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/kstack_erase.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#include <linux/unwind_deferred.h>
109
110#include <asm/pgalloc.h>
111#include <linux/uaccess.h>
112#include <asm/mmu_context.h>
113#include <asm/cacheflush.h>
114#include <asm/tlbflush.h>
115
116/* For dup_mmap(). */
117#include "../mm/internal.h"
118
119#include <trace/events/sched.h>
120
121#define CREATE_TRACE_POINTS
122#include <trace/events/task.h>
123
124#include <kunit/visibility.h>
125
126/*
127 * Minimum number of threads to boot the kernel
128 */
129#define MIN_THREADS 20
130
131/*
132 * Maximum number of threads
133 */
134#define MAX_THREADS FUTEX_TID_MASK
135
136/*
137 * Protected counters by write_lock_irq(&tasklist_lock)
138 */
139unsigned long total_forks; /* Handle normal Linux uptimes. */
140int nr_threads; /* The idle threads do not count.. */
141
142static int max_threads; /* tunable limit on nr_threads */
143
144#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
145
146static const char * const resident_page_types[] = {
147 NAMED_ARRAY_INDEX(MM_FILEPAGES),
148 NAMED_ARRAY_INDEX(MM_ANONPAGES),
149 NAMED_ARRAY_INDEX(MM_SWAPENTS),
150 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
151};
152
153DEFINE_PER_CPU(unsigned long, process_counts) = 0;
154
155__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
156
157#ifdef CONFIG_PROVE_RCU
158int lockdep_tasklist_lock_is_held(void)
159{
160 return lockdep_is_held(&tasklist_lock);
161}
162EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
163#endif /* #ifdef CONFIG_PROVE_RCU */
164
165int nr_processes(void)
166{
167 int cpu;
168 int total = 0;
169
170 for_each_possible_cpu(cpu)
171 total += per_cpu(process_counts, cpu);
172
173 return total;
174}
175
176void __weak arch_release_task_struct(struct task_struct *tsk)
177{
178}
179
180static struct kmem_cache *task_struct_cachep;
181
182static inline struct task_struct *alloc_task_struct_node(int node)
183{
184 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
185}
186
187static inline void free_task_struct(struct task_struct *tsk)
188{
189 kmem_cache_free(s: task_struct_cachep, objp: tsk);
190}
191
192#ifdef CONFIG_VMAP_STACK
193/*
194 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
195 * flush. Try to minimize the number of calls by caching stacks.
196 */
197#define NR_CACHED_STACKS 2
198static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
199/*
200 * Allocated stacks are cached and later reused by new threads, so memcg
201 * accounting is performed by the code assigning/releasing stacks to tasks.
202 * We need a zeroed memory without __GFP_ACCOUNT.
203 */
204#define GFP_VMAP_STACK (GFP_KERNEL | __GFP_ZERO)
205
206struct vm_stack {
207 struct rcu_head rcu;
208 struct vm_struct *stack_vm_area;
209};
210
211static bool try_release_thread_stack_to_cache(struct vm_struct *vm_area)
212{
213 unsigned int i;
214
215 for (i = 0; i < NR_CACHED_STACKS; i++) {
216 struct vm_struct *tmp = NULL;
217
218 if (this_cpu_try_cmpxchg(cached_stacks[i], &tmp, vm_area))
219 return true;
220 }
221 return false;
222}
223
224static void thread_stack_free_rcu(struct rcu_head *rh)
225{
226 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
227 struct vm_struct *vm_area = vm_stack->stack_vm_area;
228
229 if (try_release_thread_stack_to_cache(vm_area: vm_stack->stack_vm_area))
230 return;
231
232 vfree(addr: vm_area->addr);
233}
234
235static void thread_stack_delayed_free(struct task_struct *tsk)
236{
237 struct vm_stack *vm_stack = tsk->stack;
238
239 vm_stack->stack_vm_area = tsk->stack_vm_area;
240 call_rcu(head: &vm_stack->rcu, func: thread_stack_free_rcu);
241}
242
243static int free_vm_stack_cache(unsigned int cpu)
244{
245 struct vm_struct **cached_vm_stack_areas = per_cpu_ptr(cached_stacks, cpu);
246 int i;
247
248 for (i = 0; i < NR_CACHED_STACKS; i++) {
249 struct vm_struct *vm_area = cached_vm_stack_areas[i];
250
251 if (!vm_area)
252 continue;
253
254 vfree(addr: vm_area->addr);
255 cached_vm_stack_areas[i] = NULL;
256 }
257
258 return 0;
259}
260
261static int memcg_charge_kernel_stack(struct vm_struct *vm_area)
262{
263 int i;
264 int ret;
265 int nr_charged = 0;
266
267 BUG_ON(vm_area->nr_pages != THREAD_SIZE / PAGE_SIZE);
268
269 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
270 ret = memcg_kmem_charge_page(page: vm_area->pages[i], GFP_KERNEL, order: 0);
271 if (ret)
272 goto err;
273 nr_charged++;
274 }
275 return 0;
276err:
277 for (i = 0; i < nr_charged; i++)
278 memcg_kmem_uncharge_page(page: vm_area->pages[i], order: 0);
279 return ret;
280}
281
282static int alloc_thread_stack_node(struct task_struct *tsk, int node)
283{
284 struct vm_struct *vm_area;
285 void *stack;
286 int i;
287
288 for (i = 0; i < NR_CACHED_STACKS; i++) {
289 vm_area = this_cpu_xchg(cached_stacks[i], NULL);
290 if (!vm_area)
291 continue;
292
293 if (memcg_charge_kernel_stack(vm_area)) {
294 vfree(addr: vm_area->addr);
295 return -ENOMEM;
296 }
297
298 /* Reset stack metadata. */
299 kasan_unpoison_range(address: vm_area->addr, THREAD_SIZE);
300
301 stack = kasan_reset_tag(addr: vm_area->addr);
302
303 /* Clear stale pointers from reused stack. */
304 memset(s: stack, c: 0, THREAD_SIZE);
305
306 tsk->stack_vm_area = vm_area;
307 tsk->stack = stack;
308 return 0;
309 }
310
311 stack = __vmalloc_node(THREAD_SIZE, THREAD_ALIGN,
312 GFP_VMAP_STACK,
313 node, __builtin_return_address(0));
314 if (!stack)
315 return -ENOMEM;
316
317 vm_area = find_vm_area(addr: stack);
318 if (memcg_charge_kernel_stack(vm_area)) {
319 vfree(addr: stack);
320 return -ENOMEM;
321 }
322 /*
323 * We can't call find_vm_area() in interrupt context, and
324 * free_thread_stack() can be called in interrupt context,
325 * so cache the vm_struct.
326 */
327 tsk->stack_vm_area = vm_area;
328 stack = kasan_reset_tag(addr: stack);
329 tsk->stack = stack;
330 return 0;
331}
332
333static void free_thread_stack(struct task_struct *tsk)
334{
335 if (!try_release_thread_stack_to_cache(vm_area: tsk->stack_vm_area))
336 thread_stack_delayed_free(tsk);
337
338 tsk->stack = NULL;
339 tsk->stack_vm_area = NULL;
340}
341
342#else /* !CONFIG_VMAP_STACK */
343
344/*
345 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
346 * kmemcache based allocator.
347 */
348#if THREAD_SIZE >= PAGE_SIZE
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#else /* !(THREAD_SIZE >= PAGE_SIZE) */
381
382static struct kmem_cache *thread_stack_cache;
383
384static void thread_stack_free_rcu(struct rcu_head *rh)
385{
386 kmem_cache_free(thread_stack_cache, rh);
387}
388
389static void thread_stack_delayed_free(struct task_struct *tsk)
390{
391 struct rcu_head *rh = tsk->stack;
392
393 call_rcu(rh, thread_stack_free_rcu);
394}
395
396static int alloc_thread_stack_node(struct task_struct *tsk, int node)
397{
398 unsigned long *stack;
399 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
400 stack = kasan_reset_tag(stack);
401 tsk->stack = stack;
402 return stack ? 0 : -ENOMEM;
403}
404
405static void free_thread_stack(struct task_struct *tsk)
406{
407 thread_stack_delayed_free(tsk);
408 tsk->stack = NULL;
409}
410
411void thread_stack_cache_init(void)
412{
413 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
414 THREAD_SIZE, THREAD_SIZE, 0, 0,
415 THREAD_SIZE, NULL);
416 BUG_ON(thread_stack_cache == NULL);
417}
418
419#endif /* THREAD_SIZE >= PAGE_SIZE */
420#endif /* 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_area = 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_area->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_area;
461 int i;
462
463 vm_area = task_stack_vm_area(t: tsk);
464 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
465 memcg_kmem_uncharge_page(page: vm_area->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 Comm:%s Pid:%d\n",
590 mm, resident_page_types[i], x,
591 current->comm,
592 task_pid_nr(current));
593 }
594 }
595
596 if (mm_pgtables_bytes(mm))
597 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
598 mm_pgtables_bytes(mm));
599
600#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
601 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
602#endif
603}
604
605#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
606#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
607
608static void do_check_lazy_tlb(void *arg)
609{
610 struct mm_struct *mm = arg;
611
612 WARN_ON_ONCE(current->active_mm == mm);
613}
614
615static void do_shoot_lazy_tlb(void *arg)
616{
617 struct mm_struct *mm = arg;
618
619 if (current->active_mm == mm) {
620 WARN_ON_ONCE(current->mm);
621 current->active_mm = &init_mm;
622 switch_mm(prev: mm, next: &init_mm, current);
623 }
624}
625
626static void cleanup_lazy_tlbs(struct mm_struct *mm)
627{
628 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
629 /*
630 * In this case, lazy tlb mms are refounted and would not reach
631 * __mmdrop until all CPUs have switched away and mmdrop()ed.
632 */
633 return;
634 }
635
636 /*
637 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
638 * requires lazy mm users to switch to another mm when the refcount
639 * drops to zero, before the mm is freed. This requires IPIs here to
640 * switch kernel threads to init_mm.
641 *
642 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
643 * switch with the final userspace teardown TLB flush which leaves the
644 * mm lazy on this CPU but no others, reducing the need for additional
645 * IPIs here. There are cases where a final IPI is still required here,
646 * such as the final mmdrop being performed on a different CPU than the
647 * one exiting, or kernel threads using the mm when userspace exits.
648 *
649 * IPI overheads have not found to be expensive, but they could be
650 * reduced in a number of possible ways, for example (roughly
651 * increasing order of complexity):
652 * - The last lazy reference created by exit_mm() could instead switch
653 * to init_mm, however it's probable this will run on the same CPU
654 * immediately afterwards, so this may not reduce IPIs much.
655 * - A batch of mms requiring IPIs could be gathered and freed at once.
656 * - CPUs store active_mm where it can be remotely checked without a
657 * lock, to filter out false-positives in the cpumask.
658 * - After mm_users or mm_count reaches zero, switching away from the
659 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
660 * with some batching or delaying of the final IPIs.
661 * - A delayed freeing and RCU-like quiescing sequence based on mm
662 * switching to avoid IPIs completely.
663 */
664 on_each_cpu_mask(mask: mm_cpumask(mm), func: do_shoot_lazy_tlb, info: (void *)mm, wait: 1);
665 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
666 on_each_cpu(func: do_check_lazy_tlb, info: (void *)mm, wait: 1);
667}
668
669/*
670 * Called when the last reference to the mm
671 * is dropped: either by a lazy thread or by
672 * mmput. Free the page directory and the mm.
