Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : /*
3 : * linux/kernel/fork.c
4 : *
5 : * Copyright (C) 1991, 1992 Linus Torvalds
6 : */
7 :
8 : /*
9 : * 'fork.c' contains the help-routines for the 'fork' system call
10 : * (see also entry.S and others).
11 : * Fork is rather simple, once you get the hang of it, but the memory
12 : * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
13 : */
14 :
15 : #include <linux/anon_inodes.h>
16 : #include <linux/slab.h>
17 : #include <linux/sched/autogroup.h>
18 : #include <linux/sched/mm.h>
19 : #include <linux/sched/coredump.h>
20 : #include <linux/sched/user.h>
21 : #include <linux/sched/numa_balancing.h>
22 : #include <linux/sched/stat.h>
23 : #include <linux/sched/task.h>
24 : #include <linux/sched/task_stack.h>
25 : #include <linux/sched/cputime.h>
26 : #include <linux/seq_file.h>
27 : #include <linux/rtmutex.h>
28 : #include <linux/init.h>
29 : #include <linux/unistd.h>
30 : #include <linux/module.h>
31 : #include <linux/vmalloc.h>
32 : #include <linux/completion.h>
33 : #include <linux/personality.h>
34 : #include <linux/mempolicy.h>
35 : #include <linux/sem.h>
36 : #include <linux/file.h>
37 : #include <linux/fdtable.h>
38 : #include <linux/iocontext.h>
39 : #include <linux/key.h>
40 : #include <linux/binfmts.h>
41 : #include <linux/mman.h>
42 : #include <linux/mmu_notifier.h>
43 : #include <linux/fs.h>
44 : #include <linux/mm.h>
45 : #include <linux/vmacache.h>
46 : #include <linux/nsproxy.h>
47 : #include <linux/capability.h>
48 : #include <linux/cpu.h>
49 : #include <linux/cgroup.h>
50 : #include <linux/security.h>
51 : #include <linux/hugetlb.h>
52 : #include <linux/seccomp.h>
53 : #include <linux/swap.h>
54 : #include <linux/syscalls.h>
55 : #include <linux/jiffies.h>
56 : #include <linux/futex.h>
57 : #include <linux/compat.h>
58 : #include <linux/kthread.h>
59 : #include <linux/task_io_accounting_ops.h>
60 : #include <linux/rcupdate.h>
61 : #include <linux/ptrace.h>
62 : #include <linux/mount.h>
63 : #include <linux/audit.h>
64 : #include <linux/memcontrol.h>
65 : #include <linux/ftrace.h>
66 : #include <linux/proc_fs.h>
67 : #include <linux/profile.h>
68 : #include <linux/rmap.h>
69 : #include <linux/ksm.h>
70 : #include <linux/acct.h>
71 : #include <linux/userfaultfd_k.h>
72 : #include <linux/tsacct_kern.h>
73 : #include <linux/cn_proc.h>
74 : #include <linux/freezer.h>
75 : #include <linux/delayacct.h>
76 : #include <linux/taskstats_kern.h>
77 : #include <linux/random.h>
78 : #include <linux/tty.h>
79 : #include <linux/blkdev.h>
80 : #include <linux/fs_struct.h>
81 : #include <linux/magic.h>
82 : #include <linux/perf_event.h>
83 : #include <linux/posix-timers.h>
84 : #include <linux/user-return-notifier.h>
85 : #include <linux/oom.h>
86 : #include <linux/khugepaged.h>
87 : #include <linux/signalfd.h>
88 : #include <linux/uprobes.h>
89 : #include <linux/aio.h>
90 : #include <linux/compiler.h>
91 : #include <linux/sysctl.h>
92 : #include <linux/kcov.h>
93 : #include <linux/livepatch.h>
94 : #include <linux/thread_info.h>
95 : #include <linux/stackleak.h>
96 : #include <linux/kasan.h>
97 : #include <linux/scs.h>
98 : #include <linux/io_uring.h>
99 :
100 : #include <asm/pgalloc.h>
101 : #include <linux/uaccess.h>
102 : #include <asm/mmu_context.h>
103 : #include <asm/cacheflush.h>
104 : #include <asm/tlbflush.h>
105 :
106 : #include <trace/events/sched.h>
107 :
108 : #define CREATE_TRACE_POINTS
109 : #include <trace/events/task.h>
110 :
111 : /*
112 : * Minimum number of threads to boot the kernel
113 : */
114 : #define MIN_THREADS 20
115 :
116 : /*
117 : * Maximum number of threads
118 : */
119 : #define MAX_THREADS FUTEX_TID_MASK
120 :
121 : /*
122 : * Protected counters by write_lock_irq(&tasklist_lock)
123 : */
124 : unsigned long total_forks; /* Handle normal Linux uptimes. */
125 : int nr_threads; /* The idle threads do not count.. */
126 :
127 : static int max_threads; /* tunable limit on nr_threads */
128 :
129 : #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
130 :
131 : static const char * const resident_page_types[] = {
132 : NAMED_ARRAY_INDEX(MM_FILEPAGES),
133 : NAMED_ARRAY_INDEX(MM_ANONPAGES),
134 : NAMED_ARRAY_INDEX(MM_SWAPENTS),
135 : NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
136 : };
137 :
138 : DEFINE_PER_CPU(unsigned long, process_counts) = 0;
139 :
140 : __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
141 :
142 : #ifdef CONFIG_PROVE_RCU
143 8143 : int lockdep_tasklist_lock_is_held(void)
144 : {
145 8143 : return lockdep_is_held(&tasklist_lock);
146 : }
147 : EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
148 : #endif /* #ifdef CONFIG_PROVE_RCU */
149 :
150 153 : int nr_processes(void)
151 : {
152 153 : int cpu;
153 153 : int total = 0;
154 :
155 765 : for_each_possible_cpu(cpu)
156 612 : total += per_cpu(process_counts, cpu);
157 :
158 153 : return total;
159 : }
160 :
161 834 : void __weak arch_release_task_struct(struct task_struct *tsk)
162 : {
163 834 : }
164 :
165 : #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
166 : static struct kmem_cache *task_struct_cachep;
167 :
168 916 : static inline struct task_struct *alloc_task_struct_node(int node)
169 : {
170 916 : return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
171 : }
172 :
173 834 : static inline void free_task_struct(struct task_struct *tsk)
174 : {
175 834 : kmem_cache_free(task_struct_cachep, tsk);
176 : }
177 : #endif
178 :
179 : #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
180 :
181 : /*
182 : * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
183 : * kmemcache based allocator.
184 : */
185 : # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
186 :
187 : #ifdef CONFIG_VMAP_STACK
188 : /*
189 : * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
190 : * flush. Try to minimize the number of calls by caching stacks.
191 : */
192 : #define NR_CACHED_STACKS 2
193 : static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
194 :
195 : static int free_vm_stack_cache(unsigned int cpu)
196 : {
197 : struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
198 : int i;
199 :
200 : for (i = 0; i < NR_CACHED_STACKS; i++) {
201 : struct vm_struct *vm_stack = cached_vm_stacks[i];
202 :
203 : if (!vm_stack)
204 : continue;
205 :
206 : vfree(vm_stack->addr);
207 : cached_vm_stacks[i] = NULL;
208 : }
209 :
210 : return 0;
211 : }
212 : #endif
213 :
214 916 : static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
215 : {
216 : #ifdef CONFIG_VMAP_STACK
217 : void *stack;
218 : int i;
219 :
220 : for (i = 0; i < NR_CACHED_STACKS; i++) {
221 : struct vm_struct *s;
222 :
223 : s = this_cpu_xchg(cached_stacks[i], NULL);
224 :
225 : if (!s)
226 : continue;
227 :
228 : /* Mark stack accessible for KASAN. */
229 : kasan_unpoison_range(s->addr, THREAD_SIZE);
230 :
231 : /* Clear stale pointers from reused stack. */
232 : memset(s->addr, 0, THREAD_SIZE);
233 :
234 : tsk->stack_vm_area = s;
235 : tsk->stack = s->addr;
236 : return s->addr;
237 : }
238 :
239 : /*
240 : * Allocated stacks are cached and later reused by new threads,
241 : * so memcg accounting is performed manually on assigning/releasing
242 : * stacks to tasks. Drop __GFP_ACCOUNT.
243 : */
244 : stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
245 : VMALLOC_START, VMALLOC_END,
246 : THREADINFO_GFP & ~__GFP_ACCOUNT,
247 : PAGE_KERNEL,
248 : 0, node, __builtin_return_address(0));
249 :
250 : /*
251 : * We can't call find_vm_area() in interrupt context, and
252 : * free_thread_stack() can be called in interrupt context,
253 : * so cache the vm_struct.
254 : */
255 : if (stack) {
256 : tsk->stack_vm_area = find_vm_area(stack);
257 : tsk->stack = stack;
258 : }
259 : return stack;
260 : #else
261 916 : struct page *page = alloc_pages_node(node, THREADINFO_GFP,
262 : THREAD_SIZE_ORDER);
263 :
264 916 : if (likely(page)) {
265 916 : tsk->stack = kasan_reset_tag(page_address(page));
266 916 : return tsk->stack;
267 : }
268 : return NULL;
269 : #endif
270 : }
271 :
272 836 : static inline void free_thread_stack(struct task_struct *tsk)
273 : {
274 : #ifdef CONFIG_VMAP_STACK
275 : struct vm_struct *vm = task_stack_vm_area(tsk);
276 :
277 : if (vm) {
278 : int i;
279 :
280 : for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
281 : memcg_kmem_uncharge_page(vm->pages[i], 0);
282 :
283 : for (i = 0; i < NR_CACHED_STACKS; i++) {
284 : if (this_cpu_cmpxchg(cached_stacks[i],
285 : NULL, tsk->stack_vm_area) != NULL)
286 : continue;
287 :
288 : return;
289 : }
290 :
291 : vfree_atomic(tsk->stack);
292 : return;
293 : }
294 : #endif
295 :
296 836 : __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
297 836 : }
298 : # else
299 : static struct kmem_cache *thread_stack_cache;
300 :
301 : static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
302 : int node)
303 : {
304 : unsigned long *stack;
305 : stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
306 : stack = kasan_reset_tag(stack);
307 : tsk->stack = stack;
308 : return stack;
309 : }
310 :
311 : static void free_thread_stack(struct task_struct *tsk)
312 : {
313 : kmem_cache_free(thread_stack_cache, tsk->stack);
314 : }
315 :
316 : void thread_stack_cache_init(void)
317 : {
318 : thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
319 : THREAD_SIZE, THREAD_SIZE, 0, 0,
320 : THREAD_SIZE, NULL);
321 : BUG_ON(thread_stack_cache == NULL);
322 : }
323 : # endif
324 : #endif
325 :
326 : /* SLAB cache for signal_struct structures (tsk->signal) */
327 : static struct kmem_cache *signal_cachep;
328 :
329 : /* SLAB cache for sighand_struct structures (tsk->sighand) */
330 : struct kmem_cache *sighand_cachep;
331 :
332 : /* SLAB cache for files_struct structures (tsk->files) */
333 : struct kmem_cache *files_cachep;
334 :
335 : /* SLAB cache for fs_struct structures (tsk->fs) */
336 : struct kmem_cache *fs_cachep;
337 :
338 : /* SLAB cache for vm_area_struct structures */
339 : static struct kmem_cache *vm_area_cachep;
340 :
341 : /* SLAB cache for mm_struct structures (tsk->mm) */
342 : static struct kmem_cache *mm_cachep;
343 :
344 23143 : struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
345 : {
346 23143 : struct vm_area_struct *vma;
347 :
348 23143 : vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
349 23146 : if (vma)
350 23146 : vma_init(vma, mm);
351 23146 : return vma;
352 : }
353 :
354 66368 : struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
355 : {
356 66368 : struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
357 :
358 66369 : if (new) {
359 66369 : ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
360 66369 : ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
361 : /*
362 : * orig->shared.rb may be modified concurrently, but the clone
363 : * will be reinitialized.
364 : */
365 66369 : *new = data_race(*orig);
366 66369 : INIT_LIST_HEAD(&new->anon_vma_chain);
367 66369 : new->vm_next = new->vm_prev = NULL;
368 : }
369 66369 : return new;
370 : }
371 :
372 86616 : void vm_area_free(struct vm_area_struct *vma)
373 : {
374 86616 : kmem_cache_free(vm_area_cachep, vma);
375 0 : }
376 :
377 1752 : static void account_kernel_stack(struct task_struct *tsk, int account)
378 : {
379 1752 : void *stack = task_stack_page(tsk);
380 1752 : struct vm_struct *vm = task_stack_vm_area(tsk);
381 :
382 :
383 : /* All stack pages are in the same node. */
384 1752 : if (vm)
385 : mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
386 : account * (THREAD_SIZE / 1024));
387 : else
388 836 : mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
389 : account * (THREAD_SIZE / 1024));
390 : }
391 :
392 916 : static int memcg_charge_kernel_stack(struct task_struct *tsk)
393 : {
394 : #ifdef CONFIG_VMAP_STACK
395 : struct vm_struct *vm = task_stack_vm_area(tsk);
396 : int ret;
397 :
398 : BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
399 :
400 : if (vm) {
401 : int i;
402 :
403 : BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
404 :
405 : for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
406 : /*
407 : * If memcg_kmem_charge_page() fails, page's
408 : * memory cgroup pointer is NULL, and
409 : * memcg_kmem_uncharge_page() in free_thread_stack()
410 : * will ignore this page.
