1/*
2 * linux/kernel/fork.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * 'fork.c' contains the help-routines for the 'fork' system call
9 * (see also entry.S and others).
10 * Fork is rather simple, once you get the hang of it, but the memory
11 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12 */
13
14#include <linux/slab.h>
15#include <linux/sched/autogroup.h>
16#include <linux/sched/mm.h>
17#include <linux/sched/coredump.h>
18#include <linux/sched/user.h>
19#include <linux/sched/numa_balancing.h>
20#include <linux/sched/stat.h>
21#include <linux/sched/task.h>
22#include <linux/sched/task_stack.h>
23#include <linux/sched/cputime.h>
24#include <linux/rtmutex.h>
25#include <linux/init.h>
26#include <linux/unistd.h>
27#include <linux/module.h>
28#include <linux/vmalloc.h>
29#include <linux/completion.h>
30#include <linux/personality.h>
31#include <linux/mempolicy.h>
32#include <linux/sem.h>
33#include <linux/file.h>
34#include <linux/fdtable.h>
35#include <linux/iocontext.h>
36#include <linux/key.h>
37#include <linux/binfmts.h>
38#include <linux/mman.h>
39#include <linux/mmu_notifier.h>
40#include <linux/hmm.h>
41#include <linux/fs.h>
42#include <linux/mm.h>
43#include <linux/vmacache.h>
44#include <linux/nsproxy.h>
45#include <linux/capability.h>
46#include <linux/cpu.h>
47#include <linux/cgroup.h>
48#include <linux/security.h>
49#include <linux/hugetlb.h>
50#include <linux/seccomp.h>
51#include <linux/swap.h>
52#include <linux/syscalls.h>
53#include <linux/jiffies.h>
54#include <linux/futex.h>
55#include <linux/compat.h>
56#include <linux/kthread.h>
57#include <linux/task_io_accounting_ops.h>
58#include <linux/rcupdate.h>
59#include <linux/ptrace.h>
60#include <linux/mount.h>
61#include <linux/audit.h>
62#include <linux/memcontrol.h>
63#include <linux/ftrace.h>
64#include <linux/proc_fs.h>
65#include <linux/profile.h>
66#include <linux/rmap.h>
67#include <linux/ksm.h>
68#include <linux/acct.h>
69#include <linux/userfaultfd_k.h>
70#include <linux/tsacct_kern.h>
71#include <linux/cn_proc.h>
72#include <linux/freezer.h>
73#include <linux/delayacct.h>
74#include <linux/taskstats_kern.h>
75#include <linux/random.h>
76#include <linux/tty.h>
77#include <linux/blkdev.h>
78#include <linux/fs_struct.h>
79#include <linux/magic.h>
80#include <linux/sched/mm.h>
81#include <linux/perf_event.h>
82#include <linux/posix-timers.h>
83#include <linux/user-return-notifier.h>
84#include <linux/oom.h>
85#include <linux/khugepaged.h>
86#include <linux/signalfd.h>
87#include <linux/uprobes.h>
88#include <linux/aio.h>
89#include <linux/compiler.h>
90#include <linux/sysctl.h>
91#include <linux/kcov.h>
92#include <linux/livepatch.h>
93#include <linux/thread_info.h>
94
95#include <asm/pgtable.h>
96#include <asm/pgalloc.h>
97#include <linux/uaccess.h>
98#include <asm/mmu_context.h>
99#include <asm/cacheflush.h>
100#include <asm/tlbflush.h>
101
102#include <trace/events/sched.h>
103
104#define CREATE_TRACE_POINTS
105#include <trace/events/task.h>
106
107/*
108 * Minimum number of threads to boot the kernel
109 */
110#define MIN_THREADS 20
111
112/*
113 * Maximum number of threads
114 */
115#define MAX_THREADS FUTEX_TID_MASK
116
117/*
118 * Protected counters by write_lock_irq(&tasklist_lock)
119 */
120unsigned long total_forks; /* Handle normal Linux uptimes. */
121int nr_threads; /* The idle threads do not count.. */
122
123int max_threads; /* tunable limit on nr_threads */
124
125DEFINE_PER_CPU(unsigned long, process_counts) = 0;
126
127__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
128
129#ifdef CONFIG_PROVE_RCU
130int lockdep_tasklist_lock_is_held(void)
131{
132 return lockdep_is_held(&tasklist_lock);
133}
134EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
135#endif /* #ifdef CONFIG_PROVE_RCU */
136
137int nr_processes(void)
138{
139 int cpu;
140 int total = 0;
141
142 for_each_possible_cpu(cpu)
143 total += per_cpu(process_counts, cpu);
144
145 return total;
146}
147
148void __weak arch_release_task_struct(struct task_struct *tsk)
149{
150}
151
152#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
153static struct kmem_cache *task_struct_cachep;
154
155static inline struct task_struct *alloc_task_struct_node(int node)
156{
157 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
158}
159
160static inline void free_task_struct(struct task_struct *tsk)
161{
162 kmem_cache_free(task_struct_cachep, tsk);
163}
164#endif
165
166void __weak arch_release_thread_stack(unsigned long *stack)
167{
168}
169
170#ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
171
172/*
173 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
174 * kmemcache based allocator.
175 */
176# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
177
178#ifdef CONFIG_VMAP_STACK
179/*
180 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
181 * flush. Try to minimize the number of calls by caching stacks.
182 */
183#define NR_CACHED_STACKS 2
184static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
185
186static int free_vm_stack_cache(unsigned int cpu)
187{
188 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
189 int i;
190
191 for (i = 0; i < NR_CACHED_STACKS; i++) {
192 struct vm_struct *vm_stack = cached_vm_stacks[i];
193
194 if (!vm_stack)
195 continue;
196
197 vfree(vm_stack->addr);
198 cached_vm_stacks[i] = NULL;
199 }
200
201 return 0;
202}
203#endif
204
205static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
206{
207#ifdef CONFIG_VMAP_STACK
208 void *stack;
209 int i;
210
211 for (i = 0; i < NR_CACHED_STACKS; i++) {
212 struct vm_struct *s;
213
214 s = this_cpu_xchg(cached_stacks[i], NULL);
215
216 if (!s)
217 continue;
218
219 /* Clear stale pointers from reused stack. */
220 memset(s->addr, 0, THREAD_SIZE);
221
222 tsk->stack_vm_area = s;
223 return s->addr;
224 }
225
226 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
227 VMALLOC_START, VMALLOC_END,
228 THREADINFO_GFP,
229 PAGE_KERNEL,
230 0, node, __builtin_return_address(0));
231
232 /*
233 * We can't call find_vm_area() in interrupt context, and
234 * free_thread_stack() can be called in interrupt context,
235 * so cache the vm_struct.
236 */
237 if (stack)
238 tsk->stack_vm_area = find_vm_area(stack);
239 return stack;
240#else
241 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
242 THREAD_SIZE_ORDER);
243
244 return page ? page_address(page) : NULL;
245#endif
246}
247
248static inline void free_thread_stack(struct task_struct *tsk)
249{
250#ifdef CONFIG_VMAP_STACK
251 if (task_stack_vm_area(tsk)) {
252 int i;
253
254 for (i = 0; i < NR_CACHED_STACKS; i++) {
255 if (this_cpu_cmpxchg(cached_stacks[i],
256 NULL, tsk->stack_vm_area) != NULL)
257 continue;
258
259 return;
260 }
261
262 vfree_atomic(tsk->stack);
263 return;
264 }
265#endif
266
267 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
268}
269# else
270static struct kmem_cache *thread_stack_cache;
271
272static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
273 int node)
274{
275 return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
276}
277
278static void free_thread_stack(struct task_struct *tsk)
279{
280 kmem_cache_free(thread_stack_cache, tsk->stack);
281}
282
283void thread_stack_cache_init(void)
284{
285 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
286 THREAD_SIZE, THREAD_SIZE, 0, 0,
287 THREAD_SIZE, NULL);
288 BUG_ON(thread_stack_cache == NULL);
289}
290# endif
291#endif
292
293/* SLAB cache for signal_struct structures (tsk->signal) */
294static struct kmem_cache *signal_cachep;
295
296/* SLAB cache for sighand_struct structures (tsk->sighand) */
297struct kmem_cache *sighand_cachep;
298
299/* SLAB cache for files_struct structures (tsk->files) */
300struct kmem_cache *files_cachep;
301
302/* SLAB cache for fs_struct structures (tsk->fs) */
303struct kmem_cache *fs_cachep;
304
305/* SLAB cache for vm_area_struct structures */
306static struct kmem_cache *vm_area_cachep;
307
308/* SLAB cache for mm_struct structures (tsk->mm) */
309static struct kmem_cache *mm_cachep;
310
311struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
312{
313 struct vm_area_struct *vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
314
315 if (vma)
316 vma_init(vma, mm);
317 return vma;
318}
319
320struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
321{
322 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
323
324 if (new) {
325 *new = *orig;
326 INIT_LIST_HEAD(&new->anon_vma_chain);
327 }
328 return new;
329}
330
331void vm_area_free(struct vm_area_struct *vma)
332{
333 kmem_cache_free(vm_area_cachep, vma);
334}
335
336static void account_kernel_stack(struct task_struct *tsk, int account)
337{
338 void *stack = task_stack_page(tsk);
339 struct vm_struct *vm = task_stack_vm_area(tsk);
340
341 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
342
343 if (vm) {
344 int i;
345
346 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
347
348 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
349 mod_zone_page_state(page_zone(vm->pages[i]),
350 NR_KERNEL_STACK_KB,
351 PAGE_SIZE / 1024 * account);
352 }
353
354 /* All stack pages belong to the same memcg. */
355 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
356 account * (THREAD_SIZE / 1024));
357 } else {
358 /*
359 * All stack pages are in the same zone and belong to the
360 * same memcg.
