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/user.h>
20#include <linux/sched/numa_balancing.h>
21#include <linux/sched/stat.h>
22#include <linux/sched/task.h>
23#include <linux/sched/task_stack.h>
24#include <linux/sched/cputime.h>
25#include <linux/sched/ext.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/kmsan.h>
41#include <linux/binfmts.h>
42#include <linux/mman.h>
43#include <linux/mmu_notifier.h>
44#include <linux/fs.h>
45#include <linux/mm.h>
46#include <linux/mm_inline.h>
47#include <linux/memblock.h>
48#include <linux/nsproxy.h>
49#include <linux/capability.h>
50#include <linux/cpu.h>
51#include <linux/cgroup.h>
52#include <linux/security.h>
53#include <linux/hugetlb.h>
54#include <linux/seccomp.h>
55#include <linux/swap.h>
56#include <linux/syscalls.h>
57#include <linux/syscall_user_dispatch.h>
58#include <linux/jiffies.h>
59#include <linux/futex.h>
60#include <linux/compat.h>
61#include <linux/kthread.h>
62#include <linux/task_io_accounting_ops.h>
63#include <linux/rcupdate.h>
64#include <linux/ptrace.h>
65#include <linux/mount.h>
66#include <linux/audit.h>
67#include <linux/memcontrol.h>
68#include <linux/ftrace.h>
69#include <linux/proc_fs.h>
70#include <linux/profile.h>
71#include <linux/rmap.h>
72#include <linux/ksm.h>
73#include <linux/acct.h>
74#include <linux/userfaultfd_k.h>
75#include <linux/tsacct_kern.h>
76#include <linux/cn_proc.h>
77#include <linux/freezer.h>
78#include <linux/delayacct.h>
79#include <linux/taskstats_kern.h>
80#include <linux/tty.h>
81#include <linux/fs_struct.h>
82#include <linux/magic.h>
83#include <linux/perf_event.h>
84#include <linux/posix-timers.h>
85#include <linux/user-return-notifier.h>
86#include <linux/oom.h>
87#include <linux/khugepaged.h>
88#include <linux/signalfd.h>
89#include <linux/uprobes.h>
90#include <linux/aio.h>
91#include <linux/compiler.h>
92#include <linux/sysctl.h>
93#include <linux/kcov.h>
94#include <linux/livepatch.h>
95#include <linux/thread_info.h>
96#include <linux/stackleak.h>
97#include <linux/kasan.h>
98#include <linux/scs.h>
99#include <linux/io_uring.h>
100#include <linux/bpf.h>
101#include <linux/stackprotector.h>
102#include <linux/user_events.h>
103#include <linux/iommu.h>
104#include <linux/rseq.h>
105#include <uapi/linux/pidfd.h>
106#include <linux/pidfs.h>
107#include <linux/tick.h>
108
109#include <asm/pgalloc.h>
110#include <linux/uaccess.h>
111#include <asm/mmu_context.h>
112#include <asm/cacheflush.h>
113#include <asm/tlbflush.h>
114
115/* For dup_mmap(). */
116#include "../mm/internal.h"
117
118#include <trace/events/sched.h>
119
120#define CREATE_TRACE_POINTS
121#include <trace/events/task.h>
122
123#include <kunit/visibility.h>
124
125/*
126 * Minimum number of threads to boot the kernel
127 */
128#define MIN_THREADS 20
129
130/*
131 * Maximum number of threads
132 */
133#define MAX_THREADS FUTEX_TID_MASK
134
135/*
136 * Protected counters by write_lock_irq(&tasklist_lock)
137 */
138unsigned long total_forks; /* Handle normal Linux uptimes. */
139int nr_threads; /* The idle threads do not count.. */
140
141static int max_threads; /* tunable limit on nr_threads */
142
143#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
144
145static const char * const resident_page_types[] = {
146 NAMED_ARRAY_INDEX(MM_FILEPAGES),
147 NAMED_ARRAY_INDEX(MM_ANONPAGES),
148 NAMED_ARRAY_INDEX(MM_SWAPENTS),
149 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
150};
151
152DEFINE_PER_CPU(unsigned long, process_counts) = 0;
153
154__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
155
156#ifdef CONFIG_PROVE_RCU
157int lockdep_tasklist_lock_is_held(void)
158{
159 return lockdep_is_held(&tasklist_lock);
160}
161EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
162#endif /* #ifdef CONFIG_PROVE_RCU */
163
164int nr_processes(void)
165{
166 int cpu;
167 int total = 0;
168
169 for_each_possible_cpu(cpu)
170 total += per_cpu(process_counts, cpu);
171
172 return total;
173}
174
175void __weak arch_release_task_struct(struct task_struct *tsk)
176{
177}
178
179static struct kmem_cache *task_struct_cachep;
180
181static inline struct task_struct *alloc_task_struct_node(int node)
182{
183 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
184}
185
186static inline void free_task_struct(struct task_struct *tsk)
187{
188 kmem_cache_free(s: task_struct_cachep, objp: tsk);
189}
190
191/*
192 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
193 * kmemcache based allocator.
194 */
195# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
196
197# ifdef CONFIG_VMAP_STACK
198/*
199 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
200 * flush. Try to minimize the number of calls by caching stacks.
201 */
202#define NR_CACHED_STACKS 2
203static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
204
205struct vm_stack {
206 struct rcu_head rcu;
207 struct vm_struct *stack_vm_area;
208};
209
210static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
211{
212 unsigned int i;
213
214 for (i = 0; i < NR_CACHED_STACKS; i++) {
215 struct vm_struct *tmp = NULL;
216
217 if (this_cpu_try_cmpxchg(cached_stacks[i], &tmp, vm))
218 return true;
219 }
220 return false;
221}
222
223static void thread_stack_free_rcu(struct rcu_head *rh)
224{
225 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
226
227 if (try_release_thread_stack_to_cache(vm: vm_stack->stack_vm_area))
228 return;
229
230 vfree(addr: vm_stack);
231}
232
233static void thread_stack_delayed_free(struct task_struct *tsk)
234{
235 struct vm_stack *vm_stack = tsk->stack;
236
237 vm_stack->stack_vm_area = tsk->stack_vm_area;
238 call_rcu(head: &vm_stack->rcu, func: thread_stack_free_rcu);
239}
240
241static int free_vm_stack_cache(unsigned int cpu)
242{
243 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
244 int i;
245
246 for (i = 0; i < NR_CACHED_STACKS; i++) {
247 struct vm_struct *vm_stack = cached_vm_stacks[i];
248
249 if (!vm_stack)
250 continue;
251
252 vfree(addr: vm_stack->addr);
253 cached_vm_stacks[i] = NULL;
254 }
255
256 return 0;
257}
258
259static int memcg_charge_kernel_stack(struct vm_struct *vm)
260{
261 int i;
262 int ret;
263 int nr_charged = 0;
264
265 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
266
267 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
268 ret = memcg_kmem_charge_page(page: vm->pages[i], GFP_KERNEL, order: 0);
269 if (ret)
270 goto err;
271 nr_charged++;
272 }
273 return 0;
274err:
275 for (i = 0; i < nr_charged; i++)
276 memcg_kmem_uncharge_page(page: vm->pages[i], order: 0);
277 return ret;
278}
279
280static int alloc_thread_stack_node(struct task_struct *tsk, int node)
281{
282 struct vm_struct *vm;
283 void *stack;
284 int i;
285
286 for (i = 0; i < NR_CACHED_STACKS; i++) {
287 struct vm_struct *s;
288
289 s = this_cpu_xchg(cached_stacks[i], NULL);
290
291 if (!s)
292 continue;
293
294 /* Reset stack metadata. */
295 kasan_unpoison_range(addr: s->addr, THREAD_SIZE);
296
297 stack = kasan_reset_tag(addr: s->addr);
298
299 /* Clear stale pointers from reused stack. */
300 memset(stack, 0, THREAD_SIZE);
301
302 if (memcg_charge_kernel_stack(vm: s)) {
303 vfree(addr: s->addr);
304 return -ENOMEM;
305 }
306
307 tsk->stack_vm_area = s;
308 tsk->stack = stack;
309 return 0;
310 }
311
312 /*
313 * Allocated stacks are cached and later reused by new threads,
314 * so memcg accounting is performed manually on assigning/releasing
315 * stacks to tasks. Drop __GFP_ACCOUNT.
316 */
317 stack = __vmalloc_node(THREAD_SIZE, THREAD_ALIGN,
318 THREADINFO_GFP & ~__GFP_ACCOUNT,
319 node, __builtin_return_address(0));
320 if (!stack)
321 return -ENOMEM;
322
323 vm = find_vm_area(addr: stack);
324 if (memcg_charge_kernel_stack(vm)) {
325 vfree(addr: stack);
326 return -ENOMEM;
327 }
328 /*
329 * We can't call find_vm_area() in interrupt context, and
330 * free_thread_stack() can be called in interrupt context,
331 * so cache the vm_struct.
332 */
333 tsk->stack_vm_area = vm;
334 stack = kasan_reset_tag(addr: stack);
335 tsk->stack = stack;
336 return 0;
337}
338
339static void free_thread_stack(struct task_struct *tsk)
340{
341 if (!try_release_thread_stack_to_cache(vm: tsk->stack_vm_area))
342 thread_stack_delayed_free(tsk);
343
344 tsk->stack = NULL;
345 tsk->stack_vm_area = NULL;
346}
347
348# else /* !CONFIG_VMAP_STACK */
349
350static void thread_stack_free_rcu(struct rcu_head *rh)
351{
352 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
353}
354
355static void thread_stack_delayed_free(struct task_struct *tsk)
356{
357 struct rcu_head *rh = tsk->stack;
358
359 call_rcu(rh, thread_stack_free_rcu);
360}
361
362static int alloc_thread_stack_node(struct task_struct *tsk, int node)
363{
364 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
365 THREAD_SIZE_ORDER);
366
367 if (likely(page)) {
368 tsk->stack = kasan_reset_tag(page_address(page));
369 return 0;
370 }
371 return -ENOMEM;
372}
373
374static void free_thread_stack(struct task_struct *tsk)
375{
376 thread_stack_delayed_free(tsk);
377 tsk->stack = NULL;
378}
379
380# endif /* CONFIG_VMAP_STACK */
381# else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
382
383static struct kmem_cache *thread_stack_cache;
384
385static void thread_stack_free_rcu(struct rcu_head *rh)
386{
387 kmem_cache_free(thread_stack_cache, rh);
388}
389
390static void thread_stack_delayed_free(struct task_struct *tsk)
391{
392 struct rcu_head *rh = tsk->stack;
393
394 call_rcu(rh, thread_stack_free_rcu);
395}
396
397static int alloc_thread_stack_node(struct task_struct *tsk, int node)
398{
399 unsigned long *stack;
400 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
401 stack = kasan_reset_tag(stack);
402 tsk->stack = stack;
403 return stack ? 0 : -ENOMEM;
404}
405
406static void free_thread_stack(struct task_struct *tsk)
407{
408 thread_stack_delayed_free(tsk);
409 tsk->stack = NULL;
410}
411
412void thread_stack_cache_init(void)
413{
414 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
415 THREAD_SIZE, THREAD_SIZE, 0, 0,
416 THREAD_SIZE, NULL);
417 BUG_ON(thread_stack_cache == NULL);
418}
419
420# endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
421
422/* SLAB cache for signal_struct structures (tsk->signal) */
423static struct kmem_cache *signal_cachep;
424
425/* SLAB cache for sighand_struct structures (tsk->sighand) */
426struct kmem_cache *sighand_cachep;
427
428/* SLAB cache for files_struct structures (tsk->files) */
429struct kmem_cache *files_cachep;
430
431/* SLAB cache for fs_struct structures (tsk->fs) */
432struct kmem_cache *fs_cachep;
433
434/* SLAB cache for mm_struct structures (tsk->mm) */
435static struct kmem_cache *mm_cachep;
436
437static void account_kernel_stack(struct task_struct *tsk, int account)
438{
439 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
440 struct vm_struct *vm = task_stack_vm_area(t: tsk);
441 int i;
442
443 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
444 mod_lruvec_page_state(page: vm->pages[i], idx: NR_KERNEL_STACK_KB,
445 val: account * (PAGE_SIZE / 1024));
446 } else {
447 void *stack = task_stack_page(task: tsk);
448
449 /* All stack pages are in the same node. */
450 mod_lruvec_kmem_state(p: stack, idx: NR_KERNEL_STACK_KB,
451 val: account * (THREAD_SIZE / 1024));
452 }
453}
454
455void exit_task_stack_account(struct task_struct *tsk)
456{
457 account_kernel_stack(tsk, account: -1);
458
459 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
460 struct vm_struct *vm;
461 int i;
462
463 vm = task_stack_vm_area(t: tsk);
464 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
465 memcg_kmem_uncharge_page(page: vm->pages[i], order: 0);
466 }
467}
468
469static void release_task_stack(struct task_struct *tsk)
470{
471 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
472 return; /* Better to leak the stack than to free prematurely */
473
474 free_thread_stack(tsk);
475}
476
477#ifdef CONFIG_THREAD_INFO_IN_TASK
478void put_task_stack(struct task_struct *tsk)
479{
480 if (refcount_dec_and_test(r: &tsk->stack_refcount))
481 release_task_stack(tsk);
482}
483#endif
484
485void free_task(struct task_struct *tsk)
486{
487#ifdef CONFIG_SECCOMP
488 WARN_ON_ONCE(tsk->seccomp.filter);
489#endif
490 release_user_cpus_ptr(p: tsk);
491 scs_release(tsk);
492
493#ifndef CONFIG_THREAD_INFO_IN_TASK
494 /*
495 * The task is finally done with both the stack and thread_info,
496 * so free both.
