PageRenderTime 75ms CodeModel.GetById 34ms RepoModel.GetById 0ms app.codeStats 1ms

/kernel/fork.c

https://bitbucket.org/bradfa/linux
C | 1935 lines | 1390 code | 268 blank | 277 comment | 192 complexity | 7ff7a4118d2bafe0b83289c9b5b0f610 MD5 | raw file
Possible License(s): GPL-2.0, LGPL-2.0, AGPL-1.0
  1. /*
  2. * linux/kernel/fork.c
  3. *
  4. * Copyright (C) 1991, 1992 Linus Torvalds
  5. */
  6. /*
  7. * 'fork.c' contains the help-routines for the 'fork' system call
  8. * (see also entry.S and others).
  9. * Fork is rather simple, once you get the hang of it, but the memory
  10. * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  11. */
  12. #include <linux/slab.h>
  13. #include <linux/init.h>
  14. #include <linux/unistd.h>
  15. #include <linux/module.h>
  16. #include <linux/vmalloc.h>
  17. #include <linux/completion.h>
  18. #include <linux/personality.h>
  19. #include <linux/mempolicy.h>
  20. #include <linux/sem.h>
  21. #include <linux/file.h>
  22. #include <linux/fdtable.h>
  23. #include <linux/iocontext.h>
  24. #include <linux/key.h>
  25. #include <linux/binfmts.h>
  26. #include <linux/mman.h>
  27. #include <linux/mmu_notifier.h>
  28. #include <linux/fs.h>
  29. #include <linux/nsproxy.h>
  30. #include <linux/capability.h>
  31. #include <linux/cpu.h>
  32. #include <linux/cgroup.h>
  33. #include <linux/security.h>
  34. #include <linux/hugetlb.h>
  35. #include <linux/seccomp.h>
  36. #include <linux/swap.h>
  37. #include <linux/syscalls.h>
  38. #include <linux/jiffies.h>
  39. #include <linux/futex.h>
  40. #include <linux/compat.h>
  41. #include <linux/kthread.h>
  42. #include <linux/task_io_accounting_ops.h>
  43. #include <linux/rcupdate.h>
  44. #include <linux/ptrace.h>
  45. #include <linux/mount.h>
  46. #include <linux/audit.h>
  47. #include <linux/memcontrol.h>
  48. #include <linux/ftrace.h>
  49. #include <linux/proc_fs.h>
  50. #include <linux/profile.h>
  51. #include <linux/rmap.h>
  52. #include <linux/ksm.h>
  53. #include <linux/acct.h>
  54. #include <linux/tsacct_kern.h>
  55. #include <linux/cn_proc.h>
  56. #include <linux/freezer.h>
  57. #include <linux/delayacct.h>
  58. #include <linux/taskstats_kern.h>
  59. #include <linux/random.h>
  60. #include <linux/tty.h>
  61. #include <linux/blkdev.h>
  62. #include <linux/fs_struct.h>
  63. #include <linux/magic.h>
  64. #include <linux/perf_event.h>
  65. #include <linux/posix-timers.h>
  66. #include <linux/user-return-notifier.h>
  67. #include <linux/oom.h>
  68. #include <linux/khugepaged.h>
  69. #include <linux/signalfd.h>
  70. #include <linux/uprobes.h>
  71. #include <asm/pgtable.h>
  72. #include <asm/pgalloc.h>
  73. #include <asm/uaccess.h>
  74. #include <asm/mmu_context.h>
  75. #include <asm/cacheflush.h>
  76. #include <asm/tlbflush.h>
  77. #include <trace/events/sched.h>
  78. #define CREATE_TRACE_POINTS
  79. #include <trace/events/task.h>
  80. /*
  81. * Protected counters by write_lock_irq(&tasklist_lock)
  82. */
  83. unsigned long total_forks; /* Handle normal Linux uptimes. */
  84. int nr_threads; /* The idle threads do not count.. */
  85. int max_threads; /* tunable limit on nr_threads */
  86. DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  87. __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
  88. #ifdef CONFIG_PROVE_RCU
  89. int lockdep_tasklist_lock_is_held(void)
  90. {
  91. return lockdep_is_held(&tasklist_lock);
  92. }
  93. EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
  94. #endif /* #ifdef CONFIG_PROVE_RCU */
  95. int nr_processes(void)
  96. {
  97. int cpu;
  98. int total = 0;
  99. for_each_possible_cpu(cpu)
  100. total += per_cpu(process_counts, cpu);
  101. return total;
  102. }
  103. void __weak arch_release_task_struct(struct task_struct *tsk)
  104. {
  105. }
  106. #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
  107. static struct kmem_cache *task_struct_cachep;
  108. static inline struct task_struct *alloc_task_struct_node(int node)
  109. {
  110. return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
  111. }
  112. static inline void free_task_struct(struct task_struct *tsk)
  113. {
  114. kmem_cache_free(task_struct_cachep, tsk);
  115. }
  116. #endif
  117. void __weak arch_release_thread_info(struct thread_info *ti)
  118. {
  119. }
  120. #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
  121. /*
  122. * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
  123. * kmemcache based allocator.
  124. */
  125. # if THREAD_SIZE >= PAGE_SIZE
  126. static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
  127. int node)
  128. {
  129. struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
  130. THREAD_SIZE_ORDER);
  131. return page ? page_address(page) : NULL;
  132. }
  133. static inline void free_thread_info(struct thread_info *ti)
  134. {
  135. free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
  136. }
  137. # else
  138. static struct kmem_cache *thread_info_cache;
  139. static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
  140. int node)
  141. {
  142. return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
  143. }
  144. static void free_thread_info(struct thread_info *ti)
  145. {
  146. kmem_cache_free(thread_info_cache, ti);
  147. }
  148. void thread_info_cache_init(void)
  149. {
  150. thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
  151. THREAD_SIZE, 0, NULL);
  152. BUG_ON(thread_info_cache == NULL);
  153. }
  154. # endif
  155. #endif
  156. /* SLAB cache for signal_struct structures (tsk->signal) */
  157. static struct kmem_cache *signal_cachep;
  158. /* SLAB cache for sighand_struct structures (tsk->sighand) */
  159. struct kmem_cache *sighand_cachep;
  160. /* SLAB cache for files_struct structures (tsk->files) */
  161. struct kmem_cache *files_cachep;
  162. /* SLAB cache for fs_struct structures (tsk->fs) */
  163. struct kmem_cache *fs_cachep;
  164. /* SLAB cache for vm_area_struct structures */
  165. struct kmem_cache *vm_area_cachep;
  166. /* SLAB cache for mm_struct structures (tsk->mm) */
  167. static struct kmem_cache *mm_cachep;
  168. static void account_kernel_stack(struct thread_info *ti, int account)
  169. {
  170. struct zone *zone = page_zone(virt_to_page(ti));
  171. mod_zone_page_state(zone, NR_KERNEL_STACK, account);
  172. }
  173. void free_task(struct task_struct *tsk)
  174. {
  175. account_kernel_stack(tsk->stack, -1);
  176. arch_release_thread_info(tsk->stack);
  177. free_thread_info(tsk->stack);
  178. rt_mutex_debug_task_free(tsk);
  179. ftrace_graph_exit_task(tsk);
  180. put_seccomp_filter(tsk);
  181. arch_release_task_struct(tsk);
  182. free_task_struct(tsk);
  183. }
  184. EXPORT_SYMBOL(free_task);
  185. static inline void free_signal_struct(struct signal_struct *sig)
  186. {
  187. taskstats_tgid_free(sig);
  188. sched_autogroup_exit(sig);
  189. kmem_cache_free(signal_cachep, sig);
  190. }
  191. static inline void put_signal_struct(struct signal_struct *sig)
  192. {
  193. if (atomic_dec_and_test(&sig->sigcnt))
  194. free_signal_struct(sig);
  195. }
  196. void __put_task_struct(struct task_struct *tsk)
  197. {
  198. WARN_ON(!tsk->exit_state);
  199. WARN_ON(atomic_read(&tsk->usage));
  200. WARN_ON(tsk == current);
  201. security_task_free(tsk);
  202. exit_creds(tsk);
  203. delayacct_tsk_free(tsk);
  204. put_signal_struct(tsk->signal);
  205. if (!profile_handoff_task(tsk))
  206. free_task(tsk);
  207. }
  208. EXPORT_SYMBOL_GPL(__put_task_struct);
  209. void __init __weak arch_task_cache_init(void) { }
  210. void __init fork_init(unsigned long mempages)
  211. {
  212. #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
  213. #ifndef ARCH_MIN_TASKALIGN
  214. #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
  215. #endif
  216. /* create a slab on which task_structs can be allocated */
  217. task_struct_cachep =
  218. kmem_cache_create("task_struct", sizeof(struct task_struct),
  219. ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
  220. #endif
  221. /* do the arch specific task caches init */
  222. arch_task_cache_init();
  223. /*
  224. * The default maximum number of threads is set to a safe
  225. * value: the thread structures can take up at most half
  226. * of memory.