361 */
362 struct page *first_page = virt_to_page(stack);
363
364 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
365 THREAD_SIZE / 1024 * account);
366
367 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
368 account * (THREAD_SIZE / 1024));
369 }
370}
371
372static void release_task_stack(struct task_struct *tsk)
373{
374 if (WARN_ON(tsk->state != TASK_DEAD))
375 return; /* Better to leak the stack than to free prematurely */
376
377 account_kernel_stack(tsk, -1);
378 arch_release_thread_stack(tsk->stack);
379 free_thread_stack(tsk);
380 tsk->stack = NULL;
381#ifdef CONFIG_VMAP_STACK
382 tsk->stack_vm_area = NULL;
383#endif
384}
385
386#ifdef CONFIG_THREAD_INFO_IN_TASK
387void put_task_stack(struct task_struct *tsk)
388{
389 if (atomic_dec_and_test(&tsk->stack_refcount))
390 release_task_stack(tsk);
391}
392#endif
393
394void free_task(struct task_struct *tsk)
395{
396#ifndef CONFIG_THREAD_INFO_IN_TASK
397 /*
398 * The task is finally done with both the stack and thread_info,
399 * so free both.
400 */
401 release_task_stack(tsk);
402#else
403 /*
404 * If the task had a separate stack allocation, it should be gone
405 * by now.
406 */
407 WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
408#endif
409 rt_mutex_debug_task_free(tsk);
410 ftrace_graph_exit_task(tsk);
411 put_seccomp_filter(tsk);
412 arch_release_task_struct(tsk);
413 if (tsk->flags & PF_KTHREAD)
414 free_kthread_struct(tsk);
415 free_task_struct(tsk);
416}
417EXPORT_SYMBOL(free_task);
418
419#ifdef CONFIG_MMU
420static __latent_entropy int dup_mmap(struct mm_struct *mm,
421 struct mm_struct *oldmm)
422{
423 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
424 struct rb_node **rb_link, *rb_parent;
425 int retval;
426 unsigned long charge;
427 LIST_HEAD(uf);
428
429 uprobe_start_dup_mmap();
430 if (down_write_killable(&oldmm->mmap_sem)) {
431 retval = -EINTR;
432 goto fail_uprobe_end;
433 }
434 flush_cache_dup_mm(oldmm);
435 uprobe_dup_mmap(oldmm, mm);
436 /*
437 * Not linked in yet - no deadlock potential:
438 */
439 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
440
441 /* No ordering required: file already has been exposed. */
442 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
443
444 mm->total_vm = oldmm->total_vm;
445 mm->data_vm = oldmm->data_vm;
446 mm->exec_vm = oldmm->exec_vm;
447 mm->stack_vm = oldmm->stack_vm;
448
449 rb_link = &mm->mm_rb.rb_node;
450 rb_parent = NULL;
451 pprev = &mm->mmap;
452 retval = ksm_fork(mm, oldmm);
453 if (retval)
454 goto out;
455 retval = khugepaged_fork(mm, oldmm);
456 if (retval)
457 goto out;
458
459 prev = NULL;
460 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
461 struct file *file;
462
463 if (mpnt->vm_flags & VM_DONTCOPY) {
464 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
465 continue;
466 }
467 charge = 0;
468 /*
469 * Don't duplicate many vmas if we've been oom-killed (for
470 * example)
471 */
472 if (fatal_signal_pending(current)) {
473 retval = -EINTR;
474 goto out;
475 }
476 if (mpnt->vm_flags & VM_ACCOUNT) {
477 unsigned long len = vma_pages(mpnt);
478
479 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
480 goto fail_nomem;
481 charge = len;
482 }
483 tmp = vm_area_dup(mpnt);
484 if (!tmp)
485 goto fail_nomem;
486 retval = vma_dup_policy(mpnt, tmp);
487 if (retval)
488 goto fail_nomem_policy;
489 tmp->vm_mm = mm;
490 retval = dup_userfaultfd(tmp, &uf);
491 if (retval)
492 goto fail_nomem_anon_vma_fork;
493 if (tmp->vm_flags & VM_WIPEONFORK) {
494 /* VM_WIPEONFORK gets a clean slate in the child. */
495 tmp->anon_vma = NULL;
496 if (anon_vma_prepare(tmp))
497 goto fail_nomem_anon_vma_fork;
498 } else if (anon_vma_fork(tmp, mpnt))
499 goto fail_nomem_anon_vma_fork;
500 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
501 tmp->vm_next = tmp->vm_prev = NULL;
502 file = tmp->vm_file;
503 if (file) {
504 struct inode *inode = file_inode(file);
505 struct address_space *mapping = file->f_mapping;
506
507 get_file(file);
508 if (tmp->vm_flags & VM_DENYWRITE)
509 atomic_dec(&inode->i_writecount);
510 i_mmap_lock_write(mapping);
511 if (tmp->vm_flags & VM_SHARED)
512 atomic_inc(&mapping->i_mmap_writable);
513 flush_dcache_mmap_lock(mapping);
514 /* insert tmp into the share list, just after mpnt */
515 vma_interval_tree_insert_after(tmp, mpnt,
516 &mapping->i_mmap);
517 flush_dcache_mmap_unlock(mapping);
518 i_mmap_unlock_write(mapping);
519 }
520
521 /*
522 * Clear hugetlb-related page reserves for children. This only
523 * affects MAP_PRIVATE mappings. Faults generated by the child
524 * are not guaranteed to succeed, even if read-only
525 */
526 if (is_vm_hugetlb_page(tmp))
527 reset_vma_resv_huge_pages(tmp);
528
529 /*
530 * Link in the new vma and copy the page table entries.
531 */
532 *pprev = tmp;
533 pprev = &tmp->vm_next;
534 tmp->vm_prev = prev;
535 prev = tmp;
536
537 __vma_link_rb(mm, tmp, rb_link, rb_parent);
538 rb_link = &tmp->vm_rb.rb_right;
539 rb_parent = &tmp->vm_rb;
540
541 mm->map_count++;
542 if (!(tmp->vm_flags & VM_WIPEONFORK))
543 retval = copy_page_range(mm, oldmm, mpnt);
544
545 if (tmp->vm_ops && tmp->vm_ops->open)
546 tmp->vm_ops->open(tmp);
547
548 if (retval)
549 goto out;
550 }
551 /* a new mm has just been created */
552 arch_dup_mmap(oldmm, mm);
553 retval = 0;
554out:
555 up_write(&mm->mmap_sem);
556 flush_tlb_mm(oldmm);
557 up_write(&oldmm->mmap_sem);
558 dup_userfaultfd_complete(&uf);
559fail_uprobe_end:
560 uprobe_end_dup_mmap();
561 return retval;
562fail_nomem_anon_vma_fork:
563 mpol_put(vma_policy(tmp));
564fail_nomem_policy:
565 vm_area_free(tmp);
566fail_nomem:
567 retval = -ENOMEM;
568 vm_unacct_memory(charge);
569 goto out;
570}
571
572static inline int mm_alloc_pgd(struct mm_struct *mm)
573{
574 mm->pgd = pgd_alloc(mm);
575 if (unlikely(!mm->pgd))
576 return -ENOMEM;
577 return 0;
578}
579
580static inline void mm_free_pgd(struct mm_struct *mm)
581{
582 pgd_free(mm, mm->pgd);
583}
584#else
585static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
586{
587 down_write(&oldmm->mmap_sem);
588 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
589 up_write(&oldmm->mmap_sem);
590 return 0;
591}
592#define mm_alloc_pgd(mm) (0)
593#define mm_free_pgd(mm)
594#endif /* CONFIG_MMU */
595
596static void check_mm(struct mm_struct *mm)
597{
598 int i;
599
600 for (i = 0; i < NR_MM_COUNTERS; i++) {
601 long x = atomic_long_read(&mm->rss_stat.count[i]);
602
603 if (unlikely(x))
604 printk(KERN_ALERT "BUG: Bad rss-counter state "
605 "mm:%p idx:%d val:%ld\n", mm, i, x);
606 }
607
608 if (mm_pgtables_bytes(mm))
609 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
610 mm_pgtables_bytes(mm));
611
612#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
613 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
614#endif
615}
616
617#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
618#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
619
620/*
621 * Called when the last reference to the mm
622 * is dropped: either by a lazy thread or by
623 * mmput. Free the page directory and the mm.
624 */
625void __mmdrop(struct mm_struct *mm)
626{
627 BUG_ON(mm == &init_mm);
628 WARN_ON_ONCE(mm == current->mm);
629 WARN_ON_ONCE(mm == current->active_mm);
630 mm_free_pgd(mm);
631 destroy_context(mm);
632 hmm_mm_destroy(mm);
633 mmu_notifier_mm_destroy(mm);
634 check_mm(mm);
635 put_user_ns(mm->user_ns);
636 free_mm(mm);
637}
638EXPORT_SYMBOL_GPL(__mmdrop);
639
640static void mmdrop_async_fn(struct work_struct *work)
641{
642 struct mm_struct *mm;
643
644 mm = container_of(work, struct mm_struct, async_put_work);
645 __mmdrop(mm);
646}
647
648static void mmdrop_async(struct mm_struct *mm)
649{
650 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
651 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
652 schedule_work(&mm->async_put_work);
653 }
654}
655
656static inline void free_signal_struct(struct signal_struct *sig)
657{
658 taskstats_tgid_free(sig);
659 sched_autogroup_exit(sig);
660 /*
661 * __mmdrop is not safe to call from softirq context on x86 due to
662 * pgd_dtor so postpone it to the async context
663 */
664 if (sig->oom_mm)
665 mmdrop_async(sig->oom_mm);
666 kmem_cache_free(signal_cachep, sig);
667}
668
669static inline void put_signal_struct(struct signal_struct *sig)
670{
671 if (atomic_dec_and_test(&sig->sigcnt))
672 free_signal_struct(sig);
673}
674
675void __put_task_struct(struct task_struct *tsk)
676{
677 WARN_ON(!tsk->exit_state);
678 WARN_ON(atomic_read(&tsk->usage));
679 WARN_ON(tsk == current);
680
681 cgroup_free(tsk);
682 task_numa_free(tsk);
683 security_task_free(tsk);
684 exit_creds(tsk);
685 delayacct_tsk_free(tsk);
686 put_signal_struct(tsk->signal);
687
688 if (!profile_handoff_task(tsk))
689 free_task(tsk);
690}
691EXPORT_SYMBOL_GPL(__put_task_struct);
692
693void __init __weak arch_task_cache_init(void) { }
694
695/*
696 * set_max_threads
697 */
698static void set_max_threads(unsigned int max_threads_suggested)
699{
700 u64 threads;
701
702 /*
703 * The number of threads shall be limited such that the thread
704 * structures may only consume a small part of the available memory.
