summaryrefslogtreecommitdiff
path: root/kernel/fork.c (plain)
blob: ff9acf962da8c12f78ebee5d7a240e0716ad07bd
1/*
2 * linux/kernel/fork.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * 'fork.c' contains the help-routines for the 'fork' system call
9 * (see also entry.S and others).
10 * Fork is rather simple, once you get the hang of it, but the memory
11 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12 */
13
14#include <linux/slab.h>
15#include <linux/init.h>
16#include <linux/unistd.h>
17#include <linux/module.h>
18#include <linux/vmalloc.h>
19#include <linux/completion.h>
20#include <linux/personality.h>
21#include <linux/mempolicy.h>
22#include <linux/sem.h>
23#include <linux/file.h>
24#include <linux/fdtable.h>
25#include <linux/iocontext.h>
26#include <linux/key.h>
27#include <linux/binfmts.h>
28#include <linux/mman.h>
29#include <linux/mmu_notifier.h>
30#include <linux/fs.h>
31#include <linux/mm.h>
32#include <linux/vmacache.h>
33#include <linux/nsproxy.h>
34#include <linux/capability.h>
35#include <linux/cpu.h>
36#include <linux/cgroup.h>
37#include <linux/security.h>
38#include <linux/hugetlb.h>
39#include <linux/seccomp.h>
40#include <linux/swap.h>
41#include <linux/syscalls.h>
42#include <linux/jiffies.h>
43#include <linux/futex.h>
44#include <linux/compat.h>
45#include <linux/kthread.h>
46#include <linux/task_io_accounting_ops.h>
47#include <linux/rcupdate.h>
48#include <linux/ptrace.h>
49#include <linux/mount.h>
50#include <linux/audit.h>
51#include <linux/memcontrol.h>
52#include <linux/ftrace.h>
53#include <linux/proc_fs.h>
54#include <linux/profile.h>
55#include <linux/rmap.h>
56#include <linux/ksm.h>
57#include <linux/acct.h>
58#include <linux/tsacct_kern.h>
59#include <linux/cn_proc.h>
60#include <linux/freezer.h>
61#include <linux/kaiser.h>
62#include <linux/delayacct.h>
63#include <linux/taskstats_kern.h>
64#include <linux/random.h>
65#include <linux/tty.h>
66#include <linux/blkdev.h>
67#include <linux/fs_struct.h>
68#include <linux/magic.h>
69#include <linux/perf_event.h>
70#include <linux/posix-timers.h>
71#include <linux/user-return-notifier.h>
72#include <linux/oom.h>
73#include <linux/khugepaged.h>
74#include <linux/signalfd.h>
75#include <linux/uprobes.h>
76#include <linux/aio.h>
77#include <linux/compiler.h>
78#include <linux/sysctl.h>
79#include <linux/kcov.h>
80#include <linux/cpufreq_times.h>
81
82#include <asm/pgtable.h>
83#include <asm/pgalloc.h>
84#include <asm/uaccess.h>
85#include <asm/mmu_context.h>
86#include <asm/cacheflush.h>
87#include <asm/tlbflush.h>
88
89#ifdef CONFIG_AMLOGIC_VMAP
90#include <linux/amlogic/vmap_stack.h>
91#endif
92
93#include <trace/events/sched.h>
94
95#define CREATE_TRACE_POINTS
96#include <trace/events/task.h>
97
98/*
99 * Minimum number of threads to boot the kernel
100 */
101#define MIN_THREADS 20
102
103/*
104 * Maximum number of threads
105 */
106#define MAX_THREADS FUTEX_TID_MASK
107
108/*
109 * Protected counters by write_lock_irq(&tasklist_lock)
110 */
111unsigned long total_forks; /* Handle normal Linux uptimes. */
112int nr_threads; /* The idle threads do not count.. */
113
114int max_threads; /* tunable limit on nr_threads */
115
116DEFINE_PER_CPU(unsigned long, process_counts) = 0;
117
118__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
119
120#ifdef CONFIG_PROVE_RCU
121int lockdep_tasklist_lock_is_held(void)
122{
123 return lockdep_is_held(&tasklist_lock);
124}
125EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
126#endif /* #ifdef CONFIG_PROVE_RCU */
127
128int nr_processes(void)
129{
130 int cpu;
131 int total = 0;
132
133 for_each_possible_cpu(cpu)
134 total += per_cpu(process_counts, cpu);
135
136 return total;
137}
138
139void __weak arch_release_task_struct(struct task_struct *tsk)
140{
141}
142
143#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
144static struct kmem_cache *task_struct_cachep;
145
146static inline struct task_struct *alloc_task_struct_node(int node)
147{
148 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
149}
150
151static inline void free_task_struct(struct task_struct *tsk)
152{
153 kmem_cache_free(task_struct_cachep, tsk);
154}
155#endif
156
157void __weak arch_release_thread_stack(unsigned long *stack)
158{
159}
160
161#ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
162
163/*
164 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
165 * kmemcache based allocator.
166 */
167# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
168
169#ifdef CONFIG_VMAP_STACK
170/*
171 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
172 * flush. Try to minimize the number of calls by caching stacks.
173 */
174#define NR_CACHED_STACKS 2
175static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
176#endif
177
178static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
179{
180#ifdef CONFIG_VMAP_STACK
181 void *stack;
182 int i;
183
184 local_irq_disable();
185 for (i = 0; i < NR_CACHED_STACKS; i++) {
186 struct vm_struct *s = this_cpu_read(cached_stacks[i]);
187
188 if (!s)
189 continue;
190 this_cpu_write(cached_stacks[i], NULL);
191
192 /* Clear stale pointers from reused stack. */
193 memset(s->addr, 0, THREAD_SIZE);
194
195 tsk->stack_vm_area = s;
196 local_irq_enable();
197 return s->addr;
198 }
199 local_irq_enable();
200
201 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
202 VMALLOC_START, VMALLOC_END,
203 THREADINFO_GFP | __GFP_HIGHMEM,
204 PAGE_KERNEL,
205 0, node, __builtin_return_address(0));
206
207 /*
208 * We can't call find_vm_area() in interrupt context, and
209 * free_thread_stack() can be called in interrupt context,
210 * so cache the vm_struct.
211 */
212 if (stack)
213 tsk->stack_vm_area = find_vm_area(stack);
214 return stack;
215#else
216#ifdef CONFIG_AMLOGIC_VMAP
217 return aml_stack_alloc(node, tsk);
218#else /* CONFIG_AMLOGIC_VMAP */
219 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
220 THREAD_SIZE_ORDER);
221
222 return page ? page_address(page) : NULL;
223#endif /* CONFIG_AMLOGIC_VMAP */
224#endif
225}
226
227static inline void free_thread_stack(struct task_struct *tsk)
228{
229#ifdef CONFIG_AMLOGIC_VMAP
230 aml_stack_free(tsk);
231#else /* CONFIG_AMLOGIC_VMAP */
232 kaiser_unmap_thread_stack(tsk->stack);
233#ifdef CONFIG_VMAP_STACK
234 if (task_stack_vm_area(tsk)) {
235 unsigned long flags;
236 int i;
237
238 local_irq_save(flags);
239 for (i = 0; i < NR_CACHED_STACKS; i++) {
240 if (this_cpu_read(cached_stacks[i]))
241 continue;
242
243 this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
244 local_irq_restore(flags);
245 return;
246 }
247 local_irq_restore(flags);
248
249 vfree(tsk->stack);
250 return;
251 }
252#endif
253
254 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
255#endif /* CONFIG_AMLOGIC_VMAP */
256}
257# else
258static struct kmem_cache *thread_stack_cache;
259
260static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
261 int node)
262{
263 return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
264}
265
266static void free_thread_stack(struct task_struct *tsk)
267{
268 kmem_cache_free(thread_stack_cache, tsk->stack);
269}
270
271void thread_stack_cache_init(void)
272{
273 thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
274 THREAD_SIZE, 0, NULL);
275 BUG_ON(thread_stack_cache == NULL);
276}
277# endif
278#endif
279
280/* SLAB cache for signal_struct structures (tsk->signal) */
281static struct kmem_cache *signal_cachep;
282
283/* SLAB cache for sighand_struct structures (tsk->sighand) */
284struct kmem_cache *sighand_cachep;
285
286/* SLAB cache for files_struct structures (tsk->files) */
287struct kmem_cache *files_cachep;
288
289/* SLAB cache for fs_struct structures (tsk->fs) */
290struct kmem_cache *fs_cachep;
291
292/* SLAB cache for vm_area_struct structures */
293struct kmem_cache *vm_area_cachep;
294
295/* SLAB cache for mm_struct structures (tsk->mm) */
296static struct kmem_cache *mm_cachep;
297
298static void account_kernel_stack(struct task_struct *tsk, int account)
299{
300#ifdef CONFIG_AMLOGIC_VMAP
301 aml_account_task_stack(tsk, account);
302#else
303 void *stack = task_stack_page(tsk);
304 struct vm_struct *vm = task_stack_vm_area(tsk);
305
306 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
307
308 if (vm) {
309 int i;
310
311 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
312
313 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
314 mod_zone_page_state(page_zone(vm->pages[i]),
315 NR_KERNEL_STACK_KB,
316 PAGE_SIZE / 1024 * account);
317 }
318
319 /* All stack pages belong to the same memcg. */
320 memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
321 account * (THREAD_SIZE / 1024));
322 } else {
323 /*
324 * All stack pages are in the same zone and belong to the
325 * same memcg.