673 */
674void __mmdrop(struct mm_struct *mm)
675{
676 BUG_ON(mm == &init_mm);
677 WARN_ON_ONCE(mm == current->mm);
678
679 /* Ensure no CPUs are using this as their lazy tlb mm */
680 cleanup_lazy_tlbs(mm);
681
682 WARN_ON_ONCE(mm == current->active_mm);
683 mm_free_pgd(mm);
684 mm_free_id(mm);
685 destroy_context(mm);
686 mmu_notifier_subscriptions_destroy(mm);
687 check_mm(mm);
688 put_user_ns(ns: mm->user_ns);
689 mm_pasid_drop(mm);
690 mm_destroy_cid(mm);
691 percpu_counter_destroy_many(fbc: mm->rss_stat, nr_counters: NR_MM_COUNTERS);
692
693 free_mm(mm);
694}
695EXPORT_SYMBOL_GPL(__mmdrop);
696
697static void mmdrop_async_fn(struct work_struct *work)
698{
699 struct mm_struct *mm;
700
701 mm = container_of(work, struct mm_struct, async_put_work);
702 __mmdrop(mm);
703}
704
705static void mmdrop_async(struct mm_struct *mm)
706{
707 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
708 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
709 schedule_work(work: &mm->async_put_work);
710 }
711}
712
713static inline void free_signal_struct(struct signal_struct *sig)
714{
715 taskstats_tgid_free(sig);
716 sched_autogroup_exit(sig);
717 /*
718 * __mmdrop is not safe to call from softirq context on x86 due to
719 * pgd_dtor so postpone it to the async context
720 */
721 if (sig->oom_mm)
722 mmdrop_async(mm: sig->oom_mm);
723 kmem_cache_free(s: signal_cachep, objp: sig);
724}
725
726static inline void put_signal_struct(struct signal_struct *sig)
727{
728 if (refcount_dec_and_test(r: &sig->sigcnt))
729 free_signal_struct(sig);
730}
731
732void __put_task_struct(struct task_struct *tsk)
733{
734 WARN_ON(!tsk->exit_state);
735 WARN_ON(refcount_read(&tsk->usage));
736 WARN_ON(tsk == current);
737
738 unwind_task_free(task: tsk);
739 sched_ext_free(p: tsk);
740 io_uring_free(tsk);
741 cgroup_free(p: tsk);
742 task_numa_free(p: tsk, final: true);
743 security_task_free(task: tsk);
744 exit_creds(tsk);
745 delayacct_tsk_free(tsk);
746 put_signal_struct(sig: tsk->signal);
747 sched_core_free(tsk);
748 free_task(tsk);
749}
750EXPORT_SYMBOL_GPL(__put_task_struct);
751
752void __put_task_struct_rcu_cb(struct rcu_head *rhp)
753{
754 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
755
756 __put_task_struct(task);
757}
758EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
759
760void __init __weak arch_task_cache_init(void) { }
761
762/*
763 * set_max_threads
764 */
765static void __init set_max_threads(unsigned int max_threads_suggested)
766{
767 u64 threads;
768 unsigned long nr_pages = memblock_estimated_nr_free_pages();
769
770 /*
771 * The number of threads shall be limited such that the thread
772 * structures may only consume a small part of the available memory.
773 */
774 if (fls64(x: nr_pages) + fls64(PAGE_SIZE) > 64)
775 threads = MAX_THREADS;
776 else
777 threads = div64_u64(dividend: (u64) nr_pages * (u64) PAGE_SIZE,
778 divisor: (u64) THREAD_SIZE * 8UL);
779
780 if (threads > max_threads_suggested)
781 threads = max_threads_suggested;
782
783 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
784}
785
786#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
787/* Initialized by the architecture: */
788int arch_task_struct_size __read_mostly;
789#endif
790
791static void __init task_struct_whitelist(unsigned long *offset, unsigned long *size)
792{
793 /* Fetch thread_struct whitelist for the architecture. */
794 arch_thread_struct_whitelist(offset, size);
795
796 /*
797 * Handle zero-sized whitelist or empty thread_struct, otherwise
798 * adjust offset to position of thread_struct in task_struct.
799 */
800 if (unlikely(*size == 0))
801 *offset = 0;
802 else
803 *offset += offsetof(struct task_struct, thread);
804}
805
806void __init fork_init(void)
807{
808 int i;
809#ifndef ARCH_MIN_TASKALIGN
810#define ARCH_MIN_TASKALIGN 0
811#endif
812 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
813 unsigned long useroffset, usersize;
814
815 /* create a slab on which task_structs can be allocated */
816 task_struct_whitelist(offset: &useroffset, size: &usersize);
817 task_struct_cachep = kmem_cache_create_usercopy(name: "task_struct",
818 size: arch_task_struct_size, align,
819 SLAB_PANIC|SLAB_ACCOUNT,
820 useroffset, usersize, NULL);
821
822 /* do the arch specific task caches init */
823 arch_task_cache_init();
824
825 set_max_threads(MAX_THREADS);
826
827 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
828 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
829 init_task.signal->rlim[RLIMIT_SIGPENDING] =
830 init_task.signal->rlim[RLIMIT_NPROC];
831
832 for (i = 0; i < UCOUNT_COUNTS; i++)
833 init_user_ns.ucount_max[i] = max_threads/2;
834
835 set_userns_rlimit_max(ns: &init_user_ns, type: UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
836 set_userns_rlimit_max(ns: &init_user_ns, type: UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
837 set_userns_rlimit_max(ns: &init_user_ns, type: UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
838 set_userns_rlimit_max(ns: &init_user_ns, type: UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
839
840#ifdef CONFIG_VMAP_STACK
841 cpuhp_setup_state(state: CPUHP_BP_PREPARE_DYN, name: "fork:vm_stack_cache",
842 NULL, teardown: free_vm_stack_cache);
843#endif
844
845 scs_init();
846
847 lockdep_init_task(task: &init_task);
848 uprobes_init();
849}
850
851int __weak arch_dup_task_struct(struct task_struct *dst,
852 struct task_struct *src)
853{
854 *dst = *src;
855 return 0;
856}
857
858void set_task_stack_end_magic(struct task_struct *tsk)
859{
860 unsigned long *stackend;
861
862 stackend = end_of_stack(task: tsk);
863 *stackend = STACK_END_MAGIC; /* for overflow detection */
864}
865
866static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
867{
868 struct task_struct *tsk;
869 int err;
870
871 if (node == NUMA_NO_NODE)
872 node = tsk_fork_get_node(tsk: orig);
873 tsk = alloc_task_struct_node(node);
874 if (!tsk)
875 return NULL;
876
877 err = arch_dup_task_struct(dst: tsk, src: orig);
878 if (err)
879 goto free_tsk;
880
881 err = alloc_thread_stack_node(tsk, node);
882 if (err)
883 goto free_tsk;
884
885#ifdef CONFIG_THREAD_INFO_IN_TASK
886 refcount_set(r: &tsk->stack_refcount, n: 1);
887#endif
888 account_kernel_stack(tsk, account: 1);
889
890 err = scs_prepare(tsk, node);
891 if (err)
892 goto free_stack;
893
894#ifdef CONFIG_SECCOMP
895 /*
896 * We must handle setting up seccomp filters once we're under
897 * the sighand lock in case orig has changed between now and
898 * then. Until then, filter must be NULL to avoid messing up
899 * the usage counts on the error path calling free_task.
900 */
901 tsk->seccomp.filter = NULL;
902#endif
903
904 setup_thread_stack(tsk, orig);
905 clear_user_return_notifier(p: tsk);
906 clear_tsk_need_resched(tsk);
907 set_task_stack_end_magic(tsk);
908 clear_syscall_work_syscall_user_dispatch(tsk);
909
910#ifdef CONFIG_STACKPROTECTOR
911 tsk->stack_canary = get_random_canary();
912#endif
913 if (orig->cpus_ptr == &orig->cpus_mask)
914 tsk->cpus_ptr = &tsk->cpus_mask;
915 dup_user_cpus_ptr(dst: tsk, src: orig, node);
916
917 /*
918 * One for the user space visible state that goes away when reaped.
919 * One for the scheduler.
920 */
921 refcount_set(r: &tsk->rcu_users, n: 2);
922 /* One for the rcu users */
923 refcount_set(r: &tsk->usage, n: 1);
924#ifdef CONFIG_BLK_DEV_IO_TRACE
925 tsk->btrace_seq = 0;
926#endif
927 tsk->splice_pipe = NULL;
928 tsk->task_frag.page = NULL;
929 tsk->wake_q.next = NULL;
930 tsk->worker_private = NULL;
931
932 kcov_task_init(t: tsk);
933 kmsan_task_create(task: tsk);
934 kmap_local_fork(tsk);
935
936#ifdef CONFIG_FAULT_INJECTION
937 tsk->fail_nth = 0;
938#endif
939
940#ifdef CONFIG_BLK_CGROUP
941 tsk->throttle_disk = NULL;
942 tsk->use_memdelay = 0;
943#endif
944
945#ifdef CONFIG_ARCH_HAS_CPU_PASID
946 tsk->pasid_activated = 0;
947#endif
948
949#ifdef CONFIG_MEMCG
950 tsk->active_memcg = NULL;
951#endif
952
953#ifdef CONFIG_X86_BUS_LOCK_DETECT
954 tsk->reported_split_lock = 0;
955#endif
956
957#ifdef CONFIG_SCHED_MM_CID
958 tsk->mm_cid = -1;
959 tsk->last_mm_cid = -1;
960 tsk->mm_cid_active = 0;
961 tsk->migrate_from_cpu = -1;
962#endif
963 return tsk;
964
965free_stack:
966 exit_task_stack_account(tsk);
967 free_thread_stack(tsk);
968free_tsk:
969 free_task_struct(tsk);
970 return NULL;
971}
972
973__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
974
975static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
976
977static int __init coredump_filter_setup(char *s)
978{
979 default_dump_filter =
980 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
981 MMF_DUMP_FILTER_MASK;
982 return 1;
983}
984
985__setup("coredump_filter=", coredump_filter_setup);
986
987#include <linux/init_task.h>
988
989static void mm_init_aio(struct mm_struct *mm)
990{
991#ifdef CONFIG_AIO
992 spin_lock_init(&mm->ioctx_lock);
993 mm->ioctx_table = NULL;
994#endif
995}
996
997static __always_inline void mm_clear_owner(struct mm_struct *mm,
998 struct task_struct *p)
999{
1000#ifdef CONFIG_MEMCG
1001 if (mm->owner == p)
1002 WRITE_ONCE(mm->owner, NULL);
1003#endif
1004}
1005
1006static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1007{
1008#ifdef CONFIG_MEMCG
1009 mm->owner = p;
1010#endif
1011}
1012
1013static void mm_init_uprobes_state(struct mm_struct *mm)
1014{
1015#ifdef CONFIG_UPROBES
1016 mm->uprobes_state.xol_area = NULL;
1017 arch_uprobe_init_state(mm);
1018#endif
1019}
1020
1021static void mmap_init_lock(struct mm_struct *mm)
1022{
1023 init_rwsem(&mm->mmap_lock);
1024 mm_lock_seqcount_init(mm);
1025#ifdef CONFIG_PER_VMA_LOCK
1026 rcuwait_init(w: &mm->vma_writer_wait);
1027#endif
1028}
1029
1030static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1031 struct user_namespace *user_ns)
1032{
1033 mt_init_flags(mt: &mm->mm_mt, MM_MT_FLAGS);
1034 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1035 atomic_set(v: &mm->mm_users, i: 1);
1036 atomic_set(v: &mm->mm_count, i: 1);
1037 seqcount_init(&mm->write_protect_seq);
1038 mmap_init_lock(mm);
1039 INIT_LIST_HEAD(list: &mm->mmlist);
1040 mm_pgtables_bytes_init(mm);
1041 mm->map_count = 0;
1042 mm->locked_vm = 0;
1043 atomic64_set(v: &mm->pinned_vm, i: 0);
1044 memset(s: &mm->rss_stat, c: 0, n: sizeof(mm->rss_stat));
1045 spin_lock_init(&mm->page_table_lock);
1046 spin_lock_init(&mm->arg_lock);
1047 mm_init_cpumask(mm);
1048 mm_init_aio(mm);
1049 mm_init_owner(mm, p);
1050 mm_pasid_init(mm);
1051 RCU_INIT_POINTER(mm->exe_file, NULL);
1052 mmu_notifier_subscriptions_init(mm);
1053 init_tlb_flush_pending(mm);
1054#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
1055 mm->pmd_huge_pte = NULL;
1056#endif
1057 mm_init_uprobes_state(mm);
1058 hugetlb_count_init(mm);
1059
1060 mm_flags_clear_all(mm);
1061 if (current->mm) {
1062 unsigned long flags = __mm_flags_get_word(current->mm);
1063
1064 __mm_flags_set_word(mm, value: mmf_init_legacy_flags(flags));
1065 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1066 } else {
1067 __mm_flags_set_word(mm, value: default_dump_filter);
1068 mm->def_flags = 0;
1069 }
1070
1071 if (futex_mm_init(mm))
1072 goto fail_mm_init;
1073
1074 if (mm_alloc_pgd(mm))
1075 goto fail_nopgd;
1076
1077 if (mm_alloc_id(mm))
1078 goto fail_noid;
1079
1080 if (init_new_context(tsk: p, mm))
1081 goto fail_nocontext;
1082
1083 if (mm_alloc_cid(mm, p))
1084 goto fail_cid;
1085
1086 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1087 NR_MM_COUNTERS))
1088 goto fail_pcpu;
1089
1090 mm->user_ns = get_user_ns(ns: user_ns);
1091 lru_gen_init_mm(mm);
1092 return mm;
1093
1094fail_pcpu:
1095 mm_destroy_cid(mm);
1096fail_cid:
1097 destroy_context(mm);
1098fail_nocontext:
1099 mm_free_id(mm);
1100fail_noid:
1101 mm_free_pgd(mm);
1102fail_nopgd:
1103 futex_hash_free(mm);
1104fail_mm_init:
1105 free_mm(mm);
1106 return NULL;
1107}
1108
1109/*
1110 * Allocate and initialize an mm_struct.
