2516 }
2517 #endif
2518
2519 /*
2520  * Create a kernel thread.
2521  */
2522 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2523 {
2524         struct kernel_clone_args args = {
2525                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2526                                     CLONE_UNTRACED) & ~CSIGNAL),
2527                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2528                 .stack          = (unsigned long)fn,
2529                 .stack_size     = (unsigned long)arg,
2530         };
2531
2532         return _do_fork(&args);
2533 }
2534
2535 #ifdef __ARCH_WANT_SYS_FORK
2536 SYSCALL_DEFINE0(fork)
2537 {
2538 #ifdef CONFIG_MMU
2539         struct kernel_clone_args args = {
2540                 .exit_signal = SIGCHLD,
2541         };
2542
2543         return _do_fork(&args);
2544 #else
2545         /* can not support in nommu mode */
2546         return -EINVAL;
2547 #endif
2548 }
2549 #endif
2550
2551 #ifdef __ARCH_WANT_SYS_VFORK
2552 SYSCALL_DEFINE0(vfork)
2553 {
2554         struct kernel_clone_args args = {
2555                 .flags          = CLONE_VFORK | CLONE_VM,
2556                 .exit_signal    = SIGCHLD,
2557         };
2558
2559         return _do_fork(&args);
2560 }
2561 #endif
2562
2563 #ifdef __ARCH_WANT_SYS_CLONE
2564 #ifdef CONFIG_CLONE_BACKWARDS
2565 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2566                  int __user *, parent_tidptr,
2567                  unsigned long, tls,
2568                  int __user *, child_tidptr)
2569 #elif defined(CONFIG_CLONE_BACKWARDS2)
2570 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2571                  int __user *, parent_tidptr,
2572                  int __user *, child_tidptr,
2573                  unsigned long, tls)
2574 #elif defined(CONFIG_CLONE_BACKWARDS3)
2575 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2576                 int, stack_size,
2577                 int __user *, parent_tidptr,
2578                 int __user *, child_tidptr,
2579                 unsigned long, tls)
2580 #else
2581 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2582                  int __user *, parent_tidptr,
2583                  int __user *, child_tidptr,
2584                  unsigned long, tls)
2585 #endif
2586 {
2587         struct kernel_clone_args args = {
2588                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2589                 .pidfd          = parent_tidptr,
2590                 .child_tid      = child_tidptr,
2591                 .parent_tid     = parent_tidptr,
2592                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2593                 .stack          = newsp,
2594                 .tls            = tls,
2595         };
2596
2597         if (!legacy_clone_args_valid(&args))
2598                 return -EINVAL;
2599
2600         return _do_fork(&args);
2601 }
2602 #endif
2603
2604 #ifdef __ARCH_WANT_SYS_CLONE3
2605
2606 /*
2607  * copy_thread implementations handle CLONE_SETTLS by reading the TLS value from
2608  * the registers containing the syscall arguments for clone. This doesn't work
2609  * with clone3 since the TLS value is passed in clone_args instead.
2610  */
2611 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2612 #error clone3 requires copy_thread_tls support in arch
2613 #endif
2614
2615 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2616                                               struct clone_args __user *uargs,
2617                                               size_t usize)
2618 {
2619         int err;
2620         struct clone_args args;
2621         pid_t *kset_tid = kargs->set_tid;
2622
2623         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2624                      CLONE_ARGS_SIZE_VER0);
2625         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2626                      CLONE_ARGS_SIZE_VER1);
2627         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2628                      CLONE_ARGS_SIZE_VER2);
2629         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2630
2631         if (unlikely(usize > PAGE_SIZE))
2632                 return -E2BIG;
2633         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2634                 return -EINVAL;
2635
2636         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2637         if (err)
2638                 return err;
2639
2640         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2641                 return -EINVAL;
2642
2643         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2644                 return -EINVAL;
2645
2646         if (unlikely(args.set_tid && args.set_tid_size == 0))
2647                 return -EINVAL;
2648
2649         /*
2650          * Verify that higher 32bits of exit_signal are unset and that
2651          * it is a valid signal
2652          */
2653         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2654                      !valid_signal(args.exit_signal)))
2655                 return -EINVAL;
2656
2657         if ((args.flags & CLONE_INTO_CGROUP) &&
2658             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2659                 return -EINVAL;
2660
2661         *kargs = (struct kernel_clone_args){
2662                 .flags          = args.flags,
2663                 .pidfd          = u64_to_user_ptr(args.pidfd),
2664                 .child_tid      = u64_to_user_ptr(args.child_tid),
2665                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2666                 .exit_signal    = args.exit_signal,
2667                 .stack          = args.stack,
2668                 .stack_size     = args.stack_size,
2669                 .tls            = args.tls,
2670                 .set_tid_size   = args.set_tid_size,
2671                 .cgroup         = args.cgroup,
2672         };
2673
2674         if (args.set_tid &&
2675                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2676                         (kargs->set_tid_size * sizeof(pid_t))))
2677                 return -EFAULT;
2678
2679         kargs->set_tid = kset_tid;
2680
2681         return 0;
2682 }
2683
2684 /**
2685  * clone3_stack_valid - check and prepare stack
2686  * @kargs: kernel clone args
2687  *
2688  * Verify that the stack arguments userspace gave us are sane.