326 */
327 struct page *first_page = virt_to_page(stack);
328
329 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
330 THREAD_SIZE / 1024 * account);
331
332 memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
333 account * (THREAD_SIZE / 1024));
334 }
335#endif /* CONFIG_AMLOGIC_VMAP*/
336}
337
338static void release_task_stack(struct task_struct *tsk)
339{
340 if (WARN_ON(tsk->state != TASK_DEAD))
341 return; /* Better to leak the stack than to free prematurely */
342
343 account_kernel_stack(tsk, -1);
344 arch_release_thread_stack(tsk->stack);
345 free_thread_stack(tsk);
346 tsk->stack = NULL;
347#ifdef CONFIG_VMAP_STACK
348 tsk->stack_vm_area = NULL;
349#endif
350}
351
352#ifdef CONFIG_THREAD_INFO_IN_TASK
353void put_task_stack(struct task_struct *tsk)
354{
355 if (atomic_dec_and_test(&tsk->stack_refcount))
356 release_task_stack(tsk);
357}
358#endif
359
360void free_task(struct task_struct *tsk)
361{
362 cpufreq_task_times_exit(tsk);
363
364#ifndef CONFIG_THREAD_INFO_IN_TASK
365 /*
366 * The task is finally done with both the stack and thread_info,
367 * so free both.
368 */
369 release_task_stack(tsk);
370#else
371 /*
372 * If the task had a separate stack allocation, it should be gone
373 * by now.
374 */
375 WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
376#endif
377 rt_mutex_debug_task_free(tsk);
378 ftrace_graph_exit_task(tsk);
379 put_seccomp_filter(tsk);
380 arch_release_task_struct(tsk);
381 free_task_struct(tsk);
382}
383EXPORT_SYMBOL(free_task);
384
385static inline void free_signal_struct(struct signal_struct *sig)
386{
387 taskstats_tgid_free(sig);
388 sched_autogroup_exit(sig);
389 /*
390 * __mmdrop is not safe to call from softirq context on x86 due to
391 * pgd_dtor so postpone it to the async context
392 */
393 if (sig->oom_mm)
394 mmdrop_async(sig->oom_mm);
395 kmem_cache_free(signal_cachep, sig);
396}
397
398static inline void put_signal_struct(struct signal_struct *sig)
399{
400 if (atomic_dec_and_test(&sig->sigcnt))
401 free_signal_struct(sig);
402}
403
404void __put_task_struct(struct task_struct *tsk)
405{
406 WARN_ON(!tsk->exit_state);
407 WARN_ON(atomic_read(&tsk->usage));
408 WARN_ON(tsk == current);
409
410 cgroup_free(tsk);
411 task_numa_free(tsk);
412 security_task_free(tsk);
413 exit_creds(tsk);
414 delayacct_tsk_free(tsk);
415 put_signal_struct(tsk->signal);
416
417 if (!profile_handoff_task(tsk))
418 free_task(tsk);
419}
420EXPORT_SYMBOL_GPL(__put_task_struct);
421
422void __init __weak arch_task_cache_init(void) { }
423
424/*
425 * set_max_threads
426 */
427static void set_max_threads(unsigned int max_threads_suggested)
428{
429 u64 threads;
430
431 /*
432 * The number of threads shall be limited such that the thread
433 * structures may only consume a small part of the available memory.
434 */
435 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
436 threads = MAX_THREADS;
437 else
438 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
439 (u64) THREAD_SIZE * 8UL);
440
441 if (threads > max_threads_suggested)
442 threads = max_threads_suggested;
443
444 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
445}
446
447#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
448/* Initialized by the architecture: */
449int arch_task_struct_size __read_mostly;
450#endif
451
452void __init fork_init(void)
453{
454 int i;
455#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
456#ifndef ARCH_MIN_TASKALIGN
457#define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
458#endif
459 /* create a slab on which task_structs can be allocated */
460 task_struct_cachep = kmem_cache_create("task_struct",
461 arch_task_struct_size, ARCH_MIN_TASKALIGN,
462 SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
463#endif
464
465 /* do the arch specific task caches init */
466 arch_task_cache_init();
467
468 set_max_threads(MAX_THREADS);
469
470 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
471 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
472 init_task.signal->rlim[RLIMIT_SIGPENDING] =
473 init_task.signal->rlim[RLIMIT_NPROC];
474
475 for (i = 0; i < UCOUNT_COUNTS; i++) {
476 init_user_ns.ucount_max[i] = max_threads/2;
477 }
478}
479
480int __weak arch_dup_task_struct(struct task_struct *dst,
481 struct task_struct *src)
482{
483 *dst = *src;
484 return 0;
485}
486
487void set_task_stack_end_magic(struct task_struct *tsk)
488{
489 unsigned long *stackend;
490
491 stackend = end_of_stack(tsk);
492#ifndef CONFIG_AMLOGIC_VMAP
493 *stackend = STACK_END_MAGIC; /* for overflow detection */
494#endif
495}
496
497static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
498{
499 struct task_struct *tsk;
500 unsigned long *stack;
501 struct vm_struct *stack_vm_area;
502 int err;
503
504 if (node == NUMA_NO_NODE)
505 node = tsk_fork_get_node(orig);
506 tsk = alloc_task_struct_node(node);
507 if (!tsk)
508 return NULL;
509
510 stack = alloc_thread_stack_node(tsk, node);
511 if (!stack)
512 goto free_tsk;
513
514 stack_vm_area = task_stack_vm_area(tsk);
515
516 err = arch_dup_task_struct(tsk, orig);
517
518 /*
519 * arch_dup_task_struct() clobbers the stack-related fields. Make
520 * sure they're properly initialized before using any stack-related
521 * functions again.
522 */
523 tsk->stack = stack;
524
525 err= kaiser_map_thread_stack(tsk->stack);
526 if (err)
527 goto free_stack;
528#ifdef CONFIG_VMAP_STACK
529 tsk->stack_vm_area = stack_vm_area;
530#endif
531#ifdef CONFIG_THREAD_INFO_IN_TASK
532 atomic_set(&tsk->stack_refcount, 1);
533#endif
534
535 if (err)
536 goto free_stack;
537
538#ifdef CONFIG_SECCOMP
539 /*
540 * We must handle setting up seccomp filters once we're under
541 * the sighand lock in case orig has changed between now and
542 * then. Until then, filter must be NULL to avoid messing up
543 * the usage counts on the error path calling free_task.
544 */
545 tsk->seccomp.filter = NULL;
546#endif
547
548 setup_thread_stack(tsk, orig);
549 clear_user_return_notifier(tsk);
550 clear_tsk_need_resched(tsk);
551 set_task_stack_end_magic(tsk);
552
553#ifdef CONFIG_CC_STACKPROTECTOR
554 tsk->stack_canary = get_random_long();
555#endif
556
557 /*
558 * One for us, one for whoever does the "release_task()" (usually
559 * parent)
560 */
561 atomic_set(&tsk->usage, 2);
562#ifdef CONFIG_BLK_DEV_IO_TRACE
563 tsk->btrace_seq = 0;
564#endif
565 tsk->splice_pipe = NULL;
566 tsk->task_frag.page = NULL;
567 tsk->wake_q.next = NULL;
568
569 account_kernel_stack(tsk, 1);
570
571 kcov_task_init(tsk);
572
573 return tsk;
574
575free_stack:
576 free_thread_stack(tsk);
577free_tsk:
578 free_task_struct(tsk);
579 return NULL;
580}
581
582#ifdef CONFIG_MMU
583static __latent_entropy int dup_mmap(struct mm_struct *mm,
584 struct mm_struct *oldmm)
585{
586 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
587 struct rb_node **rb_link, *rb_parent;
588 int retval;
589 unsigned long charge;
590
591 uprobe_start_dup_mmap();
592 if (down_write_killable(&oldmm->mmap_sem)) {
593 retval = -EINTR;
594 goto fail_uprobe_end;
595 }
596 flush_cache_dup_mm(oldmm);
597 uprobe_dup_mmap(oldmm, mm);
598 /*
599 * Not linked in yet - no deadlock potential:
600 */
601 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
602
603 /* No ordering required: file already has been exposed. */
604 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
605
606 mm->total_vm = oldmm->total_vm;
607 mm->data_vm = oldmm->data_vm;
608 mm->exec_vm = oldmm->exec_vm;
609 mm->stack_vm = oldmm->stack_vm;
610
611 rb_link = &mm->mm_rb.rb_node;
612 rb_parent = NULL;
613 pprev = &mm->mmap;
614 retval = ksm_fork(mm, oldmm);
615 if (retval)
616 goto out;
617 retval = khugepaged_fork(mm, oldmm);
618 if (retval)
619 goto out;
620
621 prev = NULL;
622 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
623 struct file *file;
624
625 if (mpnt->vm_flags & VM_DONTCOPY) {
626 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
627 continue;
628 }
629 charge = 0;
630 if (mpnt->vm_flags & VM_ACCOUNT) {
631 unsigned long len = vma_pages(mpnt);
632
633 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
634 goto fail_nomem;
635 charge = len;
636 }
637 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
638 if (!tmp)
639 goto fail_nomem;
640 *tmp = *mpnt;
641 INIT_LIST_HEAD(&tmp->anon_vma_chain);
642 retval = vma_dup_policy(mpnt, tmp);
643 if (retval)
644 goto fail_nomem_policy;
645 tmp->vm_mm = mm;
646 if (anon_vma_fork(tmp, mpnt))
647 goto fail_nomem_anon_vma_fork;
648 tmp->vm_flags &=
649 ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
650 tmp->vm_next = tmp->vm_prev = NULL;
651 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
652 file = tmp->vm_file;
653 if (file) {
654 struct inode *inode = file_inode(file);
655 struct address_space *mapping = file->f_mapping;
656
657 get_file(file);
658 if (tmp->vm_flags & VM_DENYWRITE)
659 atomic_dec(&inode->i_writecount);
660 i_mmap_lock_write(mapping);
661 if (tmp->vm_flags & VM_SHARED)
662 atomic_inc(&mapping->i_mmap_writable);
663 flush_dcache_mmap_lock(mapping);
664 /* insert tmp into the share list, just after mpnt */
665 vma_interval_tree_insert_after(tmp, mpnt,
666 &mapping->i_mmap);
667 flush_dcache_mmap_unlock(mapping);
668 i_mmap_unlock_write(mapping);
669 }
670
671 /*
672 * Clear hugetlb-related page reserves for children. This only
673 * affects MAP_PRIVATE mappings. Faults generated by the child
674 * are not guaranteed to succeed, even if read-only
675 */
676 if (is_vm_hugetlb_page(tmp))
677 reset_vma_resv_huge_pages(tmp);
678
679 /*
680 * Link in the new vma and copy the page table entries.