705 */
706 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
707 threads = MAX_THREADS;
708 else
709 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
710 (u64) THREAD_SIZE * 8UL);
711
712 if (threads > max_threads_suggested)
713 threads = max_threads_suggested;
714
715 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
716}
717
718#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
719/* Initialized by the architecture: */
720int arch_task_struct_size __read_mostly;
721#endif
722
723static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
724{
725 /* Fetch thread_struct whitelist for the architecture. */
726 arch_thread_struct_whitelist(offset, size);
727
728 /*
729 * Handle zero-sized whitelist or empty thread_struct, otherwise
730 * adjust offset to position of thread_struct in task_struct.
731 */
732 if (unlikely(*size == 0))
733 *offset = 0;
734 else
735 *offset += offsetof(struct task_struct, thread);
736}
737
738void __init fork_init(void)
739{
740 int i;
741#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
742#ifndef ARCH_MIN_TASKALIGN
743#define ARCH_MIN_TASKALIGN 0
744#endif
745 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
746 unsigned long useroffset, usersize;
747
748 /* create a slab on which task_structs can be allocated */
749 task_struct_whitelist(&useroffset, &usersize);
750 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
751 arch_task_struct_size, align,
752 SLAB_PANIC|SLAB_ACCOUNT,
753 useroffset, usersize, NULL);
754#endif
755
756 /* do the arch specific task caches init */
757 arch_task_cache_init();
758
759 set_max_threads(MAX_THREADS);
760
761 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
762 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
763 init_task.signal->rlim[RLIMIT_SIGPENDING] =
764 init_task.signal->rlim[RLIMIT_NPROC];
765
766 for (i = 0; i < UCOUNT_COUNTS; i++) {
767 init_user_ns.ucount_max[i] = max_threads/2;
768 }
769
770#ifdef CONFIG_VMAP_STACK
771 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
772 NULL, free_vm_stack_cache);
773#endif
774
775 lockdep_init_task(&init_task);
776}
777
778int __weak arch_dup_task_struct(struct task_struct *dst,
779 struct task_struct *src)
780{
781 *dst = *src;
782 return 0;
783}
784
785void set_task_stack_end_magic(struct task_struct *tsk)
786{
787 unsigned long *stackend;
788
789 stackend = end_of_stack(tsk);
790 *stackend = STACK_END_MAGIC; /* for overflow detection */
791}
792
793static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
794{
795 struct task_struct *tsk;
796 unsigned long *stack;
797 struct vm_struct *stack_vm_area;
798 int err;
799
800 if (node == NUMA_NO_NODE)
801 node = tsk_fork_get_node(orig);
802 tsk = alloc_task_struct_node(node);
803 if (!tsk)
804 return NULL;
805
806 stack = alloc_thread_stack_node(tsk, node);
807 if (!stack)
808 goto free_tsk;
809
810 stack_vm_area = task_stack_vm_area(tsk);
811
812 err = arch_dup_task_struct(tsk, orig);
813
814 /*
815 * arch_dup_task_struct() clobbers the stack-related fields. Make
816 * sure they're properly initialized before using any stack-related
817 * functions again.
818 */
819 tsk->stack = stack;
820#ifdef CONFIG_VMAP_STACK
821 tsk->stack_vm_area = stack_vm_area;
822#endif
823#ifdef CONFIG_THREAD_INFO_IN_TASK
824 atomic_set(&tsk->stack_refcount, 1);
825#endif
826
827 if (err)
828 goto free_stack;
829
830#ifdef CONFIG_SECCOMP
831 /*
832 * We must handle setting up seccomp filters once we're under
833 * the sighand lock in case orig has changed between now and
834 * then. Until then, filter must be NULL to avoid messing up
835 * the usage counts on the error path calling free_task.
836 */
837 tsk->seccomp.filter = NULL;
838#endif
839
840 setup_thread_stack(tsk, orig);
841 clear_user_return_notifier(tsk);
842 clear_tsk_need_resched(tsk);
843 set_task_stack_end_magic(tsk);
844
845#ifdef CONFIG_STACKPROTECTOR
846 tsk->stack_canary = get_random_canary();
847#endif
848
849 /*
850 * One for us, one for whoever does the "release_task()" (usually
851 * parent)
852 */
853 atomic_set(&tsk->usage, 2);
854#ifdef CONFIG_BLK_DEV_IO_TRACE
855 tsk->btrace_seq = 0;
856#endif
857 tsk->splice_pipe = NULL;
858 tsk->task_frag.page = NULL;
859 tsk->wake_q.next = NULL;
860
861 account_kernel_stack(tsk, 1);
862
863 kcov_task_init(tsk);
864
865#ifdef CONFIG_FAULT_INJECTION
866 tsk->fail_nth = 0;
867#endif
868
869#ifdef CONFIG_BLK_CGROUP
870 tsk->throttle_queue = NULL;
871 tsk->use_memdelay = 0;
872#endif
873
874#ifdef CONFIG_MEMCG
875 tsk->active_memcg = NULL;
876#endif
877 return tsk;
878
879free_stack:
880 free_thread_stack(tsk);
881free_tsk:
882 free_task_struct(tsk);
883 return NULL;
884}
885
886__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
887
888static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
889
890static int __init coredump_filter_setup(char *s)
891{
892 default_dump_filter =
893 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
894 MMF_DUMP_FILTER_MASK;
895 return 1;
896}
897
898__setup("coredump_filter=", coredump_filter_setup);
899
900#include <linux/init_task.h>
901
902static void mm_init_aio(struct mm_struct *mm)
903{
904#ifdef CONFIG_AIO
905 spin_lock_init(&mm->ioctx_lock);
906 mm->ioctx_table = NULL;
907#endif
908}
909
910static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
911{
912#ifdef CONFIG_MEMCG
913 mm->owner = p;
914#endif
915}
916
917static void mm_init_uprobes_state(struct mm_struct *mm)
918{
919#ifdef CONFIG_UPROBES
920 mm->uprobes_state.xol_area = NULL;
921#endif
922}
923
924static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
925 struct user_namespace *user_ns)
926{
927 mm->mmap = NULL;
928 mm->mm_rb = RB_ROOT;
929 mm->vmacache_seqnum = 0;
930 atomic_set(&mm->mm_users, 1);
931 atomic_set(&mm->mm_count, 1);
932 init_rwsem(&mm->mmap_sem);
933 INIT_LIST_HEAD(&mm->mmlist);
934 mm->core_state = NULL;
935 mm_pgtables_bytes_init(mm);
936 mm->map_count = 0;
937 mm->locked_vm = 0;
938 mm->pinned_vm = 0;
939 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
940 spin_lock_init(&mm->page_table_lock);
941 spin_lock_init(&mm->arg_lock);
942 mm_init_cpumask(mm);
943 mm_init_aio(mm);
944 mm_init_owner(mm, p);
945 RCU_INIT_POINTER(mm->exe_file, NULL);
946 mmu_notifier_mm_init(mm);
947 hmm_mm_init(mm);
948 init_tlb_flush_pending(mm);
949#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
950 mm->pmd_huge_pte = NULL;
951#endif
952 mm_init_uprobes_state(mm);
953
954 if (current->mm) {
955 mm->flags = current->mm->flags & MMF_INIT_MASK;
956 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
957 } else {
958 mm->flags = default_dump_filter;
959 mm->def_flags = 0;
960 }
961
962 if (mm_alloc_pgd(mm))
963 goto fail_nopgd;
964
965 if (init_new_context(p, mm))
966 goto fail_nocontext;
967
968 mm->user_ns = get_user_ns(user_ns);
969 return mm;
970
971fail_nocontext:
972 mm_free_pgd(mm);
973fail_nopgd:
974 free_mm(mm);
975 return NULL;
976}
977
978/*
979 * Allocate and initialize an mm_struct.
980 */
981struct mm_struct *mm_alloc(void)
982{
983 struct mm_struct *mm;
984
985 mm = allocate_mm();
986 if (!mm)
987 return NULL;
988
989 memset(mm, 0, sizeof(*mm));
990 return mm_init(mm, current, current_user_ns());
991}
992
993static inline void __mmput(struct mm_struct *mm)
994{
995 VM_BUG_ON(atomic_read(&mm->mm_users));
996
997 uprobe_clear_state(mm);
998 exit_aio(mm);
999 ksm_exit(mm);
1000 khugepaged_exit(mm); /* must run before exit_mmap */
1001 exit_mmap(mm);
1002 mm_put_huge_zero_page(mm);
1003 set_mm_exe_file(mm, NULL);
1004 if (!list_empty(&mm->mmlist)) {
1005 spin_lock(&mmlist_lock);
1006 list_del(&mm->mmlist);
1007 spin_unlock(&mmlist_lock);
1008 }
1009 if (mm->binfmt)
1010 module_put(mm->binfmt->module);
1011 mmdrop(mm);
1012}
1013
1014/*
1015 * Decrement the use count and release all resources for an mm.