411 : */
412 : ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
413 : 0);
414 : if (ret)
415 : return ret;
416 : }
417 : }
418 : #endif
419 916 : return 0;
420 : }
421 :
422 836 : static void release_task_stack(struct task_struct *tsk)
423 : {
424 836 : if (WARN_ON(tsk->state != TASK_DEAD))
425 : return; /* Better to leak the stack than to free prematurely */
426 :
427 836 : account_kernel_stack(tsk, -1);
428 836 : free_thread_stack(tsk);
429 836 : tsk->stack = NULL;
430 : #ifdef CONFIG_VMAP_STACK
431 : tsk->stack_vm_area = NULL;
432 : #endif
433 : }
434 :
435 : #ifdef CONFIG_THREAD_INFO_IN_TASK
436 890 : void put_task_stack(struct task_struct *tsk)
437 : {
438 890 : if (refcount_dec_and_test(&tsk->stack_refcount))
439 836 : release_task_stack(tsk);
440 890 : }
441 : #endif
442 :
443 834 : void free_task(struct task_struct *tsk)
444 : {
445 834 : scs_release(tsk);
446 :
447 : #ifndef CONFIG_THREAD_INFO_IN_TASK
448 : /*
449 : * The task is finally done with both the stack and thread_info,
450 : * so free both.
451 : */
452 : release_task_stack(tsk);
453 : #else
454 : /*
455 : * If the task had a separate stack allocation, it should be gone
456 : * by now.
457 : */
458 834 : WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
459 : #endif
460 834 : rt_mutex_debug_task_free(tsk);
461 834 : ftrace_graph_exit_task(tsk);
462 834 : arch_release_task_struct(tsk);
463 834 : if (tsk->flags & PF_KTHREAD)
464 0 : free_kthread_struct(tsk);
465 834 : free_task_struct(tsk);
466 834 : }
467 : EXPORT_SYMBOL(free_task);
468 :
469 : #ifdef CONFIG_MMU
470 858 : static __latent_entropy int dup_mmap(struct mm_struct *mm,
471 : struct mm_struct *oldmm)
472 : {
473 858 : struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
474 858 : struct rb_node **rb_link, *rb_parent;
475 858 : int retval;
476 858 : unsigned long charge;
477 858 : LIST_HEAD(uf);
478 :
479 858 : uprobe_start_dup_mmap();
480 858 : if (mmap_write_lock_killable(oldmm)) {
481 0 : retval = -EINTR;
482 0 : goto fail_uprobe_end;
483 : }
484 858 : flush_cache_dup_mm(oldmm);
485 858 : uprobe_dup_mmap(oldmm, mm);
486 : /*
487 : * Not linked in yet - no deadlock potential:
488 : */
489 858 : mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
490 :
491 : /* No ordering required: file already has been exposed. */
492 858 : RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
493 :
494 858 : mm->total_vm = oldmm->total_vm;
495 858 : mm->data_vm = oldmm->data_vm;
496 858 : mm->exec_vm = oldmm->exec_vm;
497 858 : mm->stack_vm = oldmm->stack_vm;
498 :
499 858 : rb_link = &mm->mm_rb.rb_node;
500 858 : rb_parent = NULL;
501 858 : pprev = &mm->mmap;
502 858 : retval = ksm_fork(mm, oldmm);
503 858 : if (retval)
504 0 : goto out;
505 858 : retval = khugepaged_fork(mm, oldmm);
506 858 : if (retval)
507 0 : goto out;
508 :
509 858 : prev = NULL;
510 45845 : for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
511 44987 : struct file *file;
512 :
513 44987 : if (mpnt->vm_flags & VM_DONTCOPY) {
514 0 : vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
515 0 : continue;
516 : }
517 44987 : charge = 0;
518 : /*
519 : * Don't duplicate many vmas if we've been oom-killed (for
520 : * example)
521 : */
522 44987 : if (fatal_signal_pending(current)) {
523 0 : retval = -EINTR;
524 0 : goto out;
525 : }
526 44988 : if (mpnt->vm_flags & VM_ACCOUNT) {
527 20315 : unsigned long len = vma_pages(mpnt);
528 :
529 20315 : if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
530 0 : goto fail_nomem;
531 : charge = len;
532 : }
533 44988 : tmp = vm_area_dup(mpnt);
534 44988 : if (!tmp)
535 0 : goto fail_nomem;
536 44988 : retval = vma_dup_policy(mpnt, tmp);
537 44988 : if (retval)
538 0 : goto fail_nomem_policy;
539 44988 : tmp->vm_mm = mm;
540 44988 : retval = dup_userfaultfd(tmp, &uf);
541 44988 : if (retval)
542 : goto fail_nomem_anon_vma_fork;
543 44988 : if (tmp->vm_flags & VM_WIPEONFORK) {
544 : /*
545 : * VM_WIPEONFORK gets a clean slate in the child.
546 : * Don't prepare anon_vma until fault since we don't
547 : * copy page for current vma.
548 : */
549 0 : tmp->anon_vma = NULL;
550 44988 : } else if (anon_vma_fork(tmp, mpnt))
551 0 : goto fail_nomem_anon_vma_fork;
552 44988 : tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
553 44988 : file = tmp->vm_file;
554 44988 : if (file) {
555 36915 : struct inode *inode = file_inode(file);
556 36915 : struct address_space *mapping = file->f_mapping;
557 :
558 36915 : get_file(file);
559 36915 : if (tmp->vm_flags & VM_DENYWRITE)
560 8570 : put_write_access(inode);
561 36915 : i_mmap_lock_write(mapping);
562 36914 : if (tmp->vm_flags & VM_SHARED)
563 120 : mapping_allow_writable(mapping);
564 36914 : flush_dcache_mmap_lock(mapping);
565 : /* insert tmp into the share list, just after mpnt */
566 36914 : vma_interval_tree_insert_after(tmp, mpnt,
567 : &mapping->i_mmap);
568 36915 : flush_dcache_mmap_unlock(mapping);
569 36915 : i_mmap_unlock_write(mapping);
570 : }
571 :
572 : /*
573 : * Clear hugetlb-related page reserves for children. This only
574 : * affects MAP_PRIVATE mappings. Faults generated by the child
575 : * are not guaranteed to succeed, even if read-only
576 : */
577 44988 : if (is_vm_hugetlb_page(tmp))
578 44988 : reset_vma_resv_huge_pages(tmp);
579 :
580 : /*
581 : * Link in the new vma and copy the page table entries.
582 : */
583 44988 : *pprev = tmp;
584 44988 : pprev = &tmp->vm_next;
585 44988 : tmp->vm_prev = prev;
586 44988 : prev = tmp;
587 :
588 44988 : __vma_link_rb(mm, tmp, rb_link, rb_parent);
589 44988 : rb_link = &tmp->vm_rb.rb_right;
590 44988 : rb_parent = &tmp->vm_rb;
591 :
592 44988 : mm->map_count++;
593 44988 : if (!(tmp->vm_flags & VM_WIPEONFORK))
594 44988 : retval = copy_page_range(tmp, mpnt);
595 :
596 44987 : if (tmp->vm_ops && tmp->vm_ops->open)
597 0 : tmp->vm_ops->open(tmp);
598 :
599 44987 : if (retval)
600 0 : goto out;
601 : }
602 : /* a new mm has just been created */
603 858 : retval = arch_dup_mmap(oldmm, mm);
604 858 : out:
605 858 : mmap_write_unlock(mm);
606 858 : flush_tlb_mm(oldmm);
607 858 : mmap_write_unlock(oldmm);
608 858 : dup_userfaultfd_complete(&uf);
609 858 : fail_uprobe_end:
610 858 : uprobe_end_dup_mmap();
611 858 : return retval;
612 0 : fail_nomem_anon_vma_fork:
613 0 : mpol_put(vma_policy(tmp));
614 0 : fail_nomem_policy:
615 0 : vm_area_free(tmp);
616 0 : fail_nomem:
617 0 : retval = -ENOMEM;
618 0 : vm_unacct_memory(charge);
619 0 : goto out;
620 : }
621 :
622 2019 : static inline int mm_alloc_pgd(struct mm_struct *mm)
623 : {
624 4038 : mm->pgd = pgd_alloc(mm);
625 2019 : if (unlikely(!mm->pgd))
626 0 : return -ENOMEM;
627 : return 0;
628 : }
629 :
630 1992 : static inline void mm_free_pgd(struct mm_struct *mm)
631 : {
632 1992 : pgd_free(mm, mm->pgd);
633 0 : }
634 : #else
635 : static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
636 : {
637 : mmap_write_lock(oldmm);
638 : RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
639 : mmap_write_unlock(oldmm);
640 : return 0;
641 : }
642 : #define mm_alloc_pgd(mm) (0)
643 : #define mm_free_pgd(mm)
644 : #endif /* CONFIG_MMU */
645 :
646 1992 : static void check_mm(struct mm_struct *mm)
647 : {
648 1992 : int i;
649 :
650 1992 : BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
651 : "Please make sure 'struct resident_page_types[]' is updated as well");
652 :
653 9960 : for (i = 0; i < NR_MM_COUNTERS; i++) {
654 7968 : long x = atomic_long_read(&mm->rss_stat.count[i]);
655 :
656 7968 : if (unlikely(x))
657 0 : pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
658 : mm, resident_page_types[i], x);
659 : }
660 :
661 1992 : if (mm_pgtables_bytes(mm))
662 0 : pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
663 : mm_pgtables_bytes(mm));
664 :
665 : #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
666 : VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
667 : #endif
668 1992 : }
669 :
670 : #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
671 : #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
672 :
673 : /*
674 : * Called when the last reference to the mm
675 : * is dropped: either by a lazy thread or by
676 : * mmput. Free the page directory and the mm.
677 : */
678 1992 : void __mmdrop(struct mm_struct *mm)
679 : {
680 1992 : BUG_ON(mm == &init_mm);
681 1992 : WARN_ON_ONCE(mm == current->mm);
682 1992 : WARN_ON_ONCE(mm == current->active_mm);
683 1992 : mm_free_pgd(mm);
684 1992 : destroy_context(mm);
685 1992 : mmu_notifier_subscriptions_destroy(mm);
686 1992 : check_mm(mm);
687 1992 : put_user_ns(mm->user_ns);
688 1992 : free_mm(mm);
689 1992 : }
690 : EXPORT_SYMBOL_GPL(__mmdrop);
691 :
692 0 : static void mmdrop_async_fn(struct work_struct *work)
693 : {
694 0 : struct mm_struct *mm;
695 :
696 0 : mm = container_of(work, struct mm_struct, async_put_work);
697 0 : __mmdrop(mm);
698 0 : }
699 :
700 0 : static void mmdrop_async(struct mm_struct *mm)
701 : {
702 0 : if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
703 0 : INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
704 0 : schedule_work(&mm->async_put_work);
705 : }
706 0 : }
707 :
708 832 : static inline void free_signal_struct(struct signal_struct *sig)
709 : {
710 832 : taskstats_tgid_free(sig);
711 832 : sched_autogroup_exit(sig);
712 : /*
713 : * __mmdrop is not safe to call from softirq context on x86 due to
714 : * pgd_dtor so postpone it to the async context
715 : */
716 832 : if (sig->oom_mm)
717 0 : mmdrop_async(sig->oom_mm);
718 832 : kmem_cache_free(signal_cachep, sig);
719 832 : }
720 :
721 834 : static inline void put_signal_struct(struct signal_struct *sig)
722 : {
723 834 : if (refcount_dec_and_test(&sig->sigcnt))
724 832 : free_signal_struct(sig);
725 834 : }
726 :
727 834 : void __put_task_struct(struct task_struct *tsk)
728 : {
729 834 : WARN_ON(!tsk->exit_state);
730 834 : WARN_ON(refcount_read(&tsk->usage));
731 834 : WARN_ON(tsk == current);
732 :
733 834 : io_uring_free(tsk);
734 834 : cgroup_free(tsk);
735 834 : task_numa_free(tsk, true);
736 834 : security_task_free(tsk);
737 834 : exit_creds(tsk);
738 834 : delayacct_tsk_free(tsk);
739 834 : put_signal_struct(tsk->signal);
740 :
741 834 : if (!profile_handoff_task(tsk))
742 834 : free_task(tsk);
743 834 : }
744 : EXPORT_SYMBOL_GPL(__put_task_struct);
745 :
746 1 : void __init __weak arch_task_cache_init(void) { }
747 :
748 : /*
749 : * set_max_threads
750 : */
751 1 : static void set_max_threads(unsigned int max_threads_suggested)
752 : {
753 1 : u64 threads;
754 1 : unsigned long nr_pages = totalram_pages();
755 :
756 : /*
757 : * The number of threads shall be limited such that the thread
758 : * structures may only consume a small part of the available memory.
759 : */
760 1 : if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
761 : threads = MAX_THREADS;
762 : else
763 1 : threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
764 : (u64) THREAD_SIZE * 8UL);
765 :
766 1 : if (threads > max_threads_suggested)
767 : threads = max_threads_suggested;
768 :
769 1 : max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
770 1 : }
771 :
772 : #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
773 : /* Initialized by the architecture: */
774 : int arch_task_struct_size __read_mostly;
775 : #endif
776 :
777 : #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
778 1 : static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
779 : {
780 : /* Fetch thread_struct whitelist for the architecture. */
781 1 : arch_thread_struct_whitelist(offset, size);
782 :
783 : /*
784 : * Handle zero-sized whitelist or empty thread_struct, otherwise
785 : * adjust offset to position of thread_struct in task_struct.