681 */
682 *pprev = tmp;
683 pprev = &tmp->vm_next;
684 tmp->vm_prev = prev;
685 prev = tmp;
686
687 __vma_link_rb(mm, tmp, rb_link, rb_parent);
688 rb_link = &tmp->vm_rb.rb_right;
689 rb_parent = &tmp->vm_rb;
690
691 mm->map_count++;
692 retval = copy_page_range(mm, oldmm, mpnt);
693
694 if (tmp->vm_ops && tmp->vm_ops->open)
695 tmp->vm_ops->open(tmp);
696
697 if (retval)
698 goto out;
699 }
700 /* a new mm has just been created */
701 arch_dup_mmap(oldmm, mm);
702 retval = 0;
703out:
704 up_write(&mm->mmap_sem);
705 flush_tlb_mm(oldmm);
706 up_write(&oldmm->mmap_sem);
707fail_uprobe_end:
708 uprobe_end_dup_mmap();
709 return retval;
710fail_nomem_anon_vma_fork:
711 mpol_put(vma_policy(tmp));
712fail_nomem_policy:
713 kmem_cache_free(vm_area_cachep, tmp);
714fail_nomem:
715 retval = -ENOMEM;
716 vm_unacct_memory(charge);
717 goto out;
718}
719
720static inline int mm_alloc_pgd(struct mm_struct *mm)
721{
722 mm->pgd = pgd_alloc(mm);
723 if (unlikely(!mm->pgd))
724 return -ENOMEM;
725 return 0;
726}
727
728static inline void mm_free_pgd(struct mm_struct *mm)
729{
730 pgd_free(mm, mm->pgd);
731}
732#else
733static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
734{
735 down_write(&oldmm->mmap_sem);
736 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
737 up_write(&oldmm->mmap_sem);
738 return 0;
739}
740#define mm_alloc_pgd(mm) (0)
741#define mm_free_pgd(mm)
742#endif /* CONFIG_MMU */
743
744__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
745
746#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
747#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
748
749static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
750
751static int __init coredump_filter_setup(char *s)
752{
753 default_dump_filter =
754 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
755 MMF_DUMP_FILTER_MASK;
756 return 1;
757}
758
759__setup("coredump_filter=", coredump_filter_setup);
760
761#include <linux/init_task.h>
762
763static void mm_init_aio(struct mm_struct *mm)
764{
765#ifdef CONFIG_AIO
766 spin_lock_init(&mm->ioctx_lock);
767 mm->ioctx_table = NULL;
768#endif
769}
770
771static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
772{
773#ifdef CONFIG_MEMCG
774 mm->owner = p;
775#endif
776}
777
778static void mm_init_uprobes_state(struct mm_struct *mm)
779{
780#ifdef CONFIG_UPROBES
781 mm->uprobes_state.xol_area = NULL;
782#endif
783}
784
785static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
786 struct user_namespace *user_ns)
787{
788 mm->mmap = NULL;
789 mm->mm_rb = RB_ROOT;
790 mm->vmacache_seqnum = 0;
791 atomic_set(&mm->mm_users, 1);
792 atomic_set(&mm->mm_count, 1);
793 init_rwsem(&mm->mmap_sem);
794 INIT_LIST_HEAD(&mm->mmlist);
795 mm->core_state = NULL;
796 atomic_long_set(&mm->nr_ptes, 0);
797 mm_nr_pmds_init(mm);
798 mm->map_count = 0;
799 mm->locked_vm = 0;
800 mm->pinned_vm = 0;
801 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
802 spin_lock_init(&mm->page_table_lock);
803 mm_init_cpumask(mm);
804 mm_init_aio(mm);
805 mm_init_owner(mm, p);
806 RCU_INIT_POINTER(mm->exe_file, NULL);
807 mmu_notifier_mm_init(mm);
808 clear_tlb_flush_pending(mm);
809#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
810 mm->pmd_huge_pte = NULL;
811#endif
812 mm_init_uprobes_state(mm);
813
814 if (current->mm) {
815 mm->flags = current->mm->flags & MMF_INIT_MASK;
816 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
817 } else {
818 mm->flags = default_dump_filter;
819 mm->def_flags = 0;
820 }
821
822 if (mm_alloc_pgd(mm))
823 goto fail_nopgd;
824
825 if (init_new_context(p, mm))
826 goto fail_nocontext;
827
828 mm->user_ns = get_user_ns(user_ns);
829 return mm;
830
831fail_nocontext:
832 mm_free_pgd(mm);
833fail_nopgd:
834 free_mm(mm);
835 return NULL;
836}
837
838static void check_mm(struct mm_struct *mm)
839{
840 int i;
841
842 for (i = 0; i < NR_MM_COUNTERS; i++) {
843 long x = atomic_long_read(&mm->rss_stat.count[i]);
844
845 if (unlikely(x))
846 printk(KERN_ALERT "BUG: Bad rss-counter state "
847 "mm:%p idx:%d val:%ld\n", mm, i, x);
848 }
849
850 if (atomic_long_read(&mm->nr_ptes))
851 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
852 atomic_long_read(&mm->nr_ptes));
853 if (mm_nr_pmds(mm))
854 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
855 mm_nr_pmds(mm));
856
857#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
858 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
859#endif
860}
861
862/*
863 * Allocate and initialize an mm_struct.
864 */
865struct mm_struct *mm_alloc(void)
866{
867 struct mm_struct *mm;
868
869 mm = allocate_mm();
870 if (!mm)
871 return NULL;
872
873 memset(mm, 0, sizeof(*mm));
874 return mm_init(mm, current, current_user_ns());
875}
876
877/*
878 * Called when the last reference to the mm
879 * is dropped: either by a lazy thread or by
880 * mmput. Free the page directory and the mm.
881 */
882void __mmdrop(struct mm_struct *mm)
883{
884 BUG_ON(mm == &init_mm);
885 mm_free_pgd(mm);
886 destroy_context(mm);
887 mmu_notifier_mm_destroy(mm);
888 check_mm(mm);
889 put_user_ns(mm->user_ns);
890 free_mm(mm);
891}
892EXPORT_SYMBOL_GPL(__mmdrop);
893
894static inline void __mmput(struct mm_struct *mm)
895{
896 VM_BUG_ON(atomic_read(&mm->mm_users));
897
898 uprobe_clear_state(mm);
899 exit_aio(mm);
900 ksm_exit(mm);
901 khugepaged_exit(mm); /* must run before exit_mmap */
902 exit_mmap(mm);
903 mm_put_huge_zero_page(mm);
904 set_mm_exe_file(mm, NULL);
905 if (!list_empty(&mm->mmlist)) {
906 spin_lock(&mmlist_lock);
907 list_del(&mm->mmlist);
908 spin_unlock(&mmlist_lock);
909 }
910 if (mm->binfmt)
911 module_put(mm->binfmt->module);
912 set_bit(MMF_OOM_SKIP, &mm->flags);
913 mmdrop(mm);
914}
915
916/*
917 * Decrement the use count and release all resources for an mm.