1016 */
1017void mmput(struct mm_struct *mm)
1018{
1019 might_sleep();
1020
1021 if (atomic_dec_and_test(&mm->mm_users))
1022 __mmput(mm);
1023}
1024EXPORT_SYMBOL_GPL(mmput);
1025
1026#ifdef CONFIG_MMU
1027static void mmput_async_fn(struct work_struct *work)
1028{
1029 struct mm_struct *mm = container_of(work, struct mm_struct,
1030 async_put_work);
1031
1032 __mmput(mm);
1033}
1034
1035void mmput_async(struct mm_struct *mm)
1036{
1037 if (atomic_dec_and_test(&mm->mm_users)) {
1038 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1039 schedule_work(&mm->async_put_work);
1040 }
1041}
1042#endif
1043
1044/**
1045 * set_mm_exe_file - change a reference to the mm's executable file
1046 *
1047 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1048 *
1049 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1050 * invocations: in mmput() nobody alive left, in execve task is single
1051 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1052 * mm->exe_file, but does so without using set_mm_exe_file() in order
1053 * to do avoid the need for any locks.
1054 */
1055void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1056{
1057 struct file *old_exe_file;
1058
1059 /*
1060 * It is safe to dereference the exe_file without RCU as
1061 * this function is only called if nobody else can access
1062 * this mm -- see comment above for justification.
1063 */
1064 old_exe_file = rcu_dereference_raw(mm->exe_file);
1065
1066 if (new_exe_file)
1067 get_file(new_exe_file);
1068 rcu_assign_pointer(mm->exe_file, new_exe_file);
1069 if (old_exe_file)
1070 fput(old_exe_file);
1071}
1072
1073/**
1074 * get_mm_exe_file - acquire a reference to the mm's executable file
1075 *
1076 * Returns %NULL if mm has no associated executable file.
1077 * User must release file via fput().
1078 */
1079struct file *get_mm_exe_file(struct mm_struct *mm)
1080{
1081 struct file *exe_file;
1082
1083 rcu_read_lock();
1084 exe_file = rcu_dereference(mm->exe_file);
1085 if (exe_file && !get_file_rcu(exe_file))
1086 exe_file = NULL;
1087 rcu_read_unlock();
1088 return exe_file;
1089}
1090EXPORT_SYMBOL(get_mm_exe_file);
1091
1092/**
1093 * get_task_exe_file - acquire a reference to the task's executable file
1094 *
1095 * Returns %NULL if task's mm (if any) has no associated executable file or
1096 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1097 * User must release file via fput().
1098 */
1099struct file *get_task_exe_file(struct task_struct *task)
1100{
1101 struct file *exe_file = NULL;
1102 struct mm_struct *mm;
1103
1104 task_lock(task);
1105 mm = task->mm;
1106 if (mm) {
1107 if (!(task->flags & PF_KTHREAD))
1108 exe_file = get_mm_exe_file(mm);
1109 }
1110 task_unlock(task);
1111 return exe_file;
1112}
1113EXPORT_SYMBOL(get_task_exe_file);
1114
1115/**
1116 * get_task_mm - acquire a reference to the task's mm
1117 *
1118 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1119 * this kernel workthread has transiently adopted a user mm with use_mm,
1120 * to do its AIO) is not set and if so returns a reference to it, after
1121 * bumping up the use count. User must release the mm via mmput()
1122 * after use. Typically used by /proc and ptrace.
1123 */
1124struct mm_struct *get_task_mm(struct task_struct *task)
1125{
1126 struct mm_struct *mm;
1127
1128 task_lock(task);
1129 mm = task->mm;
1130 if (mm) {
1131 if (task->flags & PF_KTHREAD)
1132 mm = NULL;
1133 else
1134 mmget(mm);
1135 }
1136 task_unlock(task);
1137 return mm;
1138}
1139EXPORT_SYMBOL_GPL(get_task_mm);
1140
1141struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1142{
1143 struct mm_struct *mm;
1144 int err;
1145
1146 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1147 if (err)
1148 return ERR_PTR(err);
1149
1150 mm = get_task_mm(task);
1151 if (mm && mm != current->mm &&
1152 !ptrace_may_access(task, mode)) {
1153 mmput(mm);
1154 mm = ERR_PTR(-EACCES);
1155 }
1156 mutex_unlock(&task->signal->cred_guard_mutex);
1157
1158 return mm;
1159}
1160
1161static void complete_vfork_done(struct task_struct *tsk)
1162{
1163 struct completion *vfork;
1164
1165 task_lock(tsk);
1166 vfork = tsk->vfork_done;
1167 if (likely(vfork)) {
1168 tsk->vfork_done = NULL;
1169 complete(vfork);
1170 }
1171 task_unlock(tsk);
1172}
1173
1174static int wait_for_vfork_done(struct task_struct *child,
1175 struct completion *vfork)
1176{
1177 int killed;
1178
1179 freezer_do_not_count();
1180 killed = wait_for_completion_killable(vfork);
1181 freezer_count();
1182
1183 if (killed) {
1184 task_lock(child);
1185 child->vfork_done = NULL;
1186 task_unlock(child);
1187 }
1188
1189 put_task_struct(child);
1190 return killed;
1191}
1192
1193/* Please note the differences between mmput and mm_release.
1194 * mmput is called whenever we stop holding onto a mm_struct,
1195 * error success whatever.
1196 *
1197 * mm_release is called after a mm_struct has been removed
1198 * from the current process.
1199 *
1200 * This difference is important for error handling, when we
1201 * only half set up a mm_struct for a new process and need to restore
1202 * the old one. Because we mmput the new mm_struct before
1203 * restoring the old one. . .
1204 * Eric Biederman 10 January 1998
1205 */
1206void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1207{
1208 /* Get rid of any futexes when releasing the mm */
1209#ifdef CONFIG_FUTEX
1210 if (unlikely(tsk->robust_list)) {
1211 exit_robust_list(tsk);
1212 tsk->robust_list = NULL;
1213 }
1214#ifdef CONFIG_COMPAT
1215 if (unlikely(tsk->compat_robust_list)) {
1216 compat_exit_robust_list(tsk);
1217 tsk->compat_robust_list = NULL;
1218 }
1219#endif
1220 if (unlikely(!list_empty(&tsk->pi_state_list)))
1221 exit_pi_state_list(tsk);
1222#endif
1223
1224 uprobe_free_utask(tsk);
1225
1226 /* Get rid of any cached register state */
1227 deactivate_mm(tsk, mm);
1228
1229 /*
1230 * Signal userspace if we're not exiting with a core dump
1231 * because we want to leave the value intact for debugging
1232 * purposes.
1233 */
1234 if (tsk->clear_child_tid) {
1235 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1236 atomic_read(&mm->mm_users) > 1) {
1237 /*
1238 * We don't check the error code - if userspace has
1239 * not set up a proper pointer then tough luck.
1240 */
1241 put_user(0, tsk->clear_child_tid);
1242 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1243 1, NULL, NULL, 0, 0);
1244 }
1245 tsk->clear_child_tid = NULL;
1246 }
1247
1248 /*
1249 * All done, finally we can wake up parent and return this mm to him.
1250 * Also kthread_stop() uses this completion for synchronization.
1251 */
1252 if (tsk->vfork_done)
1253 complete_vfork_done(tsk);
1254}
1255
1256/*
1257 * Allocate a new mm structure and copy contents from the
1258 * mm structure of the passed in task structure.
1259 */
1260static struct mm_struct *dup_mm(struct task_struct *tsk)
1261{
1262 struct mm_struct *mm, *oldmm = current->mm;
1263 int err;
1264
1265 mm = allocate_mm();
1266 if (!mm)
1267 goto fail_nomem;
1268
1269 memcpy(mm, oldmm, sizeof(*mm));
1270
1271 if (!mm_init(mm, tsk, mm->user_ns))
1272 goto fail_nomem;
1273
1274 err = dup_mmap(mm, oldmm);
1275 if (err)
1276 goto free_pt;
1277
1278 mm->hiwater_rss = get_mm_rss(mm);
1279 mm->hiwater_vm = mm->total_vm;
1280
1281 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1282 goto free_pt;
1283
1284 return mm;
1285
1286free_pt:
1287 /* don't put binfmt in mmput, we haven't got module yet */
1288 mm->binfmt = NULL;
1289 mmput(mm);
1290
1291fail_nomem:
1292 return NULL;
1293}
1294
1295static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1296{
1297 struct mm_struct *mm, *oldmm;
1298 int retval;
1299
1300 tsk->min_flt = tsk->maj_flt = 0;
1301 tsk->nvcsw = tsk->nivcsw = 0;
1302#ifdef CONFIG_DETECT_HUNG_TASK
1303 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1304#endif
1305
1306 tsk->mm = NULL;
1307 tsk->active_mm = NULL;
1308
1309 /*
1310 * Are we cloning a kernel thread?
1311 *
1312 * We need to steal a active VM for that..
1313 */
1314 oldmm = current->mm;
1315 if (!oldmm)
1316 return 0;
1317
1318 /* initialize the new vmacache entries */
1319 vmacache_flush(tsk);
1320
1321 if (clone_flags & CLONE_VM) {
1322 mmget(oldmm);
1323 mm = oldmm;
1324 goto good_mm;
1325 }
1326
1327 retval = -ENOMEM;
1328 mm = dup_mm(tsk);
1329 if (!mm)
1330 goto fail_nomem;
1331
1332good_mm:
1333 tsk->mm = mm;
1334 tsk->active_mm = mm;
1335 return 0;
1336
1337fail_nomem:
1338 return retval;
1339}
1340
1341static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1342{
1343 struct fs_struct *fs = current->fs;
1344 if (clone_flags & CLONE_FS) {
1345 /* tsk->fs is already what we want */
1346 spin_lock(&fs->lock);
1347 if (fs->in_exec) {
1348 spin_unlock(&fs->lock);
1349 return -EAGAIN;
1350 }
1351 fs->users++;
1352 spin_unlock(&fs->lock);
1353 return 0;
1354 }
1355 tsk->fs = copy_fs_struct(fs);
1356 if (!tsk->fs)
1357 return -ENOMEM;
1358 return 0;
1359}
1360
1361static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1362{
1363 struct files_struct *oldf, *newf;
1364 int error = 0;
1365
1366 /*
1367 * A background process may not have any files ...