786 : */
787 1 : if (unlikely(*size == 0))
788 : *offset = 0;
789 : else
790 1 : *offset += offsetof(struct task_struct, thread);
791 : }
792 : #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
793 :
794 1 : void __init fork_init(void)
795 : {
796 1 : int i;
797 : #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
798 : #ifndef ARCH_MIN_TASKALIGN
799 : #define ARCH_MIN_TASKALIGN 0
800 : #endif
801 1 : int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
802 1 : unsigned long useroffset, usersize;
803 :
804 : /* create a slab on which task_structs can be allocated */
805 1 : task_struct_whitelist(&useroffset, &usersize);
806 1 : task_struct_cachep = kmem_cache_create_usercopy("task_struct",
807 : arch_task_struct_size, align,
808 : SLAB_PANIC|SLAB_ACCOUNT,
809 : useroffset, usersize, NULL);
810 : #endif
811 :
812 : /* do the arch specific task caches init */
813 1 : arch_task_cache_init();
814 :
815 1 : set_max_threads(MAX_THREADS);
816 :
817 1 : init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
818 1 : init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
819 1 : init_task.signal->rlim[RLIMIT_SIGPENDING] =
820 1 : init_task.signal->rlim[RLIMIT_NPROC];
821 :
822 11 : for (i = 0; i < UCOUNT_COUNTS; i++)
823 10 : init_user_ns.ucount_max[i] = max_threads/2;
824 :
825 : #ifdef CONFIG_VMAP_STACK
826 : cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
827 : NULL, free_vm_stack_cache);
828 : #endif
829 :
830 1 : scs_init();
831 :
832 1 : lockdep_init_task(&init_task);
833 1 : uprobes_init();
834 1 : }
835 :
836 0 : int __weak arch_dup_task_struct(struct task_struct *dst,
837 : struct task_struct *src)
838 : {
839 0 : *dst = *src;
840 0 : return 0;
841 : }
842 :
843 917 : void set_task_stack_end_magic(struct task_struct *tsk)
844 : {
845 917 : unsigned long *stackend;
846 :
847 917 : stackend = end_of_stack(tsk);
848 917 : *stackend = STACK_END_MAGIC; /* for overflow detection */
849 1 : }
850 :
851 916 : static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
852 : {
853 916 : struct task_struct *tsk;
854 916 : unsigned long *stack;
855 916 : struct vm_struct *stack_vm_area __maybe_unused;
856 916 : int err;
857 :
858 916 : if (node == NUMA_NO_NODE)
859 913 : node = tsk_fork_get_node(orig);
860 916 : tsk = alloc_task_struct_node(node);
861 916 : if (!tsk)
862 : return NULL;
863 :
864 916 : stack = alloc_thread_stack_node(tsk, node);
865 916 : if (!stack)
866 0 : goto free_tsk;
867 :
868 916 : if (memcg_charge_kernel_stack(tsk))
869 : goto free_stack;
870 :
871 916 : stack_vm_area = task_stack_vm_area(tsk);
872 :
873 916 : err = arch_dup_task_struct(tsk, orig);
874 :
875 : /*
876 : * arch_dup_task_struct() clobbers the stack-related fields. Make
877 : * sure they're properly initialized before using any stack-related
878 : * functions again.
879 : */
880 916 : tsk->stack = stack;
881 : #ifdef CONFIG_VMAP_STACK
882 : tsk->stack_vm_area = stack_vm_area;
883 : #endif
884 : #ifdef CONFIG_THREAD_INFO_IN_TASK
885 916 : refcount_set(&tsk->stack_refcount, 1);
886 : #endif
887 :
888 916 : if (err)
889 0 : goto free_stack;
890 :
891 916 : err = scs_prepare(tsk, node);
892 916 : if (err)
893 : goto free_stack;
894 :
895 : #ifdef CONFIG_SECCOMP
896 : /*
897 : * We must handle setting up seccomp filters once we're under
898 : * the sighand lock in case orig has changed between now and
899 : * then. Until then, filter must be NULL to avoid messing up
900 : * the usage counts on the error path calling free_task.
901 : */
902 : tsk->seccomp.filter = NULL;
903 : #endif
904 :
905 916 : setup_thread_stack(tsk, orig);
906 916 : clear_user_return_notifier(tsk);
907 916 : clear_tsk_need_resched(tsk);
908 916 : set_task_stack_end_magic(tsk);
909 916 : clear_syscall_work_syscall_user_dispatch(tsk);
910 :
911 : #ifdef CONFIG_STACKPROTECTOR
912 : tsk->stack_canary = get_random_canary();
913 : #endif
914 916 : if (orig->cpus_ptr == &orig->cpus_mask)
915 916 : tsk->cpus_ptr = &tsk->cpus_mask;
916 :
917 : /*
918 : * One for the user space visible state that goes away when reaped.
919 : * One for the scheduler.
920 : */
921 916 : refcount_set(&tsk->rcu_users, 2);
922 : /* One for the rcu users */
923 916 : refcount_set(&tsk->usage, 1);
924 : #ifdef CONFIG_BLK_DEV_IO_TRACE
925 : tsk->btrace_seq = 0;
926 : #endif
927 916 : tsk->splice_pipe = NULL;
928 916 : tsk->task_frag.page = NULL;
929 916 : tsk->wake_q.next = NULL;
930 :
931 916 : account_kernel_stack(tsk, 1);
932 :
933 916 : kcov_task_init(tsk);
934 916 : 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_queue = NULL;
942 : tsk->use_memdelay = 0;
943 : #endif
944 :
945 : #ifdef CONFIG_MEMCG
946 : tsk->active_memcg = NULL;
947 : #endif
948 916 : return tsk;
949 :
950 0 : free_stack:
951 0 : free_thread_stack(tsk);
952 0 : free_tsk:
953 0 : free_task_struct(tsk);
954 0 : return NULL;
955 : }
956 :
957 : __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
958 :
959 : static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
960 :
961 0 : static int __init coredump_filter_setup(char *s)
962 : {
963 0 : default_dump_filter =
964 0 : (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
965 : MMF_DUMP_FILTER_MASK;
966 0 : return 1;
967 : }
968 :
969 : __setup("coredump_filter=", coredump_filter_setup);
970 :
971 : #include <linux/init_task.h>
972 :
973 2019 : static void mm_init_aio(struct mm_struct *mm)
974 : {
975 : #ifdef CONFIG_AIO
976 : spin_lock_init(&mm->ioctx_lock);
977 : mm->ioctx_table = NULL;
978 : #endif
979 2019 : }
980 :
981 0 : static __always_inline void mm_clear_owner(struct mm_struct *mm,
982 : struct task_struct *p)
983 : {
984 : #ifdef CONFIG_MEMCG
985 : if (mm->owner == p)
986 : WRITE_ONCE(mm->owner, NULL);
987 : #endif
988 0 : }
989 :
990 2019 : static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
991 : {
992 : #ifdef CONFIG_MEMCG
993 : mm->owner = p;
994 : #endif
995 2019 : }
996 :
997 2019 : static void mm_init_pasid(struct mm_struct *mm)
998 : {
999 : #ifdef CONFIG_IOMMU_SUPPORT
1000 : mm->pasid = INIT_PASID;
1001 : #endif
1002 2019 : }
1003 :
1004 2019 : static void mm_init_uprobes_state(struct mm_struct *mm)
1005 : {
1006 : #ifdef CONFIG_UPROBES
1007 : mm->uprobes_state.xol_area = NULL;
1008 : #endif
1009 2019 : }
1010 :
1011 2019 : static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1012 : struct user_namespace *user_ns)
1013 : {
1014 2019 : mm->mmap = NULL;
1015 2019 : mm->mm_rb = RB_ROOT;
1016 2019 : mm->vmacache_seqnum = 0;
1017 2019 : atomic_set(&mm->mm_users, 1);
1018 2019 : atomic_set(&mm->mm_count, 1);
1019 2019 : seqcount_init(&mm->write_protect_seq);
1020 2019 : mmap_init_lock(mm);
1021 2019 : INIT_LIST_HEAD(&mm->mmlist);
1022 2019 : mm->core_state = NULL;
1023 2019 : mm_pgtables_bytes_init(mm);
1024 2019 : mm->map_count = 0;
1025 2019 : mm->locked_vm = 0;
1026 2019 : atomic_set(&mm->has_pinned, 0);
1027 2019 : atomic64_set(&mm->pinned_vm, 0);
1028 2019 : memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1029 2019 : spin_lock_init(&mm->page_table_lock);
1030 2019 : spin_lock_init(&mm->arg_lock);
1031 2019 : mm_init_cpumask(mm);
1032 2019 : mm_init_aio(mm);
1033 2019 : mm_init_owner(mm, p);
1034 2019 : mm_init_pasid(mm);
1035 2019 : RCU_INIT_POINTER(mm->exe_file, NULL);
1036 2019 : mmu_notifier_subscriptions_init(mm);
1037 2019 : init_tlb_flush_pending(mm);
1038 : #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1039 : mm->pmd_huge_pte = NULL;
1040 : #endif
1041 2019 : mm_init_uprobes_state(mm);
1042 :
1043 2019 : if (current->mm) {
1044 2017 : mm->flags = current->mm->flags & MMF_INIT_MASK;
1045 2017 : mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1046 : } else {
1047 2 : mm->flags = default_dump_filter;
1048 2 : mm->def_flags = 0;
1049 : }
1050 :
1051 2019 : if (mm_alloc_pgd(mm))
1052 0 : goto fail_nopgd;
1053 :
1054 2019 : if (init_new_context(p, mm))
1055 0 : goto fail_nocontext;
1056 :
1057 2019 : mm->user_ns = get_user_ns(user_ns);
1058 2019 : return mm;
1059 :
1060 0 : fail_nocontext:
1061 0 : mm_free_pgd(mm);
1062 0 : fail_nopgd:
1063 0 : free_mm(mm);
1064 0 : return NULL;
1065 : }
1066 :
1067 : /*
1068 : * Allocate and initialize an mm_struct.
1069 : */
1070 1161 : struct mm_struct *mm_alloc(void)
1071 : {
1072 1161 : struct mm_struct *mm;
1073 :
1074 1161 : mm = allocate_mm();
1075 1161 : if (!mm)
1076 : return NULL;
1077 :
1078 1161 : memset(mm, 0, sizeof(*mm));
1079 1161 : return mm_init(mm, current, current_user_ns());
1080 : }
1081 :
1082 1994 : static inline void __mmput(struct mm_struct *mm)
1083 : {
1084 1994 : VM_BUG_ON(atomic_read(&mm->mm_users));
1085 :
1086 1994 : uprobe_clear_state(mm);
1087 1994 : exit_aio(mm);
1088 1994 : ksm_exit(mm);
1089 1994 : khugepaged_exit(mm); /* must run before exit_mmap */
1090 1994 : exit_mmap(mm);
1091 1994 : mm_put_huge_zero_page(mm);
1092 1994 : set_mm_exe_file(mm, NULL);
1093 1994 : if (!list_empty(&mm->mmlist)) {
1094 0 : spin_lock(&mmlist_lock);
1095 0 : list_del(&mm->mmlist);
1096 0 : spin_unlock(&mmlist_lock);
1097 : }
1098 1994 : if (mm->binfmt)
1099 1994 : module_put(mm->binfmt->module);
1100 1994 : mmdrop(mm);
1101 1994 : }
1102 :
1103 : /*
1104 : * Decrement the use count and release all resources for an mm.
1105 : */
1106 2402 : void mmput(struct mm_struct *mm)
1107 : {
1108 2402 : might_sleep();
1109 :
1110 4804 : if (atomic_dec_and_test(&mm->mm_users))
1111 1994 : __mmput(mm);
1112 2402 : }
1113 : EXPORT_SYMBOL_GPL(mmput);
1114 :
1115 : #ifdef CONFIG_MMU
1116 0 : static void mmput_async_fn(struct work_struct *work)
1117 : {
1118 0 : struct mm_struct *mm = container_of(work, struct mm_struct,
1119 : async_put_work);
1120 :
1121 0 : __mmput(mm);
1122 0 : }
1123 :
1124 0 : void mmput_async(struct mm_struct *mm)
1125 : {
1126 0 : if (atomic_dec_and_test(&mm->mm_users)) {
1127 0 : INIT_WORK(&mm->async_put_work, mmput_async_fn);
1128 0 : schedule_work(&mm->async_put_work);
1129 : }
1130 0 : }
1131 : #endif
1132 :
1133 : /**
1134 : * set_mm_exe_file - change a reference to the mm's executable file
1135 : *
1136 : * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1137 : *
1138 : * Main users are mmput() and sys_execve(). Callers prevent concurrent
1139 : * invocations: in mmput() nobody alive left, in execve task is single
1140 : * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1141 : * mm->exe_file, but does so without using set_mm_exe_file() in order
1142 : * to do avoid the need for any locks.
1143 : */
1144 2545 : void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1145 : {
1146 2545 : struct file *old_exe_file;
1147 :
1148 : /*
1149 : * It is safe to dereference the exe_file without RCU as
1150 : * this function is only called if nobody else can access
1151 : * this mm -- see comment above for justification.
1152 : */
1153 2545 : old_exe_file = rcu_dereference_raw(mm->exe_file);
1154 :
1155 2545 : if (new_exe_file)
1156 551 : get_file(new_exe_file);
1157 2545 : rcu_assign_pointer(mm->exe_file, new_exe_file);
1158 2545 : if (old_exe_file)
1159 1384 : fput(old_exe_file);
1160 2545 : }
1161 :
1162 : /**
1163 : * get_mm_exe_file - acquire a reference to the mm's executable file
1164 : *
1165 : * Returns %NULL if mm has no associated executable file.
1166 : * User must release file via fput().
1167 : */
1168 904 : struct file *get_mm_exe_file(struct mm_struct *mm)
1169 : {
1170 904 : struct file *exe_file;
1171 :
1172 904 : rcu_read_lock();
1173 904 : exe_file = rcu_dereference(mm->exe_file);
1174 1807 : if (exe_file && !get_file_rcu(exe_file))
1175 0 : exe_file = NULL;
1176 904 : rcu_read_unlock();
1177 904 : return exe_file;
1178 : }
1179 : EXPORT_SYMBOL(get_mm_exe_file);
1180 :
1181 : /**
1182 : * get_task_exe_file - acquire a reference to the task's executable file
1183 : *
1184 : * Returns %NULL if task's mm (if any) has no associated executable file or
1185 : * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1186 : * User must release file via fput().