918 */
919void mmput(struct mm_struct *mm)
920{
921 might_sleep();
922
923 if (atomic_dec_and_test(&mm->mm_users))
924 __mmput(mm);
925}
926EXPORT_SYMBOL_GPL(mmput);
927
928#ifdef CONFIG_MMU
929static void mmput_async_fn(struct work_struct *work)
930{
931 struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
932 __mmput(mm);
933}
934
935void mmput_async(struct mm_struct *mm)
936{
937 if (atomic_dec_and_test(&mm->mm_users)) {
938 INIT_WORK(&mm->async_put_work, mmput_async_fn);
939 schedule_work(&mm->async_put_work);
940 }
941}
942#endif
943
944/**
945 * set_mm_exe_file - change a reference to the mm's executable file
946 *
947 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
948 *
949 * Main users are mmput() and sys_execve(). Callers prevent concurrent
950 * invocations: in mmput() nobody alive left, in execve task is single
951 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
952 * mm->exe_file, but does so without using set_mm_exe_file() in order
953 * to do avoid the need for any locks.
954 */
955void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
956{
957 struct file *old_exe_file;
958
959 /*
960 * It is safe to dereference the exe_file without RCU as
961 * this function is only called if nobody else can access
962 * this mm -- see comment above for justification.
963 */
964 old_exe_file = rcu_dereference_raw(mm->exe_file);
965
966 if (new_exe_file)
967 get_file(new_exe_file);
968 rcu_assign_pointer(mm->exe_file, new_exe_file);
969 if (old_exe_file)
970 fput(old_exe_file);
971}
972
973/**
974 * get_mm_exe_file - acquire a reference to the mm's executable file
975 *
976 * Returns %NULL if mm has no associated executable file.
977 * User must release file via fput().
978 */
979struct file *get_mm_exe_file(struct mm_struct *mm)
980{
981 struct file *exe_file;
982
983 rcu_read_lock();
984 exe_file = rcu_dereference(mm->exe_file);
985 if (exe_file && !get_file_rcu(exe_file))
986 exe_file = NULL;
987 rcu_read_unlock();
988 return exe_file;
989}
990EXPORT_SYMBOL(get_mm_exe_file);
991
992/**
993 * get_task_exe_file - acquire a reference to the task's executable file
994 *
995 * Returns %NULL if task's mm (if any) has no associated executable file or
996 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
997 * User must release file via fput().
998 */
999struct file *get_task_exe_file(struct task_struct *task)
1000{
1001 struct file *exe_file = NULL;
1002 struct mm_struct *mm;
1003
1004 task_lock(task);
1005 mm = task->mm;
1006 if (mm) {
1007 if (!(task->flags & PF_KTHREAD))
1008 exe_file = get_mm_exe_file(mm);
1009 }
1010 task_unlock(task);
1011 return exe_file;
1012}
1013EXPORT_SYMBOL(get_task_exe_file);
1014
1015/**
1016 * get_task_mm - acquire a reference to the task's mm
1017 *
1018 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1019 * this kernel workthread has transiently adopted a user mm with use_mm,
1020 * to do its AIO) is not set and if so returns a reference to it, after
1021 * bumping up the use count. User must release the mm via mmput()
1022 * after use. Typically used by /proc and ptrace.
1023 */
1024struct mm_struct *get_task_mm(struct task_struct *task)
1025{
1026 struct mm_struct *mm;
1027
1028 task_lock(task);
1029 mm = task->mm;
1030 if (mm) {
1031 if (task->flags & PF_KTHREAD)
1032 mm = NULL;
1033 else
1034 atomic_inc(&mm->mm_users);
1035 }
1036 task_unlock(task);
1037 return mm;
1038}
1039EXPORT_SYMBOL_GPL(get_task_mm);
1040
1041struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1042{
1043 struct mm_struct *mm;
1044 int err;
1045
1046 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1047 if (err)
1048 return ERR_PTR(err);
1049
1050 mm = get_task_mm(task);
1051 if (mm && mm != current->mm &&
1052 !ptrace_may_access(task, mode)) {
1053 mmput(mm);
1054 mm = ERR_PTR(-EACCES);
1055 }
1056 mutex_unlock(&task->signal->cred_guard_mutex);
1057
1058 return mm;
1059}
1060
1061static void complete_vfork_done(struct task_struct *tsk)
1062{
1063 struct completion *vfork;
1064
1065 task_lock(tsk);
1066 vfork = tsk->vfork_done;
1067 if (likely(vfork)) {
1068 tsk->vfork_done = NULL;
1069 complete(vfork);
1070 }
1071 task_unlock(tsk);
1072}
1073
1074static int wait_for_vfork_done(struct task_struct *child,
1075 struct completion *vfork)
1076{
1077 int killed;
1078
1079 freezer_do_not_count();
1080 killed = wait_for_completion_killable(vfork);
1081 freezer_count();
1082
1083 if (killed) {
1084 task_lock(child);
1085 child->vfork_done = NULL;
1086 task_unlock(child);
1087 }
1088
1089 put_task_struct(child);
1090 return killed;
1091}
1092
1093/* Please note the differences between mmput and mm_release.
1094 * mmput is called whenever we stop holding onto a mm_struct,
1095 * error success whatever.
1096 *
1097 * mm_release is called after a mm_struct has been removed
1098 * from the current process.
1099 *
1100 * This difference is important for error handling, when we
1101 * only half set up a mm_struct for a new process and need to restore
1102 * the old one. Because we mmput the new mm_struct before
1103 * restoring the old one. . .
1104 * Eric Biederman 10 January 1998
1105 */
1106void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1107{
1108 /* Get rid of any futexes when releasing the mm */
1109#ifdef CONFIG_FUTEX
1110 if (unlikely(tsk->robust_list)) {
1111 exit_robust_list(tsk);
1112 tsk->robust_list = NULL;
1113 }
1114#ifdef CONFIG_COMPAT
1115 if (unlikely(tsk->compat_robust_list)) {
1116 compat_exit_robust_list(tsk);
1117 tsk->compat_robust_list = NULL;
1118 }
1119#endif
1120 if (unlikely(!list_empty(&tsk->pi_state_list)))
1121 exit_pi_state_list(tsk);
1122#endif
1123
1124 uprobe_free_utask(tsk);
1125
1126 /* Get rid of any cached register state */
1127 deactivate_mm(tsk, mm);
1128
1129 /*
1130 * Signal userspace if we're not exiting with a core dump
1131 * because we want to leave the value intact for debugging
1132 * purposes.
1133 */
1134 if (tsk->clear_child_tid) {
1135 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1136 atomic_read(&mm->mm_users) > 1) {
1137 /*
1138 * We don't check the error code - if userspace has
1139 * not set up a proper pointer then tough luck.
1140 */
1141 put_user(0, tsk->clear_child_tid);
1142 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1143 1, NULL, NULL, 0);
1144 }
1145 tsk->clear_child_tid = NULL;
1146 }
1147
1148 /*
1149 * All done, finally we can wake up parent and return this mm to him.
1150 * Also kthread_stop() uses this completion for synchronization.
1151 */
1152 if (tsk->vfork_done)
1153 complete_vfork_done(tsk);
1154}
1155
1156/*
1157 * Allocate a new mm structure and copy contents from the
1158 * mm structure of the passed in task structure.
1159 */
1160static struct mm_struct *dup_mm(struct task_struct *tsk)
1161{
1162 struct mm_struct *mm, *oldmm = current->mm;
1163 int err;
1164
1165 mm = allocate_mm();
1166 if (!mm)
1167 goto fail_nomem;
1168
1169 memcpy(mm, oldmm, sizeof(*mm));
1170
1171 if (!mm_init(mm, tsk, mm->user_ns))
1172 goto fail_nomem;
1173
1174 err = dup_mmap(mm, oldmm);
1175 if (err)
1176 goto free_pt;
1177
1178 mm->hiwater_rss = get_mm_rss(mm);
1179 mm->hiwater_vm = mm->total_vm;
1180
1181 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1182 goto free_pt;
1183
1184 return mm;
1185
1186free_pt:
1187 /* don't put binfmt in mmput, we haven't got module yet */
1188 mm->binfmt = NULL;
1189 mmput(mm);
1190
1191fail_nomem:
1192 return NULL;
1193}
1194
1195static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1196{
1197 struct mm_struct *mm, *oldmm;
1198 int retval;
1199
1200 tsk->min_flt = tsk->maj_flt = 0;
1201 tsk->nvcsw = tsk->nivcsw = 0;
1202#ifdef CONFIG_DETECT_HUNG_TASK
1203 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1204#endif
1205
1206 tsk->mm = NULL;
1207 tsk->active_mm = NULL;
1208
1209 /*
1210 * Are we cloning a kernel thread?
1211 *
1212 * We need to steal a active VM for that..