497 */
498 release_task_stack(tsk);
499#else
500 /*
501 * If the task had a separate stack allocation, it should be gone
502 * by now.
503 */
504 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
505#endif
506 rt_mutex_debug_task_free(tsk);
507 ftrace_graph_exit_task(t: tsk);
508 arch_release_task_struct(tsk);
509 if (tsk->flags & PF_KTHREAD)
510 free_kthread_struct(k: tsk);
511 bpf_task_storage_free(task: tsk);
512 free_task_struct(tsk);
513}
514EXPORT_SYMBOL(free_task);
515
516void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
517{
518 struct file *exe_file;
519
520 exe_file = get_mm_exe_file(mm: oldmm);
521 RCU_INIT_POINTER(mm->exe_file, exe_file);
522 /*
523 * We depend on the oldmm having properly denied write access to the
524 * exe_file already.
525 */
526 if (exe_file && exe_file_deny_write_access(exe_file))
527 pr_warn_once("exe_file_deny_write_access() failed in %s\n", __func__);
528}
529
530#ifdef CONFIG_MMU
531static inline int mm_alloc_pgd(struct mm_struct *mm)
532{
533 mm->pgd = pgd_alloc(mm);
534 if (unlikely(!mm->pgd))
535 return -ENOMEM;
536 return 0;
537}
538
539static inline void mm_free_pgd(struct mm_struct *mm)
540{
541 pgd_free(mm, pgd: mm->pgd);
542}
543#else
544#define mm_alloc_pgd(mm) (0)
545#define mm_free_pgd(mm)
546#endif /* CONFIG_MMU */
547
548#ifdef CONFIG_MM_ID
549static DEFINE_IDA(mm_ida);
550
551static inline int mm_alloc_id(struct mm_struct *mm)
552{
553 int ret;
554
555 ret = ida_alloc_range(&mm_ida, MM_ID_MIN, MM_ID_MAX, GFP_KERNEL);
556 if (ret < 0)
557 return ret;
558 mm->mm_id = ret;
559 return 0;
560}
561
562static inline void mm_free_id(struct mm_struct *mm)
563{
564 const mm_id_t id = mm->mm_id;
565
566 mm->mm_id = MM_ID_DUMMY;
567 if (id == MM_ID_DUMMY)
568 return;
569 if (WARN_ON_ONCE(id < MM_ID_MIN || id > MM_ID_MAX))
570 return;
571 ida_free(&mm_ida, id);
572}
573#else /* !CONFIG_MM_ID */
574static inline int mm_alloc_id(struct mm_struct *mm) { return 0; }
575static inline void mm_free_id(struct mm_struct *mm) {}
576#endif /* CONFIG_MM_ID */
577
578static void check_mm(struct mm_struct *mm)
579{
580 int i;
581
582 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
583 "Please make sure 'struct resident_page_types[]' is updated as well");
584
585 for (i = 0; i < NR_MM_COUNTERS; i++) {
586 long x = percpu_counter_sum(fbc: &mm->rss_stat[i]);
587
588 if (unlikely(x))
589 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
590 mm, resident_page_types[i], x);
591 }
592
593 if (mm_pgtables_bytes(mm))
594 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
595 mm_pgtables_bytes(mm));
596
597#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
598 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
599#endif
600}
601
602#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
603#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
604
605static void do_check_lazy_tlb(void *arg)
606{
607 struct mm_struct *mm = arg;
608
609 WARN_ON_ONCE(current->active_mm == mm);
610}
611
612static void do_shoot_lazy_tlb(void *arg)
613{
614 struct mm_struct *mm = arg;
615
616 if (current->active_mm == mm) {
617 WARN_ON_ONCE(current->mm);
618 current->active_mm = &init_mm;
619 switch_mm(prev: mm, next: &init_mm, current);
620 }
621}
622
623static void cleanup_lazy_tlbs(struct mm_struct *mm)
624{
625 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
626 /*
627 * In this case, lazy tlb mms are refounted and would not reach
628 * __mmdrop until all CPUs have switched away and mmdrop()ed.
629 */
630 return;
631 }
632
633 /*
634 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
635 * requires lazy mm users to switch to another mm when the refcount
636 * drops to zero, before the mm is freed. This requires IPIs here to
637 * switch kernel threads to init_mm.
638 *
639 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
640 * switch with the final userspace teardown TLB flush which leaves the
641 * mm lazy on this CPU but no others, reducing the need for additional
642 * IPIs here. There are cases where a final IPI is still required here,
643 * such as the final mmdrop being performed on a different CPU than the
644 * one exiting, or kernel threads using the mm when userspace exits.
645 *
646 * IPI overheads have not found to be expensive, but they could be
647 * reduced in a number of possible ways, for example (roughly
648 * increasing order of complexity):
649 * - The last lazy reference created by exit_mm() could instead switch
650 * to init_mm, however it's probable this will run on the same CPU
651 * immediately afterwards, so this may not reduce IPIs much.
652 * - A batch of mms requiring IPIs could be gathered and freed at once.
653 * - CPUs store active_mm where it can be remotely checked without a
654 * lock, to filter out false-positives in the cpumask.
655 * - After mm_users or mm_count reaches zero, switching away from the
656 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
657 * with some batching or delaying of the final IPIs.
658 * - A delayed freeing and RCU-like quiescing sequence based on mm
659 * switching to avoid IPIs completely.
660 */
661 on_each_cpu_mask(mask: mm_cpumask(mm), func: do_shoot_lazy_tlb, info: (void *)mm, wait: 1);
662 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
663 on_each_cpu(func: do_check_lazy_tlb, info: (void *)mm, wait: 1);
664}
665
666/*
667 * Called when the last reference to the mm
668 * is dropped: either by a lazy thread or by
669 * mmput. Free the page directory and the mm.
670 */
671void __mmdrop(struct mm_struct *mm)
672{
673 BUG_ON(mm == &init_mm);
674 WARN_ON_ONCE(mm == current->mm);
675
676 /* Ensure no CPUs are using this as their lazy tlb mm */
677 cleanup_lazy_tlbs(mm);
678
679 WARN_ON_ONCE(mm == current->active_mm);
680 mm_free_pgd(mm);
681 mm_free_id(mm);
682 destroy_context(mm);
683 mmu_notifier_subscriptions_destroy(mm);
684 check_mm(mm);
685 put_user_ns(ns: mm->user_ns);
686 mm_pasid_drop(mm);
687 mm_destroy_cid(mm);
688 percpu_counter_destroy_many(fbc: mm->rss_stat, nr_counters: NR_MM_COUNTERS);
689
690 free_mm(mm);
691}
692EXPORT_SYMBOL_GPL(__mmdrop);
693
694static void mmdrop_async_fn(struct work_struct *work)
695{
696 struct mm_struct *mm;
697
698 mm = container_of(work, struct mm_struct, async_put_work);
699 __mmdrop(mm);
700}
701
702static void mmdrop_async(struct mm_struct *mm)
703{
704 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
705 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
706 schedule_work(work: &mm->async_put_work);
707 }
708}
709
710static inline void free_signal_struct(struct signal_struct *sig)
711{
712 taskstats_tgid_free(sig);
713 sched_autogroup_exit(sig);
714 /*
715 * __mmdrop is not safe to call from softirq context on x86 due to
716 * pgd_dtor so postpone it to the async context
717 */
718 if (sig->oom_mm)
719 mmdrop_async(mm: sig->oom_mm);
720 kmem_cache_free(s: signal_cachep, objp: sig);
721}
722
723static inline void put_signal_struct(struct signal_struct *sig)
724{
725 if (refcount_dec_and_test(r: &sig->sigcnt))
726 free_signal_struct(sig);
727}
728
729void __put_task_struct(struct task_struct *tsk)
730{
731 WARN_ON(!tsk->exit_state);
732 WARN_ON(refcount_read(&tsk->usage));
733 WARN_ON(tsk == current);
734
735 sched_ext_free(p: tsk);
736 io_uring_free(tsk);
737 cgroup_free(p: tsk);
738 task_numa_free(p: tsk, final: true);
739 security_task_free(task: tsk);
740 exit_creds(tsk);
741 delayacct_tsk_free(tsk);
742 put_signal_struct(sig: tsk->signal);
743 sched_core_free(tsk);
744 free_task(tsk);
745}
746EXPORT_SYMBOL_GPL(__put_task_struct);
747
748void __put_task_struct_rcu_cb(struct rcu_head *rhp)
749{
750 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
751
752 __put_task_struct(task);
753}
754EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
755
756void __init __weak arch_task_cache_init(void) { }
757
758/*
759 * set_max_threads
760 */
761static void __init set_max_threads(unsigned int max_threads_suggested)
762{
763 u64 threads;
764 unsigned long nr_pages = memblock_estimated_nr_free_pages();
765
766 /*
767 * The number of threads shall be limited such that the thread
768 * structures may only consume a small part of the available memory.
769 */
770 if (fls64(x: nr_pages) + fls64(PAGE_SIZE) > 64)
771 threads = MAX_THREADS;
772 else
773 threads = div64_u64(dividend: (u64) nr_pages * (u64) PAGE_SIZE,
774 divisor: (u64) THREAD_SIZE * 8UL);
775
776 if (threads > max_threads_suggested)
777 threads = max_threads_suggested;
778
779 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
780}
781
782#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
783/* Initialized by the architecture: */
784int arch_task_struct_size __read_mostly;
785#endif
786
787static void __init task_struct_whitelist(unsigned long *offset, unsigned long *size)
788{
789 /* Fetch thread_struct whitelist for the architecture. */
790 arch_thread_struct_whitelist(offset, size);
791
792 /*
793 * Handle zero-sized whitelist or empty thread_struct, otherwise
794 * adjust offset to position of thread_struct in task_struct.
795 */
796 if (unlikely(*size == 0))
797 *offset = 0;
798 else
799 *offset += offsetof(struct task_struct, thread);
800}
801
802void __init fork_init(void)
803{
804 int i;
805#ifndef ARCH_MIN_TASKALIGN
806#define ARCH_MIN_TASKALIGN 0
807#endif
808 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
809 unsigned long useroffset, usersize;
810
811 /* create a slab on which task_structs can be allocated */
812 task_struct_whitelist(offset: &useroffset, size: &usersize);
813 task_struct_cachep = kmem_cache_create_usercopy(name: "task_struct",
814 size: arch_task_struct_size, align,
815 SLAB_PANIC|SLAB_ACCOUNT,
816 useroffset, usersize, NULL);
817
818 /* do the arch specific task caches init */
819 arch_task_cache_init();
820
821 set_max_threads(MAX_THREADS);
822
823 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
824 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
825 init_task.signal->rlim[RLIMIT_SIGPENDING] =
826 init_task.signal->rlim[RLIMIT_NPROC];
827
828 for (i = 0; i < UCOUNT_COUNTS; i++)
829 init_user_ns.ucount_max[i] = max_threads/2;
830
831 set_userns_rlimit_max(ns: &init_user_ns, type: UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
832 set_userns_rlimit_max(ns: &init_user_ns, type: UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
833 set_userns_rlimit_max(ns: &init_user_ns, type: UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
834 set_userns_rlimit_max(ns: &init_user_ns, type: UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
835
836#ifdef CONFIG_VMAP_STACK
837 cpuhp_setup_state(state: CPUHP_BP_PREPARE_DYN, name: "fork:vm_stack_cache",
838 NULL, teardown: free_vm_stack_cache);
839#endif
840
841 scs_init();
842
843 lockdep_init_task(task: &init_task);
844 uprobes_init();
845}
846
847int __weak arch_dup_task_struct(struct task_struct *dst,
848 struct task_struct *src)
849{
850 *dst = *src;
851 return 0;
852}
853
854void set_task_stack_end_magic(struct task_struct *tsk)
855{
856 unsigned long *stackend;
857
858 stackend = end_of_stack(task: tsk);
859 *stackend = STACK_END_MAGIC; /* for overflow detection */
860}
861
862static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
863{
864 struct task_struct *tsk;
865 int err;
866
867 if (node == NUMA_NO_NODE)
868 node = tsk_fork_get_node(tsk: orig);
869 tsk = alloc_task_struct_node(node);
870 if (!tsk)
871 return NULL;
872
873 err = arch_dup_task_struct(dst: tsk, src: orig);
874 if (err)
875 goto free_tsk;
876
877 err = alloc_thread_stack_node(tsk, node);
878 if (err)
879 goto free_tsk;
880
881#ifdef CONFIG_THREAD_INFO_IN_TASK
882 refcount_set(r: &tsk->stack_refcount, n: 1);
883#endif
884 account_kernel_stack(tsk, account: 1);
885
886 err = scs_prepare(tsk, node);
887 if (err)
888 goto free_stack;
889
890#ifdef CONFIG_SECCOMP
891 /*
892 * We must handle setting up seccomp filters once we're under
893 * the sighand lock in case orig has changed between now and
894 * then. Until then, filter must be NULL to avoid messing up
895 * the usage counts on the error path calling free_task.