2689  * In addition, set the stack direction for userspace since it's easy for us to
2690  * determine.
2691  */
2692 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2693 {
2694         if (kargs->stack == 0) {
2695                 if (kargs->stack_size > 0)
2696                         return false;
2697         } else {
2698                 if (kargs->stack_size == 0)
2699                         return false;
2700
2701                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2702                         return false;
2703
2704 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2705                 kargs->stack += kargs->stack_size;
2706 #endif
2707         }
2708
2709         return true;
2710 }
2711
2712 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2713 {
2714         /* Verify that no unknown flags are passed along. */
2715         if (kargs->flags &
2716             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2717                 return false;
2718
2719         /*
2720          * - make the CLONE_DETACHED bit reuseable for clone3
2721          * - make the CSIGNAL bits reuseable for clone3
2722          */
2723         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2724                 return false;
2725
2726         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2727             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2728                 return false;
2729
2730         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2731             kargs->exit_signal)
2732                 return false;
2733
2734         if (!clone3_stack_valid(kargs))
2735                 return false;
2736
2737         return true;
2738 }
2739
2740 /**
2741  * clone3 - create a new process with specific properties
2742  * @uargs: argument structure
2743  * @size:  size of @uargs
2744  *
2745  * clone3() is the extensible successor to clone()/clone2().
2746  * It takes a struct as argument that is versioned by its size.
2747  *
2748  * Return: On success, a positive PID for the child process.
2749  *         On error, a negative errno number.
2750  */
2751 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2752 {
2753         int err;
2754
2755         struct kernel_clone_args kargs;
2756         pid_t set_tid[MAX_PID_NS_LEVEL];
2757
2758         kargs.set_tid = set_tid;
2759
2760         err = copy_clone_args_from_user(&kargs, uargs, size);
2761         if (err)
2762                 return err;
2763
2764         if (!clone3_args_valid(&kargs))
2765                 return -EINVAL;
2766
2767         return _do_fork(&kargs);
2768 }
2769 #endif
2770
2771 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2772 {
2773         struct task_struct *leader, *parent, *child;
2774         int res;
2775
2776         read_lock(&tasklist_lock);
2777         leader = top = top->group_leader;
2778 down:
2779         for_each_thread(leader, parent) {
2780                 list_for_each_entry(child, &parent->children, sibling) {
2781                         res = visitor(child, data);
2782                         if (res) {
2783                                 if (res < 0)
2784                                         goto out;
2785                                 leader = child;
2786                                 goto down;
2787                         }
2788 up:
2789                         ;
2790                 }
2791         }
2792
2793         if (leader != top) {
2794                 child = leader;
2795                 parent = child->real_parent;
2796                 leader = parent->group_leader;
2797                 goto up;
2798         }
2799 out:
2800         read_unlock(&tasklist_lock);
2801 }
2802
2803 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2804 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2805 #endif
2806
2807 static void sighand_ctor(void *data)
2808 {
2809         struct sighand_struct *sighand = data;
2810
2811         spin_lock_init(&sighand->siglock);
2812         init_waitqueue_head(&sighand->signalfd_wqh);
2813 }
2814
2815 void __init proc_caches_init(void)
2816 {
2817         unsigned int mm_size;
2818
2819         sighand_cachep = kmem_cache_create("sighand_cache",
2820                         sizeof(struct sighand_struct), 0,
2821                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2822                         SLAB_ACCOUNT, sighand_ctor);
2823         signal_cachep = kmem_cache_create("signal_cache",
2824                         sizeof(struct signal_struct), 0,
2825                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2826                         NULL);
2827         files_cachep = kmem_cache_create("files_cache",
2828                         sizeof(struct files_struct), 0,
2829                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2830                         NULL);
2831         fs_cachep = kmem_cache_create("fs_cache",
2832                         sizeof(struct fs_struct), 0,
2833                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2834                         NULL);
2835
2836         /*
2837          * The mm_cpumask is located at the end of mm_struct, and is
2838          * dynamically sized based on the maximum CPU number this system
2839          * can have, taking hotplug into account (nr_cpu_ids).