1187 : */
1188 46 : struct file *get_task_exe_file(struct task_struct *task)
1189 : {
1190 46 : struct file *exe_file = NULL;
1191 46 : struct mm_struct *mm;
1192 :
1193 46 : task_lock(task);
1194 46 : mm = task->mm;
1195 46 : if (mm) {
1196 46 : if (!(task->flags & PF_KTHREAD))
1197 46 : exe_file = get_mm_exe_file(mm);
1198 : }
1199 46 : task_unlock(task);
1200 46 : return exe_file;
1201 : }
1202 : EXPORT_SYMBOL(get_task_exe_file);
1203 :
1204 : /**
1205 : * get_task_mm - acquire a reference to the task's mm
1206 : *
1207 : * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1208 : * this kernel workthread has transiently adopted a user mm with use_mm,
1209 : * to do its AIO) is not set and if so returns a reference to it, after
1210 : * bumping up the use count. User must release the mm via mmput()
1211 : * after use. Typically used by /proc and ptrace.
1212 : */
1213 362 : struct mm_struct *get_task_mm(struct task_struct *task)
1214 : {
1215 362 : struct mm_struct *mm;
1216 :
1217 362 : task_lock(task);
1218 362 : mm = task->mm;
1219 362 : if (mm) {
1220 316 : if (task->flags & PF_KTHREAD)
1221 : mm = NULL;
1222 : else
1223 316 : mmget(mm);
1224 : }
1225 362 : task_unlock(task);
1226 362 : return mm;
1227 : }
1228 : EXPORT_SYMBOL_GPL(get_task_mm);
1229 :
1230 62 : struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1231 : {
1232 62 : struct mm_struct *mm;
1233 62 : int err;
1234 :
1235 62 : err = down_read_killable(&task->signal->exec_update_lock);
1236 62 : if (err)
1237 0 : return ERR_PTR(err);
1238 :
1239 62 : mm = get_task_mm(task);
1240 124 : if (mm && mm != current->mm &&
1241 62 : !ptrace_may_access(task, mode)) {
1242 9 : mmput(mm);
1243 9 : mm = ERR_PTR(-EACCES);
1244 : }
1245 62 : up_read(&task->signal->exec_update_lock);
1246 :
1247 62 : return mm;
1248 : }
1249 :
1250 0 : static void complete_vfork_done(struct task_struct *tsk)
1251 : {
1252 0 : struct completion *vfork;
1253 :
1254 0 : task_lock(tsk);
1255 0 : vfork = tsk->vfork_done;
1256 0 : if (likely(vfork)) {
1257 0 : tsk->vfork_done = NULL;
1258 0 : complete(vfork);
1259 : }
1260 0 : task_unlock(tsk);
1261 0 : }
1262 :
1263 0 : static int wait_for_vfork_done(struct task_struct *child,
1264 : struct completion *vfork)
1265 : {
1266 0 : int killed;
1267 :
1268 0 : freezer_do_not_count();
1269 0 : cgroup_enter_frozen();
1270 0 : killed = wait_for_completion_killable(vfork);
1271 0 : cgroup_leave_frozen(false);
1272 0 : freezer_count();
1273 :
1274 0 : if (killed) {
1275 0 : task_lock(child);
1276 0 : child->vfork_done = NULL;
1277 0 : task_unlock(child);
1278 : }
1279 :
1280 0 : put_task_struct(child);
1281 0 : return killed;
1282 : }
1283 :
1284 : /* Please note the differences between mmput and mm_release.
1285 : * mmput is called whenever we stop holding onto a mm_struct,
1286 : * error success whatever.
1287 : *
1288 : * mm_release is called after a mm_struct has been removed
1289 : * from the current process.
1290 : *
1291 : * This difference is important for error handling, when we
1292 : * only half set up a mm_struct for a new process and need to restore
1293 : * the old one. Because we mmput the new mm_struct before
1294 : * restoring the old one. . .
1295 : * Eric Biederman 10 January 1998
1296 : */
1297 1387 : static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1298 : {
1299 1387 : uprobe_free_utask(tsk);
1300 :
1301 : /* Get rid of any cached register state */
1302 1387 : deactivate_mm(tsk, mm);
1303 :
1304 : /*
1305 : * Signal userspace if we're not exiting with a core dump
1306 : * because we want to leave the value intact for debugging
1307 : * purposes.
1308 : */
1309 1387 : if (tsk->clear_child_tid) {
1310 1159 : if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1311 1159 : atomic_read(&mm->mm_users) > 1) {
1312 : /*
1313 : * We don't check the error code - if userspace has
1314 : * not set up a proper pointer then tough luck.
1315 : */
1316 2 : put_user(0, tsk->clear_child_tid);
1317 2 : do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1318 : 1, NULL, NULL, 0, 0);
1319 : }
1320 1159 : tsk->clear_child_tid = NULL;
1321 : }
1322 :
1323 : /*
1324 : * All done, finally we can wake up parent and return this mm to him.
1325 : * Also kthread_stop() uses this completion for synchronization.
1326 : */
1327 1387 : if (tsk->vfork_done)
1328 0 : complete_vfork_done(tsk);
1329 1387 : }
1330 :
1331 836 : void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1332 : {
1333 836 : futex_exit_release(tsk);
1334 836 : mm_release(tsk, mm);
1335 836 : }
1336 :
1337 551 : void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1338 : {
1339 551 : futex_exec_release(tsk);
1340 551 : mm_release(tsk, mm);
1341 551 : }
1342 :
1343 : /**
1344 : * dup_mm() - duplicates an existing mm structure
1345 : * @tsk: the task_struct with which the new mm will be associated.
1346 : * @oldmm: the mm to duplicate.
1347 : *
1348 : * Allocates a new mm structure and duplicates the provided @oldmm structure
1349 : * content into it.
1350 : *
1351 : * Return: the duplicated mm or NULL on failure.
1352 : */
1353 858 : static struct mm_struct *dup_mm(struct task_struct *tsk,
1354 : struct mm_struct *oldmm)
1355 : {
1356 858 : struct mm_struct *mm;
1357 858 : int err;
1358 :
1359 858 : mm = allocate_mm();
1360 858 : if (!mm)
1361 0 : goto fail_nomem;
1362 :
1363 858 : memcpy(mm, oldmm, sizeof(*mm));
1364 :
1365 858 : if (!mm_init(mm, tsk, mm->user_ns))
1366 0 : goto fail_nomem;
1367 :
1368 858 : err = dup_mmap(mm, oldmm);
1369 858 : if (err)
1370 0 : goto free_pt;
1371 :
1372 858 : mm->hiwater_rss = get_mm_rss(mm);
1373 858 : mm->hiwater_vm = mm->total_vm;
1374 :
1375 858 : if (mm->binfmt && !try_module_get(mm->binfmt->module))
1376 : goto free_pt;
1377 :
1378 : return mm;
1379 :
1380 0 : free_pt:
1381 : /* don't put binfmt in mmput, we haven't got module yet */
1382 0 : mm->binfmt = NULL;
1383 0 : mm_init_owner(mm, NULL);
1384 0 : mmput(mm);
1385 :
1386 : fail_nomem:
1387 : return NULL;
1388 : }
1389 :
1390 916 : static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1391 : {
1392 916 : struct mm_struct *mm, *oldmm;
1393 916 : int retval;
1394 :
1395 916 : tsk->min_flt = tsk->maj_flt = 0;
1396 916 : tsk->nvcsw = tsk->nivcsw = 0;
1397 : #ifdef CONFIG_DETECT_HUNG_TASK
1398 : tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1399 : tsk->last_switch_time = 0;
1400 : #endif
1401 :
1402 916 : tsk->mm = NULL;
1403 916 : tsk->active_mm = NULL;
1404 :
1405 : /*
1406 : * Are we cloning a kernel thread?
1407 : *
1408 : * We need to steal a active VM for that..
1409 : */
1410 916 : oldmm = current->mm;
1411 916 : if (!oldmm)
1412 : return 0;
1413 :
1414 : /* initialize the new vmacache entries */
1415 863 : vmacache_flush(tsk);
1416 :
1417 863 : if (clone_flags & CLONE_VM) {
1418 6 : mmget(oldmm);
1419 6 : mm = oldmm;
1420 6 : goto good_mm;
1421 : }
1422 :
1423 857 : retval = -ENOMEM;
1424 857 : mm = dup_mm(tsk, current->mm);
1425 857 : if (!mm)
1426 0 : goto fail_nomem;
1427 :
1428 857 : good_mm:
1429 863 : tsk->mm = mm;
1430 863 : tsk->active_mm = mm;
1431 863 : return 0;
1432 :
1433 0 : fail_nomem:
1434 0 : return retval;
1435 : }
1436 :
1437 916 : static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1438 : {
1439 916 : struct fs_struct *fs = current->fs;
1440 916 : if (clone_flags & CLONE_FS) {
1441 : /* tsk->fs is already what we want */
1442 56 : spin_lock(&fs->lock);
1443 56 : if (fs->in_exec) {
1444 0 : spin_unlock(&fs->lock);
1445 0 : return -EAGAIN;
1446 : }
1447 56 : fs->users++;
1448 56 : spin_unlock(&fs->lock);
1449 56 : return 0;
1450 : }
1451 860 : tsk->fs = copy_fs_struct(fs);
1452 860 : if (!tsk->fs)
1453 0 : return -ENOMEM;
1454 : return 0;
1455 : }
1456 :
1457 916 : static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1458 : {
1459 916 : struct files_struct *oldf, *newf;
1460 916 : int error = 0;
1461 :
1462 : /*
1463 : * A background process may not have any files ...
1464 : */
1465 916 : oldf = current->files;
1466 916 : if (!oldf)
1467 0 : goto out;
1468 :
1469 916 : if (clone_flags & CLONE_FILES) {
1470 55 : atomic_inc(&oldf->count);
1471 55 : goto out;
1472 : }
1473 :
1474 861 : newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1475 861 : if (!newf)
1476 0 : goto out;
1477 :
1478 861 : tsk->files = newf;
1479 861 : error = 0;
1480 916 : out:
1481 916 : return error;
1482 : }
1483 :
1484 916 : static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1485 : {
1486 : #ifdef CONFIG_BLOCK
1487 916 : struct io_context *ioc = current->io_context;
1488 916 : struct io_context *new_ioc;
1489 :
1490 916 : if (!ioc)
1491 : return 0;
1492 : /*
1493 : * Share io context with parent, if CLONE_IO is set
1494 : */
1495 232 : if (clone_flags & CLONE_IO) {
1496 0 : ioc_task_link(ioc);
1497 0 : tsk->io_context = ioc;
1498 232 : } else if (ioprio_valid(ioc->ioprio)) {
1499 17 : new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1500 17 : if (unlikely(!new_ioc))
1501 : return -ENOMEM;
1502 :
1503 17 : new_ioc->ioprio = ioc->ioprio;
1504 17 : put_io_context(new_ioc);
1505 : }
1506 : #endif
1507 : return 0;
1508 : }
1509 :
1510 916 : static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1511 : {
1512 916 : struct sighand_struct *sig;
1513 :
1514 916 : if (clone_flags & CLONE_SIGHAND) {
1515 6 : refcount_inc(¤t->sighand->count);
1516 6 : return 0;
1517 : }
1518 910 : sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1519 910 : RCU_INIT_POINTER(tsk->sighand, sig);
1520 910 : if (!sig)
1521 : return -ENOMEM;
1522 :
1523 910 : refcount_set(&sig->count, 1);
1524 910 : spin_lock_irq(¤t->sighand->siglock);
1525 910 : memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1526 910 : spin_unlock_irq(¤t->sighand->siglock);
1527 :
1528 : /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1529 910 : if (clone_flags & CLONE_CLEAR_SIGHAND)
1530 0 : flush_signal_handlers(tsk, 0);
1531 :
1532 : return 0;
1533 : }
1534 :
1535 836 : void __cleanup_sighand(struct sighand_struct *sighand)
1536 : {
1537 836 : if (refcount_dec_and_test(&sighand->count)) {
1538 834 : signalfd_cleanup(sighand);
1539 : /*
1540 : * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1541 : * without an RCU grace period, see __lock_task_sighand().
1542 : */
1543 834 : kmem_cache_free(sighand_cachep, sighand);
1544 : }
1545 836 : }
1546 :
1547 : /*
1548 : * Initialize POSIX timer handling for a thread group.
1549 : */
1550 910 : static void posix_cpu_timers_init_group(struct signal_struct *sig)
1551 : {
1552 910 : struct posix_cputimers *pct = &sig->posix_cputimers;
1553 910 : unsigned long cpu_limit;
1554 :
1555 910 : cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1556 910 : posix_cputimers_group_init(pct, cpu_limit);
1557 : }
1558 :
1559 916 : static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1560 : {
1561 916 : struct signal_struct *sig;
1562 :
1563 916 : if (clone_flags & CLONE_THREAD)
1564 : return 0;
1565 :
1566 910 : sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1567 910 : tsk->signal = sig;
1568 910 : if (!sig)
1569 : return -ENOMEM;
1570 :
1571 910 : sig->nr_threads = 1;
1572 910 : atomic_set(&sig->live, 1);
1573 910 : refcount_set(&sig->sigcnt, 1);
1574 :
1575 : /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1576 910 : sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1577 910 : tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1578 :
1579 910 : init_waitqueue_head(&sig->wait_chldexit);
1580 910 : sig->curr_target = tsk;
1581 910 : init_sigpending(&sig->shared_pending);
1582 910 : INIT_HLIST_HEAD(&sig->multiprocess);
1583 910 : seqlock_init(&sig->stats_lock);
1584 910 : prev_cputime_init(&sig->prev_cputime);
1585 :
1586 : #ifdef CONFIG_POSIX_TIMERS
1587 910 : INIT_LIST_HEAD(&sig->posix_timers);
1588 910 : hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1589 910 : sig->real_timer.function = it_real_fn;
1590 : #endif
1591 :
1592 910 : task_lock(current->group_leader);
1593 910 : memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1594 910 : task_unlock(current->group_leader);
1595 :
1596 910 : posix_cpu_timers_init_group(sig);
1597 :
1598 910 : tty_audit_fork(sig);
1599 910 : sched_autogroup_fork(sig);
1600 :
1601 910 : sig->oom_score_adj = current->signal->oom_score_adj;
1602 910 : sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1603 :
1604 910 : mutex_init(&sig->cred_guard_mutex);
1605 910 : init_rwsem(&sig->exec_update_lock);
1606 :
1607 910 : return 0;
1608 : }
1609 :
1610 916 : static void copy_seccomp(struct task_struct *p)
1611 : {
1612 : #ifdef CONFIG_SECCOMP
1613 : /*
1614 : * Must be called with sighand->lock held, which is common to
1615 : * all threads in the group. Holding cred_guard_mutex is not
1616 : * needed because this new task is not yet running and cannot
1617 : * be racing exec.