1368 */
1369 oldf = current->files;
1370 if (!oldf)
1371 goto out;
1372
1373 if (clone_flags & CLONE_FILES) {
1374 atomic_inc(&oldf->count);
1375 goto out;
1376 }
1377
1378 newf = dup_fd(oldf, &error);
1379 if (!newf)
1380 goto out;
1381
1382 tsk->files = newf;
1383 error = 0;
1384out:
1385 return error;
1386}
1387
1388static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1389{
1390#ifdef CONFIG_BLOCK
1391 struct io_context *ioc = current->io_context;
1392 struct io_context *new_ioc;
1393
1394 if (!ioc)
1395 return 0;
1396 /*
1397 * Share io context with parent, if CLONE_IO is set
1398 */
1399 if (clone_flags & CLONE_IO) {
1400 ioc_task_link(ioc);
1401 tsk->io_context = ioc;
1402 } else if (ioprio_valid(ioc->ioprio)) {
1403 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1404 if (unlikely(!new_ioc))
1405 return -ENOMEM;
1406
1407 new_ioc->ioprio = ioc->ioprio;
1408 put_io_context(new_ioc);
1409 }
1410#endif
1411 return 0;
1412}
1413
1414static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1415{
1416 struct sighand_struct *sig;
1417
1418 if (clone_flags & CLONE_SIGHAND) {
1419 atomic_inc(&current->sighand->count);
1420 return 0;
1421 }
1422 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1423 rcu_assign_pointer(tsk->sighand, sig);
1424 if (!sig)
1425 return -ENOMEM;
1426
1427 atomic_set(&sig->count, 1);
1428 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1429 return 0;
1430}
1431
1432void __cleanup_sighand(struct sighand_struct *sighand)
1433{
1434 if (atomic_dec_and_test(&sighand->count)) {
1435 signalfd_cleanup(sighand);
1436 /*
1437 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1438 * without an RCU grace period, see __lock_task_sighand().
1439 */
1440 kmem_cache_free(sighand_cachep, sighand);
1441 }
1442}
1443
1444#ifdef CONFIG_POSIX_TIMERS
1445/*
1446 * Initialize POSIX timer handling for a thread group.
1447 */
1448static void posix_cpu_timers_init_group(struct signal_struct *sig)
1449{
1450 unsigned long cpu_limit;
1451
1452 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1453 if (cpu_limit != RLIM_INFINITY) {
1454 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1455 sig->cputimer.running = true;
1456 }
1457
1458 /* The timer lists. */
1459 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1460 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1461 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1462}
1463#else
1464static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1465#endif
1466
1467static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1468{
1469 struct signal_struct *sig;
1470
1471 if (clone_flags & CLONE_THREAD)
1472 return 0;
1473
1474 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1475 tsk->signal = sig;
1476 if (!sig)
1477 return -ENOMEM;
1478
1479 sig->nr_threads = 1;
1480 atomic_set(&sig->live, 1);
1481 atomic_set(&sig->sigcnt, 1);
1482
1483 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1484 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1485 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1486
1487 init_waitqueue_head(&sig->wait_chldexit);
1488 sig->curr_target = tsk;
1489 init_sigpending(&sig->shared_pending);
1490 INIT_HLIST_HEAD(&sig->multiprocess);
1491 seqlock_init(&sig->stats_lock);
1492 prev_cputime_init(&sig->prev_cputime);
1493
1494#ifdef CONFIG_POSIX_TIMERS
1495 INIT_LIST_HEAD(&sig->posix_timers);
1496 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1497 sig->real_timer.function = it_real_fn;
1498#endif
1499
1500 task_lock(current->group_leader);
1501 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1502 task_unlock(current->group_leader);
1503
1504 posix_cpu_timers_init_group(sig);
1505
1506 tty_audit_fork(sig);
1507 sched_autogroup_fork(sig);
1508
1509 sig->oom_score_adj = current->signal->oom_score_adj;
1510 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1511
1512 mutex_init(&sig->cred_guard_mutex);
1513
1514 return 0;
1515}
1516
1517static void copy_seccomp(struct task_struct *p)
1518{
1519#ifdef CONFIG_SECCOMP
1520 /*
1521 * Must be called with sighand->lock held, which is common to
1522 * all threads in the group. Holding cred_guard_mutex is not
1523 * needed because this new task is not yet running and cannot
1524 * be racing exec.
1525 */
1526 assert_spin_locked(&current->sighand->siglock);
1527
1528 /* Ref-count the new filter user, and assign it. */
1529 get_seccomp_filter(current);
1530 p->seccomp = current->seccomp;
1531
1532 /*
1533 * Explicitly enable no_new_privs here in case it got set
1534 * between the task_struct being duplicated and holding the
1535 * sighand lock. The seccomp state and nnp must be in sync.
1536 */
1537 if (task_no_new_privs(current))
1538 task_set_no_new_privs(p);
1539
1540 /*
1541 * If the parent gained a seccomp mode after copying thread
1542 * flags and between before we held the sighand lock, we have
1543 * to manually enable the seccomp thread flag here.
1544 */
1545 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1546 set_tsk_thread_flag(p, TIF_SECCOMP);
1547#endif
1548}
1549
1550SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1551{
1552 current->clear_child_tid = tidptr;
1553
1554 return task_pid_vnr(current);
1555}
1556
1557static void rt_mutex_init_task(struct task_struct *p)
1558{
1559 raw_spin_lock_init(&p->pi_lock);
1560#ifdef CONFIG_RT_MUTEXES
1561 p->pi_waiters = RB_ROOT_CACHED;
1562 p->pi_top_task = NULL;
1563 p->pi_blocked_on = NULL;
1564#endif
1565}
1566
1567#ifdef CONFIG_POSIX_TIMERS
1568/*
1569 * Initialize POSIX timer handling for a single task.
1570 */
1571static void posix_cpu_timers_init(struct task_struct *tsk)
1572{
1573 tsk->cputime_expires.prof_exp = 0;
1574 tsk->cputime_expires.virt_exp = 0;
1575 tsk->cputime_expires.sched_exp = 0;
1576 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1577 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1578 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1579}
1580#else
1581static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1582#endif
1583
1584static inline void init_task_pid_links(struct task_struct *task)
1585{
1586 enum pid_type type;
1587
1588 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1589 INIT_HLIST_NODE(&task->pid_links[type]);
1590 }
1591}
1592
1593static inline void
1594init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1595{
1596 if (type == PIDTYPE_PID)
1597 task->thread_pid = pid;
1598 else
1599 task->signal->pids[type] = pid;
1600}
1601
1602static inline void rcu_copy_process(struct task_struct *p)
1603{
1604#ifdef CONFIG_PREEMPT_RCU
1605 p->rcu_read_lock_nesting = 0;
1606 p->rcu_read_unlock_special.s = 0;
1607 p->rcu_blocked_node = NULL;
1608 INIT_LIST_HEAD(&p->rcu_node_entry);
1609#endif /* #ifdef CONFIG_PREEMPT_RCU */
1610#ifdef CONFIG_TASKS_RCU
1611 p->rcu_tasks_holdout = false;
1612 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1613 p->rcu_tasks_idle_cpu = -1;
1614#endif /* #ifdef CONFIG_TASKS_RCU */
1615}
1616
1617/*
1618 * This creates a new process as a copy of the old one,
1619 * but does not actually start it yet.
1620 *
1621 * It copies the registers, and all the appropriate
1622 * parts of the process environment (as per the clone
1623 * flags). The actual kick-off is left to the caller.
1624 */
1625static __latent_entropy struct task_struct *copy_process(
1626 unsigned long clone_flags,
1627 unsigned long stack_start,
1628 unsigned long stack_size,
1629 int __user *child_tidptr,
1630 struct pid *pid,
1631 int trace,
1632 unsigned long tls,
1633 int node)
1634{
1635 int retval;
1636 struct task_struct *p;
1637 struct multiprocess_signals delayed;
1638
1639 /*
1640 * Don't allow sharing the root directory with processes in a different
1641 * namespace
1642 */
1643 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1644 return ERR_PTR(-EINVAL);
1645
1646 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1647 return ERR_PTR(-EINVAL);
1648
1649 /*
1650 * Thread groups must share signals as well, and detached threads
1651 * can only be started up within the thread group.
1652 */
1653 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1654 return ERR_PTR(-EINVAL);
1655
1656 /*
1657 * Shared signal handlers imply shared VM. By way of the above,
1658 * thread groups also imply shared VM. Blocking this case allows
1659 * for various simplifications in other code.
1660 */
1661 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1662 return ERR_PTR(-EINVAL);
1663
1664 /*
1665 * Siblings of global init remain as zombies on exit since they are
1666 * not reaped by their parent (swapper). To solve this and to avoid
1667 * multi-rooted process trees, prevent global and container-inits
1668 * from creating siblings.
1669 */
1670 if ((clone_flags & CLONE_PARENT) &&
1671 current->signal->flags & SIGNAL_UNKILLABLE)
1672 return ERR_PTR(-EINVAL);
1673
1674 /*
1675 * If the new process will be in a different pid or user namespace
1676 * do not allow it to share a thread group with the forking task.
1677 */
1678 if (clone_flags & CLONE_THREAD) {
1679 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1680 (task_active_pid_ns(current) !=
1681 current->nsproxy->pid_ns_for_children))
1682 return ERR_PTR(-EINVAL);
1683 }
1684
1685 /*
1686 * Force any signals received before this point to be delivered
1687 * before the fork happens. Collect up signals sent to multiple
1688 * processes that happen during the fork and delay them so that
1689 * they appear to happen after the fork.