1111 */
1112struct mm_struct *mm_alloc(void)
1113{
1114 struct mm_struct *mm;
1115
1116 mm = allocate_mm();
1117 if (!mm)
1118 return NULL;
1119
1120 memset(s: mm, c: 0, n: sizeof(*mm));
1121 return mm_init(mm, current, user_ns: current_user_ns());
1122}
1123EXPORT_SYMBOL_IF_KUNIT(mm_alloc);
1124
1125static inline void __mmput(struct mm_struct *mm)
1126{
1127 VM_BUG_ON(atomic_read(&mm->mm_users));
1128
1129 uprobe_clear_state(mm);
1130 exit_aio(mm);
1131 ksm_exit(mm);
1132 khugepaged_exit(mm); /* must run before exit_mmap */
1133 exit_mmap(mm);
1134 mm_put_huge_zero_folio(mm);
1135 set_mm_exe_file(mm, NULL);
1136 if (!list_empty(head: &mm->mmlist)) {
1137 spin_lock(lock: &mmlist_lock);
1138 list_del(entry: &mm->mmlist);
1139 spin_unlock(lock: &mmlist_lock);
1140 }
1141 if (mm->binfmt)
1142 module_put(module: mm->binfmt->module);
1143 lru_gen_del_mm(mm);
1144 futex_hash_free(mm);
1145 mmdrop(mm);
1146}
1147
1148/*
1149 * Decrement the use count and release all resources for an mm.
1150 */
1151void mmput(struct mm_struct *mm)
1152{
1153 might_sleep();
1154
1155 if (atomic_dec_and_test(v: &mm->mm_users))
1156 __mmput(mm);
1157}
1158EXPORT_SYMBOL_GPL(mmput);
1159
1160#if defined(CONFIG_MMU) || defined(CONFIG_FUTEX_PRIVATE_HASH)
1161static void mmput_async_fn(struct work_struct *work)
1162{
1163 struct mm_struct *mm = container_of(work, struct mm_struct,
1164 async_put_work);
1165
1166 __mmput(mm);
1167}
1168
1169void mmput_async(struct mm_struct *mm)
1170{
1171 if (atomic_dec_and_test(v: &mm->mm_users)) {
1172 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1173 schedule_work(work: &mm->async_put_work);
1174 }
1175}
1176EXPORT_SYMBOL_GPL(mmput_async);
1177#endif
1178
1179/**
1180 * set_mm_exe_file - change a reference to the mm's executable file
1181 * @mm: The mm to change.
1182 * @new_exe_file: The new file to use.
1183 *
1184 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1185 *
1186 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1187 * invocations: in mmput() nobody alive left, in execve it happens before
1188 * the new mm is made visible to anyone.
1189 *
1190 * Can only fail if new_exe_file != NULL.
1191 */
1192int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1193{
1194 struct file *old_exe_file;
1195
1196 /*
1197 * It is safe to dereference the exe_file without RCU as
1198 * this function is only called if nobody else can access
1199 * this mm -- see comment above for justification.
1200 */
1201 old_exe_file = rcu_dereference_raw(mm->exe_file);
1202
1203 if (new_exe_file) {
1204 /*
1205 * We expect the caller (i.e., sys_execve) to already denied
1206 * write access, so this is unlikely to fail.
1207 */
1208 if (unlikely(exe_file_deny_write_access(new_exe_file)))
1209 return -EACCES;
1210 get_file(f: new_exe_file);
1211 }
1212 rcu_assign_pointer(mm->exe_file, new_exe_file);
1213 if (old_exe_file) {
1214 exe_file_allow_write_access(exe_file: old_exe_file);
1215 fput(old_exe_file);
1216 }
1217 return 0;
1218}
1219
1220/**
1221 * replace_mm_exe_file - replace a reference to the mm's executable file
1222 * @mm: The mm to change.
1223 * @new_exe_file: The new file to use.
1224 *
1225 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1226 *
1227 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1228 */
1229int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1230{
1231 struct vm_area_struct *vma;
1232 struct file *old_exe_file;
1233 int ret = 0;
1234
1235 /* Forbid mm->exe_file change if old file still mapped. */
1236 old_exe_file = get_mm_exe_file(mm);
1237 if (old_exe_file) {
1238 VMA_ITERATOR(vmi, mm, 0);
1239 mmap_read_lock(mm);
1240 for_each_vma(vmi, vma) {
1241 if (!vma->vm_file)
1242 continue;
1243 if (path_equal(path1: &vma->vm_file->f_path,
1244 path2: &old_exe_file->f_path)) {
1245 ret = -EBUSY;
1246 break;
1247 }
1248 }
1249 mmap_read_unlock(mm);
1250 fput(old_exe_file);
1251 if (ret)
1252 return ret;
1253 }
1254
1255 ret = exe_file_deny_write_access(exe_file: new_exe_file);
1256 if (ret)
1257 return -EACCES;
1258 get_file(f: new_exe_file);
1259
1260 /* set the new file */
1261 mmap_write_lock(mm);
1262 old_exe_file = rcu_dereference_raw(mm->exe_file);
1263 rcu_assign_pointer(mm->exe_file, new_exe_file);
1264 mmap_write_unlock(mm);
1265
1266 if (old_exe_file) {
1267 exe_file_allow_write_access(exe_file: old_exe_file);
1268 fput(old_exe_file);
1269 }
1270 return 0;
1271}
1272
1273/**
1274 * get_mm_exe_file - acquire a reference to the mm's executable file
1275 * @mm: The mm of interest.
1276 *
1277 * Returns %NULL if mm has no associated executable file.
1278 * User must release file via fput().
1279 */
1280struct file *get_mm_exe_file(struct mm_struct *mm)
1281{
1282 struct file *exe_file;
1283
1284 rcu_read_lock();
1285 exe_file = get_file_rcu(f: &mm->exe_file);
1286 rcu_read_unlock();
1287 return exe_file;
1288}
1289
1290/**
1291 * get_task_exe_file - acquire a reference to the task's executable file
1292 * @task: The task.
1293 *
1294 * Returns %NULL if task's mm (if any) has no associated executable file or
1295 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1296 * User must release file via fput().
1297 */
1298struct file *get_task_exe_file(struct task_struct *task)
1299{
1300 struct file *exe_file = NULL;
1301 struct mm_struct *mm;
1302
1303 if (task->flags & PF_KTHREAD)
1304 return NULL;
1305
1306 task_lock(p: task);
1307 mm = task->mm;
1308 if (mm)
1309 exe_file = get_mm_exe_file(mm);
1310 task_unlock(p: task);
1311 return exe_file;
1312}
1313
1314/**
1315 * get_task_mm - acquire a reference to the task's mm
1316 * @task: The task.
1317 *
1318 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1319 * this kernel workthread has transiently adopted a user mm with use_mm,
1320 * to do its AIO) is not set and if so returns a reference to it, after
1321 * bumping up the use count. User must release the mm via mmput()
1322 * after use. Typically used by /proc and ptrace.
1323 */
1324struct mm_struct *get_task_mm(struct task_struct *task)
1325{
1326 struct mm_struct *mm;
1327
1328 if (task->flags & PF_KTHREAD)
1329 return NULL;
1330
1331 task_lock(p: task);
1332 mm = task->mm;
1333 if (mm)
1334 mmget(mm);
1335 task_unlock(p: task);
1336 return mm;
1337}
1338EXPORT_SYMBOL_GPL(get_task_mm);
1339
1340static bool may_access_mm(struct mm_struct *mm, struct task_struct *task, unsigned int mode)
1341{
1342 if (mm == current->mm)
1343 return true;
1344 if (ptrace_may_access(task, mode))
1345 return true;
1346 if ((mode & PTRACE_MODE_READ) && perfmon_capable())
1347 return true;
1348 return false;
1349}
1350
1351struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1352{
1353 struct mm_struct *mm;
1354 int err;
1355
1356 err = down_read_killable(sem: &task->signal->exec_update_lock);
1357 if (err)
1358 return ERR_PTR(error: err);
1359
1360 mm = get_task_mm(task);
1361 if (!mm) {
1362 mm = ERR_PTR(error: -ESRCH);
1363 } else if (!may_access_mm(mm, task, mode)) {
1364 mmput(mm);
1365 mm = ERR_PTR(error: -EACCES);
1366 }
1367 up_read(sem: &task->signal->exec_update_lock);
1368
1369 return mm;
1370}
1371
1372static void complete_vfork_done(struct task_struct *tsk)
1373{
1374 struct completion *vfork;
1375
1376 task_lock(p: tsk);
1377 vfork = tsk->vfork_done;
1378 if (likely(vfork)) {
1379 tsk->vfork_done = NULL;
1380 complete(vfork);
1381 }
1382 task_unlock(p: tsk);
1383}
1384
1385static int wait_for_vfork_done(struct task_struct *child,
1386 struct completion *vfork)
1387{
1388 unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1389 int killed;
1390
1391 cgroup_enter_frozen();
1392 killed = wait_for_completion_state(x: vfork, state);
1393 cgroup_leave_frozen(always_leave: false);
1394
1395 if (killed) {
1396 task_lock(p: child);
1397 child->vfork_done = NULL;
1398 task_unlock(p: child);
1399 }
1400
1401 put_task_struct(t: child);
1402 return killed;
1403}
1404
1405/* Please note the differences between mmput and mm_release.
1406 * mmput is called whenever we stop holding onto a mm_struct,
1407 * error success whatever.
1408 *
1409 * mm_release is called after a mm_struct has been removed
1410 * from the current process.
1411 *
1412 * This difference is important for error handling, when we
1413 * only half set up a mm_struct for a new process and need to restore
1414 * the old one. Because we mmput the new mm_struct before
1415 * restoring the old one. . .
1416 * Eric Biederman 10 January 1998
1417 */
1418static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1419{
1420 uprobe_free_utask(t: tsk);
1421
1422 /* Get rid of any cached register state */
1423 deactivate_mm(tsk, mm);
1424
1425 /*
1426 * Signal userspace if we're not exiting with a core dump
1427 * because we want to leave the value intact for debugging
1428 * purposes.
1429 */
1430 if (tsk->clear_child_tid) {
1431 if (atomic_read(v: &mm->mm_users) > 1) {
1432 /*
1433 * We don't check the error code - if userspace has
1434 * not set up a proper pointer then tough luck.
1435 */
1436 put_user(0, tsk->clear_child_tid);
1437 do_futex(uaddr: tsk->clear_child_tid, FUTEX_WAKE,
1438 val: 1, NULL, NULL, val2: 0, val3: 0);
1439 }
1440 tsk->clear_child_tid = NULL;
1441 }
1442
1443 /*
1444 * All done, finally we can wake up parent and return this mm to him.