1618 : */
1619 : assert_spin_locked(¤t->sighand->siglock);
1620 :
1621 : /* Ref-count the new filter user, and assign it. */
1622 : get_seccomp_filter(current);
1623 : p->seccomp = current->seccomp;
1624 :
1625 : /*
1626 : * Explicitly enable no_new_privs here in case it got set
1627 : * between the task_struct being duplicated and holding the
1628 : * sighand lock. The seccomp state and nnp must be in sync.
1629 : */
1630 : if (task_no_new_privs(current))
1631 : task_set_no_new_privs(p);
1632 :
1633 : /*
1634 : * If the parent gained a seccomp mode after copying thread
1635 : * flags and between before we held the sighand lock, we have
1636 : * to manually enable the seccomp thread flag here.
1637 : */
1638 : if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1639 : set_task_syscall_work(p, SECCOMP);
1640 : #endif
1641 916 : }
1642 :
1643 644 : SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1644 : {
1645 322 : current->clear_child_tid = tidptr;
1646 :
1647 322 : return task_pid_vnr(current);
1648 : }
1649 :
1650 916 : static void rt_mutex_init_task(struct task_struct *p)
1651 : {
1652 916 : raw_spin_lock_init(&p->pi_lock);
1653 : #ifdef CONFIG_RT_MUTEXES
1654 916 : p->pi_waiters = RB_ROOT_CACHED;
1655 916 : p->pi_top_task = NULL;
1656 916 : p->pi_blocked_on = NULL;
1657 : #endif
1658 916 : }
1659 :
1660 : static inline void init_task_pid_links(struct task_struct *task)
1661 : {
1662 : enum pid_type type;
1663 :
1664 3664 : for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1665 3664 : INIT_HLIST_NODE(&task->pid_links[type]);
1666 : }
1667 :
1668 : static inline void
1669 1832 : init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1670 : {
1671 1832 : if (type == PIDTYPE_PID)
1672 3 : task->thread_pid = pid;
1673 : else
1674 9 : task->signal->pids[type] = pid;
1675 : }
1676 :
1677 916 : static inline void rcu_copy_process(struct task_struct *p)
1678 : {
1679 : #ifdef CONFIG_PREEMPT_RCU
1680 : p->rcu_read_lock_nesting = 0;
1681 : p->rcu_read_unlock_special.s = 0;
1682 : p->rcu_blocked_node = NULL;
1683 : INIT_LIST_HEAD(&p->rcu_node_entry);
1684 : #endif /* #ifdef CONFIG_PREEMPT_RCU */
1685 : #ifdef CONFIG_TASKS_RCU
1686 : p->rcu_tasks_holdout = false;
1687 : INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1688 : p->rcu_tasks_idle_cpu = -1;
1689 : #endif /* #ifdef CONFIG_TASKS_RCU */
1690 : #ifdef CONFIG_TASKS_TRACE_RCU
1691 : p->trc_reader_nesting = 0;
1692 : p->trc_reader_special.s = 0;
1693 : INIT_LIST_HEAD(&p->trc_holdout_list);
1694 : #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1695 916 : }
1696 :
1697 0 : struct pid *pidfd_pid(const struct file *file)
1698 : {
1699 0 : if (file->f_op == &pidfd_fops)
1700 0 : return file->private_data;
1701 :
1702 0 : return ERR_PTR(-EBADF);
1703 : }
1704 :
1705 0 : static int pidfd_release(struct inode *inode, struct file *file)
1706 : {
1707 0 : struct pid *pid = file->private_data;
1708 :
1709 0 : file->private_data = NULL;
1710 0 : put_pid(pid);
1711 0 : return 0;
1712 : }
1713 :
1714 : #ifdef CONFIG_PROC_FS
1715 : /**
1716 : * pidfd_show_fdinfo - print information about a pidfd
1717 : * @m: proc fdinfo file
1718 : * @f: file referencing a pidfd
1719 : *
1720 : * Pid:
1721 : * This function will print the pid that a given pidfd refers to in the
1722 : * pid namespace of the procfs instance.
1723 : * If the pid namespace of the process is not a descendant of the pid
1724 : * namespace of the procfs instance 0 will be shown as its pid. This is
1725 : * similar to calling getppid() on a process whose parent is outside of
1726 : * its pid namespace.
1727 : *
1728 : * NSpid:
1729 : * If pid namespaces are supported then this function will also print
1730 : * the pid of a given pidfd refers to for all descendant pid namespaces
1731 : * starting from the current pid namespace of the instance, i.e. the
1732 : * Pid field and the first entry in the NSpid field will be identical.
1733 : * If the pid namespace of the process is not a descendant of the pid
1734 : * namespace of the procfs instance 0 will be shown as its first NSpid
1735 : * entry and no others will be shown.
1736 : * Note that this differs from the Pid and NSpid fields in
1737 : * /proc/<pid>/status where Pid and NSpid are always shown relative to
1738 : * the pid namespace of the procfs instance. The difference becomes
1739 : * obvious when sending around a pidfd between pid namespaces from a
1740 : * different branch of the tree, i.e. where no ancestoral relation is
1741 : * present between the pid namespaces:
1742 : * - create two new pid namespaces ns1 and ns2 in the initial pid
1743 : * namespace (also take care to create new mount namespaces in the
1744 : * new pid namespace and mount procfs)
1745 : * - create a process with a pidfd in ns1
1746 : * - send pidfd from ns1 to ns2
1747 : * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1748 : * have exactly one entry, which is 0
1749 : */
1750 0 : static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1751 : {
1752 0 : struct pid *pid = f->private_data;
1753 0 : struct pid_namespace *ns;
1754 0 : pid_t nr = -1;
1755 :
1756 0 : if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1757 0 : ns = proc_pid_ns(file_inode(m->file)->i_sb);
1758 0 : nr = pid_nr_ns(pid, ns);
1759 : }
1760 :
1761 0 : seq_put_decimal_ll(m, "Pid:\t", nr);
1762 :
1763 : #ifdef CONFIG_PID_NS
1764 : seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1765 : if (nr > 0) {
1766 : int i;
1767 :
1768 : /* If nr is non-zero it means that 'pid' is valid and that
1769 : * ns, i.e. the pid namespace associated with the procfs
1770 : * instance, is in the pid namespace hierarchy of pid.
1771 : * Start at one below the already printed level.
1772 : */
1773 : for (i = ns->level + 1; i <= pid->level; i++)
1774 : seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1775 : }
1776 : #endif
1777 0 : seq_putc(m, '\n');
1778 0 : }
1779 : #endif
1780 :
1781 : /*
1782 : * Poll support for process exit notification.
1783 : */
1784 0 : static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1785 : {
1786 0 : struct pid *pid = file->private_data;
1787 0 : __poll_t poll_flags = 0;
1788 :
1789 0 : poll_wait(file, &pid->wait_pidfd, pts);
1790 :
1791 : /*
1792 : * Inform pollers only when the whole thread group exits.
1793 : * If the thread group leader exits before all other threads in the
1794 : * group, then poll(2) should block, similar to the wait(2) family.
1795 : */
1796 0 : if (thread_group_exited(pid))
1797 0 : poll_flags = EPOLLIN | EPOLLRDNORM;
1798 :
1799 0 : return poll_flags;
1800 : }
1801 :
1802 : const struct file_operations pidfd_fops = {
1803 : .release = pidfd_release,
1804 : .poll = pidfd_poll,
1805 : #ifdef CONFIG_PROC_FS
1806 : .show_fdinfo = pidfd_show_fdinfo,
1807 : #endif
1808 : };
1809 :
1810 : static void __delayed_free_task(struct rcu_head *rhp)
1811 : {
1812 : struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1813 :
1814 : free_task(tsk);
1815 : }
1816 :
1817 0 : static __always_inline void delayed_free_task(struct task_struct *tsk)
1818 : {
1819 0 : if (IS_ENABLED(CONFIG_MEMCG))
1820 : call_rcu(&tsk->rcu, __delayed_free_task);
1821 : else
1822 0 : free_task(tsk);
1823 0 : }
1824 :
1825 916 : static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1826 : {
1827 : /* Skip if kernel thread */
1828 916 : if (!tsk->mm)
1829 : return;
1830 :
1831 : /* Skip if spawning a thread or using vfork */
1832 863 : if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1833 : return;
1834 :
1835 : /* We need to synchronize with __set_oom_adj */
1836 0 : mutex_lock(&oom_adj_mutex);
1837 0 : set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1838 : /* Update the values in case they were changed after copy_signal */
1839 0 : tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1840 0 : tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1841 0 : mutex_unlock(&oom_adj_mutex);
1842 : }
1843 :
1844 : /*
1845 : * This creates a new process as a copy of the old one,
1846 : * but does not actually start it yet.
1847 : *
1848 : * It copies the registers, and all the appropriate
1849 : * parts of the process environment (as per the clone
1850 : * flags). The actual kick-off is left to the caller.
1851 : */
1852 916 : static __latent_entropy struct task_struct *copy_process(
1853 : struct pid *pid,
1854 : int trace,
1855 : int node,
1856 : struct kernel_clone_args *args)
1857 : {
1858 916 : int pidfd = -1, retval;
1859 916 : struct task_struct *p;
1860 916 : struct multiprocess_signals delayed;
1861 916 : struct file *pidfile = NULL;
1862 916 : u64 clone_flags = args->flags;
1863 916 : struct nsproxy *nsp = current->nsproxy;
1864 :
1865 : /*
1866 : * Don't allow sharing the root directory with processes in a different
1867 : * namespace
1868 : */
1869 916 : if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1870 916 : return ERR_PTR(-EINVAL);
1871 :
1872 916 : if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1873 916 : return ERR_PTR(-EINVAL);
1874 :
1875 : /*
1876 : * Thread groups must share signals as well, and detached threads
1877 : * can only be started up within the thread group.
1878 : */
1879 916 : if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1880 916 : return ERR_PTR(-EINVAL);
1881 :
1882 : /*
1883 : * Shared signal handlers imply shared VM. By way of the above,
1884 : * thread groups also imply shared VM. Blocking this case allows
1885 : * for various simplifications in other code.
1886 : */
1887 916 : if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1888 916 : return ERR_PTR(-EINVAL);
1889 :
1890 : /*
1891 : * Siblings of global init remain as zombies on exit since they are
1892 : * not reaped by their parent (swapper). To solve this and to avoid
1893 : * multi-rooted process trees, prevent global and container-inits
1894 : * from creating siblings.
1895 : */
1896 916 : if ((clone_flags & CLONE_PARENT) &&
1897 0 : current->signal->flags & SIGNAL_UNKILLABLE)
1898 916 : return ERR_PTR(-EINVAL);
1899 :
1900 : /*
1901 : * If the new process will be in a different pid or user namespace
1902 : * do not allow it to share a thread group with the forking task.
1903 : */
1904 916 : if (clone_flags & CLONE_THREAD) {
1905 6 : if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1906 6 : (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1907 0 : return ERR_PTR(-EINVAL);
1908 : }
1909 :
1910 : /*
1911 : * If the new process will be in a different time namespace
1912 : * do not allow it to share VM or a thread group with the forking task.
1913 : */
1914 916 : if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1915 59 : if (nsp->time_ns != nsp->time_ns_for_children)
1916 916 : return ERR_PTR(-EINVAL);
1917 : }
1918 :
1919 916 : if (clone_flags & CLONE_PIDFD) {
1920 : /*
1921 : * - CLONE_DETACHED is blocked so that we can potentially
1922 : * reuse it later for CLONE_PIDFD.
1923 : * - CLONE_THREAD is blocked until someone really needs it.
1924 : */
1925 0 : if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1926 916 : return ERR_PTR(-EINVAL);
1927 : }
1928 :
1929 : /*
1930 : * Force any signals received before this point to be delivered
1931 : * before the fork happens. Collect up signals sent to multiple
1932 : * processes that happen during the fork and delay them so that
1933 : * they appear to happen after the fork.
1934 : */
1935 916 : sigemptyset(&delayed.signal);
1936 916 : INIT_HLIST_NODE(&delayed.node);
1937 :
1938 916 : spin_lock_irq(¤t->sighand->siglock);
1939 916 : if (!(clone_flags & CLONE_THREAD))
1940 910 : hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1941 916 : recalc_sigpending();
1942 916 : spin_unlock_irq(¤t->sighand->siglock);
1943 916 : retval = -ERESTARTNOINTR;
1944 916 : if (signal_pending(current))
1945 0 : goto fork_out;
1946 :
1947 916 : retval = -ENOMEM;
1948 916 : p = dup_task_struct(current, node);
1949 916 : if (!p)
1950 0 : goto fork_out;
1951 916 : if (args->io_thread)
1952 0 : p->flags |= PF_IO_WORKER;
1953 :
1954 : /*
1955 : * This _must_ happen before we call free_task(), i.e. before we jump
1956 : * to any of the bad_fork_* labels. This is to avoid freeing
1957 : * p->set_child_tid which is (ab)used as a kthread's data pointer for
1958 : * kernel threads (PF_KTHREAD).
1959 : */
1960 916 : p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1961 : /*
1962 : * Clear TID on mm_release()?