  227. */
  228. max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
  229. /*
  230. * we need to allow at least 20 threads to boot a system
  231. */
  232. if (max_threads < 20)
  233. max_threads = 20;
  234. init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
  235. init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
  236. init_task.signal->rlim[RLIMIT_SIGPENDING] =
  237. init_task.signal->rlim[RLIMIT_NPROC];
  238. }
  239. int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
  240. struct task_struct *src)
  241. {
  242. *dst = *src;
  243. return 0;
  244. }
  245. static struct task_struct *dup_task_struct(struct task_struct *orig)
  246. {
  247. struct task_struct *tsk;
  248. struct thread_info *ti;
  249. unsigned long *stackend;
  250. int node = tsk_fork_get_node(orig);
  251. int err;
  252. tsk = alloc_task_struct_node(node);
  253. if (!tsk)
  254. return NULL;
  255. ti = alloc_thread_info_node(tsk, node);
  256. if (!ti)
  257. goto free_tsk;
  258. err = arch_dup_task_struct(tsk, orig);
  259. if (err)
  260. goto free_ti;
  261. tsk->stack = ti;
  262. setup_thread_stack(tsk, orig);
  263. clear_user_return_notifier(tsk);
  264. clear_tsk_need_resched(tsk);
  265. stackend = end_of_stack(tsk);
  266. *stackend = STACK_END_MAGIC; /* for overflow detection */
  267. #ifdef CONFIG_CC_STACKPROTECTOR
  268. tsk->stack_canary = get_random_int();
  269. #endif
  270. /*
  271. * One for us, one for whoever does the "release_task()" (usually
  272. * parent)
  273. */
  274. atomic_set(&tsk->usage, 2);
  275. #ifdef CONFIG_BLK_DEV_IO_TRACE
  276. tsk->btrace_seq = 0;
  277. #endif
  278. tsk->splice_pipe = NULL;
  279. tsk->task_frag.page = NULL;
  280. account_kernel_stack(ti, 1);
  281. return tsk;
  282. free_ti:
  283. free_thread_info(ti);
  284. free_tsk:
  285. free_task_struct(tsk);
  286. return NULL;
  287. }
  288. #ifdef CONFIG_MMU
  289. static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
  290. {
  291. struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
  292. struct rb_node **rb_link, *rb_parent;
  293. int retval;
  294. unsigned long charge;
  295. struct mempolicy *pol;
  296. uprobe_start_dup_mmap();
  297. down_write(&oldmm->mmap_sem);
  298. flush_cache_dup_mm(oldmm);
  299. uprobe_dup_mmap(oldmm, mm);
  300. /*
  301. * Not linked in yet - no deadlock potential:
  302. */
  303. down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
  304. mm->locked_vm = 0;
  305. mm->mmap = NULL;
  306. mm->mmap_cache = NULL;
  307. mm->free_area_cache = oldmm->mmap_base;
  308. mm->cached_hole_size = ~0UL;
  309. mm->map_count = 0;
  310. cpumask_clear(mm_cpumask(mm));
  311. mm->mm_rb = RB_ROOT;
  312. rb_link = &mm->mm_rb.rb_node;
  313. rb_parent = NULL;
  314. pprev = &mm->mmap;
  315. retval = ksm_fork(mm, oldmm);
  316. if (retval)
  317. goto out;
  318. retval = khugepaged_fork(mm, oldmm);
  319. if (retval)
  320. goto out;
  321. prev = NULL;
  322. for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
  323. struct file *file;
  324. if (mpnt->vm_flags & VM_DONTCOPY) {
  325. vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
  326. -vma_pages(mpnt));
  327. continue;
  328. }
  329. charge = 0;
  330. if (mpnt->vm_flags & VM_ACCOUNT) {
  331. unsigned long len = vma_pages(mpnt);
  332. if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
  333. goto fail_nomem;
  334. charge = len;
  335. }
  336. tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
  337. if (!tmp)
  338. goto fail_nomem;
  339. *tmp = *mpnt;
  340. INIT_LIST_HEAD(&tmp->anon_vma_chain);
  341. pol = mpol_dup(vma_policy(mpnt));
  342. retval = PTR_ERR(pol);
  343. if (IS_ERR(pol))
  344. goto fail_nomem_policy;
  345. vma_set_policy(tmp, pol);
  346. tmp->vm_mm = mm;
  347. if (anon_vma_fork(tmp, mpnt))
  348. goto fail_nomem_anon_vma_fork;
  349. tmp->vm_flags &= ~VM_LOCKED;
  350. tmp->vm_next = tmp->vm_prev = NULL;
  351. file = tmp->vm_file;
  352. if (file) {
  353. struct inode *inode = file_inode(file);
  354. struct address_space *mapping = file->f_mapping;
  355. get_file(file);
  356. if (tmp->vm_flags & VM_DENYWRITE)
  357. atomic_dec(&inode->i_writecount);
  358. mutex_lock(&mapping->i_mmap_mutex);
  359. if (tmp->vm_flags & VM_SHARED)
  360. mapping->i_mmap_writable++;
  361. flush_dcache_mmap_lock(mapping);
  362. /* insert tmp into the share list, just after mpnt */
  363. if (unlikely(tmp->vm_flags & VM_NONLINEAR))
  364. vma_nonlinear_insert(tmp,
  365. &mapping->i_mmap_nonlinear);
  366. else
  367. vma_interval_tree_insert_after(tmp, mpnt,
  368. &mapping->i_mmap);
  369. flush_dcache_mmap_unlock(mapping);
  370. mutex_unlock(&mapping->i_mmap_mutex);
  371. }
  372. /*
  373. * Clear hugetlb-related page reserves for children. This only
  374. * affects MAP_PRIVATE mappings. Faults generated by the child
  375. * are not guaranteed to succeed, even if read-only
  376. */
  377. if (is_vm_hugetlb_page(tmp))
  378. reset_vma_resv_huge_pages(tmp);
  379. /*
  380. * Link in the new vma and copy the page table entries.