1445 * Also kthread_stop() uses this completion for synchronization.
1446 */
1447 if (tsk->vfork_done)
1448 complete_vfork_done(tsk);
1449}
1450
1451void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1452{
1453 futex_exit_release(tsk);
1454 mm_release(tsk, mm);
1455}
1456
1457void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1458{
1459 futex_exec_release(tsk);
1460 mm_release(tsk, mm);
1461}
1462
1463/**
1464 * dup_mm() - duplicates an existing mm structure
1465 * @tsk: the task_struct with which the new mm will be associated.
1466 * @oldmm: the mm to duplicate.
1467 *
1468 * Allocates a new mm structure and duplicates the provided @oldmm structure
1469 * content into it.
1470 *
1471 * Return: the duplicated mm or NULL on failure.
1472 */
1473static struct mm_struct *dup_mm(struct task_struct *tsk,
1474 struct mm_struct *oldmm)
1475{
1476 struct mm_struct *mm;
1477 int err;
1478
1479 mm = allocate_mm();
1480 if (!mm)
1481 goto fail_nomem;
1482
1483 memcpy(to: mm, from: oldmm, len: sizeof(*mm));
1484
1485 if (!mm_init(mm, p: tsk, user_ns: mm->user_ns))
1486 goto fail_nomem;
1487
1488 uprobe_start_dup_mmap();
1489 err = dup_mmap(mm, oldmm);
1490 if (err)
1491 goto free_pt;
1492 uprobe_end_dup_mmap();
1493
1494 mm->hiwater_rss = get_mm_rss(mm);
1495 mm->hiwater_vm = mm->total_vm;
1496
1497 if (mm->binfmt && !try_module_get(module: mm->binfmt->module))
1498 goto free_pt;
1499
1500 return mm;
1501
1502free_pt:
1503 /* don't put binfmt in mmput, we haven't got module yet */
1504 mm->binfmt = NULL;
1505 mm_init_owner(mm, NULL);
1506 mmput(mm);
1507 if (err)
1508 uprobe_end_dup_mmap();
1509
1510fail_nomem:
1511 return NULL;
1512}
1513
1514static int copy_mm(u64 clone_flags, struct task_struct *tsk)
1515{
1516 struct mm_struct *mm, *oldmm;
1517
1518 tsk->min_flt = tsk->maj_flt = 0;
1519 tsk->nvcsw = tsk->nivcsw = 0;
1520#ifdef CONFIG_DETECT_HUNG_TASK
1521 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1522 tsk->last_switch_time = 0;
1523#endif
1524
1525 tsk->mm = NULL;
1526 tsk->active_mm = NULL;
1527
1528 /*
1529 * Are we cloning a kernel thread?
1530 *
1531 * We need to steal a active VM for that..
1532 */
1533 oldmm = current->mm;
1534 if (!oldmm)
1535 return 0;
1536
1537 if (clone_flags & CLONE_VM) {
1538 mmget(mm: oldmm);
1539 mm = oldmm;
1540 } else {
1541 mm = dup_mm(tsk, current->mm);
1542 if (!mm)
1543 return -ENOMEM;
1544 }
1545
1546 tsk->mm = mm;
1547 tsk->active_mm = mm;
1548 sched_mm_cid_fork(t: tsk);
1549 return 0;
1550}
1551
1552static int copy_fs(u64 clone_flags, struct task_struct *tsk)
1553{
1554 struct fs_struct *fs = current->fs;
1555 if (clone_flags & CLONE_FS) {
1556 /* tsk->fs is already what we want */
1557 read_seqlock_excl(sl: &fs->seq);
1558 /* "users" and "in_exec" locked for check_unsafe_exec() */
1559 if (fs->in_exec) {
1560 read_sequnlock_excl(sl: &fs->seq);
1561 return -EAGAIN;
1562 }
1563 fs->users++;
1564 read_sequnlock_excl(sl: &fs->seq);
1565 return 0;
1566 }
1567 tsk->fs = copy_fs_struct(fs);
1568 if (!tsk->fs)
1569 return -ENOMEM;
1570 return 0;
1571}
1572
1573static int copy_files(u64 clone_flags, struct task_struct *tsk,
1574 int no_files)
1575{
1576 struct files_struct *oldf, *newf;
1577
1578 /*
1579 * A background process may not have any files ...
1580 */
1581 oldf = current->files;
1582 if (!oldf)
1583 return 0;
1584
1585 if (no_files) {
1586 tsk->files = NULL;
1587 return 0;
1588 }
1589
1590 if (clone_flags & CLONE_FILES) {
1591 atomic_inc(v: &oldf->count);
1592 return 0;
1593 }
1594
1595 newf = dup_fd(oldf, NULL);
1596 if (IS_ERR(ptr: newf))
1597 return PTR_ERR(ptr: newf);
1598
1599 tsk->files = newf;
1600 return 0;
1601}
1602
1603static int copy_sighand(u64 clone_flags, struct task_struct *tsk)
1604{
1605 struct sighand_struct *sig;
1606
1607 if (clone_flags & CLONE_SIGHAND) {
1608 refcount_inc(r: &current->sighand->count);
1609 return 0;
1610 }
1611 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1612 RCU_INIT_POINTER(tsk->sighand, sig);
1613 if (!sig)
1614 return -ENOMEM;
1615
1616 refcount_set(r: &sig->count, n: 1);
1617 spin_lock_irq(lock: &current->sighand->siglock);
1618 memcpy(to: sig->action, current->sighand->action, len: sizeof(sig->action));
1619 spin_unlock_irq(lock: &current->sighand->siglock);
1620
1621 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1622 if (clone_flags & CLONE_CLEAR_SIGHAND)
1623 flush_signal_handlers(tsk, force_default: 0);
1624
1625 return 0;
1626}
1627
1628void __cleanup_sighand(struct sighand_struct *sighand)
1629{
1630 if (refcount_dec_and_test(r: &sighand->count)) {
1631 signalfd_cleanup(sighand);
1632 /*
1633 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1634 * without an RCU grace period, see __lock_task_sighand().
1635 */
1636 kmem_cache_free(s: sighand_cachep, objp: sighand);
1637 }
1638}
1639
1640/*
1641 * Initialize POSIX timer handling for a thread group.
1642 */
1643static void posix_cpu_timers_init_group(struct signal_struct *sig)
1644{
1645 struct posix_cputimers *pct = &sig->posix_cputimers;
1646 unsigned long cpu_limit;
1647
1648 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1649 posix_cputimers_group_init(pct, cpu_limit);
1650}
1651
1652static int copy_signal(u64 clone_flags, struct task_struct *tsk)
1653{
1654 struct signal_struct *sig;
1655
1656 if (clone_flags & CLONE_THREAD)
1657 return 0;
1658
1659 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1660 tsk->signal = sig;
1661 if (!sig)
1662 return -ENOMEM;
1663
1664 sig->nr_threads = 1;
1665 sig->quick_threads = 1;
1666 atomic_set(v: &sig->live, i: 1);
1667 refcount_set(r: &sig->sigcnt, n: 1);
1668
1669 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1670 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1671 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1672
1673 init_waitqueue_head(&sig->wait_chldexit);
1674 sig->curr_target = tsk;
1675 init_sigpending(sig: &sig->shared_pending);
1676 INIT_HLIST_HEAD(&sig->multiprocess);
1677 seqlock_init(&sig->stats_lock);
1678 prev_cputime_init(prev: &sig->prev_cputime);
1679
1680#ifdef CONFIG_POSIX_TIMERS
1681 INIT_HLIST_HEAD(&sig->posix_timers);
1682 INIT_HLIST_HEAD(&sig->ignored_posix_timers);
1683 hrtimer_setup(timer: &sig->real_timer, function: it_real_fn, CLOCK_MONOTONIC, mode: HRTIMER_MODE_REL);
1684#endif
1685
1686 task_lock(current->group_leader);
1687 memcpy(to: sig->rlim, current->signal->rlim, len: sizeof sig->rlim);
1688 task_unlock(current->group_leader);
1689
1690 posix_cpu_timers_init_group(sig);
1691
1692 tty_audit_fork(sig);
1693 sched_autogroup_fork(sig);
1694
1695#ifdef CONFIG_CGROUPS
1696 init_rwsem(&sig->cgroup_threadgroup_rwsem);
1697#endif
1698
1699 sig->oom_score_adj = current->signal->oom_score_adj;
1700 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1701
1702 mutex_init(&sig->cred_guard_mutex);
1703 init_rwsem(&sig->exec_update_lock);
1704
1705 return 0;
1706}
1707
1708static void copy_seccomp(struct task_struct *p)
1709{
1710#ifdef CONFIG_SECCOMP
1711 /*
1712 * Must be called with sighand->lock held, which is common to
1713 * all threads in the group. Holding cred_guard_mutex is not
1714 * needed because this new task is not yet running and cannot
1715 * be racing exec.
1716 */
1717 assert_spin_locked(&current->sighand->siglock);
1718
1719 /* Ref-count the new filter user, and assign it. */
1720 get_seccomp_filter(current);
1721 p->seccomp = current->seccomp;
1722
1723 /*
1724 * Explicitly enable no_new_privs here in case it got set
1725 * between the task_struct being duplicated and holding the
1726 * sighand lock. The seccomp state and nnp must be in sync.
1727 */
1728 if (task_no_new_privs(current))
1729 task_set_no_new_privs(p);
1730
1731 /*
1732 * If the parent gained a seccomp mode after copying thread
1733 * flags and between before we held the sighand lock, we have
1734 * to manually enable the seccomp thread flag here.
1735 */
1736 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1737 set_task_syscall_work(p, SECCOMP);
1738#endif
1739}
1740
1741SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1742{
1743 current->clear_child_tid = tidptr;
1744
1745 return task_pid_vnr(current);
1746}
1747
1748static void rt_mutex_init_task(struct task_struct *p)
1749{
1750 raw_spin_lock_init(&p->pi_lock);
1751#ifdef CONFIG_RT_MUTEXES
1752 p->pi_waiters = RB_ROOT_CACHED;
1753 p->pi_top_task = NULL;
1754 p->pi_blocked_on = NULL;
1755#endif
1756}
1757
1758static inline void init_task_pid_links(struct task_struct *task)
1759{
1760 enum pid_type type;
1761
1762 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1763 INIT_HLIST_NODE(h: &task->pid_links[type]);
1764}
1765
1766static inline void
1767init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1768{
1769 if (type == PIDTYPE_PID)
1770 task->thread_pid = pid;
1771 else
1772 task->signal->pids[type] = pid;
1773}
1774
1775static inline void rcu_copy_process(struct task_struct *p)
1776{
1777#ifdef CONFIG_PREEMPT_RCU
1778 p->rcu_read_lock_nesting = 0;
1779 p->rcu_read_unlock_special.s = 0;
1780 p->rcu_blocked_node = NULL;
1781 INIT_LIST_HEAD(list: &p->rcu_node_entry);
1782#endif /* #ifdef CONFIG_PREEMPT_RCU */
1783#ifdef CONFIG_TASKS_RCU
1784 p->rcu_tasks_holdout = false;
1785 INIT_LIST_HEAD(list: &p->rcu_tasks_holdout_list);
1786 p->rcu_tasks_idle_cpu = -1;
1787 INIT_LIST_HEAD(list: &p->rcu_tasks_exit_list);
1788#endif /* #ifdef CONFIG_TASKS_RCU */
1789#ifdef CONFIG_TASKS_TRACE_RCU
1790 p->trc_reader_nesting = 0;
1791 p->trc_reader_special.s = 0;
1792 INIT_LIST_HEAD(list: &p->trc_holdout_list);
1793 INIT_LIST_HEAD(list: &p->trc_blkd_node);
1794#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1795}
1796
1797/**
1798 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
1799 * @pid: the struct pid for which to create a pidfd
1800 * @flags: flags of the new @pidfd
1801 * @ret_file: return the new pidfs file
1802 *
1803 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
1804 * caller's file descriptor table. The pidfd is reserved but not installed yet.