1213 */
1214 oldmm = current->mm;
1215 if (!oldmm)
1216 return 0;
1217
1218 /* initialize the new vmacache entries */
1219 vmacache_flush(tsk);
1220
1221 if (clone_flags & CLONE_VM) {
1222 atomic_inc(&oldmm->mm_users);
1223 mm = oldmm;
1224 goto good_mm;
1225 }
1226
1227 retval = -ENOMEM;
1228 mm = dup_mm(tsk);
1229 if (!mm)
1230 goto fail_nomem;
1231
1232good_mm:
1233 tsk->mm = mm;
1234 tsk->active_mm = mm;
1235 return 0;
1236
1237fail_nomem:
1238 return retval;
1239}
1240
1241static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1242{
1243 struct fs_struct *fs = current->fs;
1244 if (clone_flags & CLONE_FS) {
1245 /* tsk->fs is already what we want */
1246 spin_lock(&fs->lock);
1247 if (fs->in_exec) {
1248 spin_unlock(&fs->lock);
1249 return -EAGAIN;
1250 }
1251 fs->users++;
1252 spin_unlock(&fs->lock);
1253 return 0;
1254 }
1255 tsk->fs = copy_fs_struct(fs);
1256 if (!tsk->fs)
1257 return -ENOMEM;
1258 return 0;
1259}
1260
1261static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1262{
1263 struct files_struct *oldf, *newf;
1264 int error = 0;
1265
1266 /*
1267 * A background process may not have any files ...
1268 */
1269 oldf = current->files;
1270 if (!oldf)
1271 goto out;
1272
1273 if (clone_flags & CLONE_FILES) {
1274 atomic_inc(&oldf->count);
1275 goto out;
1276 }
1277
1278 newf = dup_fd(oldf, &error);
1279 if (!newf)
1280 goto out;
1281
1282 tsk->files = newf;
1283 error = 0;
1284out:
1285 return error;
1286}
1287
1288static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1289{
1290#ifdef CONFIG_BLOCK
1291 struct io_context *ioc = current->io_context;
1292 struct io_context *new_ioc;
1293
1294 if (!ioc)
1295 return 0;
1296 /*
1297 * Share io context with parent, if CLONE_IO is set
1298 */
1299 if (clone_flags & CLONE_IO) {
1300 ioc_task_link(ioc);
1301 tsk->io_context = ioc;
1302 } else if (ioprio_valid(ioc->ioprio)) {
1303 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1304 if (unlikely(!new_ioc))
1305 return -ENOMEM;
1306
1307 new_ioc->ioprio = ioc->ioprio;
1308 put_io_context(new_ioc);
1309 }
1310#endif
1311 return 0;
1312}
1313
1314static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1315{
1316 struct sighand_struct *sig;
1317
1318 if (clone_flags & CLONE_SIGHAND) {
1319 atomic_inc(&current->sighand->count);
1320 return 0;
1321 }
1322 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1323 rcu_assign_pointer(tsk->sighand, sig);
1324 if (!sig)
1325 return -ENOMEM;
1326
1327 atomic_set(&sig->count, 1);
1328 spin_lock_irq(&current->sighand->siglock);
1329 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1330 spin_unlock_irq(&current->sighand->siglock);
1331 return 0;
1332}
1333
1334void __cleanup_sighand(struct sighand_struct *sighand)
1335{
1336 if (atomic_dec_and_test(&sighand->count)) {
1337 signalfd_cleanup(sighand);
1338 /*
1339 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1340 * without an RCU grace period, see __lock_task_sighand().
1341 */
1342 kmem_cache_free(sighand_cachep, sighand);
1343 }
1344}
1345
1346/*
1347 * Initialize POSIX timer handling for a thread group.
1348 */
1349static void posix_cpu_timers_init_group(struct signal_struct *sig)
1350{
1351 unsigned long cpu_limit;
1352
1353 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1354 if (cpu_limit != RLIM_INFINITY) {
1355 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1356 sig->cputimer.running = true;
1357 }
1358
1359 /* The timer lists. */
1360 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1361 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1362 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1363}
1364
1365static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1366{
1367 struct signal_struct *sig;
1368
1369 if (clone_flags & CLONE_THREAD)
1370 return 0;
1371
1372 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1373 tsk->signal = sig;
1374 if (!sig)
1375 return -ENOMEM;
1376
1377 sig->nr_threads = 1;
1378 atomic_set(&sig->live, 1);
1379 atomic_set(&sig->sigcnt, 1);
1380
1381 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1382 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1383 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1384
1385 init_waitqueue_head(&sig->wait_chldexit);
1386 sig->curr_target = tsk;
1387 init_sigpending(&sig->shared_pending);
1388 INIT_LIST_HEAD(&sig->posix_timers);
1389 seqlock_init(&sig->stats_lock);
1390 prev_cputime_init(&sig->prev_cputime);
1391
1392 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1393 sig->real_timer.function = it_real_fn;
1394
1395 task_lock(current->group_leader);
1396 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1397 task_unlock(current->group_leader);
1398
1399 posix_cpu_timers_init_group(sig);
1400
1401 tty_audit_fork(sig);
1402 sched_autogroup_fork(sig);
1403
1404 sig->oom_score_adj = current->signal->oom_score_adj;
1405 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1406
1407 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1408 current->signal->is_child_subreaper;
1409
1410 mutex_init(&sig->cred_guard_mutex);
1411
1412 return 0;
1413}
1414
1415static void copy_seccomp(struct task_struct *p)
1416{
1417#ifdef CONFIG_SECCOMP
1418 /*
1419 * Must be called with sighand->lock held, which is common to
1420 * all threads in the group. Holding cred_guard_mutex is not
1421 * needed because this new task is not yet running and cannot
1422 * be racing exec.
1423 */
1424 assert_spin_locked(&current->sighand->siglock);
1425
1426 /* Ref-count the new filter user, and assign it. */
1427 get_seccomp_filter(current);
1428 p->seccomp = current->seccomp;
1429
1430 /*
1431 * Explicitly enable no_new_privs here in case it got set
1432 * between the task_struct being duplicated and holding the
1433 * sighand lock. The seccomp state and nnp must be in sync.
1434 */
1435 if (task_no_new_privs(current))
1436 task_set_no_new_privs(p);
1437
1438 /*
1439 * If the parent gained a seccomp mode after copying thread
1440 * flags and between before we held the sighand lock, we have
1441 * to manually enable the seccomp thread flag here.
1442 */
1443 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1444 set_tsk_thread_flag(p, TIF_SECCOMP);
1445#endif
1446}
1447
1448SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1449{
1450 current->clear_child_tid = tidptr;
1451
1452 return task_pid_vnr(current);
1453}
1454
1455static void rt_mutex_init_task(struct task_struct *p)
1456{
1457 raw_spin_lock_init(&p->pi_lock);
1458#ifdef CONFIG_RT_MUTEXES
1459 p->pi_waiters = RB_ROOT;
1460 p->pi_waiters_leftmost = NULL;
1461 p->pi_blocked_on = NULL;
1462#endif
1463}
1464
1465/*
1466 * Initialize POSIX timer handling for a single task.
1467 */
1468static void posix_cpu_timers_init(struct task_struct *tsk)
1469{
1470 tsk->cputime_expires.prof_exp = 0;
1471 tsk->cputime_expires.virt_exp = 0;
1472 tsk->cputime_expires.sched_exp = 0;
1473 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1474 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1475 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1476}
1477
1478static inline void
1479init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1480{
1481 task->pids[type].pid = pid;
1482}
1483
1484/*
1485 * This creates a new process as a copy of the old one,
1486 * but does not actually start it yet.
1487 *
1488 * It copies the registers, and all the appropriate
1489 * parts of the process environment (as per the clone
1490 * flags). The actual kick-off is left to the caller.
1491 */
1492static __latent_entropy struct task_struct *copy_process(
1493 unsigned long clone_flags,
1494 unsigned long stack_start,
1495 unsigned long stack_size,
1496 int __user *child_tidptr,
1497 struct pid *pid,
1498 int trace,
1499 unsigned long tls,
1500 int node)
1501{
1502 int retval;
1503 struct task_struct *p;
1504
1505 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1506 return ERR_PTR(-EINVAL);
1507
1508 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1509 return ERR_PTR(-EINVAL);
1510
1511 /*
1512 * Thread groups must share signals as well, and detached threads
1513 * can only be started up within the thread group.
1514 */
1515 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1516 return ERR_PTR(-EINVAL);
1517
1518 /*
1519 * Shared signal handlers imply shared VM. By way of the above,
1520 * thread groups also imply shared VM. Blocking this case allows
1521 * for various simplifications in other code.
1522 */
1523 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1524 return ERR_PTR(-EINVAL);
1525
1526 /*
1527 * Siblings of global init remain as zombies on exit since they are
1528 * not reaped by their parent (swapper). To solve this and to avoid
1529 * multi-rooted process trees, prevent global and container-inits
1530 * from creating siblings.
1531 */
1532 if ((clone_flags & CLONE_PARENT) &&
1533 current->signal->flags & SIGNAL_UNKILLABLE)
1534 return ERR_PTR(-EINVAL);
1535
1536 /*
1537 * If the new process will be in a different pid or user namespace
1538 * do not allow it to share a thread group with the forking task.
1539 */
1540 if (clone_flags & CLONE_THREAD) {
1541 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1542 (task_active_pid_ns(current) !=
1543 current->nsproxy->pid_ns_for_children))
1544 return ERR_PTR(-EINVAL);
1545 }
1546
1547 retval = security_task_create(clone_flags);
1548 if (retval)
1549 goto fork_out;
1550
1551 retval = -ENOMEM;
1552 p = dup_task_struct(current, node);
1553 if (!p)
1554 goto fork_out;
1555
1556 cpufreq_task_times_init(p);
1557
1558 /*
1559 * This _must_ happen before we call free_task(), i.e. before we jump
1560 * to any of the bad_fork_* labels. This is to avoid freeing
1561 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1562 * kernel threads (PF_KTHREAD).