  381. */
  382. *pprev = tmp;
  383. pprev = &tmp->vm_next;
  384. tmp->vm_prev = prev;
  385. prev = tmp;
  386. __vma_link_rb(mm, tmp, rb_link, rb_parent);
  387. rb_link = &tmp->vm_rb.rb_right;
  388. rb_parent = &tmp->vm_rb;
  389. mm->map_count++;
  390. retval = copy_page_range(mm, oldmm, mpnt);
  391. if (tmp->vm_ops && tmp->vm_ops->open)
  392. tmp->vm_ops->open(tmp);
  393. if (retval)
  394. goto out;
  395. }
  396. /* a new mm has just been created */
  397. arch_dup_mmap(oldmm, mm);
  398. retval = 0;
  399. out:
  400. up_write(&mm->mmap_sem);
  401. flush_tlb_mm(oldmm);
  402. up_write(&oldmm->mmap_sem);
  403. uprobe_end_dup_mmap();
  404. return retval;
  405. fail_nomem_anon_vma_fork:
  406. mpol_put(pol);
  407. fail_nomem_policy:
  408. kmem_cache_free(vm_area_cachep, tmp);
  409. fail_nomem:
  410. retval = -ENOMEM;
  411. vm_unacct_memory(charge);
  412. goto out;
  413. }
  414. static inline int mm_alloc_pgd(struct mm_struct *mm)
  415. {
  416. mm->pgd = pgd_alloc(mm);
  417. if (unlikely(!mm->pgd))
  418. return -ENOMEM;
  419. return 0;
  420. }
  421. static inline void mm_free_pgd(struct mm_struct *mm)
  422. {
  423. pgd_free(mm, mm->pgd);
  424. }
  425. #else
  426. #define dup_mmap(mm, oldmm) (0)
  427. #define mm_alloc_pgd(mm) (0)
  428. #define mm_free_pgd(mm)
  429. #endif /* CONFIG_MMU */
  430. __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
  431. #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
  432. #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
  433. static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
  434. static int __init coredump_filter_setup(char *s)
  435. {
  436. default_dump_filter =
  437. (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
  438. MMF_DUMP_FILTER_MASK;
  439. return 1;
  440. }
  441. __setup("coredump_filter=", coredump_filter_setup);
  442. #include <linux/init_task.h>
  443. static void mm_init_aio(struct mm_struct *mm)
  444. {
  445. #ifdef CONFIG_AIO
  446. spin_lock_init(&mm->ioctx_lock);
  447. INIT_HLIST_HEAD(&mm->ioctx_list);
  448. #endif
  449. }
  450. static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
  451. {
  452. atomic_set(&mm->mm_users, 1);
  453. atomic_set(&mm->mm_count, 1);
  454. init_rwsem(&mm->mmap_sem);
  455. INIT_LIST_HEAD(&mm->mmlist);
  456. mm->flags = (current->mm) ?
  457. (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
  458. mm->core_state = NULL;
  459. mm->nr_ptes = 0;
  460. memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
  461. spin_lock_init(&mm->page_table_lock);
  462. mm->free_area_cache = TASK_UNMAPPED_BASE;
  463. mm->cached_hole_size = ~0UL;
  464. mm_init_aio(mm);
  465. mm_init_owner(mm, p);
  466. if (likely(!mm_alloc_pgd(mm))) {
  467. mm->def_flags = 0;
  468. mmu_notifier_mm_init(mm);
  469. return mm;
  470. }
  471. free_mm(mm);
  472. return NULL;
  473. }
  474. static void check_mm(struct mm_struct *mm)
  475. {
  476. int i;
  477. for (i = 0; i < NR_MM_COUNTERS; i++) {
  478. long x = atomic_long_read(&mm->rss_stat.count[i]);
  479. if (unlikely(x))
  480. printk(KERN_ALERT "BUG: Bad rss-counter state "
  481. "mm:%p idx:%d val:%ld\n", mm, i, x);
  482. }
  483. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  484. VM_BUG_ON(mm->pmd_huge_pte);
  485. #endif
  486. }
  487. /*
  488. * Allocate and initialize an mm_struct.
  489. */
  490. struct mm_struct *mm_alloc(void)
  491. {
  492. struct mm_struct *mm;
  493. mm = allocate_mm();
  494. if (!mm)
  495. return NULL;
  496. memset(mm, 0, sizeof(*mm));
  497. mm_init_cpumask(mm);
  498. return mm_init(mm, current);
  499. }
  500. /*
  501. * Called when the last reference to the mm
  502. * is dropped: either by a lazy thread or by
  503. * mmput. Free the page directory and the mm.
  504. */
  505. void __mmdrop(struct mm_struct *mm)
  506. {
  507. BUG_ON(mm == &init_mm);
  508. mm_free_pgd(mm);
  509. destroy_context(mm);
  510. mmu_notifier_mm_destroy(mm);
  511. check_mm(mm);
  512. free_mm(mm);
  513. }
  514. EXPORT_SYMBOL_GPL(__mmdrop);
  515. /*
  516. * Decrement the use count and release all resources for an mm.
  517. */
  518. void mmput(struct mm_struct *mm)
  519. {
  520. might_sleep();
  521. if (atomic_dec_and_test(&mm->mm_users)) {
  522. uprobe_clear_state(mm);
  523. exit_aio(mm);
  524. ksm_exit(mm);
  525. khugepaged_exit(mm); /* must run before exit_mmap */
  526. exit_mmap(mm);
  527. set_mm_exe_file(mm, NULL);
  528. if (!list_empty(&mm->mmlist)) {
  529. spin_lock(&mmlist_lock);
  530. list_del(&mm->mmlist);
  531. spin_unlock(&mmlist_lock);
  532. }
  533. if (mm->binfmt)
  534. module_put(mm->binfmt->module);
  535. mmdrop(mm);
  536. }
  537. }
  538. EXPORT_SYMBOL_GPL(mmput);
  539. void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
  540. {
  541. if (new_exe_file)
  542. get_file(new_exe_file);
  543. if (mm->exe_file)
  544. fput(mm->exe_file);
  545. mm->exe_file = new_exe_file;
  546. }
  547. struct file *get_mm_exe_file(struct mm_struct *mm)
  548. {
  549. struct file *exe_file;
  550. /* We need mmap_sem to protect against races with removal of exe_file */
  551. down_read(&mm->mmap_sem);
  552. exe_file = mm->exe_file;
  553. if (exe_file)
  554. get_file(exe_file);
  555. up_read(&mm->mmap_sem);
  556. return exe_file;
  557. }
  558. static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
  559. {
  560. /* It's safe to write the exe_file pointer without exe_file_lock because
  561. * this is called during fork when the task is not yet in /proc */
  562. newmm->exe_file = get_mm_exe_file(oldmm);
  563. }
  564. /**
  565. * get_task_mm - acquire a reference to the task's mm
  566. *
  567. * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
  568. * this kernel workthread has transiently adopted a user mm with use_mm,
  569. * to do its AIO) is not set and if so returns a reference to it, after
  570. * bumping up the use count. User must release the mm via mmput()
  571. * after use. Typically used by /proc and ptrace.
  572. */
  573. struct mm_struct *get_task_mm(struct task_struct *task)
  574. {
  575. struct mm_struct *mm;
  576. task_lock(task);
  577. mm = task->mm;
  578. if (mm) {
  579. if (task->flags & PF_KTHREAD)
  580. mm = NULL;
  581. else
  582. atomic_inc(&mm->mm_users);
  583. }
  584. task_unlock(task);
  585. return mm;
  586. }
  587. EXPORT_SYMBOL_GPL(get_task_mm);
  588. struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
  589. {
  590. struct mm_struct *mm;
  591. int err;
  592. err = mutex_lock_killable(&task->signal->cred_guard_mutex);
  593. if (err)
  594. return ERR_PTR(err);
  595. mm = get_task_mm(task);
  596. if (mm && mm != current->mm &&
  597. !ptrace_may_access(task, mode)) {
  598. mmput(mm);
  599. mm = ERR_PTR(-EACCES);
  600. }
  601. mutex_unlock(&task->signal->cred_guard_mutex);
  602. return mm;
  603. }
  604. static void complete_vfork_done(struct task_struct *tsk)
  605. {
  606. struct completion *vfork;
  607. task_lock(tsk);
  608. vfork = tsk->vfork_done;
  609. if (likely(vfork)) {
  610. tsk->vfork_done = NULL;
  611. complete(vfork);
  612. }
  613. task_unlock(tsk);
  614. }
  615. static int wait_for_vfork_done(struct task_struct *child,
  616. struct completion *vfork)
  617. {
  618. int killed;
  619. freezer_do_not_count();
  620. killed = wait_for_completion_killable(vfork);
  621. freezer_count();
  622. if (killed) {
  623. task_lock(child);
  624. child->vfork_done = NULL;
  625. task_unlock(child);
  626. }
  627. put_task_struct(child);
  628. return killed;
  629. }
  630. /* Please note the differences between mmput and mm_release.