1805 *
1806 * The helper verifies that @pid is still in use, without PIDFD_THREAD the
1807 * task identified by @pid must be a thread-group leader.
1808 *
1809 * If this function returns successfully the caller is responsible to either
1810 * call fd_install() passing the returned pidfd and pidfd file as arguments in
1811 * order to install the pidfd into its file descriptor table or they must use
1812 * put_unused_fd() and fput() on the returned pidfd and pidfd file
1813 * respectively.
1814 *
1815 * This function is useful when a pidfd must already be reserved but there
1816 * might still be points of failure afterwards and the caller wants to ensure
1817 * that no pidfd is leaked into its file descriptor table.
1818 *
1819 * Return: On success, a reserved pidfd is returned from the function and a new
1820 * pidfd file is returned in the last argument to the function. On
1821 * error, a negative error code is returned from the function and the
1822 * last argument remains unchanged.
1823 */
1824int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret_file)
1825{
1826 struct file *pidfs_file;
1827
1828 /*
1829 * PIDFD_STALE is only allowed to be passed if the caller knows
1830 * that @pid is already registered in pidfs and thus
1831 * PIDFD_INFO_EXIT information is guaranteed to be available.
1832 */
1833 if (!(flags & PIDFD_STALE)) {
1834 /*
1835 * While holding the pidfd waitqueue lock removing the
1836 * task linkage for the thread-group leader pid
1837 * (PIDTYPE_TGID) isn't possible. Thus, if there's still
1838 * task linkage for PIDTYPE_PID not having thread-group
1839 * leader linkage for the pid means it wasn't a
1840 * thread-group leader in the first place.
1841 */
1842 guard(spinlock_irq)(l: &pid->wait_pidfd.lock);
1843
1844 /* Task has already been reaped. */
1845 if (!pid_has_task(pid, type: PIDTYPE_PID))
1846 return -ESRCH;
1847 /*
1848 * If this struct pid isn't used as a thread-group
1849 * leader but the caller requested to create a
1850 * thread-group leader pidfd then report ENOENT.
1851 */
1852 if (!(flags & PIDFD_THREAD) && !pid_has_task(pid, type: PIDTYPE_TGID))
1853 return -ENOENT;
1854 }
1855
1856 CLASS(get_unused_fd, pidfd)(O_CLOEXEC);
1857 if (pidfd < 0)
1858 return pidfd;
1859
1860 pidfs_file = pidfs_alloc_file(pid, flags: flags | O_RDWR);
1861 if (IS_ERR(ptr: pidfs_file))
1862 return PTR_ERR(ptr: pidfs_file);
1863
1864 *ret_file = pidfs_file;
1865 return take_fd(pidfd);
1866}
1867
1868static void __delayed_free_task(struct rcu_head *rhp)
1869{
1870 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1871
1872 free_task(tsk);
1873}
1874
1875static __always_inline void delayed_free_task(struct task_struct *tsk)
1876{
1877 if (IS_ENABLED(CONFIG_MEMCG))
1878 call_rcu(head: &tsk->rcu, func: __delayed_free_task);
1879 else
1880 free_task(tsk);
1881}
1882
1883static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1884{
1885 /* Skip if kernel thread */
1886 if (!tsk->mm)
1887 return;
1888
1889 /* Skip if spawning a thread or using vfork */
1890 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1891 return;
1892
1893 /* We need to synchronize with __set_oom_adj */
1894 mutex_lock(lock: &oom_adj_mutex);
1895 mm_flags_set(MMF_MULTIPROCESS, mm: tsk->mm);
1896 /* Update the values in case they were changed after copy_signal */
1897 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1898 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1899 mutex_unlock(lock: &oom_adj_mutex);
1900}
1901
1902#ifdef CONFIG_RV
1903static void rv_task_fork(struct task_struct *p)
1904{
1905 memset(&p->rv, 0, sizeof(p->rv));
1906}
1907#else
1908#define rv_task_fork(p) do {} while (0)
1909#endif
1910
1911static bool need_futex_hash_allocate_default(u64 clone_flags)
1912{
1913 if ((clone_flags & (CLONE_THREAD | CLONE_VM)) != (CLONE_THREAD | CLONE_VM))
1914 return false;
1915 return true;
1916}
1917
1918/*
1919 * This creates a new process as a copy of the old one,
1920 * but does not actually start it yet.
1921 *
1922 * It copies the registers, and all the appropriate
1923 * parts of the process environment (as per the clone
1924 * flags). The actual kick-off is left to the caller.
1925 */
1926__latent_entropy struct task_struct *copy_process(
1927 struct pid *pid,
1928 int trace,
1929 int node,
1930 struct kernel_clone_args *args)
1931{
1932 int pidfd = -1, retval;
1933 struct task_struct *p;
1934 struct multiprocess_signals delayed;
1935 struct file *pidfile = NULL;
1936 const u64 clone_flags = args->flags;
1937 struct nsproxy *nsp = current->nsproxy;
1938
1939 /*
1940 * Don't allow sharing the root directory with processes in a different
1941 * namespace
1942 */
1943 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1944 return ERR_PTR(error: -EINVAL);
1945
1946 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1947 return ERR_PTR(error: -EINVAL);
1948
1949 /*
1950 * Thread groups must share signals as well, and detached threads
1951 * can only be started up within the thread group.
1952 */
1953 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1954 return ERR_PTR(error: -EINVAL);
1955
1956 /*
1957 * Shared signal handlers imply shared VM. By way of the above,
1958 * thread groups also imply shared VM. Blocking this case allows
1959 * for various simplifications in other code.
1960 */
1961 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1962 return ERR_PTR(error: -EINVAL);
1963
1964 /*
1965 * Siblings of global init remain as zombies on exit since they are
1966 * not reaped by their parent (swapper). To solve this and to avoid
1967 * multi-rooted process trees, prevent global and container-inits
1968 * from creating siblings.
1969 */
1970 if ((clone_flags & CLONE_PARENT) &&
1971 current->signal->flags & SIGNAL_UNKILLABLE)
1972 return ERR_PTR(error: -EINVAL);
1973
1974 /*
1975 * If the new process will be in a different pid or user namespace
1976 * do not allow it to share a thread group with the forking task.
1977 */
1978 if (clone_flags & CLONE_THREAD) {
1979 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1980 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1981 return ERR_PTR(error: -EINVAL);
1982 }
1983
1984 if (clone_flags & CLONE_PIDFD) {
1985 /*
1986 * - CLONE_DETACHED is blocked so that we can potentially
1987 * reuse it later for CLONE_PIDFD.
1988 */
1989 if (clone_flags & CLONE_DETACHED)
1990 return ERR_PTR(error: -EINVAL);
1991 }
1992
1993 /*
1994 * Force any signals received before this point to be delivered
1995 * before the fork happens. Collect up signals sent to multiple
1996 * processes that happen during the fork and delay them so that
1997 * they appear to happen after the fork.
1998 */
1999 sigemptyset(set: &delayed.signal);
2000 INIT_HLIST_NODE(h: &delayed.node);
2001
2002 spin_lock_irq(lock: &current->sighand->siglock);
2003 if (!(clone_flags & CLONE_THREAD))
2004 hlist_add_head(n: &delayed.node, h: &current->signal->multiprocess);
2005 recalc_sigpending();
2006 spin_unlock_irq(lock: &current->sighand->siglock);
2007 retval = -ERESTARTNOINTR;
2008 if (task_sigpending(current))
2009 goto fork_out;
2010
2011 retval = -ENOMEM;
2012 p = dup_task_struct(current, node);
2013 if (!p)
2014 goto fork_out;
2015 p->flags &= ~PF_KTHREAD;
2016 if (args->kthread)
2017 p->flags |= PF_KTHREAD;
2018 if (args->user_worker) {
2019 /*
2020 * Mark us a user worker, and block any signal that isn't
2021 * fatal or STOP
2022 */
2023 p->flags |= PF_USER_WORKER;
2024 siginitsetinv(set: &p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2025 }
2026 if (args->io_thread)
2027 p->flags |= PF_IO_WORKER;
2028
2029 if (args->name)
2030 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2031
2032 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2033 /*
2034 * Clear TID on mm_release()?
2035 */
2036 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2037
2038 ftrace_graph_init_task(t: p);
2039
2040 rt_mutex_init_task(p);
2041
2042 lockdep_assert_irqs_enabled();
2043#ifdef CONFIG_PROVE_LOCKING
2044 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2045#endif
2046 retval = copy_creds(p, clone_flags);
2047 if (retval < 0)
2048 goto bad_fork_free;
2049
2050 retval = -EAGAIN;
2051 if (is_rlimit_overlimit(task_ucounts(p), type: UCOUNT_RLIMIT_NPROC, max: rlimit(RLIMIT_NPROC))) {
2052 if (p->real_cred->user != INIT_USER &&
2053 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2054 goto bad_fork_cleanup_count;
2055 }
2056 current->flags &= ~PF_NPROC_EXCEEDED;
2057
2058 /*
2059 * If multiple threads are within copy_process(), then this check
2060 * triggers too late. This doesn't hurt, the check is only there
2061 * to stop root fork bombs.
2062 */
2063 retval = -EAGAIN;
2064 if (data_race(nr_threads >= max_threads))
2065 goto bad_fork_cleanup_count;
2066
2067 delayacct_tsk_init(tsk: p); /* Must remain after dup_task_struct() */
2068 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2069 p->flags |= PF_FORKNOEXEC;
2070 INIT_LIST_HEAD(list: &p->children);
2071 INIT_LIST_HEAD(list: &p->sibling);
2072 rcu_copy_process(p);
2073 p->vfork_done = NULL;
2074 spin_lock_init(&p->alloc_lock);
2075
2076 init_sigpending(sig: &p->pending);
2077
2078 p->utime = p->stime = p->gtime = 0;
2079#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2080 p->utimescaled = p->stimescaled = 0;
2081#endif
2082 prev_cputime_init(prev: &p->prev_cputime);
2083
2084#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2085 seqcount_init(&p->vtime.seqcount);
2086 p->vtime.starttime = 0;
2087 p->vtime.state = VTIME_INACTIVE;
2088#endif
2089
2090#ifdef CONFIG_IO_URING
2091 p->io_uring = NULL;
2092#endif
2093
2094 p->default_timer_slack_ns = current->timer_slack_ns;
2095
2096#ifdef CONFIG_PSI
2097 p->psi_flags = 0;
2098#endif
2099
2100 task_io_accounting_init(ioac: &p->ioac);
2101 acct_clear_integrals(tsk: p);
2102
2103 posix_cputimers_init(pct: &p->posix_cputimers);
2104 tick_dep_init_task(tsk: p);
2105
2106 p->io_context = NULL;
2107 audit_set_context(task: p, NULL);
2108 cgroup_fork(p);
2109 if (args->kthread) {
2110 if (!set_kthread_struct(p))
2111 goto bad_fork_cleanup_delayacct;
2112 }
2113#ifdef CONFIG_NUMA
2114 p->mempolicy = mpol_dup(pol: p->mempolicy);
2115 if (IS_ERR(ptr: p->mempolicy)) {
2116 retval = PTR_ERR(ptr: p->mempolicy);
2117 p->mempolicy = NULL;
2118 goto bad_fork_cleanup_delayacct;
2119 }
2120#endif
2121#ifdef CONFIG_CPUSETS
2122 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2123 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2124#endif
2125#ifdef CONFIG_TRACE_IRQFLAGS
2126 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2127 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2128 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2129 p->softirqs_enabled = 1;
2130 p->softirq_context = 0;
2131#endif
2132
2133 p->pagefault_disabled = 0;
2134
2135 lockdep_init_task(task: p);
2136
2137 p->blocked_on = NULL; /* not blocked yet */
2138
2139#ifdef CONFIG_BCACHE
2140 p->sequential_io = 0;
2141 p->sequential_io_avg = 0;
2142#endif
2143#ifdef CONFIG_BPF_SYSCALL
2144 RCU_INIT_POINTER(p->bpf_storage, NULL);
2145 p->bpf_ctx = NULL;
2146#endif
2147
2148 unwind_task_init(task: p);
2149
2150 /* Perform scheduler related setup. Assign this task to a CPU. */
2151 retval = sched_fork(clone_flags, p);
2152 if (retval)
2153 goto bad_fork_cleanup_policy;
2154
2155 retval = perf_event_init_task(child: p, clone_flags);
2156 if (retval)
2157 goto bad_fork_sched_cancel_fork;
2158 retval = audit_alloc(task: p);
2159 if (retval)
2160 goto bad_fork_cleanup_perf;
2161 /* copy all the process information */
2162 shm_init_task(p);
2163 retval = security_task_alloc(task: p, clone_flags);
2164 if (retval)
2165 goto bad_fork_cleanup_audit;
2166 retval = copy_semundo(clone_flags, tsk: p);
2167 if (retval)
2168 goto bad_fork_cleanup_security;
2169 retval = copy_files(clone_flags, tsk: p, no_files: args->no_files);
2170 if (retval)
2171 goto bad_fork_cleanup_semundo;
2172 retval = copy_fs(clone_flags, tsk: p);
2173 if (retval)
2174 goto bad_fork_cleanup_files;
2175 retval = copy_sighand(clone_flags, tsk: p);
2176 if (retval)
2177 goto bad_fork_cleanup_fs;
2178 retval = copy_signal(clone_flags, tsk: p);
2179 if (retval)
2180 goto bad_fork_cleanup_sighand;
2181 retval = copy_mm(clone_flags, tsk: p);
2182 if (retval)
2183 goto bad_fork_cleanup_signal;
2184 retval = copy_namespaces(flags: clone_flags, tsk: p);
2185 if (retval)
2186 goto bad_fork_cleanup_mm;
2187 retval = copy_io(clone_flags, tsk: p);
2188 if (retval)
2189 goto bad_fork_cleanup_namespaces;
2190 retval = copy_thread(p, args);
2191 if (retval)
2192 goto bad_fork_cleanup_io;
2193
2194 stackleak_task_init(t: p);
2195
2196 if (pid != &init_struct_pid) {
2197 pid = alloc_pid(ns: p->nsproxy->pid_ns_for_children, set_tid: args->set_tid,
2198 set_tid_size: args->set_tid_size);
2199 if (IS_ERR(ptr: pid)) {
2200 retval = PTR_ERR(ptr: pid);
2201 goto bad_fork_cleanup_thread;
2202 }
2203 }
2204
2205 /*
2206 * This has to happen after we've potentially unshared the file
2207 * descriptor table (so that the pidfd doesn't leak into the child
2208 * if the fd table isn't shared).