1963 : */
1964 916 : p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1965 :
1966 916 : ftrace_graph_init_task(p);
1967 :
1968 916 : rt_mutex_init_task(p);
1969 :
1970 1832 : lockdep_assert_irqs_enabled();
1971 : #ifdef CONFIG_PROVE_LOCKING
1972 916 : DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1973 : #endif
1974 916 : retval = -EAGAIN;
1975 916 : if (atomic_read(&p->real_cred->user->processes) >=
1976 916 : task_rlimit(p, RLIMIT_NPROC)) {
1977 0 : if (p->real_cred->user != INIT_USER &&
1978 0 : !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1979 0 : goto bad_fork_free;
1980 : }
1981 916 : current->flags &= ~PF_NPROC_EXCEEDED;
1982 :
1983 916 : retval = copy_creds(p, clone_flags);
1984 916 : if (retval < 0)
1985 0 : goto bad_fork_free;
1986 :
1987 : /*
1988 : * If multiple threads are within copy_process(), then this check
1989 : * triggers too late. This doesn't hurt, the check is only there
1990 : * to stop root fork bombs.
1991 : */
1992 916 : retval = -EAGAIN;
1993 916 : if (data_race(nr_threads >= max_threads))
1994 0 : goto bad_fork_cleanup_count;
1995 :
1996 916 : delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1997 916 : p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1998 916 : p->flags |= PF_FORKNOEXEC;
1999 916 : INIT_LIST_HEAD(&p->children);
2000 916 : INIT_LIST_HEAD(&p->sibling);
2001 916 : rcu_copy_process(p);
2002 916 : p->vfork_done = NULL;
2003 916 : spin_lock_init(&p->alloc_lock);
2004 :
2005 916 : init_sigpending(&p->pending);
2006 :
2007 916 : p->utime = p->stime = p->gtime = 0;
2008 : #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2009 : p->utimescaled = p->stimescaled = 0;
2010 : #endif
2011 916 : prev_cputime_init(&p->prev_cputime);
2012 :
2013 : #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2014 : seqcount_init(&p->vtime.seqcount);
2015 : p->vtime.starttime = 0;
2016 : p->vtime.state = VTIME_INACTIVE;
2017 : #endif
2018 :
2019 : #ifdef CONFIG_IO_URING
2020 : p->io_uring = NULL;
2021 : #endif
2022 :
2023 : #if defined(SPLIT_RSS_COUNTING)
2024 916 : memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2025 : #endif
2026 :
2027 916 : p->default_timer_slack_ns = current->timer_slack_ns;
2028 :
2029 : #ifdef CONFIG_PSI
2030 : p->psi_flags = 0;
2031 : #endif
2032 :
2033 916 : task_io_accounting_init(&p->ioac);
2034 916 : acct_clear_integrals(p);
2035 :
2036 916 : posix_cputimers_init(&p->posix_cputimers);
2037 :
2038 916 : p->io_context = NULL;
2039 916 : audit_set_context(p, NULL);
2040 916 : cgroup_fork(p);
2041 : #ifdef CONFIG_NUMA
2042 916 : p->mempolicy = mpol_dup(p->mempolicy);
2043 916 : if (IS_ERR(p->mempolicy)) {
2044 0 : retval = PTR_ERR(p->mempolicy);
2045 0 : p->mempolicy = NULL;
2046 0 : goto bad_fork_cleanup_threadgroup_lock;
2047 : }
2048 : #endif
2049 : #ifdef CONFIG_CPUSETS
2050 : p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2051 : p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2052 : seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2053 : #endif
2054 : #ifdef CONFIG_TRACE_IRQFLAGS
2055 916 : memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2056 916 : p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2057 916 : p->irqtrace.softirq_enable_ip = _THIS_IP_;
2058 916 : p->softirqs_enabled = 1;
2059 916 : p->softirq_context = 0;
2060 : #endif
2061 :
2062 916 : p->pagefault_disabled = 0;
2063 :
2064 : #ifdef CONFIG_LOCKDEP
2065 916 : lockdep_init_task(p);
2066 : #endif
2067 :
2068 : #ifdef CONFIG_DEBUG_MUTEXES
2069 916 : p->blocked_on = NULL; /* not blocked yet */
2070 : #endif
2071 : #ifdef CONFIG_BCACHE
2072 : p->sequential_io = 0;
2073 : p->sequential_io_avg = 0;
2074 : #endif
2075 :
2076 : /* Perform scheduler related setup. Assign this task to a CPU. */
2077 916 : retval = sched_fork(clone_flags, p);
2078 916 : if (retval)
2079 0 : goto bad_fork_cleanup_policy;
2080 :
2081 916 : retval = perf_event_init_task(p);
2082 916 : if (retval)
2083 0 : goto bad_fork_cleanup_policy;
2084 916 : retval = audit_alloc(p);
2085 916 : if (retval)
2086 : goto bad_fork_cleanup_perf;
2087 : /* copy all the process information */
2088 916 : shm_init_task(p);
2089 916 : retval = security_task_alloc(p, clone_flags);
2090 916 : if (retval)
2091 0 : goto bad_fork_cleanup_audit;
2092 916 : retval = copy_semundo(clone_flags, p);
2093 916 : if (retval)
2094 0 : goto bad_fork_cleanup_security;
2095 916 : retval = copy_files(clone_flags, p);
2096 916 : if (retval)
2097 0 : goto bad_fork_cleanup_semundo;
2098 916 : retval = copy_fs(clone_flags, p);
2099 916 : if (retval)
2100 0 : goto bad_fork_cleanup_files;
2101 916 : retval = copy_sighand(clone_flags, p);
2102 916 : if (retval)
2103 0 : goto bad_fork_cleanup_fs;
2104 916 : retval = copy_signal(clone_flags, p);
2105 916 : if (retval)
2106 0 : goto bad_fork_cleanup_sighand;
2107 916 : retval = copy_mm(clone_flags, p);
2108 916 : if (retval)
2109 0 : goto bad_fork_cleanup_signal;
2110 916 : retval = copy_namespaces(clone_flags, p);
2111 916 : if (retval)
2112 0 : goto bad_fork_cleanup_mm;
2113 916 : retval = copy_io(clone_flags, p);
2114 916 : if (retval)
2115 0 : goto bad_fork_cleanup_namespaces;
2116 916 : retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2117 916 : if (retval)
2118 0 : goto bad_fork_cleanup_io;
2119 :
2120 916 : stackleak_task_init(p);
2121 :
2122 916 : if (pid != &init_struct_pid) {
2123 913 : pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2124 : args->set_tid_size);
2125 913 : if (IS_ERR(pid)) {
2126 0 : retval = PTR_ERR(pid);
2127 0 : goto bad_fork_cleanup_thread;
2128 : }
2129 : }
2130 :
2131 : /*
2132 : * This has to happen after we've potentially unshared the file
2133 : * descriptor table (so that the pidfd doesn't leak into the child
2134 : * if the fd table isn't shared).
2135 : */
2136 916 : if (clone_flags & CLONE_PIDFD) {
2137 0 : retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2138 0 : if (retval < 0)
2139 0 : goto bad_fork_free_pid;
2140 :
2141 0 : pidfd = retval;
2142 :
2143 0 : pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2144 : O_RDWR | O_CLOEXEC);
2145 0 : if (IS_ERR(pidfile)) {
2146 0 : put_unused_fd(pidfd);
2147 0 : retval = PTR_ERR(pidfile);
2148 0 : goto bad_fork_free_pid;
2149 : }
2150 0 : get_pid(pid); /* held by pidfile now */
2151 :
2152 0 : retval = put_user(pidfd, args->pidfd);
2153 0 : if (retval)
2154 0 : goto bad_fork_put_pidfd;
2155 : }
2156 :
2157 : #ifdef CONFIG_BLOCK
2158 916 : p->plug = NULL;
2159 : #endif
2160 916 : futex_init_task(p);
2161 :
2162 : /*
2163 : * sigaltstack should be cleared when sharing the same VM
2164 : */
2165 916 : if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2166 59 : sas_ss_reset(p);
2167 :
2168 : /*
2169 : * Syscall tracing and stepping should be turned off in the
2170 : * child regardless of CLONE_PTRACE.
2171 : */
2172 916 : user_disable_single_step(p);
2173 916 : clear_task_syscall_work(p, SYSCALL_TRACE);
2174 : #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2175 916 : clear_task_syscall_work(p, SYSCALL_EMU);
2176 : #endif
2177 916 : clear_tsk_latency_tracing(p);
2178 :
2179 : /* ok, now we should be set up.. */
2180 916 : p->pid = pid_nr(pid);
2181 916 : if (clone_flags & CLONE_THREAD) {
2182 6 : p->group_leader = current->group_leader;
2183 6 : p->tgid = current->tgid;
2184 : } else {
2185 910 : p->group_leader = p;
2186 910 : p->tgid = p->pid;
2187 : }
2188 :
2189 916 : p->nr_dirtied = 0;
2190 916 : p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2191 916 : p->dirty_paused_when = 0;
2192 :
2193 916 : p->pdeath_signal = 0;
2194 916 : INIT_LIST_HEAD(&p->thread_group);
2195 916 : p->task_works = NULL;
2196 :
2197 : #ifdef CONFIG_KRETPROBES
2198 : p->kretprobe_instances.first = NULL;
2199 : #endif
2200 :
2201 : /*
2202 : * Ensure that the cgroup subsystem policies allow the new process to be
2203 : * forked. It should be noted that the new process's css_set can be changed
2204 : * between here and cgroup_post_fork() if an organisation operation is in
2205 : * progress.
2206 : */
2207 916 : retval = cgroup_can_fork(p, args);
2208 916 : if (retval)
2209 0 : goto bad_fork_put_pidfd;
2210 :
2211 : /*
2212 : * From this point on we must avoid any synchronous user-space
2213 : * communication until we take the tasklist-lock. In particular, we do
2214 : * not want user-space to be able to predict the process start-time by
2215 : * stalling fork(2) after we recorded the start_time but before it is
2216 : * visible to the system.
2217 : */
2218 :
2219 916 : p->start_time = ktime_get_ns();
2220 916 : p->start_boottime = ktime_get_boottime_ns();
2221 :
2222 : /*
2223 : * Make it visible to the rest of the system, but dont wake it up yet.
2224 : * Need tasklist lock for parent etc handling!
2225 : */
2226 916 : write_lock_irq(&tasklist_lock);
2227 :
2228 : /* CLONE_PARENT re-uses the old parent */
2229 916 : if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2230 6 : p->real_parent = current->real_parent;
2231 6 : p->parent_exec_id = current->parent_exec_id;
2232 6 : if (clone_flags & CLONE_THREAD)
2233 6 : p->exit_signal = -1;
2234 : else
2235 0 : p->exit_signal = current->group_leader->exit_signal;
2236 : } else {
2237 910 : p->real_parent = current;
2238 910 : p->parent_exec_id = current->self_exec_id;
2239 910 : p->exit_signal = args->exit_signal;
2240 : }
2241 :
2242 916 : klp_copy_process(p);
2243 :
2244 916 : spin_lock(¤t->sighand->siglock);
2245 :
2246 : /*
2247 : * Copy seccomp details explicitly here, in case they were changed
2248 : * before holding sighand lock.
2249 : */
2250 916 : copy_seccomp(p);
2251 :
2252 916 : rseq_fork(p, clone_flags);
2253 :
2254 : /* Don't start children in a dying pid namespace */
2255 1832 : if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2256 0 : retval = -ENOMEM;
2257 0 : goto bad_fork_cancel_cgroup;
2258 : }
2259 :
2260 : /* Let kill terminate clone/fork in the middle */
2261 916 : if (fatal_signal_pending(current)) {
2262 0 : retval = -EINTR;
2263 0 : goto bad_fork_cancel_cgroup;
2264 : }
2265 :
2266 : /* past the last point of failure */
2267 916 : if (pidfile)
2268 0 : fd_install(pidfd, pidfile);
2269 :
2270 4580 : init_task_pid_links(p);
2271 916 : if (likely(p->pid)) {
2272 1826 : ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2273 :
2274 913 : init_task_pid(p, PIDTYPE_PID, pid);
2275 913 : if (thread_group_leader(p)) {
2276 907 : init_task_pid(p, PIDTYPE_TGID, pid);
2277 907 : init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2278 907 : init_task_pid(p, PIDTYPE_SID, task_session(current));
2279 :
2280 907 : if (is_child_reaper(pid)) {
2281 1 : ns_of_pid(pid)->child_reaper = p;
2282 1 : p->signal->flags |= SIGNAL_UNKILLABLE;
2283 : }
2284 907 : p->signal->shared_pending.signal = delayed.signal;
2285 907 : p->signal->tty = tty_kref_get(current->signal->tty);
2286 : /*
2287 : * Inherit has_child_subreaper flag under the same
2288 : * tasklist_lock with adding child to the process tree
2289 : * for propagate_has_child_subreaper optimization.