  631. * mmput is called whenever we stop holding onto a mm_struct,
  632. * error success whatever.
  633. *
  634. * mm_release is called after a mm_struct has been removed
  635. * from the current process.
  636. *
  637. * This difference is important for error handling, when we
  638. * only half set up a mm_struct for a new process and need to restore
  639. * the old one. Because we mmput the new mm_struct before
  640. * restoring the old one. . .
  641. * Eric Biederman 10 January 1998
  642. */
  643. void mm_release(struct task_struct *tsk, struct mm_struct *mm)
  644. {
  645. /* Get rid of any futexes when releasing the mm */
  646. #ifdef CONFIG_FUTEX
  647. if (unlikely(tsk->robust_list)) {
  648. exit_robust_list(tsk);
  649. tsk->robust_list = NULL;
  650. }
  651. #ifdef CONFIG_COMPAT
  652. if (unlikely(tsk->compat_robust_list)) {
  653. compat_exit_robust_list(tsk);
  654. tsk->compat_robust_list = NULL;
  655. }
  656. #endif
  657. if (unlikely(!list_empty(&tsk->pi_state_list)))
  658. exit_pi_state_list(tsk);
  659. #endif
  660. uprobe_free_utask(tsk);
  661. /* Get rid of any cached register state */
  662. deactivate_mm(tsk, mm);
  663. /*
  664. * If we're exiting normally, clear a user-space tid field if
  665. * requested. We leave this alone when dying by signal, to leave
  666. * the value intact in a core dump, and to save the unnecessary
  667. * trouble, say, a killed vfork parent shouldn't touch this mm.
  668. * Userland only wants this done for a sys_exit.
  669. */
  670. if (tsk->clear_child_tid) {
  671. if (!(tsk->flags & PF_SIGNALED) &&
  672. atomic_read(&mm->mm_users) > 1) {
  673. /*
  674. * We don't check the error code - if userspace has
  675. * not set up a proper pointer then tough luck.
  676. */
  677. put_user(0, tsk->clear_child_tid);
  678. sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
  679. 1, NULL, NULL, 0);
  680. }
  681. tsk->clear_child_tid = NULL;
  682. }
  683. /*
  684. * All done, finally we can wake up parent and return this mm to him.
  685. * Also kthread_stop() uses this completion for synchronization.
  686. */
  687. if (tsk->vfork_done)
  688. complete_vfork_done(tsk);
  689. }
  690. /*
  691. * Allocate a new mm structure and copy contents from the
  692. * mm structure of the passed in task structure.
  693. */
  694. struct mm_struct *dup_mm(struct task_struct *tsk)
  695. {
  696. struct mm_struct *mm, *oldmm = current->mm;
  697. int err;
  698. if (!oldmm)
  699. return NULL;
  700. mm = allocate_mm();
  701. if (!mm)
  702. goto fail_nomem;
  703. memcpy(mm, oldmm, sizeof(*mm));
  704. mm_init_cpumask(mm);
  705. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  706. mm->pmd_huge_pte = NULL;
  707. #endif
  708. #ifdef CONFIG_NUMA_BALANCING
  709. mm->first_nid = NUMA_PTE_SCAN_INIT;
  710. #endif
  711. if (!mm_init(mm, tsk))
  712. goto fail_nomem;
  713. if (init_new_context(tsk, mm))
  714. goto fail_nocontext;
  715. dup_mm_exe_file(oldmm, mm);
  716. err = dup_mmap(mm, oldmm);
  717. if (err)
  718. goto free_pt;
  719. mm->hiwater_rss = get_mm_rss(mm);
  720. mm->hiwater_vm = mm->total_vm;
  721. if (mm->binfmt && !try_module_get(mm->binfmt->module))
  722. goto free_pt;
  723. return mm;
  724. free_pt:
  725. /* don't put binfmt in mmput, we haven't got module yet */
  726. mm->binfmt = NULL;
  727. mmput(mm);
  728. fail_nomem:
  729. return NULL;
  730. fail_nocontext:
  731. /*
  732. * If init_new_context() failed, we cannot use mmput() to free the mm
  733. * because it calls destroy_context()
  734. */
  735. mm_free_pgd(mm);
  736. free_mm(mm);
  737. return NULL;
  738. }
  739. static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
  740. {
  741. struct mm_struct *mm, *oldmm;
  742. int retval;
  743. tsk->min_flt = tsk->maj_flt = 0;
  744. tsk->nvcsw = tsk->nivcsw = 0;
  745. #ifdef CONFIG_DETECT_HUNG_TASK
  746. tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
  747. #endif
  748. tsk->mm = NULL;
  749. tsk->active_mm = NULL;
  750. /*
  751. * Are we cloning a kernel thread?
  752. *
  753. * We need to steal a active VM for that..
  754. */
  755. oldmm = current->mm;
  756. if (!oldmm)
  757. return 0;
  758. if (clone_flags & CLONE_VM) {
  759. atomic_inc(&oldmm->mm_users);
  760. mm = oldmm;
  761. goto good_mm;
  762. }
  763. retval = -ENOMEM;
  764. mm = dup_mm(tsk);
  765. if (!mm)
  766. goto fail_nomem;
  767. good_mm:
  768. tsk->mm = mm;
  769. tsk->active_mm = mm;
  770. return 0;
  771. fail_nomem:
  772. return retval;
  773. }
  774. static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
  775. {
  776. struct fs_struct *fs = current->fs;
  777. if (clone_flags & CLONE_FS) {
  778. /* tsk->fs is already what we want */
  779. spin_lock(&fs->lock);
  780. if (fs->in_exec) {
  781. spin_unlock(&fs->lock);
  782. return -EAGAIN;
  783. }
  784. fs->users++;
  785. spin_unlock(&fs->lock);
  786. return 0;
  787. }
  788. tsk->fs = copy_fs_struct(fs);
  789. if (!tsk->fs)
  790. return -ENOMEM;
  791. return 0;
  792. }
  793. static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
  794. {
  795. struct files_struct *oldf, *newf;
  796. int error = 0;
  797. /*
  798. * A background process may not have any files ...
  799. */
  800. oldf = current->files;
  801. if (!oldf)
  802. goto out;
  803. if (clone_flags & CLONE_FILES) {
  804. atomic_inc(&oldf->count);
  805. goto out;
  806. }
  807. newf = dup_fd(oldf, &error);
  808. if (!newf)
  809. goto out;
  810. tsk->files = newf;
  811. error = 0;
  812. out:
  813. return error;
  814. }
  815. static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
  816. {
  817. #ifdef CONFIG_BLOCK
  818. struct io_context *ioc = current->io_context;
  819. struct io_context *new_ioc;
  820. if (!ioc)
  821. return 0;
  822. /*
  823. * Share io context with parent, if CLONE_IO is set
  824. */
  825. if (clone_flags & CLONE_IO) {
  826. ioc_task_link(ioc);
  827. tsk->io_context = ioc;
  828. } else if (ioprio_valid(ioc->ioprio)) {
  829. new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
  830. if (unlikely(!new_ioc))
  831. return -ENOMEM;
  832. new_ioc->ioprio = ioc->ioprio;
  833. put_io_context(new_ioc);
  834. }
  835. #endif
  836. return 0;
  837. }
  838. static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
  839. {
  840. struct sighand_struct *sig;
  841. if (clone_flags & CLONE_SIGHAND) {
  842. atomic_inc(&current->sighand->count);
  843. return 0;
  844. }
  845. sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
  846. rcu_assign_pointer(tsk->sighand, sig);
  847. if (!sig)
  848. return -ENOMEM;
  849. atomic_set(&sig->count, 1);
  850. memcpy(sig->action, current->sighand->action, sizeof(sig->action));
  851. return 0;
  852. }
  853. void __cleanup_sighand(struct sighand_struct *sighand)
  854. {
  855. if (atomic_dec_and_test(&sighand->count)) {
  856. signalfd_cleanup(sighand);
  857. kmem_cache_free(sighand_cachep, sighand);
  858. }
  859. }
  860. /*
  861. * Initialize POSIX timer handling for a thread group.