896 */
897 tsk->seccomp.filter = NULL;
898#endif
899
900 setup_thread_stack(tsk, orig);
901 clear_user_return_notifier(p: tsk);
902 clear_tsk_need_resched(tsk);
903 set_task_stack_end_magic(tsk);
904 clear_syscall_work_syscall_user_dispatch(tsk);
905
906#ifdef CONFIG_STACKPROTECTOR
907 tsk->stack_canary = get_random_canary();
908#endif
909 if (orig->cpus_ptr == &orig->cpus_mask)
910 tsk->cpus_ptr = &tsk->cpus_mask;
911 dup_user_cpus_ptr(dst: tsk, src: orig, node);
912
913 /*
914 * One for the user space visible state that goes away when reaped.
915 * One for the scheduler.
916 */
917 refcount_set(r: &tsk->rcu_users, n: 2);
918 /* One for the rcu users */
919 refcount_set(r: &tsk->usage, n: 1);
920#ifdef CONFIG_BLK_DEV_IO_TRACE
921 tsk->btrace_seq = 0;
922#endif
923 tsk->splice_pipe = NULL;
924 tsk->task_frag.page = NULL;
925 tsk->wake_q.next = NULL;
926 tsk->worker_private = NULL;
927
928 kcov_task_init(t: tsk);
929 kmsan_task_create(task: tsk);
930 kmap_local_fork(tsk);
931
932#ifdef CONFIG_FAULT_INJECTION
933 tsk->fail_nth = 0;
934#endif
935
936#ifdef CONFIG_BLK_CGROUP
937 tsk->throttle_disk = NULL;
938 tsk->use_memdelay = 0;
939#endif
940
941#ifdef CONFIG_ARCH_HAS_CPU_PASID
942 tsk->pasid_activated = 0;
943#endif
944
945#ifdef CONFIG_MEMCG
946 tsk->active_memcg = NULL;
947#endif
948
949#ifdef CONFIG_X86_BUS_LOCK_DETECT
950 tsk->reported_split_lock = 0;
951#endif
952
953#ifdef CONFIG_SCHED_MM_CID
954 tsk->mm_cid = -1;
955 tsk->last_mm_cid = -1;
956 tsk->mm_cid_active = 0;
957 tsk->migrate_from_cpu = -1;
958#endif
959 return tsk;
960
961free_stack:
962 exit_task_stack_account(tsk);
963 free_thread_stack(tsk);
964free_tsk:
965 free_task_struct(tsk);
966 return NULL;
967}
968
969__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
970
971static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
972
973static int __init coredump_filter_setup(char *s)
974{
975 default_dump_filter =
976 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
977 MMF_DUMP_FILTER_MASK;
978 return 1;
979}
980
981__setup("coredump_filter=", coredump_filter_setup);
982
983#include <linux/init_task.h>
984
985static void mm_init_aio(struct mm_struct *mm)
986{
987#ifdef CONFIG_AIO
988 spin_lock_init(&mm->ioctx_lock);
989 mm->ioctx_table = NULL;
990#endif
991}
992
993static __always_inline void mm_clear_owner(struct mm_struct *mm,
994 struct task_struct *p)
995{
996#ifdef CONFIG_MEMCG
997 if (mm->owner == p)
998 WRITE_ONCE(mm->owner, NULL);
999#endif
1000}
1001
1002static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1003{
1004#ifdef CONFIG_MEMCG
1005 mm->owner = p;
1006#endif
1007}
1008
1009static void mm_init_uprobes_state(struct mm_struct *mm)
1010{
1011#ifdef CONFIG_UPROBES
1012 mm->uprobes_state.xol_area = NULL;
1013#endif
1014}
1015
1016static void mmap_init_lock(struct mm_struct *mm)
1017{
1018 init_rwsem(&mm->mmap_lock);
1019 mm_lock_seqcount_init(mm);
1020#ifdef CONFIG_PER_VMA_LOCK
1021 rcuwait_init(w: &mm->vma_writer_wait);
1022#endif
1023}
1024
1025static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1026 struct user_namespace *user_ns)
1027{
1028 mt_init_flags(mt: &mm->mm_mt, MM_MT_FLAGS);
1029 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1030 atomic_set(v: &mm->mm_users, i: 1);
1031 atomic_set(v: &mm->mm_count, i: 1);
1032 seqcount_init(&mm->write_protect_seq);
1033 mmap_init_lock(mm);
1034 INIT_LIST_HEAD(list: &mm->mmlist);
1035 mm_pgtables_bytes_init(mm);
1036 mm->map_count = 0;
1037 mm->locked_vm = 0;
1038 atomic64_set(v: &mm->pinned_vm, i: 0);
1039 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1040 spin_lock_init(&mm->page_table_lock);
1041 spin_lock_init(&mm->arg_lock);
1042 mm_init_cpumask(mm);
1043 mm_init_aio(mm);
1044 mm_init_owner(mm, p);
1045 mm_pasid_init(mm);
1046 RCU_INIT_POINTER(mm->exe_file, NULL);
1047 mmu_notifier_subscriptions_init(mm);
1048 init_tlb_flush_pending(mm);
1049 futex_mm_init(mm);
1050#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
1051 mm->pmd_huge_pte = NULL;
1052#endif
1053 mm_init_uprobes_state(mm);
1054 hugetlb_count_init(mm);
1055
1056 if (current->mm) {
1057 mm->flags = mmf_init_flags(current->mm->flags);
1058 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1059 } else {
1060 mm->flags = default_dump_filter;
1061 mm->def_flags = 0;
1062 }
1063
1064 if (mm_alloc_pgd(mm))
1065 goto fail_nopgd;
1066
1067 if (mm_alloc_id(mm))
1068 goto fail_noid;
1069
1070 if (init_new_context(tsk: p, mm))
1071 goto fail_nocontext;
1072
1073 if (mm_alloc_cid(mm, p))
1074 goto fail_cid;
1075
1076 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1077 NR_MM_COUNTERS))
1078 goto fail_pcpu;
1079
1080 mm->user_ns = get_user_ns(ns: user_ns);
1081 lru_gen_init_mm(mm);
1082 return mm;
1083
1084fail_pcpu:
1085 mm_destroy_cid(mm);
1086fail_cid:
1087 destroy_context(mm);
1088fail_nocontext:
1089 mm_free_id(mm);
1090fail_noid:
1091 mm_free_pgd(mm);
1092fail_nopgd:
1093 free_mm(mm);
1094 return NULL;
1095}
1096
1097/*
1098 * Allocate and initialize an mm_struct.
1099 */
1100struct mm_struct *mm_alloc(void)
1101{
1102 struct mm_struct *mm;
1103
1104 mm = allocate_mm();
1105 if (!mm)
1106 return NULL;
1107
1108 memset(mm, 0, sizeof(*mm));
1109 return mm_init(mm, current, current_user_ns());
1110}
1111EXPORT_SYMBOL_IF_KUNIT(mm_alloc);
1112
1113static inline void __mmput(struct mm_struct *mm)
1114{
1115 VM_BUG_ON(atomic_read(&mm->mm_users));
1116
1117 uprobe_clear_state(mm);
1118 exit_aio(mm);
1119 ksm_exit(mm);
1120 khugepaged_exit(mm); /* must run before exit_mmap */
1121 exit_mmap(mm);
1122 mm_put_huge_zero_folio(mm);
1123 set_mm_exe_file(mm, NULL);
1124 if (!list_empty(head: &mm->mmlist)) {
1125 spin_lock(lock: &mmlist_lock);
1126 list_del(entry: &mm->mmlist);
1127 spin_unlock(lock: &mmlist_lock);
1128 }
1129 if (mm->binfmt)
1130 module_put(module: mm->binfmt->module);
1131 lru_gen_del_mm(mm);
1132 futex_hash_free(mm);
1133 mmdrop(mm);
1134}
1135
1136/*
1137 * Decrement the use count and release all resources for an mm.
1138 */
1139void mmput(struct mm_struct *mm)
1140{
1141 might_sleep();
1142
1143 if (atomic_dec_and_test(v: &mm->mm_users))
1144 __mmput(mm);
1145}
1146EXPORT_SYMBOL_GPL(mmput);
1147
1148#ifdef CONFIG_MMU
1149static void mmput_async_fn(struct work_struct *work)
1150{
1151 struct mm_struct *mm = container_of(work, struct mm_struct,
1152 async_put_work);
1153
1154 __mmput(mm);
1155}
1156
1157void mmput_async(struct mm_struct *mm)
1158{
1159 if (atomic_dec_and_test(v: &mm->mm_users)) {
1160 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1161 schedule_work(work: &mm->async_put_work);
1162 }
1163}
1164EXPORT_SYMBOL_GPL(mmput_async);
1165#endif
1166
1167/**
1168 * set_mm_exe_file - change a reference to the mm's executable file
1169 * @mm: The mm to change.
1170 * @new_exe_file: The new file to use.
1171 *
1172 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1173 *
1174 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1175 * invocations: in mmput() nobody alive left, in execve it happens before
1176 * the new mm is made visible to anyone.
1177 *
1178 * Can only fail if new_exe_file != NULL.
1179 */
1180int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1181{
1182 struct file *old_exe_file;
1183
1184 /*
1185 * It is safe to dereference the exe_file without RCU as
1186 * this function is only called if nobody else can access
1187 * this mm -- see comment above for justification.
1188 */
1189 old_exe_file = rcu_dereference_raw(mm->exe_file);
1190
1191 if (new_exe_file) {
1192 /*
1193 * We expect the caller (i.e., sys_execve) to already denied
1194 * write access, so this is unlikely to fail.
1195 */
1196 if (unlikely(exe_file_deny_write_access(new_exe_file)))
1197 return -EACCES;
1198 get_file(f: new_exe_file);
1199 }
1200 rcu_assign_pointer(mm->exe_file, new_exe_file);
1201 if (old_exe_file) {
1202 exe_file_allow_write_access(exe_file: old_exe_file);
1203 fput(old_exe_file);
1204 }
1205 return 0;
1206}
1207
1208/**
1209 * replace_mm_exe_file - replace a reference to the mm's executable file
1210 * @mm: The mm to change.
1211 * @new_exe_file: The new file to use.
1212 *
1213 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1214 *
1215 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1216 */
1217int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1218{
1219 struct vm_area_struct *vma;
1220 struct file *old_exe_file;
1221 int ret = 0;
1222
1223 /* Forbid mm->exe_file change if old file still mapped. */
1224 old_exe_file = get_mm_exe_file(mm);
1225 if (old_exe_file) {
1226 VMA_ITERATOR(vmi, mm, 0);
1227 mmap_read_lock(mm);
1228 for_each_vma(vmi, vma) {
1229 if (!vma->vm_file)
1230 continue;
1231 if (path_equal(path1: &vma->vm_file->f_path,
1232 path2: &old_exe_file->f_path)) {
1233 ret = -EBUSY;
1234 break;
1235 }
1236 }
1237 mmap_read_unlock(mm);
1238 fput(old_exe_file);
1239 if (ret)
1240 return ret;
1241 }
1242
1243 ret = exe_file_deny_write_access(exe_file: new_exe_file);
1244 if (ret)
1245 return -EACCES;
1246 get_file(f: new_exe_file);
1247
1248 /* set the new file */
1249 mmap_write_lock(mm);
1250 old_exe_file = rcu_dereference_raw(mm->exe_file);
1251 rcu_assign_pointer(mm->exe_file, new_exe_file);
1252 mmap_write_unlock(mm);
1253
1254 if (old_exe_file) {
1255 exe_file_allow_write_access(exe_file: old_exe_file);
1256 fput(old_exe_file);
1257 }
1258 return 0;
1259}
1260
1261/**
1262 * get_mm_exe_file - acquire a reference to the mm's executable file
1263 * @mm: The mm of interest.
1264 *
1265 * Returns %NULL if mm has no associated executable file.
1266 * User must release file via fput().