2840          */
2841         mm_size = sizeof(struct mm_struct) + cpumask_size();
2842
2843         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2844                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2845                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2846                         offsetof(struct mm_struct, saved_auxv),
2847                         sizeof_field(struct mm_struct, saved_auxv),
2848                         NULL);
2849         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2850         mmap_init();
2851         nsproxy_cache_init();
2852 }
2853
2854 /*
2855  * Check constraints on flags passed to the unshare system call.
2856  */
2857 static int check_unshare_flags(unsigned long unshare_flags)
2858 {
2859         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2860                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2861                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2862                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2863                                 CLONE_NEWTIME))
2864                 return -EINVAL;
2865         /*
2866          * Not implemented, but pretend it works if there is nothing
2867          * to unshare.  Note that unsharing the address space or the
2868          * signal handlers also need to unshare the signal queues (aka
2869          * CLONE_THREAD).
2870          */
2871         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2872                 if (!thread_group_empty(current))
2873                         return -EINVAL;
2874         }
2875         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2876                 if (refcount_read(&current->sighand->count) > 1)
2877                         return -EINVAL;
2878         }
2879         if (unshare_flags & CLONE_VM) {
2880                 if (!current_is_single_threaded())
2881                         return -EINVAL;
2882         }
2883
2884         return 0;
2885 }
2886
2887 /*
2888  * Unshare the filesystem structure if it is being shared
2889  */
2890 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2891 {
2892         struct fs_struct *fs = current->fs;
2893
2894         if (!(unshare_flags & CLONE_FS) || !fs)
2895                 return 0;
2896
2897         /* don't need lock here; in the worst case we'll do useless copy */
2898         if (fs->users == 1)
2899                 return 0;
2900
2901         *new_fsp = copy_fs_struct(fs);
2902         if (!*new_fsp)
2903                 return -ENOMEM;
2904
2905         return 0;
2906 }
2907
2908 /*
2909  * Unshare file descriptor table if it is being shared
2910  */
2911 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2912 {
2913         struct files_struct *fd = current->files;
2914         int error = 0;
2915
2916         if ((unshare_flags & CLONE_FILES) &&
2917             (fd && atomic_read(&fd->count) > 1)) {
2918                 *new_fdp = dup_fd(fd, &error);
2919                 if (!*new_fdp)
2920                         return error;
2921         }
2922
2923         return 0;
2924 }
2925
2926 /*
2927  * unshare allows a process to 'unshare' part of the process
2928  * context which was originally shared using clone.  copy_*
2929  * functions used by do_fork() cannot be used here directly
2930  * because they modify an inactive task_struct that is being
2931  * constructed. Here we are modifying the current, active,
2932  * task_struct.
2933  */
2934 int ksys_unshare(unsigned long unshare_flags)
2935 {
2936         struct fs_struct *fs, *new_fs = NULL;
2937         struct files_struct *fd, *new_fd = NULL;
2938         struct cred *new_cred = NULL;
2939         struct nsproxy *new_nsproxy = NULL;
2940         int do_sysvsem = 0;
2941         int err;
2942
2943         /*
2944          * If unsharing a user namespace must also unshare the thread group
2945          * and unshare the filesystem root and working directories.
2946          */
2947         if (unshare_flags & CLONE_NEWUSER)
2948                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2949         /*
2950          * If unsharing vm, must also unshare signal handlers.
2951          */
2952         if (unshare_flags & CLONE_VM)
2953                 unshare_flags |= CLONE_SIGHAND;
2954         /*
2955          * If unsharing a signal handlers, must also unshare the signal queues.