1563 */
1564 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1565 /*
1566 * Clear TID on mm_release()?
1567 */
1568 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1569
1570 ftrace_graph_init_task(p);
1571
1572 rt_mutex_init_task(p);
1573
1574#ifdef CONFIG_PROVE_LOCKING
1575 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1576 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1577#endif
1578 retval = -EAGAIN;
1579 if (atomic_read(&p->real_cred->user->processes) >=
1580 task_rlimit(p, RLIMIT_NPROC)) {
1581 if (p->real_cred->user != INIT_USER &&
1582 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1583 goto bad_fork_free;
1584 }
1585 current->flags &= ~PF_NPROC_EXCEEDED;
1586
1587 retval = copy_creds(p, clone_flags);
1588 if (retval < 0)
1589 goto bad_fork_free;
1590
1591 /*
1592 * If multiple threads are within copy_process(), then this check
1593 * triggers too late. This doesn't hurt, the check is only there
1594 * to stop root fork bombs.
1595 */
1596 retval = -EAGAIN;
1597 if (nr_threads >= max_threads)
1598 goto bad_fork_cleanup_count;
1599
1600 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1601 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1602 p->flags |= PF_FORKNOEXEC;
1603 INIT_LIST_HEAD(&p->children);
1604 INIT_LIST_HEAD(&p->sibling);
1605 rcu_copy_process(p);
1606 p->vfork_done = NULL;
1607 spin_lock_init(&p->alloc_lock);
1608
1609 init_sigpending(&p->pending);
1610
1611 p->utime = p->stime = p->gtime = 0;
1612 p->utimescaled = p->stimescaled = 0;
1613 prev_cputime_init(&p->prev_cputime);
1614
1615#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1616 seqcount_init(&p->vtime_seqcount);
1617 p->vtime_snap = 0;
1618 p->vtime_snap_whence = VTIME_INACTIVE;
1619#endif
1620
1621#if defined(SPLIT_RSS_COUNTING)
1622 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1623#endif
1624
1625 p->default_timer_slack_ns = current->timer_slack_ns;
1626
1627#ifdef CONFIG_PSI
1628 p->psi_flags = 0;
1629#endif
1630
1631 task_io_accounting_init(&p->ioac);
1632 acct_clear_integrals(p);
1633
1634 posix_cpu_timers_init(p);
1635
1636 p->io_context = NULL;
1637 p->audit_context = NULL;
1638 cgroup_fork(p);
1639#ifdef CONFIG_NUMA
1640 p->mempolicy = mpol_dup(p->mempolicy);
1641 if (IS_ERR(p->mempolicy)) {
1642 retval = PTR_ERR(p->mempolicy);
1643 p->mempolicy = NULL;
1644 goto bad_fork_cleanup_threadgroup_lock;
1645 }
1646#endif
1647#ifdef CONFIG_CPUSETS
1648 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1649 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1650 seqcount_init(&p->mems_allowed_seq);
1651#endif
1652#ifdef CONFIG_TRACE_IRQFLAGS
1653 p->irq_events = 0;
1654 p->hardirqs_enabled = 0;
1655 p->hardirq_enable_ip = 0;
1656 p->hardirq_enable_event = 0;
1657 p->hardirq_disable_ip = _THIS_IP_;
1658 p->hardirq_disable_event = 0;
1659 p->softirqs_enabled = 1;
1660 p->softirq_enable_ip = _THIS_IP_;
1661 p->softirq_enable_event = 0;
1662 p->softirq_disable_ip = 0;
1663 p->softirq_disable_event = 0;
1664 p->hardirq_context = 0;
1665 p->softirq_context = 0;
1666#endif
1667
1668 p->pagefault_disabled = 0;
1669
1670#ifdef CONFIG_LOCKDEP
1671 p->lockdep_depth = 0; /* no locks held yet */
1672 p->curr_chain_key = 0;
1673 p->lockdep_recursion = 0;
1674#endif
1675
1676#ifdef CONFIG_DEBUG_MUTEXES
1677 p->blocked_on = NULL; /* not blocked yet */
1678#endif
1679#ifdef CONFIG_BCACHE
1680 p->sequential_io = 0;
1681 p->sequential_io_avg = 0;
1682#endif
1683
1684 /* Perform scheduler related setup. Assign this task to a CPU. */
1685 retval = sched_fork(clone_flags, p);
1686 if (retval)
1687 goto bad_fork_cleanup_policy;
1688
1689 retval = perf_event_init_task(p);
1690 if (retval)
1691 goto bad_fork_cleanup_policy;
1692 retval = audit_alloc(p);
1693 if (retval)
1694 goto bad_fork_cleanup_perf;
1695 /* copy all the process information */
1696 shm_init_task(p);
1697 retval = copy_semundo(clone_flags, p);
1698 if (retval)
1699 goto bad_fork_cleanup_audit;
1700 retval = copy_files(clone_flags, p);
1701 if (retval)
1702 goto bad_fork_cleanup_semundo;
1703 retval = copy_fs(clone_flags, p);
1704 if (retval)
1705 goto bad_fork_cleanup_files;
1706 retval = copy_sighand(clone_flags, p);
1707 if (retval)
1708 goto bad_fork_cleanup_fs;
1709 retval = copy_signal(clone_flags, p);
1710 if (retval)
1711 goto bad_fork_cleanup_sighand;
1712 retval = copy_mm(clone_flags, p);
1713 if (retval)
1714 goto bad_fork_cleanup_signal;
1715 retval = copy_namespaces(clone_flags, p);
1716 if (retval)
1717 goto bad_fork_cleanup_mm;
1718 retval = copy_io(clone_flags, p);
1719 if (retval)
1720 goto bad_fork_cleanup_namespaces;
1721 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1722 if (retval)
1723 goto bad_fork_cleanup_io;
1724
1725 if (pid != &init_struct_pid) {
1726 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1727 if (IS_ERR(pid)) {
1728 retval = PTR_ERR(pid);
1729 goto bad_fork_cleanup_thread;
1730 }
1731 }
1732
1733#ifdef CONFIG_BLOCK
1734 p->plug = NULL;
1735#endif
1736#ifdef CONFIG_FUTEX
1737 p->robust_list = NULL;
1738#ifdef CONFIG_COMPAT
1739 p->compat_robust_list = NULL;
1740#endif
1741 INIT_LIST_HEAD(&p->pi_state_list);
1742 p->pi_state_cache = NULL;
1743#endif
1744 /*
1745 * sigaltstack should be cleared when sharing the same VM
1746 */
1747 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1748 sas_ss_reset(p);
1749
1750 /*
1751 * Syscall tracing and stepping should be turned off in the
1752 * child regardless of CLONE_PTRACE.
1753 */
1754 user_disable_single_step(p);
1755 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1756#ifdef TIF_SYSCALL_EMU
1757 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1758#endif
1759 clear_all_latency_tracing(p);
1760
1761 /* ok, now we should be set up.. */
1762 p->pid = pid_nr(pid);
1763 if (clone_flags & CLONE_THREAD) {
1764 p->exit_signal = -1;
1765 p->group_leader = current->group_leader;
1766 p->tgid = current->tgid;
1767 } else {
1768 if (clone_flags & CLONE_PARENT)
1769 p->exit_signal = current->group_leader->exit_signal;
1770 else
1771 p->exit_signal = (clone_flags & CSIGNAL);
1772 p->group_leader = p;
1773 p->tgid = p->pid;
1774 }
1775
1776 p->nr_dirtied = 0;
1777 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1778 p->dirty_paused_when = 0;
1779
1780 p->pdeath_signal = 0;
1781 INIT_LIST_HEAD(&p->thread_group);
1782 p->task_works = NULL;
1783
1784 threadgroup_change_begin(current);
1785 /*
1786 * Ensure that the cgroup subsystem policies allow the new process to be
1787 * forked. It should be noted the the new process's css_set can be changed
1788 * between here and cgroup_post_fork() if an organisation operation is in
1789 * progress.
1790 */
1791 retval = cgroup_can_fork(p);
1792 if (retval)
1793 goto bad_fork_free_pid;
1794
1795 /*
1796 * From this point on we must avoid any synchronous user-space
1797 * communication until we take the tasklist-lock. In particular, we do
1798 * not want user-space to be able to predict the process start-time by
1799 * stalling fork(2) after we recorded the start_time but before it is
1800 * visible to the system.
1801 */
1802
1803 p->start_time = ktime_get_ns();
1804 p->real_start_time = ktime_get_boot_ns();
1805
1806 /*
1807 * Make it visible to the rest of the system, but dont wake it up yet.
1808 * Need tasklist lock for parent etc handling!
1809 */
1810 write_lock_irq(&tasklist_lock);
1811
1812 /* CLONE_PARENT re-uses the old parent */
1813 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1814 p->real_parent = current->real_parent;
1815 p->parent_exec_id = current->parent_exec_id;
1816 } else {
1817 p->real_parent = current;
1818 p->parent_exec_id = current->self_exec_id;
1819 }
1820
1821 spin_lock(&current->sighand->siglock);
1822
1823 /*
1824 * Copy seccomp details explicitly here, in case they were changed
1825 * before holding sighand lock.