1267 */
1268struct file *get_mm_exe_file(struct mm_struct *mm)
1269{
1270 struct file *exe_file;
1271
1272 rcu_read_lock();
1273 exe_file = get_file_rcu(f: &mm->exe_file);
1274 rcu_read_unlock();
1275 return exe_file;
1276}
1277
1278/**
1279 * get_task_exe_file - acquire a reference to the task's executable file
1280 * @task: The task.
1281 *
1282 * Returns %NULL if task's mm (if any) has no associated executable file or
1283 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1284 * User must release file via fput().
1285 */
1286struct file *get_task_exe_file(struct task_struct *task)
1287{
1288 struct file *exe_file = NULL;
1289 struct mm_struct *mm;
1290
1291 if (task->flags & PF_KTHREAD)
1292 return NULL;
1293
1294 task_lock(p: task);
1295 mm = task->mm;
1296 if (mm)
1297 exe_file = get_mm_exe_file(mm);
1298 task_unlock(p: task);
1299 return exe_file;
1300}
1301
1302/**
1303 * get_task_mm - acquire a reference to the task's mm
1304 * @task: The task.
1305 *
1306 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1307 * this kernel workthread has transiently adopted a user mm with use_mm,
1308 * to do its AIO) is not set and if so returns a reference to it, after
1309 * bumping up the use count. User must release the mm via mmput()
1310 * after use. Typically used by /proc and ptrace.
1311 */
1312struct mm_struct *get_task_mm(struct task_struct *task)
1313{
1314 struct mm_struct *mm;
1315
1316 if (task->flags & PF_KTHREAD)
1317 return NULL;
1318
1319 task_lock(p: task);
1320 mm = task->mm;
1321 if (mm)
1322 mmget(mm);
1323 task_unlock(p: task);
1324 return mm;
1325}
1326EXPORT_SYMBOL_GPL(get_task_mm);
1327
1328static bool may_access_mm(struct mm_struct *mm, struct task_struct *task, unsigned int mode)
1329{
1330 if (mm == current->mm)
1331 return true;
1332 if (ptrace_may_access(task, mode))
1333 return true;
1334 if ((mode & PTRACE_MODE_READ) && perfmon_capable())
1335 return true;
1336 return false;
1337}
1338
1339struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1340{
1341 struct mm_struct *mm;
1342 int err;
1343
1344 err = down_read_killable(sem: &task->signal->exec_update_lock);
1345 if (err)
1346 return ERR_PTR(error: err);
1347
1348 mm = get_task_mm(task);
1349 if (!mm) {
1350 mm = ERR_PTR(error: -ESRCH);
1351 } else if (!may_access_mm(mm, task, mode)) {
1352 mmput(mm);
1353 mm = ERR_PTR(error: -EACCES);
1354 }
1355 up_read(sem: &task->signal->exec_update_lock);
1356
1357 return mm;
1358}
1359
1360static void complete_vfork_done(struct task_struct *tsk)
1361{
1362 struct completion *vfork;
1363
1364 task_lock(p: tsk);
1365 vfork = tsk->vfork_done;
1366 if (likely(vfork)) {
1367 tsk->vfork_done = NULL;
1368 complete(vfork);
1369 }
1370 task_unlock(p: tsk);
1371}
1372
1373static int wait_for_vfork_done(struct task_struct *child,
1374 struct completion *vfork)
1375{
1376 unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1377 int killed;
1378
1379 cgroup_enter_frozen();
1380 killed = wait_for_completion_state(x: vfork, state);
1381 cgroup_leave_frozen(always_leave: false);
1382
1383 if (killed) {
1384 task_lock(p: child);
1385 child->vfork_done = NULL;
1386 task_unlock(p: child);
1387 }
1388
1389 put_task_struct(t: child);
1390 return killed;
1391}
1392
1393/* Please note the differences between mmput and mm_release.
1394 * mmput is called whenever we stop holding onto a mm_struct,
1395 * error success whatever.
1396 *
1397 * mm_release is called after a mm_struct has been removed
1398 * from the current process.
1399 *
1400 * This difference is important for error handling, when we
1401 * only half set up a mm_struct for a new process and need to restore
1402 * the old one. Because we mmput the new mm_struct before
1403 * restoring the old one. . .
1404 * Eric Biederman 10 January 1998
1405 */
1406static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1407{
1408 uprobe_free_utask(t: tsk);
1409
1410 /* Get rid of any cached register state */
1411 deactivate_mm(tsk, mm);
1412
1413 /*
1414 * Signal userspace if we're not exiting with a core dump
1415 * because we want to leave the value intact for debugging
1416 * purposes.
1417 */
1418 if (tsk->clear_child_tid) {
1419 if (atomic_read(v: &mm->mm_users) > 1) {
1420 /*
1421 * We don't check the error code - if userspace has
1422 * not set up a proper pointer then tough luck.
1423 */
1424 put_user(0, tsk->clear_child_tid);
1425 do_futex(uaddr: tsk->clear_child_tid, FUTEX_WAKE,
1426 val: 1, NULL, NULL, val2: 0, val3: 0);
1427 }
1428 tsk->clear_child_tid = NULL;
1429 }
1430
1431 /*
1432 * All done, finally we can wake up parent and return this mm to him.
1433 * Also kthread_stop() uses this completion for synchronization.
1434 */
1435 if (tsk->vfork_done)
1436 complete_vfork_done(tsk);
1437}
1438
1439void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1440{
1441 futex_exit_release(tsk);
1442 mm_release(tsk, mm);
1443}
1444
1445void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1446{
1447 futex_exec_release(tsk);
1448 mm_release(tsk, mm);
1449}
1450
1451/**
1452 * dup_mm() - duplicates an existing mm structure
1453 * @tsk: the task_struct with which the new mm will be associated.
1454 * @oldmm: the mm to duplicate.
1455 *
1456 * Allocates a new mm structure and duplicates the provided @oldmm structure
1457 * content into it.
1458 *
1459 * Return: the duplicated mm or NULL on failure.
1460 */
1461static struct mm_struct *dup_mm(struct task_struct *tsk,
1462 struct mm_struct *oldmm)
1463{
1464 struct mm_struct *mm;
1465 int err;
1466
1467 mm = allocate_mm();
1468 if (!mm)
1469 goto fail_nomem;
1470
1471 memcpy(mm, oldmm, sizeof(*mm));
1472
1473 if (!mm_init(mm, p: tsk, user_ns: mm->user_ns))
1474 goto fail_nomem;
1475
1476 uprobe_start_dup_mmap();
1477 err = dup_mmap(mm, oldmm);
1478 if (err)
1479 goto free_pt;
1480 uprobe_end_dup_mmap();
1481
1482 mm->hiwater_rss = get_mm_rss(mm);
1483 mm->hiwater_vm = mm->total_vm;
1484
1485 if (mm->binfmt && !try_module_get(module: mm->binfmt->module))
1486 goto free_pt;
1487
1488 return mm;
1489
1490free_pt:
1491 /* don't put binfmt in mmput, we haven't got module yet */
1492 mm->binfmt = NULL;
1493 mm_init_owner(mm, NULL);
1494 mmput(mm);
1495 if (err)
1496 uprobe_end_dup_mmap();
1497
1498fail_nomem:
1499 return NULL;
1500}
1501
1502static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1503{
1504 struct mm_struct *mm, *oldmm;
1505
1506 tsk->min_flt = tsk->maj_flt = 0;
1507 tsk->nvcsw = tsk->nivcsw = 0;
1508#ifdef CONFIG_DETECT_HUNG_TASK
1509 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1510 tsk->last_switch_time = 0;
1511#endif
1512
1513 tsk->mm = NULL;
1514 tsk->active_mm = NULL;
1515
1516 /*
1517 * Are we cloning a kernel thread?
1518 *
1519 * We need to steal a active VM for that..
1520 */
1521 oldmm = current->mm;
1522 if (!oldmm)
1523 return 0;
1524
1525 if (clone_flags & CLONE_VM) {
1526 mmget(mm: oldmm);
1527 mm = oldmm;
1528 } else {
1529 mm = dup_mm(tsk, current->mm);
1530 if (!mm)
1531 return -ENOMEM;
1532 }
1533
1534 tsk->mm = mm;
1535 tsk->active_mm = mm;
1536 sched_mm_cid_fork(t: tsk);
1537 return 0;
1538}
1539
1540static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1541{
1542 struct fs_struct *fs = current->fs;
1543 if (clone_flags & CLONE_FS) {
1544 /* tsk->fs is already what we want */
1545 spin_lock(lock: &fs->lock);
1546 /* "users" and "in_exec" locked for check_unsafe_exec() */
1547 if (fs->in_exec) {
1548 spin_unlock(lock: &fs->lock);
1549 return -EAGAIN;
1550 }
1551 fs->users++;
1552 spin_unlock(lock: &fs->lock);
1553 return 0;
1554 }
1555 tsk->fs = copy_fs_struct(fs);
1556 if (!tsk->fs)
1557 return -ENOMEM;
1558 return 0;
1559}
1560
1561static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1562 int no_files)
1563{
1564 struct files_struct *oldf, *newf;
1565
1566 /*
1567 * A background process may not have any files ...
1568 */
1569 oldf = current->files;
1570 if (!oldf)
1571 return 0;
1572
1573 if (no_files) {
1574 tsk->files = NULL;
1575 return 0;
1576 }
1577
1578 if (clone_flags & CLONE_FILES) {
1579 atomic_inc(v: &oldf->count);
1580 return 0;
1581 }
1582
1583 newf = dup_fd(oldf, NULL);
1584 if (IS_ERR(ptr: newf))
1585 return PTR_ERR(ptr: newf);
1586
1587 tsk->files = newf;
1588 return 0;
1589}
1590
1591static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1592{
1593 struct sighand_struct *sig;
1594
1595 if (clone_flags & CLONE_SIGHAND) {
1596 refcount_inc(r: &current->sighand->count);
1597 return 0;
1598 }
1599 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1600 RCU_INIT_POINTER(tsk->sighand, sig);
1601 if (!sig)
1602 return -ENOMEM;
1603
1604 refcount_set(r: &sig->count, n: 1);
1605 spin_lock_irq(lock: &current->sighand->siglock);
1606 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1607 spin_unlock_irq(lock: &current->sighand->siglock);
1608
1609 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1610 if (clone_flags & CLONE_CLEAR_SIGHAND)
1611 flush_signal_handlers(tsk, force_default: 0);
1612
1613 return 0;
1614}
1615
1616void __cleanup_sighand(struct sighand_struct *sighand)
1617{
1618 if (refcount_dec_and_test(r: &sighand->count)) {
1619 signalfd_cleanup(sighand);
1620 /*
1621 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1622 * without an RCU grace period, see __lock_task_sighand().
1623 */
1624 kmem_cache_free(s: sighand_cachep, objp: sighand);
1625 }
1626}
1627
1628/*
1629 * Initialize POSIX timer handling for a thread group.
1630 */
1631static void posix_cpu_timers_init_group(struct signal_struct *sig)
1632{
1633 struct posix_cputimers *pct = &sig->posix_cputimers;
1634 unsigned long cpu_limit;
1635
1636 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1637 posix_cputimers_group_init(pct, cpu_limit);
1638}
1639
1640static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1641{
1642 struct signal_struct *sig;
1643
1644 if (clone_flags & CLONE_THREAD)
1645 return 0;
1646
1647 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1648 tsk->signal = sig;
1649 if (!sig)
1650 return -ENOMEM;
1651
1652 sig->nr_threads = 1;
1653 sig->quick_threads = 1;
1654 atomic_set(v: &sig->live, i: 1);
1655 refcount_set(r: &sig->sigcnt, n: 1);
1656
1657 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1658 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1659 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1660
1661 init_waitqueue_head(&sig->wait_chldexit);
1662 sig->curr_target = tsk;
1663 init_sigpending(sig: &sig->shared_pending);
1664 INIT_HLIST_HEAD(&sig->multiprocess);
1665 seqlock_init(&sig->stats_lock);
1666 prev_cputime_init(prev: &sig->prev_cputime);
1667
1668#ifdef CONFIG_POSIX_TIMERS
1669 INIT_HLIST_HEAD(&sig->posix_timers);
1670 INIT_HLIST_HEAD(&sig->ignored_posix_timers);
1671 hrtimer_setup(timer: &sig->real_timer, function: it_real_fn, CLOCK_MONOTONIC, mode: HRTIMER_MODE_REL);
1672#endif
1673
1674 task_lock(current->group_leader);
1675 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1676 task_unlock(current->group_leader);
1677
1678 posix_cpu_timers_init_group(sig);
1679
1680 tty_audit_fork(sig);
1681 sched_autogroup_fork(sig);
1682
1683 sig->oom_score_adj = current->signal->oom_score_adj;
1684 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1685
1686 mutex_init(&sig->cred_guard_mutex);
1687 init_rwsem(&sig->exec_update_lock);
1688
1689 return 0;
1690}
1691
1692static void copy_seccomp(struct task_struct *p)
1693{
1694#ifdef CONFIG_SECCOMP
1695 /*
1696 * Must be called with sighand->lock held, which is common to
1697 * all threads in the group. Holding cred_guard_mutex is not
1698 * needed because this new task is not yet running and cannot
1699 * be racing exec.