  862. */
  863. static void posix_cpu_timers_init_group(struct signal_struct *sig)
  864. {
  865. unsigned long cpu_limit;
  866. /* Thread group counters. */
  867. thread_group_cputime_init(sig);
  868. cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
  869. if (cpu_limit != RLIM_INFINITY) {
  870. sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
  871. sig->cputimer.running = 1;
  872. }
  873. /* The timer lists. */
  874. INIT_LIST_HEAD(&sig->cpu_timers[0]);
  875. INIT_LIST_HEAD(&sig->cpu_timers[1]);
  876. INIT_LIST_HEAD(&sig->cpu_timers[2]);
  877. }
  878. static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
  879. {
  880. struct signal_struct *sig;
  881. if (clone_flags & CLONE_THREAD)
  882. return 0;
  883. sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
  884. tsk->signal = sig;
  885. if (!sig)
  886. return -ENOMEM;
  887. sig->nr_threads = 1;
  888. atomic_set(&sig->live, 1);
  889. atomic_set(&sig->sigcnt, 1);
  890. init_waitqueue_head(&sig->wait_chldexit);
  891. sig->curr_target = tsk;
  892. init_sigpending(&sig->shared_pending);
  893. INIT_LIST_HEAD(&sig->posix_timers);
  894. hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
  895. sig->real_timer.function = it_real_fn;
  896. task_lock(current->group_leader);
  897. memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
  898. task_unlock(current->group_leader);
  899. posix_cpu_timers_init_group(sig);
  900. tty_audit_fork(sig);
  901. sched_autogroup_fork(sig);
  902. #ifdef CONFIG_CGROUPS
  903. init_rwsem(&sig->group_rwsem);
  904. #endif
  905. sig->oom_score_adj = current->signal->oom_score_adj;
  906. sig->oom_score_adj_min = current->signal->oom_score_adj_min;
  907. sig->has_child_subreaper = current->signal->has_child_subreaper ||
  908. current->signal->is_child_subreaper;
  909. mutex_init(&sig->cred_guard_mutex);
  910. return 0;
  911. }
  912. static void copy_flags(unsigned long clone_flags, struct task_struct *p)
  913. {
  914. unsigned long new_flags = p->flags;
  915. new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
  916. new_flags |= PF_FORKNOEXEC;
  917. p->flags = new_flags;
  918. }
  919. SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
  920. {
  921. current->clear_child_tid = tidptr;
  922. return task_pid_vnr(current);
  923. }
  924. static void rt_mutex_init_task(struct task_struct *p)
  925. {
  926. raw_spin_lock_init(&p->pi_lock);
  927. #ifdef CONFIG_RT_MUTEXES
  928. plist_head_init(&p->pi_waiters);
  929. p->pi_blocked_on = NULL;
  930. #endif
  931. }
  932. #ifdef CONFIG_MM_OWNER
  933. void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
  934. {
  935. mm->owner = p;
  936. }
  937. #endif /* CONFIG_MM_OWNER */
  938. /*
  939. * Initialize POSIX timer handling for a single task.
  940. */
  941. static void posix_cpu_timers_init(struct task_struct *tsk)
  942. {
  943. tsk->cputime_expires.prof_exp = 0;
  944. tsk->cputime_expires.virt_exp = 0;
  945. tsk->cputime_expires.sched_exp = 0;
  946. INIT_LIST_HEAD(&tsk->cpu_timers[0]);
  947. INIT_LIST_HEAD(&tsk->cpu_timers[1]);
  948. INIT_LIST_HEAD(&tsk->cpu_timers[2]);
  949. }
  950. /*
  951. * This creates a new process as a copy of the old one,
  952. * but does not actually start it yet.
  953. *
  954. * It copies the registers, and all the appropriate
  955. * parts of the process environment (as per the clone
  956. * flags). The actual kick-off is left to the caller.
  957. */
  958. static struct task_struct *copy_process(unsigned long clone_flags,
  959. unsigned long stack_start,
  960. unsigned long stack_size,
  961. int __user *child_tidptr,
  962. struct pid *pid,
  963. int trace)
  964. {
  965. int retval;
  966. struct task_struct *p;
  967. if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
  968. return ERR_PTR(-EINVAL);
  969. if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
  970. return ERR_PTR(-EINVAL);
  971. /*
  972. * Thread groups must share signals as well, and detached threads
  973. * can only be started up within the thread group.
  974. */
  975. if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
  976. return ERR_PTR(-EINVAL);
  977. /*
  978. * Shared signal handlers imply shared VM. By way of the above,
  979. * thread groups also imply shared VM. Blocking this case allows
  980. * for various simplifications in other code.
  981. */
  982. if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
  983. return ERR_PTR(-EINVAL);
  984. /*
  985. * Siblings of global init remain as zombies on exit since they are
  986. * not reaped by their parent (swapper). To solve this and to avoid
  987. * multi-rooted process trees, prevent global and container-inits
  988. * from creating siblings.
  989. */
  990. if ((clone_flags & CLONE_PARENT) &&
  991. current->signal->flags & SIGNAL_UNKILLABLE)
  992. return ERR_PTR(-EINVAL);
  993. /*
  994. * If the new process will be in a different pid namespace
  995. * don't allow the creation of threads.
  996. */
  997. if ((clone_flags & (CLONE_VM|CLONE_NEWPID)) &&
  998. (task_active_pid_ns(current) != current->nsproxy->pid_ns))
  999. return ERR_PTR(-EINVAL);
  1000. retval = security_task_create(clone_flags);
  1001. if (retval)
  1002. goto fork_out;
  1003. retval = -ENOMEM;
  1004. p = dup_task_struct(current);
  1005. if (!p)
  1006. goto fork_out;
  1007. ftrace_graph_init_task(p);
  1008. get_seccomp_filter(p);
  1009. rt_mutex_init_task(p);
  1010. #ifdef CONFIG_PROVE_LOCKING
  1011. DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
  1012. DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
  1013. #endif
  1014. retval = -EAGAIN;
  1015. if (atomic_read(&p->real_cred->user->processes) >=
  1016. task_rlimit(p, RLIMIT_NPROC)) {
  1017. if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
  1018. p->real_cred->user != INIT_USER)
  1019. goto bad_fork_free;
  1020. }
  1021. current->flags &= ~PF_NPROC_EXCEEDED;
  1022. retval = copy_creds(p, clone_flags);
  1023. if (retval < 0)
  1024. goto bad_fork_free;
  1025. /*
  1026. * If multiple threads are within copy_process(), then this check
  1027. * triggers too late. This doesn't hurt, the check is only there
  1028. * to stop root fork bombs.