1690 */
1691 sigemptyset(&delayed.signal);
1692 INIT_HLIST_NODE(&delayed.node);
1693
1694 spin_lock_irq(&current->sighand->siglock);
1695 if (!(clone_flags & CLONE_THREAD))
1696 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1697 recalc_sigpending();
1698 spin_unlock_irq(&current->sighand->siglock);
1699 retval = -ERESTARTNOINTR;
1700 if (signal_pending(current))
1701 goto fork_out;
1702
1703 retval = -ENOMEM;
1704 p = dup_task_struct(current, node);
1705 if (!p)
1706 goto fork_out;
1707
1708 /*
1709 * This _must_ happen before we call free_task(), i.e. before we jump
1710 * to any of the bad_fork_* labels. This is to avoid freeing
1711 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1712 * kernel threads (PF_KTHREAD).
1713 */
1714 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1715 /*
1716 * Clear TID on mm_release()?
1717 */
1718 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1719
1720 ftrace_graph_init_task(p);
1721
1722 rt_mutex_init_task(p);
1723
1724#ifdef CONFIG_PROVE_LOCKING
1725 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1726 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1727#endif
1728 retval = -EAGAIN;
1729 if (atomic_read(&p->real_cred->user->processes) >=
1730 task_rlimit(p, RLIMIT_NPROC)) {
1731 if (p->real_cred->user != INIT_USER &&
1732 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1733 goto bad_fork_free;
1734 }
1735 current->flags &= ~PF_NPROC_EXCEEDED;
1736
1737 retval = copy_creds(p, clone_flags);
1738 if (retval < 0)
1739 goto bad_fork_free;
1740
1741 /*
1742 * If multiple threads are within copy_process(), then this check
1743 * triggers too late. This doesn't hurt, the check is only there
1744 * to stop root fork bombs.
1745 */
1746 retval = -EAGAIN;
1747 if (nr_threads >= max_threads)
1748 goto bad_fork_cleanup_count;
1749
1750 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1751 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1752 p->flags |= PF_FORKNOEXEC;
1753 INIT_LIST_HEAD(&p->children);
1754 INIT_LIST_HEAD(&p->sibling);
1755 rcu_copy_process(p);
1756 p->vfork_done = NULL;
1757 spin_lock_init(&p->alloc_lock);
1758
1759 init_sigpending(&p->pending);
1760
1761 p->utime = p->stime = p->gtime = 0;
1762#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1763 p->utimescaled = p->stimescaled = 0;
1764#endif
1765 prev_cputime_init(&p->prev_cputime);
1766
1767#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1768 seqcount_init(&p->vtime.seqcount);
1769 p->vtime.starttime = 0;
1770 p->vtime.state = VTIME_INACTIVE;
1771#endif
1772
1773#if defined(SPLIT_RSS_COUNTING)
1774 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1775#endif
1776
1777 p->default_timer_slack_ns = current->timer_slack_ns;
1778
1779 task_io_accounting_init(&p->ioac);
1780 acct_clear_integrals(p);
1781
1782 posix_cpu_timers_init(p);
1783
1784 p->start_time = ktime_get_ns();
1785 p->real_start_time = ktime_get_boot_ns();
1786 p->io_context = NULL;
1787 audit_set_context(p, NULL);
1788 cgroup_fork(p);
1789#ifdef CONFIG_NUMA
1790 p->mempolicy = mpol_dup(p->mempolicy);
1791 if (IS_ERR(p->mempolicy)) {
1792 retval = PTR_ERR(p->mempolicy);
1793 p->mempolicy = NULL;
1794 goto bad_fork_cleanup_threadgroup_lock;
1795 }
1796#endif
1797#ifdef CONFIG_CPUSETS
1798 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1799 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1800 seqcount_init(&p->mems_allowed_seq);
1801#endif
1802#ifdef CONFIG_TRACE_IRQFLAGS
1803 p->irq_events = 0;
1804 p->hardirqs_enabled = 0;
1805 p->hardirq_enable_ip = 0;
1806 p->hardirq_enable_event = 0;
1807 p->hardirq_disable_ip = _THIS_IP_;
1808 p->hardirq_disable_event = 0;
1809 p->softirqs_enabled = 1;
1810 p->softirq_enable_ip = _THIS_IP_;
1811 p->softirq_enable_event = 0;
1812 p->softirq_disable_ip = 0;
1813 p->softirq_disable_event = 0;
1814 p->hardirq_context = 0;
1815 p->softirq_context = 0;
1816#endif
1817
1818 p->pagefault_disabled = 0;
1819
1820#ifdef CONFIG_LOCKDEP
1821 p->lockdep_depth = 0; /* no locks held yet */
1822 p->curr_chain_key = 0;
1823 p->lockdep_recursion = 0;
1824 lockdep_init_task(p);
1825#endif
1826
1827#ifdef CONFIG_DEBUG_MUTEXES
1828 p->blocked_on = NULL; /* not blocked yet */
1829#endif
1830#ifdef CONFIG_BCACHE
1831 p->sequential_io = 0;
1832 p->sequential_io_avg = 0;
1833#endif
1834
1835 /* Perform scheduler related setup. Assign this task to a CPU. */
1836 retval = sched_fork(clone_flags, p);
1837 if (retval)
1838 goto bad_fork_cleanup_policy;
1839
1840 retval = perf_event_init_task(p);
1841 if (retval)
1842 goto bad_fork_cleanup_policy;
1843 retval = audit_alloc(p);
1844 if (retval)
1845 goto bad_fork_cleanup_perf;
1846 /* copy all the process information */
1847 shm_init_task(p);
1848 retval = security_task_alloc(p, clone_flags);
1849 if (retval)
1850 goto bad_fork_cleanup_audit;
1851 retval = copy_semundo(clone_flags, p);
1852 if (retval)
1853 goto bad_fork_cleanup_security;
1854 retval = copy_files(clone_flags, p);
1855 if (retval)
1856 goto bad_fork_cleanup_semundo;
1857 retval = copy_fs(clone_flags, p);
1858 if (retval)
1859 goto bad_fork_cleanup_files;
1860 retval = copy_sighand(clone_flags, p);
1861 if (retval)
1862 goto bad_fork_cleanup_fs;
1863 retval = copy_signal(clone_flags, p);
1864 if (retval)
1865 goto bad_fork_cleanup_sighand;
1866 retval = copy_mm(clone_flags, p);
1867 if (retval)
1868 goto bad_fork_cleanup_signal;
1869 retval = copy_namespaces(clone_flags, p);
1870 if (retval)
1871 goto bad_fork_cleanup_mm;
1872 retval = copy_io(clone_flags, p);
1873 if (retval)
1874 goto bad_fork_cleanup_namespaces;
1875 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1876 if (retval)
1877 goto bad_fork_cleanup_io;
1878
1879 if (pid != &init_struct_pid) {
1880 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1881 if (IS_ERR(pid)) {
1882 retval = PTR_ERR(pid);
1883 goto bad_fork_cleanup_thread;
1884 }
1885 }
1886
1887#ifdef CONFIG_BLOCK
1888 p->plug = NULL;
1889#endif
1890#ifdef CONFIG_FUTEX
1891 p->robust_list = NULL;
1892#ifdef CONFIG_COMPAT
1893 p->compat_robust_list = NULL;
1894#endif
1895 INIT_LIST_HEAD(&p->pi_state_list);
1896 p->pi_state_cache = NULL;
1897#endif
1898 /*
1899 * sigaltstack should be cleared when sharing the same VM
1900 */
1901 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1902 sas_ss_reset(p);
1903
1904 /*
1905 * Syscall tracing and stepping should be turned off in the
1906 * child regardless of CLONE_PTRACE.
1907 */
1908 user_disable_single_step(p);
1909 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1910#ifdef TIF_SYSCALL_EMU
1911 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1912#endif
1913 clear_all_latency_tracing(p);
1914
1915 /* ok, now we should be set up.. */
1916 p->pid = pid_nr(pid);
1917 if (clone_flags & CLONE_THREAD) {
1918 p->exit_signal = -1;
1919 p->group_leader = current->group_leader;
1920 p->tgid = current->tgid;
1921 } else {
1922 if (clone_flags & CLONE_PARENT)
1923 p->exit_signal = current->group_leader->exit_signal;
1924 else
1925 p->exit_signal = (clone_flags & CSIGNAL);
1926 p->group_leader = p;
1927 p->tgid = p->pid;
1928 }
1929
1930 p->nr_dirtied = 0;
1931 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1932 p->dirty_paused_when = 0;
1933
1934 p->pdeath_signal = 0;
1935 INIT_LIST_HEAD(&p->thread_group);
1936 p->task_works = NULL;
1937
1938 cgroup_threadgroup_change_begin(current);
1939 /*
1940 * Ensure that the cgroup subsystem policies allow the new process to be
1941 * forked. It should be noted the the new process's css_set can be changed
1942 * between here and cgroup_post_fork() if an organisation operation is in
1943 * progress.
1944 */
1945 retval = cgroup_can_fork(p);
1946 if (retval)
1947 goto bad_fork_free_pid;
1948
1949 /*
1950 * Make it visible to the rest of the system, but dont wake it up yet.
1951 * Need tasklist lock for parent etc handling!
1952 */
1953 write_lock_irq(&tasklist_lock);
1954
1955 /* CLONE_PARENT re-uses the old parent */
1956 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1957 p->real_parent = current->real_parent;
1958 p->parent_exec_id = current->parent_exec_id;
1959 } else {
1960 p->real_parent = current;
1961 p->parent_exec_id = current->self_exec_id;
1962 }
1963
1964 klp_copy_process(p);
1965
1966 spin_lock(&current->sighand->siglock);
1967
1968 /*
1969 * Copy seccomp details explicitly here, in case they were changed
1970 * before holding sighand lock.