1700 */
1701 assert_spin_locked(&current->sighand->siglock);
1702
1703 /* Ref-count the new filter user, and assign it. */
1704 get_seccomp_filter(current);
1705 p->seccomp = current->seccomp;
1706
1707 /*
1708 * Explicitly enable no_new_privs here in case it got set
1709 * between the task_struct being duplicated and holding the
1710 * sighand lock. The seccomp state and nnp must be in sync.
1711 */
1712 if (task_no_new_privs(current))
1713 task_set_no_new_privs(p);
1714
1715 /*
1716 * If the parent gained a seccomp mode after copying thread
1717 * flags and between before we held the sighand lock, we have
1718 * to manually enable the seccomp thread flag here.
1719 */
1720 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1721 set_task_syscall_work(p, SECCOMP);
1722#endif
1723}
1724
1725SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1726{
1727 current->clear_child_tid = tidptr;
1728
1729 return task_pid_vnr(current);
1730}
1731
1732static void rt_mutex_init_task(struct task_struct *p)
1733{
1734 raw_spin_lock_init(&p->pi_lock);
1735#ifdef CONFIG_RT_MUTEXES
1736 p->pi_waiters = RB_ROOT_CACHED;
1737 p->pi_top_task = NULL;
1738 p->pi_blocked_on = NULL;
1739#endif
1740}
1741
1742static inline void init_task_pid_links(struct task_struct *task)
1743{
1744 enum pid_type type;
1745
1746 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1747 INIT_HLIST_NODE(h: &task->pid_links[type]);
1748}
1749
1750static inline void
1751init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1752{
1753 if (type == PIDTYPE_PID)
1754 task->thread_pid = pid;
1755 else
1756 task->signal->pids[type] = pid;
1757}
1758
1759static inline void rcu_copy_process(struct task_struct *p)
1760{
1761#ifdef CONFIG_PREEMPT_RCU
1762 p->rcu_read_lock_nesting = 0;
1763 p->rcu_read_unlock_special.s = 0;
1764 p->rcu_blocked_node = NULL;
1765 INIT_LIST_HEAD(list: &p->rcu_node_entry);
1766#endif /* #ifdef CONFIG_PREEMPT_RCU */
1767#ifdef CONFIG_TASKS_RCU
1768 p->rcu_tasks_holdout = false;
1769 INIT_LIST_HEAD(list: &p->rcu_tasks_holdout_list);
1770 p->rcu_tasks_idle_cpu = -1;
1771 INIT_LIST_HEAD(list: &p->rcu_tasks_exit_list);
1772#endif /* #ifdef CONFIG_TASKS_RCU */
1773#ifdef CONFIG_TASKS_TRACE_RCU
1774 p->trc_reader_nesting = 0;
1775 p->trc_reader_special.s = 0;
1776 INIT_LIST_HEAD(list: &p->trc_holdout_list);
1777 INIT_LIST_HEAD(list: &p->trc_blkd_node);
1778#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1779}
1780
1781/**
1782 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
1783 * @pid: the struct pid for which to create a pidfd
1784 * @flags: flags of the new @pidfd
1785 * @ret_file: return the new pidfs file
1786 *
1787 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
1788 * caller's file descriptor table. The pidfd is reserved but not installed yet.
1789 *
1790 * The helper verifies that @pid is still in use, without PIDFD_THREAD the
1791 * task identified by @pid must be a thread-group leader.
1792 *
1793 * If this function returns successfully the caller is responsible to either
1794 * call fd_install() passing the returned pidfd and pidfd file as arguments in
1795 * order to install the pidfd into its file descriptor table or they must use
1796 * put_unused_fd() and fput() on the returned pidfd and pidfd file
1797 * respectively.
1798 *
1799 * This function is useful when a pidfd must already be reserved but there
1800 * might still be points of failure afterwards and the caller wants to ensure
1801 * that no pidfd is leaked into its file descriptor table.
1802 *
1803 * Return: On success, a reserved pidfd is returned from the function and a new
1804 * pidfd file is returned in the last argument to the function. On
1805 * error, a negative error code is returned from the function and the
1806 * last argument remains unchanged.
1807 */
1808int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret_file)
1809{
1810 struct file *pidfs_file;
1811
1812 /*
1813 * PIDFD_STALE is only allowed to be passed if the caller knows
1814 * that @pid is already registered in pidfs and thus
1815 * PIDFD_INFO_EXIT information is guaranteed to be available.
1816 */
1817 if (!(flags & PIDFD_STALE)) {
1818 /*
1819 * While holding the pidfd waitqueue lock removing the
1820 * task linkage for the thread-group leader pid
1821 * (PIDTYPE_TGID) isn't possible. Thus, if there's still
1822 * task linkage for PIDTYPE_PID not having thread-group
1823 * leader linkage for the pid means it wasn't a
1824 * thread-group leader in the first place.
1825 */
1826 guard(spinlock_irq)(l: &pid->wait_pidfd.lock);
1827
1828 /* Task has already been reaped. */
1829 if (!pid_has_task(pid, type: PIDTYPE_PID))
1830 return -ESRCH;
1831 /*
1832 * If this struct pid isn't used as a thread-group
1833 * leader but the caller requested to create a
1834 * thread-group leader pidfd then report ENOENT.
1835 */
1836 if (!(flags & PIDFD_THREAD) && !pid_has_task(pid, type: PIDTYPE_TGID))
1837 return -ENOENT;
1838 }
1839
1840 CLASS(get_unused_fd, pidfd)(O_CLOEXEC);
1841 if (pidfd < 0)
1842 return pidfd;
1843
1844 pidfs_file = pidfs_alloc_file(pid, flags: flags | O_RDWR);
1845 if (IS_ERR(ptr: pidfs_file))
1846 return PTR_ERR(ptr: pidfs_file);
1847
1848 *ret_file = pidfs_file;
1849 return take_fd(pidfd);
1850}
1851
1852static void __delayed_free_task(struct rcu_head *rhp)
1853{
1854 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1855
1856 free_task(tsk);
1857}
1858
1859static __always_inline void delayed_free_task(struct task_struct *tsk)
1860{
1861 if (IS_ENABLED(CONFIG_MEMCG))
1862 call_rcu(head: &tsk->rcu, func: __delayed_free_task);
1863 else
1864 free_task(tsk);
1865}
1866
1867static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1868{
1869 /* Skip if kernel thread */
1870 if (!tsk->mm)
1871 return;
1872
1873 /* Skip if spawning a thread or using vfork */
1874 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1875 return;
1876
1877 /* We need to synchronize with __set_oom_adj */
1878 mutex_lock(&oom_adj_mutex);
1879 set_bit(MMF_MULTIPROCESS, addr: &tsk->mm->flags);
1880 /* Update the values in case they were changed after copy_signal */
1881 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1882 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1883 mutex_unlock(lock: &oom_adj_mutex);
1884}
1885
1886#ifdef CONFIG_RV
1887static void rv_task_fork(struct task_struct *p)
1888{
1889 int i;
1890
1891 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
1892 p->rv[i].da_mon.monitoring = false;
1893}
1894#else
1895#define rv_task_fork(p) do {} while (0)
1896#endif
1897
1898static bool need_futex_hash_allocate_default(u64 clone_flags)
1899{
1900 if ((clone_flags & (CLONE_THREAD | CLONE_VM)) != (CLONE_THREAD | CLONE_VM))
1901 return false;
1902 return true;
1903}
1904
1905/*
1906 * This creates a new process as a copy of the old one,
1907 * but does not actually start it yet.
1908 *
1909 * It copies the registers, and all the appropriate
1910 * parts of the process environment (as per the clone
1911 * flags). The actual kick-off is left to the caller.
1912 */
1913__latent_entropy struct task_struct *copy_process(
1914 struct pid *pid,
1915 int trace,
1916 int node,
1917 struct kernel_clone_args *args)
1918{
1919 int pidfd = -1, retval;
1920 struct task_struct *p;
1921 struct multiprocess_signals delayed;
1922 struct file *pidfile = NULL;
1923 const u64 clone_flags = args->flags;
1924 struct nsproxy *nsp = current->nsproxy;
1925
1926 /*
1927 * Don't allow sharing the root directory with processes in a different
1928 * namespace
1929 */
1930 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1931 return ERR_PTR(error: -EINVAL);
1932
1933 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1934 return ERR_PTR(error: -EINVAL);
1935
1936 /*
1937 * Thread groups must share signals as well, and detached threads
1938 * can only be started up within the thread group.
1939 */
1940 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1941 return ERR_PTR(error: -EINVAL);
1942
1943 /*
1944 * Shared signal handlers imply shared VM. By way of the above,
1945 * thread groups also imply shared VM. Blocking this case allows
1946 * for various simplifications in other code.
1947 */
1948 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1949 return ERR_PTR(error: -EINVAL);
1950
1951 /*
1952 * Siblings of global init remain as zombies on exit since they are
1953 * not reaped by their parent (swapper). To solve this and to avoid
1954 * multi-rooted process trees, prevent global and container-inits
1955 * from creating siblings.
1956 */
1957 if ((clone_flags & CLONE_PARENT) &&
1958 current->signal->flags & SIGNAL_UNKILLABLE)
1959 return ERR_PTR(error: -EINVAL);
1960
1961 /*
1962 * If the new process will be in a different pid or user namespace
1963 * do not allow it to share a thread group with the forking task.
1964 */
1965 if (clone_flags & CLONE_THREAD) {
1966 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1967 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1968 return ERR_PTR(error: -EINVAL);
1969 }
1970
1971 if (clone_flags & CLONE_PIDFD) {
1972 /*
1973 * - CLONE_DETACHED is blocked so that we can potentially
1974 * reuse it later for CLONE_PIDFD.
1975 */
1976 if (clone_flags & CLONE_DETACHED)
1977 return ERR_PTR(error: -EINVAL);
1978 }
1979
1980 /*
1981 * Force any signals received before this point to be delivered
1982 * before the fork happens. Collect up signals sent to multiple
1983 * processes that happen during the fork and delay them so that
1984 * they appear to happen after the fork.
1985 */
1986 sigemptyset(set: &delayed.signal);
1987 INIT_HLIST_NODE(h: &delayed.node);
1988
1989 spin_lock_irq(lock: &current->sighand->siglock);
1990 if (!(clone_flags & CLONE_THREAD))
1991 hlist_add_head(n: &delayed.node, h: &current->signal->multiprocess);
1992 recalc_sigpending();
1993 spin_unlock_irq(lock: &current->sighand->siglock);
1994 retval = -ERESTARTNOINTR;
1995 if (task_sigpending(current))
1996 goto fork_out;
1997
1998 retval = -ENOMEM;
1999 p = dup_task_struct(current, node);
2000 if (!p)
2001 goto fork_out;
2002 p->flags &= ~PF_KTHREAD;
2003 if (args->kthread)
2004 p->flags |= PF_KTHREAD;
2005 if (args->user_worker) {
2006 /*
2007 * Mark us a user worker, and block any signal that isn't
2008 * fatal or STOP
2009 */
2010 p->flags |= PF_USER_WORKER;
2011 siginitsetinv(set: &p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2012 }
2013 if (args->io_thread)
2014 p->flags |= PF_IO_WORKER;
2015
2016 if (args->name)
2017 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2018
2019 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2020 /*
2021 * Clear TID on mm_release()?
2022 */
2023 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2024
2025 ftrace_graph_init_task(t: p);
2026
2027 rt_mutex_init_task(p);
2028
2029 lockdep_assert_irqs_enabled();
2030#ifdef CONFIG_PROVE_LOCKING
2031 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2032#endif
2033 retval = copy_creds(p, clone_flags);
2034 if (retval < 0)
2035 goto bad_fork_free;
2036
2037 retval = -EAGAIN;
2038 if (is_rlimit_overlimit(task_ucounts(p), type: UCOUNT_RLIMIT_NPROC, max: rlimit(RLIMIT_NPROC))) {
2039 if (p->real_cred->user != INIT_USER &&
2040 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2041 goto bad_fork_cleanup_count;
2042 }
2043 current->flags &= ~PF_NPROC_EXCEEDED;
2044
2045 /*
2046 * If multiple threads are within copy_process(), then this check
2047 * triggers too late. This doesn't hurt, the check is only there
2048 * to stop root fork bombs.