2290 : */
2291 907 : p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2292 : p->real_parent->signal->is_child_subreaper;
2293 907 : list_add_tail(&p->sibling, &p->real_parent->children);
2294 907 : list_add_tail_rcu(&p->tasks, &init_task.tasks);
2295 907 : attach_pid(p, PIDTYPE_TGID);
2296 907 : attach_pid(p, PIDTYPE_PGID);
2297 907 : attach_pid(p, PIDTYPE_SID);
2298 907 : __this_cpu_inc(process_counts);
2299 : } else {
2300 6 : current->signal->nr_threads++;
2301 6 : atomic_inc(¤t->signal->live);
2302 6 : refcount_inc(¤t->signal->sigcnt);
2303 6 : task_join_group_stop(p);
2304 12 : list_add_tail_rcu(&p->thread_group,
2305 6 : &p->group_leader->thread_group);
2306 919 : list_add_tail_rcu(&p->thread_node,
2307 6 : &p->signal->thread_head);
2308 : }
2309 913 : attach_pid(p, PIDTYPE_PID);
2310 913 : nr_threads++;
2311 : }
2312 916 : total_forks++;
2313 916 : hlist_del_init(&delayed.node);
2314 916 : spin_unlock(¤t->sighand->siglock);
2315 916 : syscall_tracepoint_update(p);
2316 916 : write_unlock_irq(&tasklist_lock);
2317 :
2318 916 : proc_fork_connector(p);
2319 916 : sched_post_fork(p);
2320 916 : cgroup_post_fork(p, args);
2321 916 : perf_event_fork(p);
2322 :
2323 916 : trace_task_newtask(p, clone_flags);
2324 916 : uprobe_copy_process(p, clone_flags);
2325 :
2326 916 : copy_oom_score_adj(clone_flags, p);
2327 :
2328 916 : return p;
2329 :
2330 0 : bad_fork_cancel_cgroup:
2331 0 : spin_unlock(¤t->sighand->siglock);
2332 0 : write_unlock_irq(&tasklist_lock);
2333 0 : cgroup_cancel_fork(p, args);
2334 0 : bad_fork_put_pidfd:
2335 0 : if (clone_flags & CLONE_PIDFD) {
2336 0 : fput(pidfile);
2337 0 : put_unused_fd(pidfd);
2338 : }
2339 0 : bad_fork_free_pid:
2340 0 : if (pid != &init_struct_pid)
2341 0 : free_pid(pid);
2342 0 : bad_fork_cleanup_thread:
2343 0 : exit_thread(p);
2344 0 : bad_fork_cleanup_io:
2345 0 : if (p->io_context)
2346 0 : exit_io_context(p);
2347 0 : bad_fork_cleanup_namespaces:
2348 0 : exit_task_namespaces(p);
2349 0 : bad_fork_cleanup_mm:
2350 0 : if (p->mm) {
2351 0 : mm_clear_owner(p->mm, p);
2352 0 : mmput(p->mm);
2353 : }
2354 0 : bad_fork_cleanup_signal:
2355 0 : if (!(clone_flags & CLONE_THREAD))
2356 0 : free_signal_struct(p->signal);
2357 0 : bad_fork_cleanup_sighand:
2358 0 : __cleanup_sighand(p->sighand);
2359 0 : bad_fork_cleanup_fs:
2360 0 : exit_fs(p); /* blocking */
2361 0 : bad_fork_cleanup_files:
2362 0 : exit_files(p); /* blocking */
2363 0 : bad_fork_cleanup_semundo:
2364 0 : exit_sem(p);
2365 0 : bad_fork_cleanup_security:
2366 0 : security_task_free(p);
2367 0 : bad_fork_cleanup_audit:
2368 0 : audit_free(p);
2369 0 : bad_fork_cleanup_perf:
2370 0 : perf_event_free_task(p);
2371 0 : bad_fork_cleanup_policy:
2372 0 : lockdep_free_task(p);
2373 : #ifdef CONFIG_NUMA
2374 0 : mpol_put(p->mempolicy);
2375 0 : bad_fork_cleanup_threadgroup_lock:
2376 : #endif
2377 0 : delayacct_tsk_free(p);
2378 0 : bad_fork_cleanup_count:
2379 0 : atomic_dec(&p->cred->user->processes);
2380 0 : exit_creds(p);
2381 0 : bad_fork_free:
2382 0 : p->state = TASK_DEAD;
2383 0 : put_task_stack(p);
2384 0 : delayed_free_task(p);
2385 0 : fork_out:
2386 0 : spin_lock_irq(¤t->sighand->siglock);
2387 0 : hlist_del_init(&delayed.node);
2388 0 : spin_unlock_irq(¤t->sighand->siglock);
2389 0 : return ERR_PTR(retval);
2390 : }
2391 :
2392 3 : static inline void init_idle_pids(struct task_struct *idle)
2393 : {
2394 3 : enum pid_type type;
2395 :
2396 15 : for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2397 12 : INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2398 24 : init_task_pid(idle, type, &init_struct_pid);
2399 : }
2400 3 : }
2401 :
2402 3 : struct task_struct *fork_idle(int cpu)
2403 : {
2404 3 : struct task_struct *task;
2405 3 : struct kernel_clone_args args = {
2406 : .flags = CLONE_VM,
2407 : };
2408 :
2409 3 : task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2410 3 : if (!IS_ERR(task)) {
2411 3 : init_idle_pids(task);
2412 3 : init_idle(task, cpu);
2413 : }
2414 :
2415 3 : return task;
2416 : }
2417 :
2418 1 : struct mm_struct *copy_init_mm(void)
2419 : {
2420 1 : return dup_mm(NULL, &init_mm);
2421 : }
2422 :
2423 : /*
2424 : * This is like kernel_clone(), but shaved down and tailored to just
2425 : * creating io_uring workers. It returns a created task, or an error pointer.
2426 : * The returned task is inactive, and the caller must fire it up through
2427 : * wake_up_new_task(p). All signals are blocked in the created task.
2428 : */
2429 0 : struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2430 : {
2431 0 : unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2432 : CLONE_IO;
2433 0 : struct kernel_clone_args args = {
2434 : .flags = ((lower_32_bits(flags) | CLONE_VM |
2435 : CLONE_UNTRACED) & ~CSIGNAL),
2436 : .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2437 0 : .stack = (unsigned long)fn,
2438 0 : .stack_size = (unsigned long)arg,
2439 : .io_thread = 1,
2440 : };
2441 0 : struct task_struct *tsk;
2442 :
2443 0 : tsk = copy_process(NULL, 0, node, &args);
2444 0 : if (!IS_ERR(tsk)) {
2445 0 : sigfillset(&tsk->blocked);
2446 0 : sigdelsetmask(&tsk->blocked, sigmask(SIGKILL));
2447 0 : tsk->flags |= PF_NOFREEZE;
2448 : }
2449 0 : return tsk;
2450 : }
2451 :
2452 : /*
2453 : * Ok, this is the main fork-routine.
2454 : *
2455 : * It copies the process, and if successful kick-starts
2456 : * it and waits for it to finish using the VM if required.
2457 : *
2458 : * args->exit_signal is expected to be checked for sanity by the caller.
2459 : */
2460 913 : pid_t kernel_clone(struct kernel_clone_args *args)
2461 : {
2462 913 : u64 clone_flags = args->flags;
2463 913 : struct completion vfork;
2464 913 : struct pid *pid;
2465 913 : struct task_struct *p;
2466 913 : int trace = 0;
2467 913 : pid_t nr;
2468 :
2469 : /*
2470 : * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2471 : * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2472 : * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2473 : * field in struct clone_args and it still doesn't make sense to have
2474 : * them both point at the same memory location. Performing this check
2475 : * here has the advantage that we don't need to have a separate helper
2476 : * to check for legacy clone().
2477 : */
2478 913 : if ((args->flags & CLONE_PIDFD) &&
2479 0 : (args->flags & CLONE_PARENT_SETTID) &&
2480 0 : (args->pidfd == args->parent_tid))
2481 : return -EINVAL;
2482 :
2483 : /*
2484 : * Determine whether and which event to report to ptracer. When
2485 : * called from kernel_thread or CLONE_UNTRACED is explicitly
2486 : * requested, no event is reported; otherwise, report if the event
2487 : * for the type of forking is enabled.
2488 : */
2489 913 : if (!(clone_flags & CLONE_UNTRACED)) {
2490 863 : if (clone_flags & CLONE_VFORK)
2491 : trace = PTRACE_EVENT_VFORK;
2492 863 : else if (args->exit_signal != SIGCHLD)
2493 : trace = PTRACE_EVENT_CLONE;
2494 : else
2495 857 : trace = PTRACE_EVENT_FORK;
2496 :
2497 863 : if (likely(!ptrace_event_enabled(current, trace)))
2498 863 : trace = 0;
2499 : }
2500 :
2501 913 : p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2502 913 : add_latent_entropy();
2503 :
2504 913 : if (IS_ERR(p))
2505 0 : return PTR_ERR(p);
2506 :
2507 : /*
2508 : * Do this prior waking up the new thread - the thread pointer
2509 : * might get invalid after that point, if the thread exits quickly.
2510 : */
2511 913 : trace_sched_process_fork(current, p);
2512 :
2513 913 : pid = get_task_pid(p, PIDTYPE_PID);
2514 913 : nr = pid_vnr(pid);
2515 :
2516 913 : if (clone_flags & CLONE_PARENT_SETTID)
2517 10 : put_user(nr, args->parent_tid);
2518 :
2519 913 : if (clone_flags & CLONE_VFORK) {
2520 0 : p->vfork_done = &vfork;
2521 0 : init_completion(&vfork);
2522 0 : get_task_struct(p);
2523 : }
2524 :
2525 913 : wake_up_new_task(p);
2526 :
2527 : /* forking complete and child started to run, tell ptracer */
2528 913 : if (unlikely(trace))
2529 0 : ptrace_event_pid(trace, pid);
2530 :
2531 913 : if (clone_flags & CLONE_VFORK) {
2532 0 : if (!wait_for_vfork_done(p, &vfork))
2533 0 : ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2534 : }
2535 :
2536 913 : put_pid(pid);
2537 913 : return nr;
2538 : }
2539 :
2540 : /*
2541 : * Create a kernel thread.
2542 : */
2543 50 : pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2544 : {
2545 50 : struct kernel_clone_args args = {
2546 50 : .flags = ((lower_32_bits(flags) | CLONE_VM |
2547 50 : CLONE_UNTRACED) & ~CSIGNAL),
2548 50 : .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2549 50 : .stack = (unsigned long)fn,
2550 50 : .stack_size = (unsigned long)arg,
2551 : };
2552 :
2553 50 : return kernel_clone(&args);
2554 : }
2555 :
2556 : #ifdef __ARCH_WANT_SYS_FORK
2557 0 : SYSCALL_DEFINE0(fork)
2558 : {
2559 : #ifdef CONFIG_MMU
2560 0 : struct kernel_clone_args args = {
2561 : .exit_signal = SIGCHLD,
2562 : };
2563 :
2564 0 : return kernel_clone(&args);
2565 : #else
2566 : /* can not support in nommu mode */
2567 : return -EINVAL;
2568 : #endif
2569 : }
2570 : #endif
2571 :
2572 : #ifdef __ARCH_WANT_SYS_VFORK
2573 0 : SYSCALL_DEFINE0(vfork)
2574 : {
2575 0 : struct kernel_clone_args args = {
2576 : .flags = CLONE_VFORK | CLONE_VM,
2577 : .exit_signal = SIGCHLD,
2578 : };
2579 :
2580 0 : return kernel_clone(&args);
2581 : }
2582 : #endif
2583 :
2584 : #ifdef __ARCH_WANT_SYS_CLONE
2585 : #ifdef CONFIG_CLONE_BACKWARDS
2586 : SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2587 : int __user *, parent_tidptr,
2588 : unsigned long, tls,
2589 : int __user *, child_tidptr)
2590 : #elif defined(CONFIG_CLONE_BACKWARDS2)
2591 : SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2592 : int __user *, parent_tidptr,
2593 : int __user *, child_tidptr,
2594 : unsigned long, tls)
2595 : #elif defined(CONFIG_CLONE_BACKWARDS3)
2596 : SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2597 : int, stack_size,
2598 : int __user *, parent_tidptr,
2599 : int __user *, child_tidptr,
2600 : unsigned long, tls)
2601 : #else
2602 1726 : SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2603 : int __user *, parent_tidptr,
2604 : int __user *, child_tidptr,
2605 : unsigned long, tls)
2606 : #endif
2607 : {
2608 863 : struct kernel_clone_args args = {
2609 863 : .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2610 : .pidfd = parent_tidptr,
2611 : .child_tid = child_tidptr,
2612 : .parent_tid = parent_tidptr,
2613 863 : .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2614 : .stack = newsp,
2615 : .tls = tls,
2616 : };
2617 :
2618 863 : return kernel_clone(&args);
2619 : }
2620 : #endif
2621 :
2622 : #ifdef __ARCH_WANT_SYS_CLONE3
2623 :
2624 0 : noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2625 : struct clone_args __user *uargs,
2626 : size_t usize)
2627 : {
2628 0 : int err;
2629 0 : struct clone_args args;
2630 0 : pid_t *kset_tid = kargs->set_tid;
2631 :
2632 0 : BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2633 : CLONE_ARGS_SIZE_VER0);
2634 0 : BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2635 : CLONE_ARGS_SIZE_VER1);
2636 0 : BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2637 : CLONE_ARGS_SIZE_VER2);
2638 0 : BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2639 :
2640 0 : if (unlikely(usize > PAGE_SIZE))
2641 : return -E2BIG;
2642 0 : if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2643 : return -EINVAL;
2644 :
2645 0 : err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2646 0 : if (err)
2647 0 : return err;
2648 :
2649 0 : if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2650 : return -EINVAL;
2651 :
2652 0 : if (unlikely(!args.set_tid && args.set_tid_size > 0))
2653 : return -EINVAL;
2654 :
2655 0 : if (unlikely(args.set_tid && args.set_tid_size == 0))
2656 : return -EINVAL;
2657 :
2658 : /*
2659 : * Verify that higher 32bits of exit_signal are unset and that
2660 : * it is a valid signal
2661 : */
2662 0 : if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2663 : !valid_signal(args.exit_signal)))
2664 : return -EINVAL;
2665 :
2666 0 : if ((args.flags & CLONE_INTO_CGROUP) &&
2667 0 : (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2668 : return -EINVAL;
2669 :
2670 0 : *kargs = (struct kernel_clone_args){
2671 : .flags = args.flags,
2672 0 : .pidfd = u64_to_user_ptr(args.pidfd),
2673 0 : .child_tid = u64_to_user_ptr(args.child_tid),
2674 0 : .parent_tid = u64_to_user_ptr(args.parent_tid),
2675 : .exit_signal = args.exit_signal,
2676 0 : .stack = args.stack,
2677 0 : .stack_size = args.stack_size,
2678 0 : .tls = args.tls,
2679 : .set_tid_size = args.set_tid_size,
2680 0 : .cgroup = args.cgroup,
2681 : };
2682 :
2683 0 : if (args.set_tid &&
2684 0 : copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2685 : (kargs->set_tid_size * sizeof(pid_t))))
2686 : return -EFAULT;
2687 :
2688 0 : kargs->set_tid = kset_tid;
2689 :
2690 0 : return 0;
2691 : }
2692 :
2693 : /**
2694 : * clone3_stack_valid - check and prepare stack
2695 : * @kargs: kernel clone args
2696 : *
2697 : * Verify that the stack arguments userspace gave us are sane.