1826 */
1827 copy_seccomp(p);
1828
1829 /*
1830 * Process group and session signals need to be delivered to just the
1831 * parent before the fork or both the parent and the child after the
1832 * fork. Restart if a signal comes in before we add the new process to
1833 * it's process group.
1834 * A fatal signal pending means that current will exit, so the new
1835 * thread can't slip out of an OOM kill (or normal SIGKILL).
1836 */
1837 recalc_sigpending();
1838 if (signal_pending(current)) {
1839 retval = -ERESTARTNOINTR;
1840 goto bad_fork_cancel_cgroup;
1841 }
1842 if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1843 retval = -ENOMEM;
1844 goto bad_fork_cancel_cgroup;
1845 }
1846
1847 if (likely(p->pid)) {
1848 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1849
1850 init_task_pid(p, PIDTYPE_PID, pid);
1851 if (thread_group_leader(p)) {
1852 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1853 init_task_pid(p, PIDTYPE_SID, task_session(current));
1854
1855 if (is_child_reaper(pid)) {
1856 ns_of_pid(pid)->child_reaper = p;
1857 p->signal->flags |= SIGNAL_UNKILLABLE;
1858 }
1859
1860 p->signal->leader_pid = pid;
1861 p->signal->tty = tty_kref_get(current->signal->tty);
1862 list_add_tail(&p->sibling, &p->real_parent->children);
1863 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1864 attach_pid(p, PIDTYPE_PGID);
1865 attach_pid(p, PIDTYPE_SID);
1866 __this_cpu_inc(process_counts);
1867 } else {
1868 current->signal->nr_threads++;
1869 atomic_inc(&current->signal->live);
1870 atomic_inc(&current->signal->sigcnt);
1871 list_add_tail_rcu(&p->thread_group,
1872 &p->group_leader->thread_group);
1873 list_add_tail_rcu(&p->thread_node,
1874 &p->signal->thread_head);
1875 }
1876 attach_pid(p, PIDTYPE_PID);
1877 nr_threads++;
1878 }
1879
1880 total_forks++;
1881 spin_unlock(&current->sighand->siglock);
1882 syscall_tracepoint_update(p);
1883 write_unlock_irq(&tasklist_lock);
1884
1885 proc_fork_connector(p);
1886 cgroup_post_fork(p);
1887 threadgroup_change_end(current);
1888 perf_event_fork(p);
1889
1890 trace_task_newtask(p, clone_flags);
1891 uprobe_copy_process(p, clone_flags);
1892
1893 return p;
1894
1895bad_fork_cancel_cgroup:
1896 spin_unlock(&current->sighand->siglock);
1897 write_unlock_irq(&tasklist_lock);
1898 cgroup_cancel_fork(p);
1899bad_fork_free_pid:
1900 threadgroup_change_end(current);
1901 if (pid != &init_struct_pid)
1902 free_pid(pid);
1903bad_fork_cleanup_thread:
1904 exit_thread(p);
1905bad_fork_cleanup_io:
1906 if (p->io_context)
1907 exit_io_context(p);
1908bad_fork_cleanup_namespaces:
1909 exit_task_namespaces(p);
1910bad_fork_cleanup_mm:
1911 if (p->mm)
1912 mmput(p->mm);
1913bad_fork_cleanup_signal:
1914 if (!(clone_flags & CLONE_THREAD))
1915 free_signal_struct(p->signal);
1916bad_fork_cleanup_sighand:
1917 __cleanup_sighand(p->sighand);
1918bad_fork_cleanup_fs:
1919 exit_fs(p); /* blocking */
1920bad_fork_cleanup_files:
1921 exit_files(p); /* blocking */
1922bad_fork_cleanup_semundo:
1923 exit_sem(p);
1924bad_fork_cleanup_audit:
1925 audit_free(p);
1926bad_fork_cleanup_perf:
1927 perf_event_free_task(p);
1928bad_fork_cleanup_policy:
1929#ifdef CONFIG_NUMA
1930 mpol_put(p->mempolicy);
1931bad_fork_cleanup_threadgroup_lock:
1932#endif
1933 delayacct_tsk_free(p);
1934bad_fork_cleanup_count:
1935 atomic_dec(&p->cred->user->processes);
1936 exit_creds(p);
1937bad_fork_free:
1938 p->state = TASK_DEAD;
1939 put_task_stack(p);
1940 free_task(p);
1941fork_out:
1942 return ERR_PTR(retval);
1943}
1944
1945static inline void init_idle_pids(struct pid_link *links)
1946{
1947 enum pid_type type;
1948
1949 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1950 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1951 links[type].pid = &init_struct_pid;
1952 }
1953}
1954
1955struct task_struct *fork_idle(int cpu)
1956{
1957 struct task_struct *task;
1958 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1959 cpu_to_node(cpu));
1960 if (!IS_ERR(task)) {
1961 init_idle_pids(task->pids);
1962 init_idle(task, cpu);
1963 }
1964
1965 return task;
1966}
1967
1968/*
1969 * Ok, this is the main fork-routine.
1970 *
1971 * It copies the process, and if successful kick-starts
1972 * it and waits for it to finish using the VM if required.
1973 */
1974long _do_fork(unsigned long clone_flags,
1975 unsigned long stack_start,
1976 unsigned long stack_size,
1977 int __user *parent_tidptr,
1978 int __user *child_tidptr,
1979 unsigned long tls)
1980{
1981 struct task_struct *p;
1982 int trace = 0;
1983 long nr;
1984
1985 /*
1986 * Determine whether and which event to report to ptracer. When
1987 * called from kernel_thread or CLONE_UNTRACED is explicitly
1988 * requested, no event is reported; otherwise, report if the event
1989 * for the type of forking is enabled.
1990 */
1991 if (!(clone_flags & CLONE_UNTRACED)) {
1992 if (clone_flags & CLONE_VFORK)
1993 trace = PTRACE_EVENT_VFORK;
1994 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1995 trace = PTRACE_EVENT_CLONE;
1996 else
1997 trace = PTRACE_EVENT_FORK;
1998
1999 if (likely(!ptrace_event_enabled(current, trace)))
2000 trace = 0;
2001 }
2002
2003 p = copy_process(clone_flags, stack_start, stack_size,
2004 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2005 add_latent_entropy();
2006 /*
2007 * Do this prior waking up the new thread - the thread pointer
2008 * might get invalid after that point, if the thread exits quickly.
2009 */
2010 if (!IS_ERR(p)) {
2011 struct completion vfork;
2012 struct pid *pid;
2013
2014 cpufreq_task_times_alloc(p);
2015
2016 trace_sched_process_fork(current, p);
2017
2018 pid = get_task_pid(p, PIDTYPE_PID);
2019 nr = pid_vnr(pid);
2020
2021 if (clone_flags & CLONE_PARENT_SETTID)
2022 put_user(nr, parent_tidptr);
2023
2024 if (clone_flags & CLONE_VFORK) {
2025 p->vfork_done = &vfork;
2026 init_completion(&vfork);
2027 get_task_struct(p);
2028 }
2029
2030 wake_up_new_task(p);
2031
2032 /* forking complete and child started to run, tell ptracer */
2033 if (unlikely(trace))
2034 ptrace_event_pid(trace, pid);
2035
2036 if (clone_flags & CLONE_VFORK) {
2037 if (!wait_for_vfork_done(p, &vfork))
2038 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2039 }
2040
2041 put_pid(pid);
2042 } else {
2043 nr = PTR_ERR(p);
2044 }
2045 return nr;
2046}
2047
2048#ifndef CONFIG_HAVE_COPY_THREAD_TLS
2049/* For compatibility with architectures that call do_fork directly rather than
2050 * using the syscall entry points below. */
2051long do_fork(unsigned long clone_flags,
2052 unsigned long stack_start,
2053 unsigned long stack_size,
2054 int __user *parent_tidptr,
2055 int __user *child_tidptr)
2056{
2057 return _do_fork(clone_flags, stack_start, stack_size,
2058 parent_tidptr, child_tidptr, 0);
2059}
2060#endif
2061
2062/*
2063 * Create a kernel thread.