  1029. */
  1030. retval = -EAGAIN;
  1031. if (nr_threads >= max_threads)
  1032. goto bad_fork_cleanup_count;
  1033. if (!try_module_get(task_thread_info(p)->exec_domain->module))
  1034. goto bad_fork_cleanup_count;
  1035. p->did_exec = 0;
  1036. delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
  1037. copy_flags(clone_flags, p);
  1038. INIT_LIST_HEAD(&p->children);
  1039. INIT_LIST_HEAD(&p->sibling);
  1040. rcu_copy_process(p);
  1041. p->vfork_done = NULL;
  1042. spin_lock_init(&p->alloc_lock);
  1043. init_sigpending(&p->pending);
  1044. p->utime = p->stime = p->gtime = 0;
  1045. p->utimescaled = p->stimescaled = 0;
  1046. #ifndef CONFIG_VIRT_CPU_ACCOUNTING
  1047. p->prev_cputime.utime = p->prev_cputime.stime = 0;
  1048. #endif
  1049. #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
  1050. seqlock_init(&p->vtime_seqlock);
  1051. p->vtime_snap = 0;
  1052. p->vtime_snap_whence = VTIME_SLEEPING;
  1053. #endif
  1054. #if defined(SPLIT_RSS_COUNTING)
  1055. memset(&p->rss_stat, 0, sizeof(p->rss_stat));
  1056. #endif
  1057. p->default_timer_slack_ns = current->timer_slack_ns;
  1058. task_io_accounting_init(&p->ioac);
  1059. acct_clear_integrals(p);
  1060. posix_cpu_timers_init(p);
  1061. do_posix_clock_monotonic_gettime(&p->start_time);
  1062. p->real_start_time = p->start_time;
  1063. monotonic_to_bootbased(&p->real_start_time);
  1064. p->io_context = NULL;
  1065. p->audit_context = NULL;
  1066. if (clone_flags & CLONE_THREAD)
  1067. threadgroup_change_begin(current);
  1068. cgroup_fork(p);
  1069. #ifdef CONFIG_NUMA
  1070. p->mempolicy = mpol_dup(p->mempolicy);
  1071. if (IS_ERR(p->mempolicy)) {
  1072. retval = PTR_ERR(p->mempolicy);
  1073. p->mempolicy = NULL;
  1074. goto bad_fork_cleanup_cgroup;
  1075. }
  1076. mpol_fix_fork_child_flag(p);
  1077. #endif
  1078. #ifdef CONFIG_CPUSETS
  1079. p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
  1080. p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
  1081. seqcount_init(&p->mems_allowed_seq);
  1082. #endif
  1083. #ifdef CONFIG_TRACE_IRQFLAGS
  1084. p->irq_events = 0;
  1085. p->hardirqs_enabled = 0;
  1086. p->hardirq_enable_ip = 0;
  1087. p->hardirq_enable_event = 0;
  1088. p->hardirq_disable_ip = _THIS_IP_;
  1089. p->hardirq_disable_event = 0;
  1090. p->softirqs_enabled = 1;
  1091. p->softirq_enable_ip = _THIS_IP_;
  1092. p->softirq_enable_event = 0;
  1093. p->softirq_disable_ip = 0;
  1094. p->softirq_disable_event = 0;
  1095. p->hardirq_context = 0;
  1096. p->softirq_context = 0;
  1097. #endif
  1098. #ifdef CONFIG_LOCKDEP
  1099. p->lockdep_depth = 0; /* no locks held yet */
  1100. p->curr_chain_key = 0;
  1101. p->lockdep_recursion = 0;
  1102. #endif
  1103. #ifdef CONFIG_DEBUG_MUTEXES
  1104. p->blocked_on = NULL; /* not blocked yet */
  1105. #endif
  1106. #ifdef CONFIG_MEMCG
  1107. p->memcg_batch.do_batch = 0;
  1108. p->memcg_batch.memcg = NULL;
  1109. #endif
  1110. /* Perform scheduler related setup. Assign this task to a CPU. */
  1111. sched_fork(p);
  1112. retval = perf_event_init_task(p);
  1113. if (retval)
  1114. goto bad_fork_cleanup_policy;
  1115. retval = audit_alloc(p);
  1116. if (retval)
  1117. goto bad_fork_cleanup_policy;
  1118. /* copy all the process information */
  1119. retval = copy_semundo(clone_flags, p);
  1120. if (retval)
  1121. goto bad_fork_cleanup_audit;
  1122. retval = copy_files(clone_flags, p);
  1123. if (retval)
  1124. goto bad_fork_cleanup_semundo;
  1125. retval = copy_fs(clone_flags, p);
  1126. if (retval)
  1127. goto bad_fork_cleanup_files;
  1128. retval = copy_sighand(clone_flags, p);
  1129. if (retval)
  1130. goto bad_fork_cleanup_fs;
  1131. retval = copy_signal(clone_flags, p);
  1132. if (retval)
  1133. goto bad_fork_cleanup_sighand;
  1134. retval = copy_mm(clone_flags, p);
  1135. if (retval)
  1136. goto bad_fork_cleanup_signal;
  1137. retval = copy_namespaces(clone_flags, p);
  1138. if (retval)
  1139. goto bad_fork_cleanup_mm;
  1140. retval = copy_io(clone_flags, p);
  1141. if (retval)
  1142. goto bad_fork_cleanup_namespaces;
  1143. retval = copy_thread(clone_flags, stack_start, stack_size, p);
  1144. if (retval)
  1145. goto bad_fork_cleanup_io;
  1146. if (pid != &init_struct_pid) {
  1147. retval = -ENOMEM;
  1148. pid = alloc_pid(p->nsproxy->pid_ns);
  1149. if (!pid)
  1150. goto bad_fork_cleanup_io;
  1151. }
  1152. p->pid = pid_nr(pid);
  1153. p->tgid = p->pid;
  1154. if (clone_flags & CLONE_THREAD)
  1155. p->tgid = current->tgid;
  1156. p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
  1157. /*
  1158. * Clear TID on mm_release()?
  1159. */
  1160. p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
  1161. #ifdef CONFIG_BLOCK
  1162. p->plug = NULL;
  1163. #endif
  1164. #ifdef CONFIG_FUTEX
  1165. p->robust_list = NULL;
  1166. #ifdef CONFIG_COMPAT
  1167. p->compat_robust_list = NULL;
  1168. #endif
  1169. INIT_LIST_HEAD(&p->pi_state_list);
  1170. p->pi_state_cache = NULL;
  1171. #endif
  1172. uprobe_copy_process(p);
  1173. /*
  1174. * sigaltstack should be cleared when sharing the same VM
  1175. */
  1176. if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
  1177. p->sas_ss_sp = p->sas_ss_size = 0;
  1178. /*
  1179. * Syscall tracing and stepping should be turned off in the
  1180. * child regardless of CLONE_PTRACE.
  1181. */
  1182. user_disable_single_step(p);
  1183. clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
  1184. #ifdef TIF_SYSCALL_EMU
  1185. clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
  1186. #endif
  1187. clear_all_latency_tracing(p);
  1188. /* ok, now we should be set up.. */
  1189. if (clone_flags & CLONE_THREAD)
  1190. p->exit_signal = -1;
  1191. else if (clone_flags & CLONE_PARENT)
  1192. p->exit_signal = current->group_leader->exit_signal;
  1193. else
  1194. p->exit_signal = (clone_flags & CSIGNAL);
  1195. p->pdeath_signal = 0;
  1196. p->exit_state = 0;
  1197. p->nr_dirtied = 0;
  1198. p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
  1199. p->dirty_paused_when = 0;
  1200. /*
  1201. * Ok, make it visible to the rest of the system.
  1202. * We dont wake it up yet.
  1203. */
  1204. p->group_leader = p;
  1205. INIT_LIST_HEAD(&p->thread_group);
  1206. p->task_works = NULL;
  1207. /* Need tasklist lock for parent etc handling! */
  1208. write_lock_irq(&tasklist_lock);
  1209. /* CLONE_PARENT re-uses the old parent */
  1210. if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
  1211. p->real_parent = current->real_parent;
  1212. p->parent_exec_id = current->parent_exec_id;
  1213. } else {
  1214. p->real_parent = current;
  1215. p->parent_exec_id = current->self_exec_id;
  1216. }
  1217. spin_lock(&current->sighand->siglock);
  1218. /*
  1219. * Process group and session signals need to be delivered to just the
  1220. * parent before the fork or both the parent and the child after the
  1221. * fork. Restart if a signal comes in before we add the new process to
  1222. * it's process group.