2209 */
2210 if (clone_flags & CLONE_PIDFD) {
2211 int flags = (clone_flags & CLONE_THREAD) ? PIDFD_THREAD : 0;
2212
2213 /*
2214 * Note that no task has been attached to @pid yet indicate
2215 * that via CLONE_PIDFD.
2216 */
2217 retval = pidfd_prepare(pid, flags: flags | PIDFD_STALE, ret_file: &pidfile);
2218 if (retval < 0)
2219 goto bad_fork_free_pid;
2220 pidfd = retval;
2221
2222 retval = put_user(pidfd, args->pidfd);
2223 if (retval)
2224 goto bad_fork_put_pidfd;
2225 }
2226
2227#ifdef CONFIG_BLOCK
2228 p->plug = NULL;
2229#endif
2230 futex_init_task(tsk: p);
2231
2232 /*
2233 * sigaltstack should be cleared when sharing the same VM
2234 */
2235 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2236 sas_ss_reset(p);
2237
2238 /*
2239 * Syscall tracing and stepping should be turned off in the
2240 * child regardless of CLONE_PTRACE.
2241 */
2242 user_disable_single_step(p);
2243 clear_task_syscall_work(p, SYSCALL_TRACE);
2244#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2245 clear_task_syscall_work(p, SYSCALL_EMU);
2246#endif
2247 clear_tsk_latency_tracing(p);
2248
2249 /* ok, now we should be set up.. */
2250 p->pid = pid_nr(pid);
2251 if (clone_flags & CLONE_THREAD) {
2252 p->group_leader = current->group_leader;
2253 p->tgid = current->tgid;
2254 } else {
2255 p->group_leader = p;
2256 p->tgid = p->pid;
2257 }
2258
2259 p->nr_dirtied = 0;
2260 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2261 p->dirty_paused_when = 0;
2262
2263 p->pdeath_signal = 0;
2264 p->task_works = NULL;
2265 clear_posix_cputimers_work(p);
2266
2267#ifdef CONFIG_KRETPROBES
2268 p->kretprobe_instances.first = NULL;
2269#endif
2270#ifdef CONFIG_RETHOOK
2271 p->rethooks.first = NULL;
2272#endif
2273
2274 /*
2275 * Ensure that the cgroup subsystem policies allow the new process to be
2276 * forked. It should be noted that the new process's css_set can be changed
2277 * between here and cgroup_post_fork() if an organisation operation is in
2278 * progress.
2279 */
2280 retval = cgroup_can_fork(p, kargs: args);
2281 if (retval)
2282 goto bad_fork_put_pidfd;
2283
2284 /*
2285 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2286 * the new task on the correct runqueue. All this *before* the task
2287 * becomes visible.
2288 *
2289 * This isn't part of ->can_fork() because while the re-cloning is
2290 * cgroup specific, it unconditionally needs to place the task on a
2291 * runqueue.
2292 */
2293 retval = sched_cgroup_fork(p, kargs: args);
2294 if (retval)
2295 goto bad_fork_cancel_cgroup;
2296
2297 /*
2298 * Allocate a default futex hash for the user process once the first
2299 * thread spawns.
2300 */
2301 if (need_futex_hash_allocate_default(clone_flags)) {
2302 retval = futex_hash_allocate_default();
2303 if (retval)
2304 goto bad_fork_cancel_cgroup;
2305 /*
2306 * If we fail beyond this point we don't free the allocated
2307 * futex hash map. We assume that another thread will be created
2308 * and makes use of it. The hash map will be freed once the main
2309 * thread terminates.
2310 */
2311 }
2312 /*
2313 * From this point on we must avoid any synchronous user-space
2314 * communication until we take the tasklist-lock. In particular, we do
2315 * not want user-space to be able to predict the process start-time by
2316 * stalling fork(2) after we recorded the start_time but before it is
2317 * visible to the system.
2318 */
2319
2320 p->start_time = ktime_get_ns();
2321 p->start_boottime = ktime_get_boottime_ns();
2322
2323 /*
2324 * Make it visible to the rest of the system, but dont wake it up yet.
2325 * Need tasklist lock for parent etc handling!
2326 */
2327 write_lock_irq(&tasklist_lock);
2328
2329 /* CLONE_PARENT re-uses the old parent */
2330 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2331 p->real_parent = current->real_parent;
2332 p->parent_exec_id = current->parent_exec_id;
2333 if (clone_flags & CLONE_THREAD)
2334 p->exit_signal = -1;
2335 else
2336 p->exit_signal = current->group_leader->exit_signal;
2337 } else {
2338 p->real_parent = current;
2339 p->parent_exec_id = current->self_exec_id;
2340 p->exit_signal = args->exit_signal;
2341 }
2342
2343 klp_copy_process(child: p);
2344
2345 sched_core_fork(p);
2346
2347 spin_lock(lock: &current->sighand->siglock);
2348
2349 rv_task_fork(p);
2350
2351 rseq_fork(t: p, clone_flags);
2352
2353 /* Don't start children in a dying pid namespace */
2354 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2355 retval = -ENOMEM;
2356 goto bad_fork_core_free;
2357 }
2358
2359 /* Let kill terminate clone/fork in the middle */
2360 if (fatal_signal_pending(current)) {
2361 retval = -EINTR;
2362 goto bad_fork_core_free;
2363 }
2364
2365 /* No more failure paths after this point. */
2366
2367 /*
2368 * Copy seccomp details explicitly here, in case they were changed
2369 * before holding sighand lock.
2370 */
2371 copy_seccomp(p);
2372
2373 init_task_pid_links(task: p);
2374 if (likely(p->pid)) {
2375 ptrace_init_task(child: p, ptrace: (clone_flags & CLONE_PTRACE) || trace);
2376
2377 init_task_pid(task: p, type: PIDTYPE_PID, pid);
2378 if (thread_group_leader(p)) {
2379 init_task_pid(task: p, type: PIDTYPE_TGID, pid);
2380 init_task_pid(task: p, type: PIDTYPE_PGID, pid: task_pgrp(current));
2381 init_task_pid(task: p, type: PIDTYPE_SID, pid: task_session(current));
2382
2383 if (is_child_reaper(pid)) {
2384 ns_of_pid(pid)->child_reaper = p;
2385 p->signal->flags |= SIGNAL_UNKILLABLE;
2386 }
2387 p->signal->shared_pending.signal = delayed.signal;
2388 p->signal->tty = tty_kref_get(current->signal->tty);
2389 /*
2390 * Inherit has_child_subreaper flag under the same
2391 * tasklist_lock with adding child to the process tree
2392 * for propagate_has_child_subreaper optimization.
2393 */
2394 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2395 p->real_parent->signal->is_child_subreaper;
2396 list_add_tail(new: &p->sibling, head: &p->real_parent->children);
2397 list_add_tail_rcu(new: &p->tasks, head: &init_task.tasks);
2398 attach_pid(task: p, PIDTYPE_TGID);
2399 attach_pid(task: p, PIDTYPE_PGID);
2400 attach_pid(task: p, PIDTYPE_SID);
2401 __this_cpu_inc(process_counts);
2402 } else {
2403 current->signal->nr_threads++;
2404 current->signal->quick_threads++;
2405 atomic_inc(v: &current->signal->live);
2406 refcount_inc(r: &current->signal->sigcnt);
2407 task_join_group_stop(task: p);
2408 list_add_tail_rcu(new: &p->thread_node,
2409 head: &p->signal->thread_head);
2410 }
2411 attach_pid(task: p, PIDTYPE_PID);
2412 nr_threads++;
2413 }
2414 total_forks++;
2415 hlist_del_init(n: &delayed.node);
2416 spin_unlock(lock: &current->sighand->siglock);
2417 syscall_tracepoint_update(p);
2418 write_unlock_irq(&tasklist_lock);
2419
2420 if (pidfile)
2421 fd_install(fd: pidfd, file: pidfile);
2422
2423 proc_fork_connector(task: p);
2424 sched_post_fork(p);
2425 cgroup_post_fork(p, kargs: args);
2426 perf_event_fork(tsk: p);
2427
2428 trace_task_newtask(task: p, clone_flags);
2429 uprobe_copy_process(t: p, flags: clone_flags);
2430 user_events_fork(t: p, clone_flags);
2431
2432 copy_oom_score_adj(clone_flags, tsk: p);
2433
2434 return p;
2435
2436bad_fork_core_free:
2437 sched_core_free(tsk: p);
2438 spin_unlock(lock: &current->sighand->siglock);
2439 write_unlock_irq(&tasklist_lock);
2440bad_fork_cancel_cgroup:
2441 cgroup_cancel_fork(p, kargs: args);
2442bad_fork_put_pidfd:
2443 if (clone_flags & CLONE_PIDFD) {
2444 fput(pidfile);
2445 put_unused_fd(fd: pidfd);
2446 }
2447bad_fork_free_pid:
2448 if (pid != &init_struct_pid)
2449 free_pid(pid);
2450bad_fork_cleanup_thread:
2451 exit_thread(tsk: p);
2452bad_fork_cleanup_io:
2453 if (p->io_context)
2454 exit_io_context(task: p);
2455bad_fork_cleanup_namespaces:
2456 exit_task_namespaces(tsk: p);
2457bad_fork_cleanup_mm:
2458 if (p->mm) {
2459 mm_clear_owner(mm: p->mm, p);
2460 mmput(p->mm);
2461 }
2462bad_fork_cleanup_signal:
2463 if (!(clone_flags & CLONE_THREAD))
2464 free_signal_struct(sig: p->signal);
2465bad_fork_cleanup_sighand:
2466 __cleanup_sighand(sighand: p->sighand);
2467bad_fork_cleanup_fs:
2468 exit_fs(p); /* blocking */
2469bad_fork_cleanup_files:
2470 exit_files(p); /* blocking */
2471bad_fork_cleanup_semundo:
2472 exit_sem(tsk: p);
2473bad_fork_cleanup_security:
2474 security_task_free(task: p);
2475bad_fork_cleanup_audit:
2476 audit_free(task: p);
2477bad_fork_cleanup_perf:
2478 perf_event_free_task(task: p);
2479bad_fork_sched_cancel_fork:
2480 sched_cancel_fork(p);
2481bad_fork_cleanup_policy:
2482 lockdep_free_task(task: p);
2483#ifdef CONFIG_NUMA
2484 mpol_put(pol: p->mempolicy);
2485#endif
2486bad_fork_cleanup_delayacct:
2487 delayacct_tsk_free(tsk: p);
2488bad_fork_cleanup_count:
2489 dec_rlimit_ucounts(task_ucounts(p), type: UCOUNT_RLIMIT_NPROC, v: 1);
2490 exit_creds(p);
2491bad_fork_free:
2492 WRITE_ONCE(p->__state, TASK_DEAD);
2493 exit_task_stack_account(tsk: p);
2494 put_task_stack(tsk: p);
2495 delayed_free_task(tsk: p);
2496fork_out:
2497 spin_lock_irq(lock: &current->sighand->siglock);
2498 hlist_del_init(n: &delayed.node);
2499 spin_unlock_irq(lock: &current->sighand->siglock);
2500 return ERR_PTR(error: retval);
2501}
2502
2503static inline void init_idle_pids(struct task_struct *idle)
2504{
2505 enum pid_type type;
2506
2507 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2508 INIT_HLIST_NODE(h: &idle->pid_links[type]); /* not really needed */
2509 init_task_pid(task: idle, type, pid: &init_struct_pid);
2510 }
2511}
2512
2513static int idle_dummy(void *dummy)
2514{
2515 /* This function is never called */
2516 return 0;
2517}
2518
2519struct task_struct * __init fork_idle(int cpu)
2520{
2521 struct task_struct *task;
2522 struct kernel_clone_args args = {
2523 .flags = CLONE_VM,
2524 .fn = &idle_dummy,
2525 .fn_arg = NULL,
2526 .kthread = 1,
2527 .idle = 1,
2528 };
2529
2530 task = copy_process(pid: &init_struct_pid, trace: 0, node: cpu_to_node(cpu), args: &args);
2531 if (!IS_ERR(ptr: task)) {
2532 init_idle_pids(idle: task);
2533 init_idle(idle: task, cpu);
2534 }
2535
2536 return task;
2537}
2538
2539/*
2540 * This is like kernel_clone(), but shaved down and tailored to just
2541 * creating io_uring workers. It returns a created task, or an error pointer.