2064 */
2065pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2066{
2067 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2068 (unsigned long)arg, NULL, NULL, 0);
2069}
2070
2071#ifdef __ARCH_WANT_SYS_FORK
2072SYSCALL_DEFINE0(fork)
2073{
2074#ifdef CONFIG_MMU
2075 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2076#else
2077 /* can not support in nommu mode */
2078 return -EINVAL;
2079#endif
2080}
2081#endif
2082
2083#ifdef __ARCH_WANT_SYS_VFORK
2084SYSCALL_DEFINE0(vfork)
2085{
2086 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2087 0, NULL, NULL, 0);
2088}
2089#endif
2090
2091#ifdef __ARCH_WANT_SYS_CLONE
2092#ifdef CONFIG_CLONE_BACKWARDS
2093SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2094 int __user *, parent_tidptr,
2095 unsigned long, tls,
2096 int __user *, child_tidptr)
2097#elif defined(CONFIG_CLONE_BACKWARDS2)
2098SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2099 int __user *, parent_tidptr,
2100 int __user *, child_tidptr,
2101 unsigned long, tls)
2102#elif defined(CONFIG_CLONE_BACKWARDS3)
2103SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2104 int, stack_size,
2105 int __user *, parent_tidptr,
2106 int __user *, child_tidptr,
2107 unsigned long, tls)
2108#else
2109SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2110 int __user *, parent_tidptr,
2111 int __user *, child_tidptr,
2112 unsigned long, tls)
2113#endif
2114{
2115 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2116}
2117#endif
2118
2119#ifndef ARCH_MIN_MMSTRUCT_ALIGN
2120#define ARCH_MIN_MMSTRUCT_ALIGN 0
2121#endif
2122
2123static void sighand_ctor(void *data)
2124{
2125 struct sighand_struct *sighand = data;
2126
2127 spin_lock_init(&sighand->siglock);
2128 init_waitqueue_head(&sighand->signalfd_wqh);
2129}
2130
2131void __init proc_caches_init(void)
2132{
2133 sighand_cachep = kmem_cache_create("sighand_cache",
2134 sizeof(struct sighand_struct), 0,
2135 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2136 SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2137 signal_cachep = kmem_cache_create("signal_cache",
2138 sizeof(struct signal_struct), 0,
2139 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2140 NULL);
2141 files_cachep = kmem_cache_create("files_cache",
2142 sizeof(struct files_struct), 0,
2143 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2144 NULL);
2145 fs_cachep = kmem_cache_create("fs_cache",
2146 sizeof(struct fs_struct), 0,
2147 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2148 NULL);
2149 /*
2150 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2151 * whole struct cpumask for the OFFSTACK case. We could change
2152 * this to *only* allocate as much of it as required by the
2153 * maximum number of CPU's we can ever have. The cpumask_allocation
2154 * is at the end of the structure, exactly for that reason.
2155 */
2156 mm_cachep = kmem_cache_create("mm_struct",
2157 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2158 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2159 NULL);
2160 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2161 mmap_init();
2162 nsproxy_cache_init();
2163}
2164
2165/*
2166 * Check constraints on flags passed to the unshare system call.
2167 */
2168static int check_unshare_flags(unsigned long unshare_flags)
2169{
2170 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2171 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2172 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2173 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2174 return -EINVAL;
2175 /*
2176 * Not implemented, but pretend it works if there is nothing
2177 * to unshare. Note that unsharing the address space or the
2178 * signal handlers also need to unshare the signal queues (aka
2179 * CLONE_THREAD).
2180 */
2181 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2182 if (!thread_group_empty(current))
2183 return -EINVAL;
2184 }
2185 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2186 if (atomic_read(&current->sighand->count) > 1)
2187 return -EINVAL;
2188 }
2189 if (unshare_flags & CLONE_VM) {
2190 if (!current_is_single_threaded())
2191 return -EINVAL;
2192 }
2193
2194 return 0;
2195}
2196
2197/*
2198 * Unshare the filesystem structure if it is being shared
2199 */
2200static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2201{
2202 struct fs_struct *fs = current->fs;
2203
2204 if (!(unshare_flags & CLONE_FS) || !fs)
2205 return 0;
2206
2207 /* don't need lock here; in the worst case we'll do useless copy */
2208 if (fs->users == 1)
2209 return 0;
2210
2211 *new_fsp = copy_fs_struct(fs);
2212 if (!*new_fsp)
2213 return -ENOMEM;
2214
2215 return 0;
2216}
2217
2218/*
2219 * Unshare file descriptor table if it is being shared
2220 */
2221static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2222{
2223 struct files_struct *fd = current->files;
2224 int error = 0;
2225
2226 if ((unshare_flags & CLONE_FILES) &&
2227 (fd && atomic_read(&fd->count) > 1)) {
2228 *new_fdp = dup_fd(fd, &error);
2229 if (!*new_fdp)
2230 return error;
2231 }
2232
2233 return 0;
2234}
2235
2236/*
2237 * unshare allows a process to 'unshare' part of the process
2238 * context which was originally shared using clone. copy_*
2239 * functions used by do_fork() cannot be used here directly
2240 * because they modify an inactive task_struct that is being
2241 * constructed. Here we are modifying the current, active,
2242 * task_struct.
2243 */
2244SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2245{
2246 struct fs_struct *fs, *new_fs = NULL;
2247 struct files_struct *fd, *new_fd = NULL;
2248 struct cred *new_cred = NULL;
2249 struct nsproxy *new_nsproxy = NULL;
2250 int do_sysvsem = 0;
2251 int err;
2252
2253 /*
2254 * If unsharing a user namespace must also unshare the thread group
2255 * and unshare the filesystem root and working directories.
2256 */
2257 if (unshare_flags & CLONE_NEWUSER)
2258 unshare_flags |= CLONE_THREAD | CLONE_FS;
2259 /*
2260 * If unsharing vm, must also unshare signal handlers.
2261 */
2262 if (unshare_flags & CLONE_VM)
2263 unshare_flags |= CLONE_SIGHAND;
2264 /*
2265 * If unsharing a signal handlers, must also unshare the signal queues.
2266 */
2267 if (unshare_flags & CLONE_SIGHAND)
2268 unshare_flags |= CLONE_THREAD;
2269 /*
2270 * If unsharing namespace, must also unshare filesystem information.
2271 */
2272 if (unshare_flags & CLONE_NEWNS)
2273 unshare_flags |= CLONE_FS;
2274
2275 err = check_unshare_flags(unshare_flags);
2276 if (err)
2277 goto bad_unshare_out;
2278 /*
2279 * CLONE_NEWIPC must also detach from the undolist: after switching
2280 * to a new ipc namespace, the semaphore arrays from the old
2281 * namespace are unreachable.
2282 */
2283 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2284 do_sysvsem = 1;
2285 err = unshare_fs(unshare_flags, &new_fs);
2286 if (err)
2287 goto bad_unshare_out;
2288 err = unshare_fd(unshare_flags, &new_fd);
2289 if (err)
2290 goto bad_unshare_cleanup_fs;
2291 err = unshare_userns(unshare_flags, &new_cred);
2292 if (err)
2293 goto bad_unshare_cleanup_fd;
2294 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2295 new_cred, new_fs);
2296 if (err)
2297 goto bad_unshare_cleanup_cred;
2298
2299 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2300 if (do_sysvsem) {
2301 /*
2302 * CLONE_SYSVSEM is equivalent to sys_exit().
2303 */
2304 exit_sem(current);
2305 }
2306 if (unshare_flags & CLONE_NEWIPC) {
2307 /* Orphan segments in old ns (see sem above). */
2308 exit_shm(current);
2309 shm_init_task(current);
2310 }
2311
2312 if (new_nsproxy)
2313 switch_task_namespaces(current, new_nsproxy);
2314
2315 task_lock(current);
2316
2317 if (new_fs) {
2318 fs = current->fs;
2319 spin_lock(&fs->lock);
2320 current->fs = new_fs;
2321 if (--fs->users)
2322 new_fs = NULL;
2323 else
2324 new_fs = fs;
2325 spin_unlock(&fs->lock);
2326 }
2327
2328 if (new_fd) {
2329 fd = current->files;
2330 current->files = new_fd;
2331 new_fd = fd;
2332 }
2333
2334 task_unlock(current);
2335
2336 if (new_cred) {
2337 /* Install the new user namespace */
2338 commit_creds(new_cred);
2339 new_cred = NULL;
2340 }
2341 }
2342
2343bad_unshare_cleanup_cred:
2344 if (new_cred)
2345 put_cred(new_cred);
2346bad_unshare_cleanup_fd:
2347 if (new_fd)
2348 put_files_struct(new_fd);
2349
2350bad_unshare_cleanup_fs:
2351 if (new_fs)
2352 free_fs_struct(new_fs);
2353
2354bad_unshare_out:
2355 return err;
2356}
2357
2358/*
2359 * Helper to unshare the files of the current task.
2360 * We don't want to expose copy_files internals to
2361 * the exec layer of the kernel.
2362 */
2363
2364int unshare_files(struct files_struct **displaced)
2365{
2366 struct task_struct *task = current;
2367 struct files_struct *copy = NULL;
2368 int error;
2369
2370 error = unshare_fd(CLONE_FILES, &copy);
2371 if (error || !copy) {
2372 *displaced = NULL;
2373 return error;
2374 }
2375 *displaced = task->files;
2376 task_lock(task);
2377 task->files = copy;
2378 task_unlock(task);
2379 return 0;
2380}
2381
2382int sysctl_max_threads(struct ctl_table *table, int write,
2383 void __user *buffer, size_t *lenp, loff_t *ppos)
2384{
2385 struct ctl_table t;
2386 int ret;
2387 int threads = max_threads;
2388 int min = MIN_THREADS;
2389 int max = MAX_THREADS;
2390
2391 t = *table;
2392 t.data = &threads;
2393 t.extra1 = &min;
2394 t.extra2 = &max;
2395
2396 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2397 if (ret || !write)
2398 return ret;
2399
2400 set_max_threads(threads);
2401
2402 return 0;
2403}
2404
generated by cgit at 2024-05-18 17:53:08 (GMT)