  1223. * A fatal signal pending means that current will exit, so the new
  1224. * thread can't slip out of an OOM kill (or normal SIGKILL).
  1225. */
  1226. recalc_sigpending();
  1227. if (signal_pending(current)) {
  1228. spin_unlock(&current->sighand->siglock);
  1229. write_unlock_irq(&tasklist_lock);
  1230. retval = -ERESTARTNOINTR;
  1231. goto bad_fork_free_pid;
  1232. }
  1233. if (clone_flags & CLONE_THREAD) {
  1234. current->signal->nr_threads++;
  1235. atomic_inc(&current->signal->live);
  1236. atomic_inc(&current->signal->sigcnt);
  1237. p->group_leader = current->group_leader;
  1238. list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
  1239. }
  1240. if (likely(p->pid)) {
  1241. ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
  1242. if (thread_group_leader(p)) {
  1243. if (is_child_reaper(pid)) {
  1244. ns_of_pid(pid)->child_reaper = p;
  1245. p->signal->flags |= SIGNAL_UNKILLABLE;
  1246. }
  1247. p->signal->leader_pid = pid;
  1248. p->signal->tty = tty_kref_get(current->signal->tty);
  1249. attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
  1250. attach_pid(p, PIDTYPE_SID, task_session(current));
  1251. list_add_tail(&p->sibling, &p->real_parent->children);
  1252. list_add_tail_rcu(&p->tasks, &init_task.tasks);
  1253. __this_cpu_inc(process_counts);
  1254. }
  1255. attach_pid(p, PIDTYPE_PID, pid);
  1256. nr_threads++;
  1257. }
  1258. total_forks++;
  1259. spin_unlock(&current->sighand->siglock);
  1260. write_unlock_irq(&tasklist_lock);
  1261. proc_fork_connector(p);
  1262. cgroup_post_fork(p);
  1263. if (clone_flags & CLONE_THREAD)
  1264. threadgroup_change_end(current);
  1265. perf_event_fork(p);
  1266. trace_task_newtask(p, clone_flags);
  1267. return p;
  1268. bad_fork_free_pid:
  1269. if (pid != &init_struct_pid)
  1270. free_pid(pid);
  1271. bad_fork_cleanup_io:
  1272. if (p->io_context)
  1273. exit_io_context(p);
  1274. bad_fork_cleanup_namespaces:
  1275. exit_task_namespaces(p);
  1276. bad_fork_cleanup_mm:
  1277. if (p->mm)
  1278. mmput(p->mm);
  1279. bad_fork_cleanup_signal:
  1280. if (!(clone_flags & CLONE_THREAD))
  1281. free_signal_struct(p->signal);
  1282. bad_fork_cleanup_sighand:
  1283. __cleanup_sighand(p->sighand);
  1284. bad_fork_cleanup_fs:
  1285. exit_fs(p); /* blocking */
  1286. bad_fork_cleanup_files:
  1287. exit_files(p); /* blocking */
  1288. bad_fork_cleanup_semundo:
  1289. exit_sem(p);
  1290. bad_fork_cleanup_audit:
  1291. audit_free(p);
  1292. bad_fork_cleanup_policy:
  1293. perf_event_free_task(p);
  1294. #ifdef CONFIG_NUMA
  1295. mpol_put(p->mempolicy);
  1296. bad_fork_cleanup_cgroup:
  1297. #endif
  1298. if (clone_flags & CLONE_THREAD)
  1299. threadgroup_change_end(current);
  1300. cgroup_exit(p, 0);
  1301. delayacct_tsk_free(p);
  1302. module_put(task_thread_info(p)->exec_domain->module);
  1303. bad_fork_cleanup_count:
  1304. atomic_dec(&p->cred->user->processes);
  1305. exit_creds(p);
  1306. bad_fork_free:
  1307. free_task(p);
  1308. fork_out:
  1309. return ERR_PTR(retval);
  1310. }
  1311. static inline void init_idle_pids(struct pid_link *links)
  1312. {
  1313. enum pid_type type;
  1314. for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
  1315. INIT_HLIST_NODE(&links[type].node); /* not really needed */
  1316. links[type].pid = &init_struct_pid;
  1317. }
  1318. }
  1319. struct task_struct * __cpuinit fork_idle(int cpu)
  1320. {
  1321. struct task_struct *task;
  1322. task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
  1323. if (!IS_ERR(task)) {
  1324. init_idle_pids(task->pids);
  1325. init_idle(task, cpu);
  1326. }
  1327. return task;
  1328. }
  1329. /*
  1330. * Ok, this is the main fork-routine.
  1331. *
  1332. * It copies the process, and if successful kick-starts
  1333. * it and waits for it to finish using the VM if required.
  1334. */
  1335. long do_fork(unsigned long clone_flags,
  1336. unsigned long stack_start,
  1337. unsigned long stack_size,
  1338. int __user *parent_tidptr,
  1339. int __user *child_tidptr)
  1340. {
  1341. struct task_struct *p;
  1342. int trace = 0;
  1343. long nr;
  1344. /*
  1345. * Do some preliminary argument and permissions checking before we
  1346. * actually start allocating stuff
  1347. */
  1348. if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
  1349. if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
  1350. return -EINVAL;
  1351. }
  1352. /*
  1353. * Determine whether and which event to report to ptracer. When
  1354. * called from kernel_thread or CLONE_UNTRACED is explicitly
  1355. * requested, no event is reported; otherwise, report if the event
  1356. * for the type of forking is enabled.
  1357. */
  1358. if (!(clone_flags & CLONE_UNTRACED)) {
  1359. if (clone_flags & CLONE_VFORK)
  1360. trace = PTRACE_EVENT_VFORK;
  1361. else if ((clone_flags & CSIGNAL) != SIGCHLD)
  1362. trace = PTRACE_EVENT_CLONE;
  1363. else
  1364. trace = PTRACE_EVENT_FORK;
  1365. if (likely(!ptrace_event_enabled(current, trace)))
  1366. trace = 0;
  1367. }
  1368. p = copy_process(clone_flags, stack_start, stack_size,
  1369. child_tidptr, NULL, trace);
  1370. /*
  1371. * Do this prior waking up the new thread - the thread pointer
  1372. * might get invalid after that point, if the thread exits quickly.
  1373. */
  1374. if (!IS_ERR(p)) {
  1375. struct completion vfork;
  1376. trace_sched_process_fork(current, p);
  1377. nr = task_pid_vnr(p);
  1378. if (clone_flags & CLONE_PARENT_SETTID)
  1379. put_user(nr, parent_tidptr);
  1380. if (clone_flags & CLONE_VFORK) {
  1381. p->vfork_done = &vfork;
  1382. init_completion(&vfork);
  1383. get_task_struct(p);
  1384. }
  1385. wake_up_new_task(p);
  1386. /* forking complete and child started to run, tell ptracer */
  1387. if (unlikely(trace))
  1388. ptrace_event(trace, nr);
  1389. if (clone_flags & CLONE_VFORK) {
  1390. if (!wait_for_vfork_done(p, &vfork))
  1391. ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
  1392. }
  1393. } else {
  1394. nr = PTR_ERR(p);
  1395. }
  1396. return nr;
  1397. }
  1398. /*
  1399. * Create a kernel thread.