2049 */
2050 retval = -EAGAIN;
2051 if (data_race(nr_threads >= max_threads))
2052 goto bad_fork_cleanup_count;
2053
2054 delayacct_tsk_init(tsk: p); /* Must remain after dup_task_struct() */
2055 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2056 p->flags |= PF_FORKNOEXEC;
2057 INIT_LIST_HEAD(list: &p->children);
2058 INIT_LIST_HEAD(list: &p->sibling);
2059 rcu_copy_process(p);
2060 p->vfork_done = NULL;
2061 spin_lock_init(&p->alloc_lock);
2062
2063 init_sigpending(sig: &p->pending);
2064
2065 p->utime = p->stime = p->gtime = 0;
2066#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2067 p->utimescaled = p->stimescaled = 0;
2068#endif
2069 prev_cputime_init(prev: &p->prev_cputime);
2070
2071#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2072 seqcount_init(&p->vtime.seqcount);
2073 p->vtime.starttime = 0;
2074 p->vtime.state = VTIME_INACTIVE;
2075#endif
2076
2077#ifdef CONFIG_IO_URING
2078 p->io_uring = NULL;
2079#endif
2080
2081 p->default_timer_slack_ns = current->timer_slack_ns;
2082
2083#ifdef CONFIG_PSI
2084 p->psi_flags = 0;
2085#endif
2086
2087 task_io_accounting_init(ioac: &p->ioac);
2088 acct_clear_integrals(tsk: p);
2089
2090 posix_cputimers_init(pct: &p->posix_cputimers);
2091 tick_dep_init_task(tsk: p);
2092
2093 p->io_context = NULL;
2094 audit_set_context(task: p, NULL);
2095 cgroup_fork(p);
2096 if (args->kthread) {
2097 if (!set_kthread_struct(p))
2098 goto bad_fork_cleanup_delayacct;
2099 }
2100#ifdef CONFIG_NUMA
2101 p->mempolicy = mpol_dup(pol: p->mempolicy);
2102 if (IS_ERR(ptr: p->mempolicy)) {
2103 retval = PTR_ERR(ptr: p->mempolicy);
2104 p->mempolicy = NULL;
2105 goto bad_fork_cleanup_delayacct;
2106 }
2107#endif
2108#ifdef CONFIG_CPUSETS
2109 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2110 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2111#endif
2112#ifdef CONFIG_TRACE_IRQFLAGS
2113 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2114 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2115 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2116 p->softirqs_enabled = 1;
2117 p->softirq_context = 0;
2118#endif
2119
2120 p->pagefault_disabled = 0;
2121
2122#ifdef CONFIG_LOCKDEP
2123 lockdep_init_task(task: p);
2124#endif
2125
2126#ifdef CONFIG_DEBUG_MUTEXES
2127 p->blocked_on = NULL; /* not blocked yet */
2128#endif
2129#ifdef CONFIG_BCACHE
2130 p->sequential_io = 0;
2131 p->sequential_io_avg = 0;
2132#endif
2133#ifdef CONFIG_BPF_SYSCALL
2134 RCU_INIT_POINTER(p->bpf_storage, NULL);
2135 p->bpf_ctx = NULL;
2136#endif
2137
2138 /* Perform scheduler related setup. Assign this task to a CPU. */
2139 retval = sched_fork(clone_flags, p);
2140 if (retval)
2141 goto bad_fork_cleanup_policy;
2142
2143 retval = perf_event_init_task(child: p, clone_flags);
2144 if (retval)
2145 goto bad_fork_sched_cancel_fork;
2146 retval = audit_alloc(task: p);
2147 if (retval)
2148 goto bad_fork_cleanup_perf;
2149 /* copy all the process information */
2150 shm_init_task(p);
2151 retval = security_task_alloc(task: p, clone_flags);
2152 if (retval)
2153 goto bad_fork_cleanup_audit;
2154 retval = copy_semundo(clone_flags, tsk: p);
2155 if (retval)
2156 goto bad_fork_cleanup_security;
2157 retval = copy_files(clone_flags, tsk: p, no_files: args->no_files);
2158 if (retval)
2159 goto bad_fork_cleanup_semundo;
2160 retval = copy_fs(clone_flags, tsk: p);
2161 if (retval)
2162 goto bad_fork_cleanup_files;
2163 retval = copy_sighand(clone_flags, tsk: p);
2164 if (retval)
2165 goto bad_fork_cleanup_fs;
2166 retval = copy_signal(clone_flags, tsk: p);
2167 if (retval)
2168 goto bad_fork_cleanup_sighand;
2169 retval = copy_mm(clone_flags, tsk: p);
2170 if (retval)
2171 goto bad_fork_cleanup_signal;
2172 retval = copy_namespaces(flags: clone_flags, tsk: p);
2173 if (retval)
2174 goto bad_fork_cleanup_mm;
2175 retval = copy_io(clone_flags, tsk: p);
2176 if (retval)
2177 goto bad_fork_cleanup_namespaces;
2178 retval = copy_thread(p, args);
2179 if (retval)
2180 goto bad_fork_cleanup_io;
2181
2182 stackleak_task_init(t: p);
2183
2184 if (pid != &init_struct_pid) {
2185 pid = alloc_pid(ns: p->nsproxy->pid_ns_for_children, set_tid: args->set_tid,
2186 set_tid_size: args->set_tid_size);
2187 if (IS_ERR(ptr: pid)) {
2188 retval = PTR_ERR(ptr: pid);
2189 goto bad_fork_cleanup_thread;
2190 }
2191 }
2192
2193 /*
2194 * This has to happen after we've potentially unshared the file
2195 * descriptor table (so that the pidfd doesn't leak into the child
2196 * if the fd table isn't shared).
2197 */
2198 if (clone_flags & CLONE_PIDFD) {
2199 int flags = (clone_flags & CLONE_THREAD) ? PIDFD_THREAD : 0;
2200
2201 /*
2202 * Note that no task has been attached to @pid yet indicate
2203 * that via CLONE_PIDFD.
2204 */
2205 retval = pidfd_prepare(pid, flags: flags | PIDFD_STALE, ret_file: &pidfile);
2206 if (retval < 0)
2207 goto bad_fork_free_pid;
2208 pidfd = retval;
2209
2210 retval = put_user(pidfd, args->pidfd);
2211 if (retval)
2212 goto bad_fork_put_pidfd;
2213 }
2214
2215#ifdef CONFIG_BLOCK
2216 p->plug = NULL;
2217#endif
2218 futex_init_task(tsk: p);
2219
2220 /*
2221 * sigaltstack should be cleared when sharing the same VM
2222 */
2223 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2224 sas_ss_reset(p);
2225
2226 /*
2227 * Syscall tracing and stepping should be turned off in the
2228 * child regardless of CLONE_PTRACE.
2229 */
2230 user_disable_single_step(p);
2231 clear_task_syscall_work(p, SYSCALL_TRACE);
2232#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2233 clear_task_syscall_work(p, SYSCALL_EMU);
2234#endif
2235 clear_tsk_latency_tracing(p);
2236
2237 /* ok, now we should be set up.. */
2238 p->pid = pid_nr(pid);
2239 if (clone_flags & CLONE_THREAD) {
2240 p->group_leader = current->group_leader;
2241 p->tgid = current->tgid;
2242 } else {
2243 p->group_leader = p;
2244 p->tgid = p->pid;
2245 }
2246
2247 p->nr_dirtied = 0;
2248 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2249 p->dirty_paused_when = 0;
2250
2251 p->pdeath_signal = 0;
2252 p->task_works = NULL;
2253 clear_posix_cputimers_work(p);
2254
2255#ifdef CONFIG_KRETPROBES
2256 p->kretprobe_instances.first = NULL;
2257#endif
2258#ifdef CONFIG_RETHOOK
2259 p->rethooks.first = NULL;
2260#endif
2261
2262 /*
2263 * Ensure that the cgroup subsystem policies allow the new process to be
2264 * forked. It should be noted that the new process's css_set can be changed
2265 * between here and cgroup_post_fork() if an organisation operation is in
2266 * progress.
2267 */
2268 retval = cgroup_can_fork(p, kargs: args);
2269 if (retval)
2270 goto bad_fork_put_pidfd;
2271
2272 /*
2273 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2274 * the new task on the correct runqueue. All this *before* the task
2275 * becomes visible.
2276 *
2277 * This isn't part of ->can_fork() because while the re-cloning is
2278 * cgroup specific, it unconditionally needs to place the task on a
2279 * runqueue.
2280 */
2281 retval = sched_cgroup_fork(p, kargs: args);
2282 if (retval)
2283 goto bad_fork_cancel_cgroup;
2284
2285 /*
2286 * Allocate a default futex hash for the user process once the first
2287 * thread spawns.
2288 */
2289 if (need_futex_hash_allocate_default(clone_flags)) {
2290 retval = futex_hash_allocate_default();
2291 if (retval)
2292 goto bad_fork_core_free;
2293 /*
2294 * If we fail beyond this point we don't free the allocated
2295 * futex hash map. We assume that another thread will be created
2296 * and makes use of it. The hash map will be freed once the main
2297 * thread terminates.
2298 */
2299 }
2300 /*
2301 * From this point on we must avoid any synchronous user-space
2302 * communication until we take the tasklist-lock. In particular, we do
2303 * not want user-space to be able to predict the process start-time by
2304 * stalling fork(2) after we recorded the start_time but before it is
2305 * visible to the system.
2306 */
2307
2308 p->start_time = ktime_get_ns();
2309 p->start_boottime = ktime_get_boottime_ns();
2310
2311 /*
2312 * Make it visible to the rest of the system, but dont wake it up yet.
2313 * Need tasklist lock for parent etc handling!
2314 */
2315 write_lock_irq(&tasklist_lock);
2316
2317 /* CLONE_PARENT re-uses the old parent */
2318 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2319 p->real_parent = current->real_parent;
2320 p->parent_exec_id = current->parent_exec_id;
2321 if (clone_flags & CLONE_THREAD)
2322 p->exit_signal = -1;
2323 else
2324 p->exit_signal = current->group_leader->exit_signal;
2325 } else {
2326 p->real_parent = current;
2327 p->parent_exec_id = current->self_exec_id;
2328 p->exit_signal = args->exit_signal;
2329 }
2330
2331 klp_copy_process(child: p);
2332
2333 sched_core_fork(p);
2334
2335 spin_lock(lock: &current->sighand->siglock);
2336
2337 rv_task_fork(p);
2338
2339 rseq_fork(t: p, clone_flags);
2340
2341 /* Don't start children in a dying pid namespace */
2342 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2343 retval = -ENOMEM;
2344 goto bad_fork_core_free;
2345 }
2346
2347 /* Let kill terminate clone/fork in the middle */
2348 if (fatal_signal_pending(current)) {
2349 retval = -EINTR;
2350 goto bad_fork_core_free;
2351 }
2352
2353 /* No more failure paths after this point. */
2354
2355 /*
2356 * Copy seccomp details explicitly here, in case they were changed
2357 * before holding sighand lock.
2358 */
2359 copy_seccomp(p);
2360
2361 init_task_pid_links(task: p);
2362 if (likely(p->pid)) {
2363 ptrace_init_task(child: p, ptrace: (clone_flags & CLONE_PTRACE) || trace);
2364
2365 init_task_pid(task: p, type: PIDTYPE_PID, pid);
2366 if (thread_group_leader(p)) {
2367 init_task_pid(task: p, type: PIDTYPE_TGID, pid);
2368 init_task_pid(task: p, type: PIDTYPE_PGID, pid: task_pgrp(current));
2369 init_task_pid(task: p, type: PIDTYPE_SID, pid: task_session(current));
2370
2371 if (is_child_reaper(pid)) {
2372 ns_of_pid(pid)->child_reaper = p;
2373 p->signal->flags |= SIGNAL_UNKILLABLE;
2374 }
2375 p->signal->shared_pending.signal = delayed.signal;
2376 p->signal->tty = tty_kref_get(current->signal->tty);
2377 /*
2378 * Inherit has_child_subreaper flag under the same
2379 * tasklist_lock with adding child to the process tree
2380 * for propagate_has_child_subreaper optimization.