2698 : * In addition, set the stack direction for userspace since it's easy for us to
2699 : * determine.
2700 : */
2701 0 : static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2702 : {
2703 0 : if (kargs->stack == 0) {
2704 0 : if (kargs->stack_size > 0)
2705 0 : return false;
2706 : } else {
2707 0 : if (kargs->stack_size == 0)
2708 : return false;
2709 :
2710 0 : if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2711 : return false;
2712 :
2713 : #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2714 0 : kargs->stack += kargs->stack_size;
2715 : #endif
2716 : }
2717 :
2718 : return true;
2719 : }
2720 :
2721 0 : static bool clone3_args_valid(struct kernel_clone_args *kargs)
2722 : {
2723 : /* Verify that no unknown flags are passed along. */
2724 0 : if (kargs->flags &
2725 : ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2726 : return false;
2727 :
2728 : /*
2729 : * - make the CLONE_DETACHED bit reuseable for clone3
2730 : * - make the CSIGNAL bits reuseable for clone3
2731 : */
2732 0 : if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2733 : return false;
2734 :
2735 0 : if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2736 : (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2737 : return false;
2738 :
2739 0 : if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2740 0 : kargs->exit_signal)
2741 : return false;
2742 :
2743 0 : if (!clone3_stack_valid(kargs))
2744 0 : return false;
2745 :
2746 : return true;
2747 : }
2748 :
2749 : /**
2750 : * clone3 - create a new process with specific properties
2751 : * @uargs: argument structure
2752 : * @size: size of @uargs
2753 : *
2754 : * clone3() is the extensible successor to clone()/clone2().
2755 : * It takes a struct as argument that is versioned by its size.
2756 : *
2757 : * Return: On success, a positive PID for the child process.
2758 : * On error, a negative errno number.
2759 : */
2760 0 : SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2761 : {
2762 0 : int err;
2763 :
2764 0 : struct kernel_clone_args kargs;
2765 0 : pid_t set_tid[MAX_PID_NS_LEVEL];
2766 :
2767 0 : kargs.set_tid = set_tid;
2768 :
2769 0 : err = copy_clone_args_from_user(&kargs, uargs, size);
2770 0 : if (err)
2771 0 : return err;
2772 :
2773 0 : if (!clone3_args_valid(&kargs))
2774 : return -EINVAL;
2775 :
2776 0 : return kernel_clone(&kargs);
2777 : }
2778 : #endif
2779 :
2780 1 : void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2781 : {
2782 1 : struct task_struct *leader, *parent, *child;
2783 1 : int res;
2784 :
2785 1 : read_lock(&tasklist_lock);
2786 1 : leader = top = top->group_leader;
2787 2 : down:
2788 4 : for_each_thread(leader, parent) {
2789 3 : list_for_each_entry(child, &parent->children, sibling) {
2790 1 : res = visitor(child, data);
2791 1 : if (res) {
2792 1 : if (res < 0)
2793 0 : goto out;
2794 1 : leader = child;
2795 1 : goto down;
2796 : }
2797 0 : up:
2798 1 : ;
2799 : }
2800 : }
2801 :
2802 2 : if (leader != top) {
2803 1 : child = leader;
2804 1 : parent = child->real_parent;
2805 1 : leader = parent->group_leader;
2806 1 : goto up;
2807 : }
2808 1 : out:
2809 1 : read_unlock(&tasklist_lock);
2810 1 : }
2811 :
2812 : #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2813 : #define ARCH_MIN_MMSTRUCT_ALIGN 0
2814 : #endif
2815 :
2816 117 : static void sighand_ctor(void *data)
2817 : {
2818 117 : struct sighand_struct *sighand = data;
2819 :
2820 117 : spin_lock_init(&sighand->siglock);
2821 117 : init_waitqueue_head(&sighand->signalfd_wqh);
2822 117 : }
2823 :
2824 1 : void __init proc_caches_init(void)
2825 : {
2826 1 : unsigned int mm_size;
2827 :
2828 1 : sighand_cachep = kmem_cache_create("sighand_cache",
2829 : sizeof(struct sighand_struct), 0,
2830 : SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2831 : SLAB_ACCOUNT, sighand_ctor);
2832 1 : signal_cachep = kmem_cache_create("signal_cache",
2833 : sizeof(struct signal_struct), 0,
2834 : SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2835 : NULL);
2836 1 : files_cachep = kmem_cache_create("files_cache",
2837 : sizeof(struct files_struct), 0,
2838 : SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2839 : NULL);
2840 1 : fs_cachep = kmem_cache_create("fs_cache",
2841 : sizeof(struct fs_struct), 0,
2842 : SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2843 : NULL);
2844 :
2845 : /*
2846 : * The mm_cpumask is located at the end of mm_struct, and is
2847 : * dynamically sized based on the maximum CPU number this system
2848 : * can have, taking hotplug into account (nr_cpu_ids).
2849 : */
2850 1 : mm_size = sizeof(struct mm_struct) + cpumask_size();
2851 :
2852 1 : mm_cachep = kmem_cache_create_usercopy("mm_struct",
2853 : mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2854 : SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2855 : offsetof(struct mm_struct, saved_auxv),
2856 : sizeof_field(struct mm_struct, saved_auxv),
2857 : NULL);
2858 1 : vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2859 1 : mmap_init();
2860 1 : nsproxy_cache_init();
2861 1 : }
2862 :
2863 : /*
2864 : * Check constraints on flags passed to the unshare system call.
2865 : */
2866 50 : static int check_unshare_flags(unsigned long unshare_flags)
2867 : {
2868 50 : if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2869 : CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2870 : CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2871 : CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2872 : CLONE_NEWTIME))
2873 : return -EINVAL;
2874 : /*
2875 : * Not implemented, but pretend it works if there is nothing
2876 : * to unshare. Note that unsharing the address space or the
2877 : * signal handlers also need to unshare the signal queues (aka
2878 : * CLONE_THREAD).
2879 : */
2880 50 : if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2881 0 : if (!thread_group_empty(current))
2882 : return -EINVAL;
2883 : }
2884 50 : if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2885 0 : if (refcount_read(¤t->sighand->count) > 1)
2886 : return -EINVAL;
2887 : }
2888 50 : if (unshare_flags & CLONE_VM) {
2889 0 : if (!current_is_single_threaded())
2890 0 : return -EINVAL;
2891 : }
2892 :
2893 : return 0;
2894 : }
2895 :
2896 : /*
2897 : * Unshare the filesystem structure if it is being shared
2898 : */
2899 50 : static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2900 : {
2901 50 : struct fs_struct *fs = current->fs;
2902 :
2903 50 : if (!(unshare_flags & CLONE_FS) || !fs)
2904 : return 0;
2905 :
2906 : /* don't need lock here; in the worst case we'll do useless copy */
2907 50 : if (fs->users == 1)
2908 : return 0;
2909 :
2910 1 : *new_fsp = copy_fs_struct(fs);
2911 1 : if (!*new_fsp)
2912 0 : return -ENOMEM;
2913 :
2914 : return 0;
2915 : }
2916 :
2917 : /*
2918 : * Unshare file descriptor table if it is being shared
2919 : */
2920 601 : int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2921 : struct files_struct **new_fdp)
2922 : {
2923 601 : struct files_struct *fd = current->files;
2924 601 : int error = 0;
2925 :
2926 601 : if ((unshare_flags & CLONE_FILES) &&
2927 551 : (fd && atomic_read(&fd->count) > 1)) {
2928 0 : *new_fdp = dup_fd(fd, max_fds, &error);
2929 0 : if (!*new_fdp)
2930 0 : return error;
2931 : }
2932 :
2933 : return 0;
2934 : }
2935 :
2936 : /*
2937 : * unshare allows a process to 'unshare' part of the process
2938 : * context which was originally shared using clone. copy_*
2939 : * functions used by kernel_clone() cannot be used here directly
2940 : * because they modify an inactive task_struct that is being
2941 : * constructed. Here we are modifying the current, active,
2942 : * task_struct.
2943 : */
2944 50 : int ksys_unshare(unsigned long unshare_flags)
2945 : {
2946 50 : struct fs_struct *fs, *new_fs = NULL;
2947 50 : struct files_struct *fd, *new_fd = NULL;
2948 50 : struct cred *new_cred = NULL;
2949 50 : struct nsproxy *new_nsproxy = NULL;
2950 50 : int do_sysvsem = 0;
2951 50 : int err;
2952 :
2953 : /*
2954 : * If unsharing a user namespace must also unshare the thread group
2955 : * and unshare the filesystem root and working directories.
2956 : */
2957 50 : if (unshare_flags & CLONE_NEWUSER)
2958 0 : unshare_flags |= CLONE_THREAD | CLONE_FS;
2959 : /*
2960 : * If unsharing vm, must also unshare signal handlers.
2961 : */
2962 50 : if (unshare_flags & CLONE_VM)
2963 0 : unshare_flags |= CLONE_SIGHAND;
2964 : /*
2965 : * If unsharing a signal handlers, must also unshare the signal queues.
2966 : */
2967 50 : if (unshare_flags & CLONE_SIGHAND)
2968 0 : unshare_flags |= CLONE_THREAD;
2969 : /*
2970 : * If unsharing namespace, must also unshare filesystem information.
2971 : */
2972 50 : if (unshare_flags & CLONE_NEWNS)
2973 50 : unshare_flags |= CLONE_FS;
2974 :
2975 50 : err = check_unshare_flags(unshare_flags);
2976 50 : if (err)
2977 0 : goto bad_unshare_out;
2978 : /*
2979 : * CLONE_NEWIPC must also detach from the undolist: after switching
2980 : * to a new ipc namespace, the semaphore arrays from the old
2981 : * namespace are unreachable.
2982 : */
2983 50 : if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2984 0 : do_sysvsem = 1;
2985 50 : err = unshare_fs(unshare_flags, &new_fs);
2986 50 : if (err)
2987 0 : goto bad_unshare_out;
2988 50 : err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2989 50 : if (err)
2990 0 : goto bad_unshare_cleanup_fs;
2991 50 : err = unshare_userns(unshare_flags, &new_cred);
2992 50 : if (err)
2993 0 : goto bad_unshare_cleanup_fd;
2994 50 : err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2995 : new_cred, new_fs);
2996 50 : if (err)
2997 0 : goto bad_unshare_cleanup_cred;
2998 :
2999 50 : if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3000 50 : if (do_sysvsem) {
3001 : /*
3002 : * CLONE_SYSVSEM is equivalent to sys_exit().
3003 : */
3004 0 : exit_sem(current);
3005 : }
3006 50 : if (unshare_flags & CLONE_NEWIPC) {
3007 : /* Orphan segments in old ns (see sem above). */
3008 0 : exit_shm(current);
3009 0 : shm_init_task(current);
3010 : }
3011 :
3012 50 : if (new_nsproxy)
3013 50 : switch_task_namespaces(current, new_nsproxy);
3014 :
3015 50 : task_lock(current);
3016 :
3017 50 : if (new_fs) {
3018 1 : fs = current->fs;
3019 1 : spin_lock(&fs->lock);
3020 1 : current->fs = new_fs;
3021 1 : if (--fs->users)
3022 1 : new_fs = NULL;
3023 : else
3024 0 : new_fs = fs;
3025 1 : spin_unlock(&fs->lock);
3026 : }
3027 :
3028 50 : if (new_fd) {
3029 0 : fd = current->files;
3030 0 : current->files = new_fd;
3031 0 : new_fd = fd;
3032 : }
3033 :
3034 50 : task_unlock(current);
3035 :
3036 50 : if (new_cred) {
3037 : /* Install the new user namespace */
3038 : commit_creds(new_cred);
3039 : new_cred = NULL;
3040 : }
3041 : }
3042 :
3043 50 : perf_event_namespaces(current);
3044 :
3045 : bad_unshare_cleanup_cred:
3046 : if (new_cred)
3047 : put_cred(new_cred);
3048 50 : bad_unshare_cleanup_fd:
3049 50 : if (new_fd)
3050 0 : put_files_struct(new_fd);
3051 :
3052 50 : bad_unshare_cleanup_fs:
3053 50 : if (new_fs)
3054 0 : free_fs_struct(new_fs);
3055 :
3056 50 : bad_unshare_out:
3057 50 : return err;
3058 : }
3059 :
3060 98 : SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3061 : {
3062 49 : return ksys_unshare(unshare_flags);
3063 : }
3064 :
3065 : /*
3066 : * Helper to unshare the files of the current task.
3067 : * We don't want to expose copy_files internals to
3068 : * the exec layer of the kernel.
3069 : */
3070 :
3071 551 : int unshare_files(void)
3072 : {
3073 551 : struct task_struct *task = current;
3074 551 : struct files_struct *old, *copy = NULL;
3075 551 : int error;
3076 :
3077 551 : error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3078 551 : if (error || !copy)
3079 : return error;
3080 :
3081 0 : old = task->files;
3082 0 : task_lock(task);
3083 0 : task->files = copy;
3084 0 : task_unlock(task);
3085 0 : put_files_struct(old);
3086 0 : return 0;
3087 : }
3088 :
3089 8 : int sysctl_max_threads(struct ctl_table *table, int write,
3090 : void *buffer, size_t *lenp, loff_t *ppos)
3091 : {
3092 8 : struct ctl_table t;
3093 8 : int ret;
3094 8 : int threads = max_threads;
3095 8 : int min = 1;
3096 8 : int max = MAX_THREADS;
3097 :
3098 8 : t = *table;
3099 8 : t.data = &threads;
3100 8 : t.extra1 = &min;
3101 8 : t.extra2 = &max;
3102 :
3103 8 : ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3104 8 : if (ret || !write)
3105 : return ret;
3106 :
3107 0 : max_threads = threads;
3108 :
3109 0 : return 0;
3110 : }
|