1971 */
1972 copy_seccomp(p);
1973
1974 rseq_fork(p, clone_flags);
1975
1976 /* Don't start children in a dying pid namespace */
1977 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
1978 retval = -ENOMEM;
1979 goto bad_fork_cancel_cgroup;
1980 }
1981
1982 /* Let kill terminate clone/fork in the middle */
1983 if (fatal_signal_pending(current)) {
1984 retval = -EINTR;
1985 goto bad_fork_cancel_cgroup;
1986 }
1987
1988
1989 init_task_pid_links(p);
1990 if (likely(p->pid)) {
1991 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1992
1993 init_task_pid(p, PIDTYPE_PID, pid);
1994 if (thread_group_leader(p)) {
1995 init_task_pid(p, PIDTYPE_TGID, pid);
1996 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1997 init_task_pid(p, PIDTYPE_SID, task_session(current));
1998
1999 if (is_child_reaper(pid)) {
2000 ns_of_pid(pid)->child_reaper = p;
2001 p->signal->flags |= SIGNAL_UNKILLABLE;
2002 }
2003 p->signal->shared_pending.signal = delayed.signal;
2004 p->signal->tty = tty_kref_get(current->signal->tty);
2005 /*
2006 * Inherit has_child_subreaper flag under the same
2007 * tasklist_lock with adding child to the process tree
2008 * for propagate_has_child_subreaper optimization.
2009 */
2010 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2011 p->real_parent->signal->is_child_subreaper;
2012 list_add_tail(&p->sibling, &p->real_parent->children);
2013 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2014 attach_pid(p, PIDTYPE_TGID);
2015 attach_pid(p, PIDTYPE_PGID);
2016 attach_pid(p, PIDTYPE_SID);
2017 __this_cpu_inc(process_counts);
2018 } else {
2019 current->signal->nr_threads++;
2020 atomic_inc(&current->signal->live);
2021 atomic_inc(&current->signal->sigcnt);
2022 task_join_group_stop(p);
2023 list_add_tail_rcu(&p->thread_group,
2024 &p->group_leader->thread_group);
2025 list_add_tail_rcu(&p->thread_node,
2026 &p->signal->thread_head);
2027 }
2028 attach_pid(p, PIDTYPE_PID);
2029 nr_threads++;
2030 }
2031 total_forks++;
2032 hlist_del_init(&delayed.node);
2033 spin_unlock(&current->sighand->siglock);
2034 syscall_tracepoint_update(p);
2035 write_unlock_irq(&tasklist_lock);
2036
2037 proc_fork_connector(p);
2038 cgroup_post_fork(p);
2039 cgroup_threadgroup_change_end(current);
2040 perf_event_fork(p);
2041
2042 trace_task_newtask(p, clone_flags);
2043 uprobe_copy_process(p, clone_flags);
2044
2045 return p;
2046
2047bad_fork_cancel_cgroup:
2048 spin_unlock(&current->sighand->siglock);
2049 write_unlock_irq(&tasklist_lock);
2050 cgroup_cancel_fork(p);
2051bad_fork_free_pid:
2052 cgroup_threadgroup_change_end(current);
2053 if (pid != &init_struct_pid)
2054 free_pid(pid);
2055bad_fork_cleanup_thread:
2056 exit_thread(p);
2057bad_fork_cleanup_io:
2058 if (p->io_context)
2059 exit_io_context(p);
2060bad_fork_cleanup_namespaces:
2061 exit_task_namespaces(p);
2062bad_fork_cleanup_mm:
2063 if (p->mm)
2064 mmput(p->mm);
2065bad_fork_cleanup_signal:
2066 if (!(clone_flags & CLONE_THREAD))
2067 free_signal_struct(p->signal);
2068bad_fork_cleanup_sighand:
2069 __cleanup_sighand(p->sighand);
2070bad_fork_cleanup_fs:
2071 exit_fs(p); /* blocking */
2072bad_fork_cleanup_files:
2073 exit_files(p); /* blocking */
2074bad_fork_cleanup_semundo:
2075 exit_sem(p);
2076bad_fork_cleanup_security:
2077 security_task_free(p);
2078bad_fork_cleanup_audit:
2079 audit_free(p);
2080bad_fork_cleanup_perf:
2081 perf_event_free_task(p);
2082bad_fork_cleanup_policy:
2083 lockdep_free_task(p);
2084#ifdef CONFIG_NUMA
2085 mpol_put(p->mempolicy);
2086bad_fork_cleanup_threadgroup_lock:
2087#endif
2088 delayacct_tsk_free(p);
2089bad_fork_cleanup_count:
2090 atomic_dec(&p->cred->user->processes);
2091 exit_creds(p);
2092bad_fork_free:
2093 p->state = TASK_DEAD;
2094 put_task_stack(p);
2095 free_task(p);
2096fork_out:
2097 spin_lock_irq(&current->sighand->siglock);
2098 hlist_del_init(&delayed.node);
2099 spin_unlock_irq(&current->sighand->siglock);
2100 return ERR_PTR(retval);
2101}
2102
2103static inline void init_idle_pids(struct task_struct *idle)
2104{
2105 enum pid_type type;
2106
2107 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2108 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2109 init_task_pid(idle, type, &init_struct_pid);
2110 }
2111}
2112
2113struct task_struct *fork_idle(int cpu)
2114{
2115 struct task_struct *task;
2116 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2117 cpu_to_node(cpu));
2118 if (!IS_ERR(task)) {
2119 init_idle_pids(task);
2120 init_idle(task, cpu);
2121 }
2122
2123 return task;
2124}
2125
2126/*
2127 * Ok, this is the main fork-routine.
2128 *
2129 * It copies the process, and if successful kick-starts
2130 * it and waits for it to finish using the VM if required.
2131 */
2132long _do_fork(unsigned long clone_flags,
2133 unsigned long stack_start,
2134 unsigned long stack_size,
2135 int __user *parent_tidptr,
2136 int __user *child_tidptr,
2137 unsigned long tls)
2138{
2139 struct completion vfork;
2140 struct pid *pid;
2141 struct task_struct *p;
2142 int trace = 0;
2143 long nr;
2144
2145 /*
2146 * Determine whether and which event to report to ptracer. When
2147 * called from kernel_thread or CLONE_UNTRACED is explicitly
2148 * requested, no event is reported; otherwise, report if the event
2149 * for the type of forking is enabled.
2150 */
2151 if (!(clone_flags & CLONE_UNTRACED)) {
2152 if (clone_flags & CLONE_VFORK)
2153 trace = PTRACE_EVENT_VFORK;
2154 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2155 trace = PTRACE_EVENT_CLONE;
2156 else
2157 trace = PTRACE_EVENT_FORK;
2158
2159 if (likely(!ptrace_event_enabled(current, trace)))
2160 trace = 0;
2161 }
2162
2163 p = copy_process(clone_flags, stack_start, stack_size,
2164 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2165 add_latent_entropy();
2166
2167 if (IS_ERR(p))
2168 return PTR_ERR(p);
2169
2170 /*
2171 * Do this prior waking up the new thread - the thread pointer
2172 * might get invalid after that point, if the thread exits quickly.
2173 */
2174 trace_sched_process_fork(current, p);
2175
2176 pid = get_task_pid(p, PIDTYPE_PID);
2177 nr = pid_vnr(pid);
2178
2179 if (clone_flags & CLONE_PARENT_SETTID)
2180 put_user(nr, parent_tidptr);
2181
2182 if (clone_flags & CLONE_VFORK) {
2183 p->vfork_done = &vfork;
2184 init_completion(&vfork);
2185 get_task_struct(p);
2186 }
2187
2188 wake_up_new_task(p);
2189
2190 /* forking complete and child started to run, tell ptracer */
2191 if (unlikely(trace))
2192 ptrace_event_pid(trace, pid);
2193
2194 if (clone_flags & CLONE_VFORK) {
2195 if (!wait_for_vfork_done(p, &vfork))
2196 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2197 }
2198
2199 put_pid(pid);
2200 return nr;
2201}
2202
2203#ifndef CONFIG_HAVE_COPY_THREAD_TLS
2204/* For compatibility with architectures that call do_fork directly rather than
2205 * using the syscall entry points below. */
2206long do_fork(unsigned long clone_flags,
2207 unsigned long stack_start,
2208 unsigned long stack_size,
2209 int __user *parent_tidptr,
2210 int __user *child_tidptr)
2211{
2212 return _do_fork(clone_flags, stack_start, stack_size,
2213 parent_tidptr, child_tidptr, 0);
2214}
2215#endif
2216
2217/*
2218 * Create a kernel thread.