2381 */
2382 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2383 p->real_parent->signal->is_child_subreaper;
2384 list_add_tail(new: &p->sibling, head: &p->real_parent->children);
2385 list_add_tail_rcu(new: &p->tasks, head: &init_task.tasks);
2386 attach_pid(task: p, PIDTYPE_TGID);
2387 attach_pid(task: p, PIDTYPE_PGID);
2388 attach_pid(task: p, PIDTYPE_SID);
2389 __this_cpu_inc(process_counts);
2390 } else {
2391 current->signal->nr_threads++;
2392 current->signal->quick_threads++;
2393 atomic_inc(v: &current->signal->live);
2394 refcount_inc(r: &current->signal->sigcnt);
2395 task_join_group_stop(task: p);
2396 list_add_tail_rcu(new: &p->thread_node,
2397 head: &p->signal->thread_head);
2398 }
2399 attach_pid(task: p, PIDTYPE_PID);
2400 nr_threads++;
2401 }
2402 total_forks++;
2403 hlist_del_init(n: &delayed.node);
2404 spin_unlock(lock: &current->sighand->siglock);
2405 syscall_tracepoint_update(p);
2406 write_unlock_irq(&tasklist_lock);
2407
2408 if (pidfile)
2409 fd_install(fd: pidfd, file: pidfile);
2410
2411 proc_fork_connector(task: p);
2412 sched_post_fork(p);
2413 cgroup_post_fork(p, kargs: args);
2414 perf_event_fork(tsk: p);
2415
2416 trace_task_newtask(task: p, clone_flags);
2417 uprobe_copy_process(t: p, flags: clone_flags);
2418 user_events_fork(t: p, clone_flags);
2419
2420 copy_oom_score_adj(clone_flags, tsk: p);
2421
2422 return p;
2423
2424bad_fork_core_free:
2425 sched_core_free(tsk: p);
2426 spin_unlock(lock: &current->sighand->siglock);
2427 write_unlock_irq(&tasklist_lock);
2428bad_fork_cancel_cgroup:
2429 cgroup_cancel_fork(p, kargs: args);
2430bad_fork_put_pidfd:
2431 if (clone_flags & CLONE_PIDFD) {
2432 fput(pidfile);
2433 put_unused_fd(fd: pidfd);
2434 }
2435bad_fork_free_pid:
2436 if (pid != &init_struct_pid)
2437 free_pid(pid);
2438bad_fork_cleanup_thread:
2439 exit_thread(tsk: p);
2440bad_fork_cleanup_io:
2441 if (p->io_context)
2442 exit_io_context(task: p);
2443bad_fork_cleanup_namespaces:
2444 exit_task_namespaces(tsk: p);
2445bad_fork_cleanup_mm:
2446 if (p->mm) {
2447 mm_clear_owner(mm: p->mm, p);
2448 mmput(p->mm);
2449 }
2450bad_fork_cleanup_signal:
2451 if (!(clone_flags & CLONE_THREAD))
2452 free_signal_struct(sig: p->signal);
2453bad_fork_cleanup_sighand:
2454 __cleanup_sighand(sighand: p->sighand);
2455bad_fork_cleanup_fs:
2456 exit_fs(p); /* blocking */
2457bad_fork_cleanup_files:
2458 exit_files(p); /* blocking */
2459bad_fork_cleanup_semundo:
2460 exit_sem(tsk: p);
2461bad_fork_cleanup_security:
2462 security_task_free(task: p);
2463bad_fork_cleanup_audit:
2464 audit_free(task: p);
2465bad_fork_cleanup_perf:
2466 perf_event_free_task(task: p);
2467bad_fork_sched_cancel_fork:
2468 sched_cancel_fork(p);
2469bad_fork_cleanup_policy:
2470 lockdep_free_task(task: p);
2471#ifdef CONFIG_NUMA
2472 mpol_put(pol: p->mempolicy);
2473#endif
2474bad_fork_cleanup_delayacct:
2475 delayacct_tsk_free(tsk: p);
2476bad_fork_cleanup_count:
2477 dec_rlimit_ucounts(task_ucounts(p), type: UCOUNT_RLIMIT_NPROC, v: 1);
2478 exit_creds(p);
2479bad_fork_free:
2480 WRITE_ONCE(p->__state, TASK_DEAD);
2481 exit_task_stack_account(tsk: p);
2482 put_task_stack(tsk: p);
2483 delayed_free_task(tsk: p);
2484fork_out:
2485 spin_lock_irq(lock: &current->sighand->siglock);
2486 hlist_del_init(n: &delayed.node);
2487 spin_unlock_irq(lock: &current->sighand->siglock);
2488 return ERR_PTR(error: retval);
2489}
2490
2491static inline void init_idle_pids(struct task_struct *idle)
2492{
2493 enum pid_type type;
2494
2495 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2496 INIT_HLIST_NODE(h: &idle->pid_links[type]); /* not really needed */
2497 init_task_pid(task: idle, type, pid: &init_struct_pid);
2498 }
2499}
2500
2501static int idle_dummy(void *dummy)
2502{
2503 /* This function is never called */
2504 return 0;
2505}
2506
2507struct task_struct * __init fork_idle(int cpu)
2508{
2509 struct task_struct *task;
2510 struct kernel_clone_args args = {
2511 .flags = CLONE_VM,
2512 .fn = &idle_dummy,
2513 .fn_arg = NULL,
2514 .kthread = 1,
2515 .idle = 1,
2516 };
2517
2518 task = copy_process(pid: &init_struct_pid, trace: 0, cpu_to_node(cpu), args: &args);
2519 if (!IS_ERR(ptr: task)) {
2520 init_idle_pids(idle: task);
2521 init_idle(idle: task, cpu);
2522 }
2523
2524 return task;
2525}
2526
2527/*
2528 * This is like kernel_clone(), but shaved down and tailored to just
2529 * creating io_uring workers. It returns a created task, or an error pointer.
2530 * The returned task is inactive, and the caller must fire it up through
2531 * wake_up_new_task(p). All signals are blocked in the created task.
2532 */
2533struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2534{
2535 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2536 CLONE_IO;
2537 struct kernel_clone_args args = {
2538 .flags = ((lower_32_bits(flags) | CLONE_VM |
2539 CLONE_UNTRACED) & ~CSIGNAL),
2540 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2541 .fn = fn,
2542 .fn_arg = arg,
2543 .io_thread = 1,
2544 .user_worker = 1,
2545 };
2546
2547 return copy_process(NULL, trace: 0, node, args: &args);
2548}
2549
2550/*
2551 * Ok, this is the main fork-routine.
2552 *
2553 * It copies the process, and if successful kick-starts
2554 * it and waits for it to finish using the VM if required.
2555 *
2556 * args->exit_signal is expected to be checked for sanity by the caller.
2557 */
2558pid_t kernel_clone(struct kernel_clone_args *args)
2559{
2560 u64 clone_flags = args->flags;
2561 struct completion vfork;
2562 struct pid *pid;
2563 struct task_struct *p;
2564 int trace = 0;
2565 pid_t nr;
2566
2567 /*
2568 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2569 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2570 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2571 * field in struct clone_args and it still doesn't make sense to have
2572 * them both point at the same memory location. Performing this check
2573 * here has the advantage that we don't need to have a separate helper
2574 * to check for legacy clone().
2575 */
2576 if ((clone_flags & CLONE_PIDFD) &&
2577 (clone_flags & CLONE_PARENT_SETTID) &&
2578 (args->pidfd == args->parent_tid))
2579 return -EINVAL;
2580
2581 /*
2582 * Determine whether and which event to report to ptracer. When
2583 * called from kernel_thread or CLONE_UNTRACED is explicitly
2584 * requested, no event is reported; otherwise, report if the event
2585 * for the type of forking is enabled.
2586 */
2587 if (!(clone_flags & CLONE_UNTRACED)) {
2588 if (clone_flags & CLONE_VFORK)
2589 trace = PTRACE_EVENT_VFORK;
2590 else if (args->exit_signal != SIGCHLD)
2591 trace = PTRACE_EVENT_CLONE;
2592 else
2593 trace = PTRACE_EVENT_FORK;
2594
2595 if (likely(!ptrace_event_enabled(current, trace)))
2596 trace = 0;
2597 }
2598
2599 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2600 add_latent_entropy();
2601
2602 if (IS_ERR(ptr: p))
2603 return PTR_ERR(ptr: p);
2604
2605 /*
2606 * Do this prior waking up the new thread - the thread pointer
2607 * might get invalid after that point, if the thread exits quickly.
2608 */
2609 trace_sched_process_fork(current, child: p);
2610
2611 pid = get_task_pid(task: p, type: PIDTYPE_PID);
2612 nr = pid_vnr(pid);
2613
2614 if (clone_flags & CLONE_PARENT_SETTID)
2615 put_user(nr, args->parent_tid);
2616
2617 if (clone_flags & CLONE_VFORK) {
2618 p->vfork_done = &vfork;
2619 init_completion(x: &vfork);
2620 get_task_struct(t: p);
2621 }
2622
2623 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2624 /* lock the task to synchronize with memcg migration */
2625 task_lock(p);
2626 lru_gen_add_mm(mm: p->mm);
2627 task_unlock(p);
2628 }
2629
2630 wake_up_new_task(tsk: p);
2631
2632 /* forking complete and child started to run, tell ptracer */
2633 if (unlikely(trace))
2634 ptrace_event_pid(event: trace, pid);
2635
2636 if (clone_flags & CLONE_VFORK) {
2637 if (!wait_for_vfork_done(child: p, vfork: &vfork))
2638 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2639 }
2640
2641 put_pid(pid);
2642 return nr;
2643}
2644
2645/*
2646 * Create a kernel thread.
2647 */
2648pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2649 unsigned long flags)
2650{
2651 struct kernel_clone_args args = {
2652 .flags = ((lower_32_bits(flags) | CLONE_VM |
2653 CLONE_UNTRACED) & ~CSIGNAL),
2654 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2655 .fn = fn,
2656 .fn_arg = arg,
2657 .name = name,
2658 .kthread = 1,
2659 };
2660
2661 return kernel_clone(args: &args);
2662}
2663
2664/*
2665 * Create a user mode thread.
2666 */
2667pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2668{
2669 struct kernel_clone_args args = {
2670 .flags = ((lower_32_bits(flags) | CLONE_VM |
2671 CLONE_UNTRACED) & ~CSIGNAL),
2672 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2673 .fn = fn,
2674 .fn_arg = arg,
2675 };
2676
2677 return kernel_clone(args: &args);
2678}
2679
2680#ifdef __ARCH_WANT_SYS_FORK
2681SYSCALL_DEFINE0(fork)
2682{
2683#ifdef CONFIG_MMU
2684 struct kernel_clone_args args = {
2685 .exit_signal = SIGCHLD,
2686 };
2687
2688 return kernel_clone(args: &args);
2689#else
2690 /* can not support in nommu mode */
2691 return -EINVAL;
2692#endif
2693}
2694#endif
2695
2696#ifdef __ARCH_WANT_SYS_VFORK
2697SYSCALL_DEFINE0(vfork)
2698{
2699 struct kernel_clone_args args = {
2700 .flags = CLONE_VFORK | CLONE_VM,
2701 .exit_signal = SIGCHLD,
2702 };
2703
2704 return kernel_clone(args: &args);
2705}
2706#endif
2707
2708#ifdef __ARCH_WANT_SYS_CLONE
2709#ifdef CONFIG_CLONE_BACKWARDS
2710SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2711 int __user *, parent_tidptr,
2712 unsigned long, tls,
2713 int __user *, child_tidptr)
2714#elif defined(CONFIG_CLONE_BACKWARDS2)
2715SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2716 int __user *, parent_tidptr,
2717 int __user *, child_tidptr,
2718 unsigned long, tls)
2719#elif defined(CONFIG_CLONE_BACKWARDS3)
2720SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2721 int, stack_size,
2722 int __user *, parent_tidptr,
2723 int __user *, child_tidptr,
2724 unsigned long, tls)
2725#else
2726SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2727 int __user *, parent_tidptr,
2728 int __user *, child_tidptr,
2729 unsigned long, tls)
2730#endif
2731{
2732 struct kernel_clone_args args = {
2733 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2734 .pidfd = parent_tidptr,
2735 .child_tid = child_tidptr,
2736 .parent_tid = parent_tidptr,
2737 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2738 .stack = newsp,
2739 .tls = tls,
2740 };
2741
2742 return kernel_clone(args: &args);
2743}
2744#endif
2745
2746noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2747 struct clone_args __user *uargs,
2748 size_t usize)
2749{
2750 int err;
2751 struct clone_args args;
2752 pid_t *kset_tid = kargs->set_tid;
2753
2754 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2755 CLONE_ARGS_SIZE_VER0);
2756 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2757 CLONE_ARGS_SIZE_VER1);
2758 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2759 CLONE_ARGS_SIZE_VER2);
2760 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2761
2762 if (unlikely(usize > PAGE_SIZE))
2763 return -E2BIG;
2764 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2765 return -EINVAL;
2766
2767 err = copy_struct_from_user(dst: &args, ksize: sizeof(args), src: uargs, usize);
2768 if (err)
2769 return err;
2770
2771 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2772 return -EINVAL;
2773
2774 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2775 return -EINVAL;
2776
2777 if (unlikely(args.set_tid && args.set_tid_size == 0))
2778 return -EINVAL;
2779
2780 /*
2781 * Verify that higher 32bits of exit_signal are unset and that
2782 * it is a valid signal
2783 */
2784 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2785 !valid_signal(args.exit_signal)))
2786 return -EINVAL;
2787
2788 if ((args.flags & CLONE_INTO_CGROUP) &&
2789 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2790 return -EINVAL;
2791
2792 *kargs = (struct kernel_clone_args){
2793 .flags = args.flags,
2794 .pidfd = u64_to_user_ptr(args.pidfd),
2795 .child_tid = u64_to_user_ptr(args.child_tid),
2796 .parent_tid = u64_to_user_ptr(args.parent_tid),
2797 .exit_signal = args.exit_signal,
2798 .stack = args.stack,
2799 .stack_size = args.stack_size,
2800 .tls = args.tls,
2801 .set_tid_size = args.set_tid_size,
2802 .cgroup = args.cgroup,
2803 };
2804
2805 if (args.set_tid &&
2806 copy_from_user(to: kset_tid, u64_to_user_ptr(args.set_tid),
2807 n: (kargs->set_tid_size * sizeof(pid_t))))
2808 return -EFAULT;
2809
2810 kargs->set_tid = kset_tid;
2811
2812 return 0;
2813}
2814
2815/**
2816 * clone3_stack_valid - check and prepare stack
2817 * @kargs: kernel clone args
2818 *
2819 * Verify that the stack arguments userspace gave us are sane.