2956          */
2957         if (unshare_flags & CLONE_SIGHAND)
2958                 unshare_flags |= CLONE_THREAD;
2959         /*
2960          * If unsharing namespace, must also unshare filesystem information.
2961          */
2962         if (unshare_flags & CLONE_NEWNS)
2963                 unshare_flags |= CLONE_FS;
2964
2965         err = check_unshare_flags(unshare_flags);
2966         if (err)
2967                 goto bad_unshare_out;
2968         /*
2969          * CLONE_NEWIPC must also detach from the undolist: after switching
2970          * to a new ipc namespace, the semaphore arrays from the old
2971          * namespace are unreachable.
2972          */
2973         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2974                 do_sysvsem = 1;
2975         err = unshare_fs(unshare_flags, &new_fs);
2976         if (err)
2977                 goto bad_unshare_out;
2978         err = unshare_fd(unshare_flags, &new_fd);
2979         if (err)
2980                 goto bad_unshare_cleanup_fs;
2981         err = unshare_userns(unshare_flags, &new_cred);
2982         if (err)
2983                 goto bad_unshare_cleanup_fd;
2984         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2985                                          new_cred, new_fs);
2986         if (err)
2987                 goto bad_unshare_cleanup_cred;
2988
2989         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2990                 if (do_sysvsem) {
2991                         /*
2992                          * CLONE_SYSVSEM is equivalent to sys_exit().
2993                          */
2994                         exit_sem(current);
2995                 }
2996                 if (unshare_flags & CLONE_NEWIPC) {
2997                         /* Orphan segments in old ns (see sem above). */
2998                         exit_shm(current);
2999                         shm_init_task(current);
3000                 }
3001
3002                 if (new_nsproxy)
3003                         switch_task_namespaces(current, new_nsproxy);
3004
3005                 task_lock(current);
3006
3007                 if (new_fs) {
3008                         fs = current->fs;
3009                         spin_lock(&fs->lock);
3010                         current->fs = new_fs;
3011                         if (--fs->users)
3012                                 new_fs = NULL;
3013                         else
3014                                 new_fs = fs;
3015                         spin_unlock(&fs->lock);
3016                 }
3017
3018                 if (new_fd) {
3019                         fd = current->files;
3020                         current->files = new_fd;
3021                         new_fd = fd;
3022                 }
3023
3024                 task_unlock(current);
3025
3026                 if (new_cred) {
3027                         /* Install the new user namespace */
3028                         commit_creds(new_cred);
3029                         new_cred = NULL;
3030                 }
3031         }
3032
3033         perf_event_namespaces(current);
3034
3035 bad_unshare_cleanup_cred:
3036         if (new_cred)
3037                 put_cred(new_cred);
3038 bad_unshare_cleanup_fd:
3039         if (new_fd)
3040                 put_files_struct(new_fd);
3041
3042 bad_unshare_cleanup_fs:
3043         if (new_fs)
3044                 free_fs_struct(new_fs);
3045
3046 bad_unshare_out:
3047         return err;
3048 }
3049
3050 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3051 {
3052         return ksys_unshare(unshare_flags);
3053 }
3054
3055 /*
3056  *      Helper to unshare the files of the current task.
3057  *      We don't want to expose copy_files internals to
3058  *      the exec layer of the kernel.
3059  */
3060
3061 int unshare_files(struct files_struct **displaced)
3062 {
3063         struct task_struct *task = current;
3064         struct files_struct *copy = NULL;
3065         int error;
3066
3067         error = unshare_fd(CLONE_FILES, &copy);
3068         if (error || !copy) {
3069                 *displaced = NULL;
3070                 return error;
3071         }
3072         *displaced = task->files;
3073         task_lock(task);
3074         task->files = copy;
3075         task_unlock(task);
3076         return 0;
3077 }
3078
3079 int sysctl_max_threads(struct ctl_table *table, int write,
3080                        void __user *buffer, size_t *lenp, loff_t *ppos)
3081 {
3082         struct ctl_table t;
3083         int ret;
3084         int threads = max_threads;
3085         int min = 1;
3086         int max = MAX_THREADS;
3087
3088         t = *table;
3089         t.data = &threads;
3090         t.extra1 = &min;
3091         t.extra2 = &max;
3092
3093         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3094         if (ret || !write)
3095                 return ret;
3096
3097         max_threads = threads;
3098
3099         return 0;
3100 }