2219 */
2220pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2221{
2222 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2223 (unsigned long)arg, NULL, NULL, 0);
2224}
2225
2226#ifdef __ARCH_WANT_SYS_FORK
2227SYSCALL_DEFINE0(fork)
2228{
2229#ifdef CONFIG_MMU
2230 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2231#else
2232 /* can not support in nommu mode */
2233 return -EINVAL;
2234#endif
2235}
2236#endif
2237
2238#ifdef __ARCH_WANT_SYS_VFORK
2239SYSCALL_DEFINE0(vfork)
2240{
2241 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2242 0, NULL, NULL, 0);
2243}
2244#endif
2245
2246#ifdef __ARCH_WANT_SYS_CLONE
2247#ifdef CONFIG_CLONE_BACKWARDS
2248SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2249 int __user *, parent_tidptr,
2250 unsigned long, tls,
2251 int __user *, child_tidptr)
2252#elif defined(CONFIG_CLONE_BACKWARDS2)
2253SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2254 int __user *, parent_tidptr,
2255 int __user *, child_tidptr,
2256 unsigned long, tls)
2257#elif defined(CONFIG_CLONE_BACKWARDS3)
2258SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2259 int, stack_size,
2260 int __user *, parent_tidptr,
2261 int __user *, child_tidptr,
2262 unsigned long, tls)
2263#else
2264SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2265 int __user *, parent_tidptr,
2266 int __user *, child_tidptr,
2267 unsigned long, tls)
2268#endif
2269{
2270 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2271}
2272#endif
2273
2274void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2275{
2276 struct task_struct *leader, *parent, *child;
2277 int res;
2278
2279 read_lock(&tasklist_lock);
2280 leader = top = top->group_leader;
2281down:
2282 for_each_thread(leader, parent) {
2283 list_for_each_entry(child, &parent->children, sibling) {
2284 res = visitor(child, data);
2285 if (res) {
2286 if (res < 0)
2287 goto out;
2288 leader = child;
2289 goto down;
2290 }
2291up:
2292 ;
2293 }
2294 }
2295
2296 if (leader != top) {
2297 child = leader;
2298 parent = child->real_parent;
2299 leader = parent->group_leader;
2300 goto up;
2301 }
2302out:
2303 read_unlock(&tasklist_lock);
2304}
2305
2306#ifndef ARCH_MIN_MMSTRUCT_ALIGN
2307#define ARCH_MIN_MMSTRUCT_ALIGN 0
2308#endif
2309
2310static void sighand_ctor(void *data)
2311{
2312 struct sighand_struct *sighand = data;
2313
2314 spin_lock_init(&sighand->siglock);
2315 init_waitqueue_head(&sighand->signalfd_wqh);
2316}
2317
2318void __init proc_caches_init(void)
2319{
2320 unsigned int mm_size;
2321
2322 sighand_cachep = kmem_cache_create("sighand_cache",
2323 sizeof(struct sighand_struct), 0,
2324 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2325 SLAB_ACCOUNT, sighand_ctor);
2326 signal_cachep = kmem_cache_create("signal_cache",
2327 sizeof(struct signal_struct), 0,
2328 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2329 NULL);
2330 files_cachep = kmem_cache_create("files_cache",
2331 sizeof(struct files_struct), 0,
2332 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2333 NULL);
2334 fs_cachep = kmem_cache_create("fs_cache",
2335 sizeof(struct fs_struct), 0,
2336 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2337 NULL);
2338
2339 /*
2340 * The mm_cpumask is located at the end of mm_struct, and is
2341 * dynamically sized based on the maximum CPU number this system
2342 * can have, taking hotplug into account (nr_cpu_ids).
2343 */
2344 mm_size = sizeof(struct mm_struct) + cpumask_size();
2345
2346 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2347 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2348 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2349 offsetof(struct mm_struct, saved_auxv),
2350 sizeof_field(struct mm_struct, saved_auxv),
2351 NULL);
2352 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2353 mmap_init();
2354 nsproxy_cache_init();
2355}
2356
2357/*
2358 * Check constraints on flags passed to the unshare system call.
2359 */
2360static int check_unshare_flags(unsigned long unshare_flags)
2361{
2362 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2363 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2364 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2365 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2366 return -EINVAL;
2367 /*
2368 * Not implemented, but pretend it works if there is nothing
2369 * to unshare. Note that unsharing the address space or the
2370 * signal handlers also need to unshare the signal queues (aka
2371 * CLONE_THREAD).
2372 */
2373 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2374 if (!thread_group_empty(current))
2375 return -EINVAL;
2376 }
2377 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2378 if (atomic_read(&current->sighand->count) > 1)
2379 return -EINVAL;
2380 }
2381 if (unshare_flags & CLONE_VM) {
2382 if (!current_is_single_threaded())
2383 return -EINVAL;
2384 }
2385
2386 return 0;
2387}
2388
2389/*
2390 * Unshare the filesystem structure if it is being shared
2391 */
2392static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2393{
2394 struct fs_struct *fs = current->fs;
2395
2396 if (!(unshare_flags & CLONE_FS) || !fs)
2397 return 0;
2398
2399 /* don't need lock here; in the worst case we'll do useless copy */
2400 if (fs->users == 1)
2401 return 0;
2402
2403 *new_fsp = copy_fs_struct(fs);
2404 if (!*new_fsp)
2405 return -ENOMEM;
2406
2407 return 0;
2408}
2409
2410/*
2411 * Unshare file descriptor table if it is being shared
2412 */
2413static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2414{
2415 struct files_struct *fd = current->files;
2416 int error = 0;
2417
2418 if ((unshare_flags & CLONE_FILES) &&
2419 (fd && atomic_read(&fd->count) > 1)) {
2420 *new_fdp = dup_fd(fd, &error);
2421 if (!*new_fdp)
2422 return error;
2423 }
2424
2425 return 0;
2426}
2427
2428/*
2429 * unshare allows a process to 'unshare' part of the process
2430 * context which was originally shared using clone. copy_*
2431 * functions used by do_fork() cannot be used here directly
2432 * because they modify an inactive task_struct that is being
2433 * constructed. Here we are modifying the current, active,
2434 * task_struct.
2435 */
2436int ksys_unshare(unsigned long unshare_flags)
2437{
2438 struct fs_struct *fs, *new_fs = NULL;
2439 struct files_struct *fd, *new_fd = NULL;
2440 struct cred *new_cred = NULL;
2441 struct nsproxy *new_nsproxy = NULL;
2442 int do_sysvsem = 0;
2443 int err;
2444
2445 /*
2446 * If unsharing a user namespace must also unshare the thread group
2447 * and unshare the filesystem root and working directories.
2448 */
2449 if (unshare_flags & CLONE_NEWUSER)
2450 unshare_flags |= CLONE_THREAD | CLONE_FS;
2451 /*
2452 * If unsharing vm, must also unshare signal handlers.
2453 */
2454 if (unshare_flags & CLONE_VM)
2455 unshare_flags |= CLONE_SIGHAND;
2456 /*
2457 * If unsharing a signal handlers, must also unshare the signal queues.
2458 */
2459 if (unshare_flags & CLONE_SIGHAND)
2460 unshare_flags |= CLONE_THREAD;
2461 /*
2462 * If unsharing namespace, must also unshare filesystem information.
2463 */
2464 if (unshare_flags & CLONE_NEWNS)
2465 unshare_flags |= CLONE_FS;
2466
2467 err = check_unshare_flags(unshare_flags);
2468 if (err)
2469 goto bad_unshare_out;
2470 /*
2471 * CLONE_NEWIPC must also detach from the undolist: after switching
2472 * to a new ipc namespace, the semaphore arrays from the old
2473 * namespace are unreachable.
2474 */
2475 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2476 do_sysvsem = 1;
2477 err = unshare_fs(unshare_flags, &new_fs);
2478 if (err)
2479 goto bad_unshare_out;
2480 err = unshare_fd(unshare_flags, &new_fd);
2481 if (err)
2482 goto bad_unshare_cleanup_fs;
2483 err = unshare_userns(unshare_flags, &new_cred);
2484 if (err)
2485 goto bad_unshare_cleanup_fd;
2486 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2487 new_cred, new_fs);
2488 if (err)
2489 goto bad_unshare_cleanup_cred;
2490
2491 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2492 if (do_sysvsem) {
2493 /*
2494 * CLONE_SYSVSEM is equivalent to sys_exit().
2495 */
2496 exit_sem(current);
2497 }
2498 if (unshare_flags & CLONE_NEWIPC) {
2499 /* Orphan segments in old ns (see sem above). */
2500 exit_shm(current);
2501 shm_init_task(current);
2502 }
2503
2504 if (new_nsproxy)
2505 switch_task_namespaces(current, new_nsproxy);
2506
2507 task_lock(current);
2508
2509 if (new_fs) {
2510 fs = current->fs;
2511 spin_lock(&fs->lock);
2512 current->fs = new_fs;
2513 if (--fs->users)
2514 new_fs = NULL;
2515 else
2516 new_fs = fs;
2517 spin_unlock(&fs->lock);
2518 }
2519
2520 if (new_fd) {
2521 fd = current->files;
2522 current->files = new_fd;
2523 new_fd = fd;
2524 }
2525
2526 task_unlock(current);
2527
2528 if (new_cred) {
2529 /* Install the new user namespace */
2530 commit_creds(new_cred);
2531 new_cred = NULL;
2532 }
2533 }
2534
2535 perf_event_namespaces(current);
2536
2537bad_unshare_cleanup_cred:
2538 if (new_cred)
2539 put_cred(new_cred);
2540bad_unshare_cleanup_fd:
2541 if (new_fd)
2542 put_files_struct(new_fd);
2543
2544bad_unshare_cleanup_fs:
2545 if (new_fs)
2546 free_fs_struct(new_fs);
2547
2548bad_unshare_out:
2549 return err;
2550}
2551
2552SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2553{
2554 return ksys_unshare(unshare_flags);
2555}
2556
2557/*
2558 * Helper to unshare the files of the current task.
2559 * We don't want to expose copy_files internals to
2560 * the exec layer of the kernel.
2561 */
2562
2563int unshare_files(struct files_struct **displaced)
2564{
2565 struct task_struct *task = current;
2566 struct files_struct *copy = NULL;
2567 int error;
2568
2569 error = unshare_fd(CLONE_FILES, &copy);
2570 if (error || !copy) {
2571 *displaced = NULL;
2572 return error;
2573 }
2574 *displaced = task->files;
2575 task_lock(task);
2576 task->files = copy;
2577 task_unlock(task);
2578 return 0;
2579}
2580
2581int sysctl_max_threads(struct ctl_table *table, int write,
2582 void __user *buffer, size_t *lenp, loff_t *ppos)
2583{
2584 struct ctl_table t;
2585 int ret;
2586 int threads = max_threads;
2587 int min = MIN_THREADS;
2588 int max = MAX_THREADS;
2589
2590 t = *table;
2591 t.data = &threads;
2592 t.extra1 = &min;
2593 t.extra2 = &max;
2594
2595 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2596 if (ret || !write)
2597 return ret;
2598
2599 set_max_threads(threads);
2600
2601 return 0;
2602}
2603