2820 * In addition, set the stack direction for userspace since it's easy for us to
2821 * determine.
2822 */
2823static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2824{
2825 if (kargs->stack == 0) {
2826 if (kargs->stack_size > 0)
2827 return false;
2828 } else {
2829 if (kargs->stack_size == 0)
2830 return false;
2831
2832 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2833 return false;
2834
2835#if !defined(CONFIG_STACK_GROWSUP)
2836 kargs->stack += kargs->stack_size;
2837#endif
2838 }
2839
2840 return true;
2841}
2842
2843static bool clone3_args_valid(struct kernel_clone_args *kargs)
2844{
2845 /* Verify that no unknown flags are passed along. */
2846 if (kargs->flags &
2847 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2848 return false;
2849
2850 /*
2851 * - make the CLONE_DETACHED bit reusable for clone3
2852 * - make the CSIGNAL bits reusable for clone3
2853 */
2854 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
2855 return false;
2856
2857 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2858 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2859 return false;
2860
2861 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2862 kargs->exit_signal)
2863 return false;
2864
2865 if (!clone3_stack_valid(kargs))
2866 return false;
2867
2868 return true;
2869}
2870
2871/**
2872 * sys_clone3 - create a new process with specific properties
2873 * @uargs: argument structure
2874 * @size: size of @uargs
2875 *
2876 * clone3() is the extensible successor to clone()/clone2().
2877 * It takes a struct as argument that is versioned by its size.
2878 *
2879 * Return: On success, a positive PID for the child process.
2880 * On error, a negative errno number.
2881 */
2882SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2883{
2884 int err;
2885
2886 struct kernel_clone_args kargs;
2887 pid_t set_tid[MAX_PID_NS_LEVEL];
2888
2889#ifdef __ARCH_BROKEN_SYS_CLONE3
2890#warning clone3() entry point is missing, please fix
2891 return -ENOSYS;
2892#endif
2893
2894 kargs.set_tid = set_tid;
2895
2896 err = copy_clone_args_from_user(kargs: &kargs, uargs, usize: size);
2897 if (err)
2898 return err;
2899
2900 if (!clone3_args_valid(kargs: &kargs))
2901 return -EINVAL;
2902
2903 return kernel_clone(args: &kargs);
2904}
2905
2906void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2907{
2908 struct task_struct *leader, *parent, *child;
2909 int res;
2910
2911 read_lock(&tasklist_lock);
2912 leader = top = top->group_leader;
2913down:
2914 for_each_thread(leader, parent) {
2915 list_for_each_entry(child, &parent->children, sibling) {
2916 res = visitor(child, data);
2917 if (res) {
2918 if (res < 0)
2919 goto out;
2920 leader = child;
2921 goto down;
2922 }
2923up:
2924 ;
2925 }
2926 }
2927
2928 if (leader != top) {
2929 child = leader;
2930 parent = child->real_parent;
2931 leader = parent->group_leader;
2932 goto up;
2933 }
2934out:
2935 read_unlock(&tasklist_lock);
2936}
2937
2938#ifndef ARCH_MIN_MMSTRUCT_ALIGN
2939#define ARCH_MIN_MMSTRUCT_ALIGN 0
2940#endif
2941
2942static void sighand_ctor(void *data)
2943{
2944 struct sighand_struct *sighand = data;
2945
2946 spin_lock_init(&sighand->siglock);
2947 init_waitqueue_head(&sighand->signalfd_wqh);
2948}
2949
2950void __init mm_cache_init(void)
2951{
2952 unsigned int mm_size;
2953
2954 /*
2955 * The mm_cpumask is located at the end of mm_struct, and is
2956 * dynamically sized based on the maximum CPU number this system
2957 * can have, taking hotplug into account (nr_cpu_ids).
2958 */
2959 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
2960
2961 mm_cachep = kmem_cache_create_usercopy(name: "mm_struct",
2962 size: mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2963 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2964 offsetof(struct mm_struct, saved_auxv),
2965 sizeof_field(struct mm_struct, saved_auxv),
2966 NULL);
2967}
2968
2969void __init proc_caches_init(void)
2970{
2971 sighand_cachep = kmem_cache_create("sighand_cache",
2972 sizeof(struct sighand_struct), 0,
2973 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2974 SLAB_ACCOUNT, sighand_ctor);
2975 signal_cachep = kmem_cache_create("signal_cache",
2976 sizeof(struct signal_struct), 0,
2977 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2978 NULL);
2979 files_cachep = kmem_cache_create("files_cache",
2980 sizeof(struct files_struct), 0,
2981 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2982 NULL);
2983 fs_cachep = kmem_cache_create("fs_cache",
2984 sizeof(struct fs_struct), 0,
2985 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2986 NULL);
2987 mmap_init();
2988 nsproxy_cache_init();
2989}
2990
2991/*
2992 * Check constraints on flags passed to the unshare system call.
2993 */
2994static int check_unshare_flags(unsigned long unshare_flags)
2995{
2996 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2997 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2998 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2999 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3000 CLONE_NEWTIME))
3001 return -EINVAL;
3002 /*
3003 * Not implemented, but pretend it works if there is nothing
3004 * to unshare. Note that unsharing the address space or the
3005 * signal handlers also need to unshare the signal queues (aka
3006 * CLONE_THREAD).
3007 */
3008 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3009 if (!thread_group_empty(current))
3010 return -EINVAL;
3011 }
3012 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3013 if (refcount_read(r: &current->sighand->count) > 1)
3014 return -EINVAL;
3015 }
3016 if (unshare_flags & CLONE_VM) {
3017 if (!current_is_single_threaded())
3018 return -EINVAL;
3019 }
3020
3021 return 0;
3022}
3023
3024/*
3025 * Unshare the filesystem structure if it is being shared
3026 */
3027static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3028{
3029 struct fs_struct *fs = current->fs;
3030
3031 if (!(unshare_flags & CLONE_FS) || !fs)
3032 return 0;
3033
3034 /* don't need lock here; in the worst case we'll do useless copy */
3035 if (fs->users == 1)
3036 return 0;
3037
3038 *new_fsp = copy_fs_struct(fs);
3039 if (!*new_fsp)
3040 return -ENOMEM;
3041
3042 return 0;
3043}
3044
3045/*
3046 * Unshare file descriptor table if it is being shared
3047 */
3048static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
3049{
3050 struct files_struct *fd = current->files;
3051
3052 if ((unshare_flags & CLONE_FILES) &&
3053 (fd && atomic_read(v: &fd->count) > 1)) {
3054 fd = dup_fd(fd, NULL);
3055 if (IS_ERR(ptr: fd))
3056 return PTR_ERR(ptr: fd);
3057 *new_fdp = fd;
3058 }
3059
3060 return 0;
3061}
3062
3063/*
3064 * unshare allows a process to 'unshare' part of the process
3065 * context which was originally shared using clone. copy_*
3066 * functions used by kernel_clone() cannot be used here directly
3067 * because they modify an inactive task_struct that is being
3068 * constructed. Here we are modifying the current, active,
3069 * task_struct.
3070 */
3071int ksys_unshare(unsigned long unshare_flags)
3072{
3073 struct fs_struct *fs, *new_fs = NULL;
3074 struct files_struct *new_fd = NULL;
3075 struct cred *new_cred = NULL;
3076 struct nsproxy *new_nsproxy = NULL;
3077 int do_sysvsem = 0;
3078 int err;
3079
3080 /*
3081 * If unsharing a user namespace must also unshare the thread group
3082 * and unshare the filesystem root and working directories.
3083 */
3084 if (unshare_flags & CLONE_NEWUSER)
3085 unshare_flags |= CLONE_THREAD | CLONE_FS;
3086 /*
3087 * If unsharing vm, must also unshare signal handlers.
3088 */
3089 if (unshare_flags & CLONE_VM)
3090 unshare_flags |= CLONE_SIGHAND;
3091 /*
3092 * If unsharing a signal handlers, must also unshare the signal queues.
3093 */
3094 if (unshare_flags & CLONE_SIGHAND)
3095 unshare_flags |= CLONE_THREAD;
3096 /*
3097 * If unsharing namespace, must also unshare filesystem information.
3098 */
3099 if (unshare_flags & CLONE_NEWNS)
3100 unshare_flags |= CLONE_FS;
3101
3102 err = check_unshare_flags(unshare_flags);
3103 if (err)
3104 goto bad_unshare_out;
3105 /*
3106 * CLONE_NEWIPC must also detach from the undolist: after switching
3107 * to a new ipc namespace, the semaphore arrays from the old
3108 * namespace are unreachable.
3109 */
3110 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3111 do_sysvsem = 1;
3112 err = unshare_fs(unshare_flags, new_fsp: &new_fs);
3113 if (err)
3114 goto bad_unshare_out;
3115 err = unshare_fd(unshare_flags, new_fdp: &new_fd);
3116 if (err)
3117 goto bad_unshare_cleanup_fs;
3118 err = unshare_userns(unshare_flags, new_cred: &new_cred);
3119 if (err)
3120 goto bad_unshare_cleanup_fd;
3121 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3122 new_cred, new_fs);
3123 if (err)
3124 goto bad_unshare_cleanup_cred;
3125
3126 if (new_cred) {
3127 err = set_cred_ucounts(new_cred);
3128 if (err)
3129 goto bad_unshare_cleanup_cred;
3130 }
3131
3132 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3133 if (do_sysvsem) {
3134 /*
3135 * CLONE_SYSVSEM is equivalent to sys_exit().
3136 */
3137 exit_sem(current);
3138 }
3139 if (unshare_flags & CLONE_NEWIPC) {
3140 /* Orphan segments in old ns (see sem above). */
3141 exit_shm(current);
3142 shm_init_task(current);
3143 }
3144
3145 if (new_nsproxy)
3146 switch_task_namespaces(current, new: new_nsproxy);
3147
3148 task_lock(current);
3149
3150 if (new_fs) {
3151 fs = current->fs;
3152 spin_lock(lock: &fs->lock);
3153 current->fs = new_fs;
3154 if (--fs->users)
3155 new_fs = NULL;
3156 else
3157 new_fs = fs;
3158 spin_unlock(lock: &fs->lock);
3159 }
3160
3161 if (new_fd)
3162 swap(current->files, new_fd);
3163
3164 task_unlock(current);
3165
3166 if (new_cred) {
3167 /* Install the new user namespace */
3168 commit_creds(new_cred);
3169 new_cred = NULL;
3170 }
3171 }
3172
3173 perf_event_namespaces(current);
3174
3175bad_unshare_cleanup_cred:
3176 if (new_cred)
3177 put_cred(cred: new_cred);
3178bad_unshare_cleanup_fd:
3179 if (new_fd)
3180 put_files_struct(fs: new_fd);
3181
3182bad_unshare_cleanup_fs:
3183 if (new_fs)
3184 free_fs_struct(new_fs);
3185
3186bad_unshare_out:
3187 return err;
3188}
3189
3190SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3191{
3192 return ksys_unshare(unshare_flags);
3193}
3194
3195/*
3196 * Helper to unshare the files of the current task.
3197 * We don't want to expose copy_files internals to
3198 * the exec layer of the kernel.
3199 */
3200
3201int unshare_files(void)
3202{
3203 struct task_struct *task = current;
3204 struct files_struct *old, *copy = NULL;
3205 int error;
3206
3207 error = unshare_fd(CLONE_FILES, new_fdp: &copy);
3208 if (error || !copy)
3209 return error;
3210
3211 old = task->files;
3212 task_lock(p: task);
3213 task->files = copy;
3214 task_unlock(p: task);
3215 put_files_struct(fs: old);
3216 return 0;
3217}
3218
3219int sysctl_max_threads(const struct ctl_table *table, int write,
3220 void *buffer, size_t *lenp, loff_t *ppos)
3221{
3222 struct ctl_table t;
3223 int ret;
3224 int threads = max_threads;
3225 int min = 1;
3226 int max = MAX_THREADS;
3227
3228 t = *table;
3229 t.data = &threads;
3230 t.extra1 = &min;
3231 t.extra2 = &max;
3232
3233 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3234 if (ret || !write)
3235 return ret;
3236
3237 max_threads = threads;
3238
3239 return 0;
3240}
3241

Provided by KDAB

Privacy Policy
Improve your Profiling and Debugging skills
Find out more

source code of linux/kernel/fork.c