2542 * The returned task is inactive, and the caller must fire it up through
2543 * wake_up_new_task(p). All signals are blocked in the created task.
2544 */
2545struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2546{
2547 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2548 CLONE_IO|CLONE_VM|CLONE_UNTRACED;
2549 struct kernel_clone_args args = {
2550 .flags = flags,
2551 .fn = fn,
2552 .fn_arg = arg,
2553 .io_thread = 1,
2554 .user_worker = 1,
2555 };
2556
2557 return copy_process(NULL, trace: 0, node, args: &args);
2558}
2559
2560/*
2561 * Ok, this is the main fork-routine.
2562 *
2563 * It copies the process, and if successful kick-starts
2564 * it and waits for it to finish using the VM if required.
2565 *
2566 * args->exit_signal is expected to be checked for sanity by the caller.
2567 */
2568pid_t kernel_clone(struct kernel_clone_args *args)
2569{
2570 u64 clone_flags = args->flags;
2571 struct completion vfork;
2572 struct pid *pid;
2573 struct task_struct *p;
2574 int trace = 0;
2575 pid_t nr;
2576
2577 /*
2578 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2579 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2580 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2581 * field in struct clone_args and it still doesn't make sense to have
2582 * them both point at the same memory location. Performing this check
2583 * here has the advantage that we don't need to have a separate helper
2584 * to check for legacy clone().
2585 */
2586 if ((clone_flags & CLONE_PIDFD) &&
2587 (clone_flags & CLONE_PARENT_SETTID) &&
2588 (args->pidfd == args->parent_tid))
2589 return -EINVAL;
2590
2591 /*
2592 * Determine whether and which event to report to ptracer. When
2593 * called from kernel_thread or CLONE_UNTRACED is explicitly
2594 * requested, no event is reported; otherwise, report if the event
2595 * for the type of forking is enabled.
2596 */
2597 if (!(clone_flags & CLONE_UNTRACED)) {
2598 if (clone_flags & CLONE_VFORK)
2599 trace = PTRACE_EVENT_VFORK;
2600 else if (args->exit_signal != SIGCHLD)
2601 trace = PTRACE_EVENT_CLONE;
2602 else
2603 trace = PTRACE_EVENT_FORK;
2604
2605 if (likely(!ptrace_event_enabled(current, trace)))
2606 trace = 0;
2607 }
2608
2609 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2610 add_latent_entropy();
2611
2612 if (IS_ERR(ptr: p))
2613 return PTR_ERR(ptr: p);
2614
2615 /*
2616 * Do this prior waking up the new thread - the thread pointer
2617 * might get invalid after that point, if the thread exits quickly.
2618 */
2619 trace_sched_process_fork(current, child: p);
2620
2621 pid = get_task_pid(task: p, type: PIDTYPE_PID);
2622 nr = pid_vnr(pid);
2623
2624 if (clone_flags & CLONE_PARENT_SETTID)
2625 put_user(nr, args->parent_tid);
2626
2627 if (clone_flags & CLONE_VFORK) {
2628 p->vfork_done = &vfork;
2629 init_completion(x: &vfork);
2630 get_task_struct(t: p);
2631 }
2632
2633 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2634 /* lock the task to synchronize with memcg migration */
2635 task_lock(p);
2636 lru_gen_add_mm(mm: p->mm);
2637 task_unlock(p);
2638 }
2639
2640 wake_up_new_task(tsk: p);
2641
2642 /* forking complete and child started to run, tell ptracer */
2643 if (unlikely(trace))
2644 ptrace_event_pid(event: trace, pid);
2645
2646 if (clone_flags & CLONE_VFORK) {
2647 if (!wait_for_vfork_done(child: p, vfork: &vfork))
2648 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2649 }
2650
2651 put_pid(pid);
2652 return nr;
2653}
2654
2655/*
2656 * Create a kernel thread.
2657 */
2658pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2659 unsigned long flags)
2660{
2661 struct kernel_clone_args args = {
2662 .flags = ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2663 .exit_signal = (flags & CSIGNAL),
2664 .fn = fn,
2665 .fn_arg = arg,
2666 .name = name,
2667 .kthread = 1,
2668 };
2669
2670 return kernel_clone(args: &args);
2671}
2672
2673/*
2674 * Create a user mode thread.
2675 */
2676pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2677{
2678 struct kernel_clone_args args = {
2679 .flags = ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2680 .exit_signal = (flags & CSIGNAL),
2681 .fn = fn,
2682 .fn_arg = arg,
2683 };
2684
2685 return kernel_clone(args: &args);
2686}
2687
2688#ifdef __ARCH_WANT_SYS_FORK
2689SYSCALL_DEFINE0(fork)
2690{
2691#ifdef CONFIG_MMU
2692 struct kernel_clone_args args = {
2693 .exit_signal = SIGCHLD,
2694 };
2695
2696 return kernel_clone(args: &args);
2697#else
2698 /* can not support in nommu mode */
2699 return -EINVAL;
2700#endif
2701}
2702#endif
2703
2704#ifdef __ARCH_WANT_SYS_VFORK
2705SYSCALL_DEFINE0(vfork)
2706{
2707 struct kernel_clone_args args = {
2708 .flags = CLONE_VFORK | CLONE_VM,
2709 .exit_signal = SIGCHLD,
2710 };
2711
2712 return kernel_clone(args: &args);
2713}
2714#endif
2715
2716#ifdef __ARCH_WANT_SYS_CLONE
2717#ifdef CONFIG_CLONE_BACKWARDS
2718SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2719 int __user *, parent_tidptr,
2720 unsigned long, tls,
2721 int __user *, child_tidptr)
2722#elif defined(CONFIG_CLONE_BACKWARDS2)
2723SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2724 int __user *, parent_tidptr,
2725 int __user *, child_tidptr,
2726 unsigned long, tls)
2727#elif defined(CONFIG_CLONE_BACKWARDS3)
2728SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2729 int, stack_size,
2730 int __user *, parent_tidptr,
2731 int __user *, child_tidptr,
2732 unsigned long, tls)
2733#else
2734SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2735 int __user *, parent_tidptr,
2736 int __user *, child_tidptr,
2737 unsigned long, tls)
2738#endif
2739{
2740 struct kernel_clone_args args = {
2741 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2742 .pidfd = parent_tidptr,
2743 .child_tid = child_tidptr,
2744 .parent_tid = parent_tidptr,
2745 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2746 .stack = newsp,
2747 .tls = tls,
2748 };
2749
2750 return kernel_clone(args: &args);
2751}
2752#endif
2753
2754static noinline int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2755 struct clone_args __user *uargs,
2756 size_t usize)
2757{
2758 int err;
2759 struct clone_args args;
2760 pid_t *kset_tid = kargs->set_tid;
2761
2762 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2763 CLONE_ARGS_SIZE_VER0);
2764 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2765 CLONE_ARGS_SIZE_VER1);
2766 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2767 CLONE_ARGS_SIZE_VER2);
2768 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2769
2770 if (unlikely(usize > PAGE_SIZE))
2771 return -E2BIG;
2772 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2773 return -EINVAL;
2774
2775 err = copy_struct_from_user(dst: &args, ksize: sizeof(args), src: uargs, usize);
2776 if (err)
2777 return err;
2778
2779 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2780 return -EINVAL;
2781
2782 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2783 return -EINVAL;
2784
2785 if (unlikely(args.set_tid && args.set_tid_size == 0))
2786 return -EINVAL;
2787
2788 /*
2789 * Verify that higher 32bits of exit_signal are unset and that
2790 * it is a valid signal
2791 */
2792 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2793 !valid_signal(args.exit_signal)))
2794 return -EINVAL;
2795
2796 if ((args.flags & CLONE_INTO_CGROUP) &&
2797 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2798 return -EINVAL;
2799
2800 *kargs = (struct kernel_clone_args){
2801 .flags = args.flags,
2802 .pidfd = u64_to_user_ptr(args.pidfd),
2803 .child_tid = u64_to_user_ptr(args.child_tid),
2804 .parent_tid = u64_to_user_ptr(args.parent_tid),
2805 .exit_signal = args.exit_signal,
2806 .stack = args.stack,
2807 .stack_size = args.stack_size,
2808 .tls = args.tls,
2809 .set_tid_size = args.set_tid_size,
2810 .cgroup = args.cgroup,
2811 };
2812
2813 if (args.set_tid &&
2814 copy_from_user(to: kset_tid, u64_to_user_ptr(args.set_tid),
2815 n: (kargs->set_tid_size * sizeof(pid_t))))
2816 return -EFAULT;
2817
2818 kargs->set_tid = kset_tid;
2819
2820 return 0;
2821}
2822
2823/**
2824 * clone3_stack_valid - check and prepare stack
2825 * @kargs: kernel clone args
2826 *
2827 * Verify that the stack arguments userspace gave us are sane.
2828 * In addition, set the stack direction for userspace since it's easy for us to
2829 * determine.
2830 */
2831static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2832{
2833 if (kargs->stack == 0) {
2834 if (kargs->stack_size > 0)
2835 return false;
2836 } else {
2837 if (kargs->stack_size == 0)
2838 return false;
2839
2840 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2841 return false;
2842
2843#if !defined(CONFIG_STACK_GROWSUP)
2844 kargs->stack += kargs->stack_size;
2845#endif
2846 }
2847
2848 return true;
2849}
2850
2851static bool clone3_args_valid(struct kernel_clone_args *kargs)
2852{
2853 /* Verify that no unknown flags are passed along. */
2854 if (kargs->flags &
2855 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2856 return false;
2857
2858 /*
2859 * - make the CLONE_DETACHED bit reusable for clone3
2860 * - make the CSIGNAL bits reusable for clone3
2861 */
2862 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
2863 return false;
2864
2865 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2866 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2867 return false;
2868
2869 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2870 kargs->exit_signal)
2871 return false;
2872
2873 if (!clone3_stack_valid(kargs))
2874 return false;
2875
2876 return true;
2877}
2878
2879/**
2880 * sys_clone3 - create a new process with specific properties
2881 * @uargs: argument structure
2882 * @size: size of @uargs
2883 *
2884 * clone3() is the extensible successor to clone()/clone2().
2885 * It takes a struct as argument that is versioned by its size.
2886 *
2887 * Return: On success, a positive PID for the child process.
2888 * On error, a negative errno number.
2889 */
2890SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2891{
2892 int err;
2893
2894 struct kernel_clone_args kargs;
2895 pid_t set_tid[MAX_PID_NS_LEVEL];
2896
2897#ifdef __ARCH_BROKEN_SYS_CLONE3
2898#warning clone3() entry point is missing, please fix
2899 return -ENOSYS;
2900#endif
2901
2902 kargs.set_tid = set_tid;
2903
2904 err = copy_clone_args_from_user(kargs: &kargs, uargs, usize: size);
2905 if (err)
2906 return err;
2907
2908 if (!clone3_args_valid(kargs: &kargs))
2909 return -EINVAL;
2910
2911 return kernel_clone(args: &kargs);
2912}
2913
2914void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2915{
2916 struct task_struct *leader, *parent, *child;
2917 int res;
2918
2919 read_lock(&tasklist_lock);
2920 leader = top = top->group_leader;
2921down:
2922 for_each_thread(leader, parent) {
2923 list_for_each_entry(child, &parent->children, sibling) {
2924 res = visitor(child, data);
2925 if (res) {
2926 if (res < 0)
2927 goto out;
2928 leader = child;
2929 goto down;
2930 }
2931up:
2932 ;
2933 }
2934 }
2935
2936 if (leader != top) {
2937 child = leader;
2938 parent = child->real_parent;
2939 leader = parent->group_leader;
2940 goto up;
2941 }
2942out:
2943 read_unlock(&tasklist_lock);
2944}
2945
2946#ifndef ARCH_MIN_MMSTRUCT_ALIGN
2947#define ARCH_MIN_MMSTRUCT_ALIGN 0
2948#endif
2949
2950static void sighand_ctor(void *data)
2951{
2952 struct sighand_struct *sighand = data;
2953
2954 spin_lock_init(&sighand->siglock);
2955 init_waitqueue_head(&sighand->signalfd_wqh);
2956}
2957
2958void __init mm_cache_init(void)
2959{
2960 unsigned int mm_size;
2961
2962 /*
2963 * The mm_cpumask is located at the end of mm_struct, and is
2964 * dynamically sized based on the maximum CPU number this system
2965 * can have, taking hotplug into account (nr_cpu_ids).
2966 */
2967 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
2968
2969 mm_cachep = kmem_cache_create_usercopy(name: "mm_struct",
2970 size: mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2971 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2972 offsetof(struct mm_struct, saved_auxv),
2973 sizeof_field(struct mm_struct, saved_auxv),
2974 NULL);
2975}
2976
2977void __init proc_caches_init(void)
2978{
2979 sighand_cachep = kmem_cache_create("sighand_cache",
2980 sizeof(struct sighand_struct), 0,
2981 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2982 SLAB_ACCOUNT, sighand_ctor);
2983 signal_cachep = kmem_cache_create("signal_cache",
2984 sizeof(struct signal_struct), 0,
2985 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2986 NULL);
2987 files_cachep = kmem_cache_create("files_cache",
2988 sizeof(struct files_struct), 0,
2989 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2990 NULL);
2991 fs_cachep = kmem_cache_create("fs_cache",
2992 sizeof(struct fs_struct), 0,
2993 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2994 NULL);
2995 mmap_init();
2996 nsproxy_cache_init();
2997}
2998
2999/*
3000 * Check constraints on flags passed to the unshare system call.
3001 */
3002static int check_unshare_flags(unsigned long unshare_flags)
3003{
3004 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3005 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3006 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3007 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3008 CLONE_NEWTIME))
3009 return -EINVAL;
3010 /*
3011 * Not implemented, but pretend it works if there is nothing
3012 * to unshare. Note that unsharing the address space or the
3013 * signal handlers also need to unshare the signal queues (aka
3014 * CLONE_THREAD).
3015 */
3016 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3017 if (!thread_group_empty(current))
3018 return -EINVAL;
3019 }
3020 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3021 if (refcount_read(r: &current->sighand->count) > 1)
3022 return -EINVAL;
3023 }
3024 if (unshare_flags & CLONE_VM) {
3025 if (!current_is_single_threaded())
3026 return -EINVAL;
3027 }
3028
3029 return 0;
3030}
3031
3032/*
3033 * Unshare the filesystem structure if it is being shared
3034 */
3035static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3036{
3037 struct fs_struct *fs = current->fs;
3038
3039 if (!(unshare_flags & CLONE_FS) || !fs)
3040 return 0;
3041
3042 /* don't need lock here; in the worst case we'll do useless copy */
3043 if (fs->users == 1)
3044 return 0;
3045
3046 *new_fsp = copy_fs_struct(fs);
3047 if (!*new_fsp)
3048 return -ENOMEM;
3049
3050 return 0;
3051}
3052
3053/*
3054 * Unshare file descriptor table if it is being shared
3055 */
3056static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
3057{
3058 struct files_struct *fd = current->files;
3059
3060 if ((unshare_flags & CLONE_FILES) &&
3061 (fd && atomic_read(v: &fd->count) > 1)) {
3062 fd = dup_fd(fd, NULL);
3063 if (IS_ERR(ptr: fd))
3064 return PTR_ERR(ptr: fd);
3065 *new_fdp = fd;
3066 }
3067
3068 return 0;
3069}
3070
3071/*
3072 * unshare allows a process to 'unshare' part of the process
3073 * context which was originally shared using clone. copy_*
3074 * functions used by kernel_clone() cannot be used here directly
3075 * because they modify an inactive task_struct that is being
3076 * constructed. Here we are modifying the current, active,
3077 * task_struct.
3078 */
3079int ksys_unshare(unsigned long unshare_flags)
3080{
3081 struct fs_struct *fs, *new_fs = NULL;
3082 struct files_struct *new_fd = NULL;
3083 struct cred *new_cred = NULL;
3084 struct nsproxy *new_nsproxy = NULL;
3085 int do_sysvsem = 0;
3086 int err;
3087
3088 /*
3089 * If unsharing a user namespace must also unshare the thread group
3090 * and unshare the filesystem root and working directories.
3091 */
3092 if (unshare_flags & CLONE_NEWUSER)
3093 unshare_flags |= CLONE_THREAD | CLONE_FS;
3094 /*
3095 * If unsharing vm, must also unshare signal handlers.
3096 */
3097 if (unshare_flags & CLONE_VM)
3098 unshare_flags |= CLONE_SIGHAND;
3099 /*
3100 * If unsharing a signal handlers, must also unshare the signal queues.
3101 */
3102 if (unshare_flags & CLONE_SIGHAND)
3103 unshare_flags |= CLONE_THREAD;
3104 /*
3105 * If unsharing namespace, must also unshare filesystem information.
3106 */
3107 if (unshare_flags & CLONE_NEWNS)
3108 unshare_flags |= CLONE_FS;
3109
3110 err = check_unshare_flags(unshare_flags);
3111 if (err)
3112 goto bad_unshare_out;
3113 /*
3114 * CLONE_NEWIPC must also detach from the undolist: after switching
3115 * to a new ipc namespace, the semaphore arrays from the old
3116 * namespace are unreachable.
3117 */
3118 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3119 do_sysvsem = 1;
3120 err = unshare_fs(unshare_flags, new_fsp: &new_fs);
3121 if (err)
3122 goto bad_unshare_out;
3123 err = unshare_fd(unshare_flags, new_fdp: &new_fd);
3124 if (err)
3125 goto bad_unshare_cleanup_fs;
3126 err = unshare_userns(unshare_flags, new_cred: &new_cred);
3127 if (err)
3128 goto bad_unshare_cleanup_fd;
3129 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3130 new_cred, new_fs);
3131 if (err)
3132 goto bad_unshare_cleanup_cred;
3133
3134 if (new_cred) {
3135 err = set_cred_ucounts(new_cred);
3136 if (err)
3137 goto bad_unshare_cleanup_cred;
3138 }
3139
3140 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3141 if (do_sysvsem) {
3142 /*
3143 * CLONE_SYSVSEM is equivalent to sys_exit().
3144 */
3145 exit_sem(current);
3146 }
3147 if (unshare_flags & CLONE_NEWIPC) {
3148 /* Orphan segments in old ns (see sem above). */
3149 exit_shm(current);
3150 shm_init_task(current);
3151 }
3152
3153 if (new_nsproxy)
3154 switch_task_namespaces(current, new: new_nsproxy);
3155
3156 task_lock(current);
3157
3158 if (new_fs) {
3159 fs = current->fs;
3160 read_seqlock_excl(sl: &fs->seq);
3161 current->fs = new_fs;
3162 if (--fs->users)
3163 new_fs = NULL;
3164 else
3165 new_fs = fs;
3166 read_sequnlock_excl(sl: &fs->seq);
3167 }
3168
3169 if (new_fd)
3170 swap(current->files, new_fd);
3171
3172 task_unlock(current);
3173
3174 if (new_cred) {
3175 /* Install the new user namespace */
3176 commit_creds(new_cred);
3177 new_cred = NULL;
3178 }
3179 }
3180
3181 perf_event_namespaces(current);
3182
3183bad_unshare_cleanup_cred:
3184 if (new_cred)
3185 put_cred(cred: new_cred);
3186bad_unshare_cleanup_fd:
3187 if (new_fd)
3188 put_files_struct(fs: new_fd);
3189
3190bad_unshare_cleanup_fs:
3191 if (new_fs)
3192 free_fs_struct(new_fs);
3193
3194bad_unshare_out:
3195 return err;
3196}
3197
3198SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3199{
3200 return ksys_unshare(unshare_flags);
3201}
3202
3203/*
3204 * Helper to unshare the files of the current task.
3205 * We don't want to expose copy_files internals to
3206 * the exec layer of the kernel.
3207 */
3208
3209int unshare_files(void)
3210{
3211 struct task_struct *task = current;
3212 struct files_struct *old, *copy = NULL;
3213 int error;
3214
3215 error = unshare_fd(CLONE_FILES, new_fdp: &copy);
3216 if (error || !copy)
3217 return error;
3218
3219 old = task->files;
3220 task_lock(p: task);
3221 task->files = copy;
3222 task_unlock(p: task);
3223 put_files_struct(fs: old);
3224 return 0;
3225}
3226
3227static int sysctl_max_threads(const struct ctl_table *table, int write,
3228 void *buffer, size_t *lenp, loff_t *ppos)
3229{
3230 struct ctl_table t;
3231 int ret;
3232 int threads = max_threads;
3233 int min = 1;
3234 int max = MAX_THREADS;
3235
3236 t = *table;
3237 t.data = &threads;
3238 t.extra1 = &min;
3239 t.extra2 = &max;
3240
3241 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3242 if (ret || !write)
3243 return ret;
3244
3245 max_threads = threads;
3246
3247 return 0;
3248}
3249
3250static const struct ctl_table fork_sysctl_table[] = {
3251 {
3252 .procname = "threads-max",
3253 .data = NULL,
3254 .maxlen = sizeof(int),
3255 .mode = 0644,
3256 .proc_handler = sysctl_max_threads,
3257 },
3258};
3259
3260static int __init init_fork_sysctl(void)
3261{
3262 register_sysctl_init("kernel", fork_sysctl_table);
3263 return 0;
3264}
3265
3266subsys_initcall(init_fork_sysctl);
3267