  1400. */
  1401. pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
  1402. {
  1403. return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
  1404. (unsigned long)arg, NULL, NULL);
  1405. }
  1406. #ifdef __ARCH_WANT_SYS_FORK
  1407. SYSCALL_DEFINE0(fork)
  1408. {
  1409. #ifdef CONFIG_MMU
  1410. return do_fork(SIGCHLD, 0, 0, NULL, NULL);
  1411. #else
  1412. /* can not support in nommu mode */
  1413. return(-EINVAL);
  1414. #endif
  1415. }
  1416. #endif
  1417. #ifdef __ARCH_WANT_SYS_VFORK
  1418. SYSCALL_DEFINE0(vfork)
  1419. {
  1420. return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
  1421. 0, NULL, NULL);
  1422. }
  1423. #endif
  1424. #ifdef __ARCH_WANT_SYS_CLONE
  1425. #ifdef CONFIG_CLONE_BACKWARDS
  1426. SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
  1427. int __user *, parent_tidptr,
  1428. int, tls_val,
  1429. int __user *, child_tidptr)
  1430. #elif defined(CONFIG_CLONE_BACKWARDS2)
  1431. SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
  1432. int __user *, parent_tidptr,
  1433. int __user *, child_tidptr,
  1434. int, tls_val)
  1435. #else
  1436. SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
  1437. int __user *, parent_tidptr,
  1438. int __user *, child_tidptr,
  1439. int, tls_val)
  1440. #endif
  1441. {
  1442. long ret = do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
  1443. asmlinkage_protect(5, ret, clone_flags, newsp,
  1444. parent_tidptr, child_tidptr, tls_val);
  1445. return ret;
  1446. }
  1447. #endif
  1448. #ifndef ARCH_MIN_MMSTRUCT_ALIGN
  1449. #define ARCH_MIN_MMSTRUCT_ALIGN 0
  1450. #endif
  1451. static void sighand_ctor(void *data)
  1452. {
  1453. struct sighand_struct *sighand = data;
  1454. spin_lock_init(&sighand->siglock);
  1455. init_waitqueue_head(&sighand->signalfd_wqh);
  1456. }
  1457. void __init proc_caches_init(void)
  1458. {
  1459. sighand_cachep = kmem_cache_create("sighand_cache",
  1460. sizeof(struct sighand_struct), 0,
  1461. SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
  1462. SLAB_NOTRACK, sighand_ctor);
  1463. signal_cachep = kmem_cache_create("signal_cache",
  1464. sizeof(struct signal_struct), 0,
  1465. SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
  1466. files_cachep = kmem_cache_create("files_cache",
  1467. sizeof(struct files_struct), 0,
  1468. SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
  1469. fs_cachep = kmem_cache_create("fs_cache",
  1470. sizeof(struct fs_struct), 0,
  1471. SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
  1472. /*
  1473. * FIXME! The "sizeof(struct mm_struct)" currently includes the
  1474. * whole struct cpumask for the OFFSTACK case. We could change
  1475. * this to *only* allocate as much of it as required by the
  1476. * maximum number of CPU's we can ever have. The cpumask_allocation
  1477. * is at the end of the structure, exactly for that reason.
  1478. */
  1479. mm_cachep = kmem_cache_create("mm_struct",
  1480. sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
  1481. SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
  1482. vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
  1483. mmap_init();
  1484. nsproxy_cache_init();
  1485. }
  1486. /*
  1487. * Check constraints on flags passed to the unshare system call.
  1488. */
  1489. static int check_unshare_flags(unsigned long unshare_flags)
  1490. {
  1491. if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
  1492. CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
  1493. CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
  1494. CLONE_NEWUSER|CLONE_NEWPID))
  1495. return -EINVAL;
  1496. /*
  1497. * Not implemented, but pretend it works if there is nothing to
  1498. * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
  1499. * needs to unshare vm.
  1500. */
  1501. if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
  1502. /* FIXME: get_task_mm() increments ->mm_users */
  1503. if (atomic_read(&current->mm->mm_users) > 1)
  1504. return -EINVAL;
  1505. }
  1506. return 0;
  1507. }
  1508. /*
  1509. * Unshare the filesystem structure if it is being shared
  1510. */
  1511. static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
  1512. {
  1513. struct fs_struct *fs = current->fs;
  1514. if (!(unshare_flags & CLONE_FS) || !fs)
  1515. return 0;
  1516. /* don't need lock here; in the worst case we'll do useless copy */
  1517. if (fs->users == 1)
  1518. return 0;
  1519. *new_fsp = copy_fs_struct(fs);
  1520. if (!*new_fsp)
  1521. return -ENOMEM;
  1522. return 0;
  1523. }
  1524. /*
  1525. * Unshare file descriptor table if it is being shared
  1526. */
  1527. static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
  1528. {
  1529. struct files_struct *fd = current->files;
  1530. int error = 0;
  1531. if ((unshare_flags & CLONE_FILES) &&
  1532. (fd && atomic_read(&fd->count) > 1)) {
  1533. *new_fdp = dup_fd(fd, &error);
  1534. if (!*new_fdp)
  1535. return error;
  1536. }
  1537. return 0;
  1538. }
  1539. /*
  1540. * unshare allows a process to 'unshare' part of the process
  1541. * context which was originally shared using clone. copy_*
  1542. * functions used by do_fork() cannot be used here directly
  1543. * because they modify an inactive task_struct that is being
  1544. * constructed. Here we are modifying the current, active,
  1545. * task_struct.
  1546. */
  1547. SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
  1548. {
  1549. struct fs_struct *fs, *new_fs = NULL;
  1550. struct files_struct *fd, *new_fd = NULL;
  1551. struct cred *new_cred = NULL;
  1552. struct nsproxy *new_nsproxy = NULL;
  1553. int do_sysvsem = 0;
  1554. int err;
  1555. /*
  1556. * If unsharing a user namespace must also unshare the thread.
  1557. */
  1558. if (unshare_flags & CLONE_NEWUSER)
  1559. unshare_flags |= CLONE_THREAD | CLONE_FS;
  1560. /*
  1561. * If unsharing a pid namespace must also unshare the thread.
  1562. */
  1563. if (unshare_flags & CLONE_NEWPID)
  1564. unshare_flags |= CLONE_THREAD;
  1565. /*
  1566. * If unsharing a thread from a thread group, must also unshare vm.
  1567. */
  1568. if (unshare_flags & CLONE_THREAD)
  1569. unshare_flags |= CLONE_VM;
  1570. /*
  1571. * If unsharing vm, must also unshare signal handlers.
  1572. */
  1573. if (unshare_flags & CLONE_VM)
  1574. unshare_flags |= CLONE_SIGHAND;
  1575. /*
  1576. * If unsharing namespace, must also unshare filesystem information.
  1577. */
  1578. if (unshare_flags & CLONE_NEWNS)
  1579. unshare_flags |= CLONE_FS;
  1580. err = check_unshare_flags(unshare_flags);
  1581. if (err)
  1582. goto bad_unshare_out;
  1583. /*
  1584. * CLONE_NEWIPC must also detach from the undolist: after switching
  1585. * to a new ipc namespace, the semaphore arrays from the old
  1586. * namespace are unreachable.
  1587. */
  1588. if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
  1589. do_sysvsem = 1;
  1590. err = unshare_fs(unshare_flags, &new_fs);
  1591. if (err)
  1592. goto bad_unshare_out;
  1593. err = unshare_fd(unshare_flags, &new_fd);
  1594. if (err)
  1595. goto bad_unshare_cleanup_fs;
  1596. err = unshare_userns(unshare_flags, &new_cred);
  1597. if (err)
  1598. goto bad_unshare_cleanup_fd;
  1599. err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
  1600. new_cred, new_fs);
  1601. if (err)
  1602. goto bad_unshare_cleanup_cred;
  1603. if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
  1604. if (do_sysvsem) {
  1605. /*
  1606. * CLONE_SYSVSEM is equivalent to sys_exit().
  1607. */
  1608. exit_sem(current);
  1609. }
  1610. if (new_nsproxy)
  1611. switch_task_namespaces(current, new_nsproxy);
  1612. task_lock(current);
  1613. if (new_fs) {
  1614. fs = current->fs;
  1615. spin_lock(&fs->lock);
  1616. current->fs = new_fs;
  1617. if (--fs->users)
  1618. new_fs = NULL;
  1619. else
  1620. new_fs = fs;
  1621. spin_unlock(&fs->lock);
  1622. }
  1623. if (new_fd) {
  1624. fd = current->files;
  1625. current->files = new_fd;
  1626. new_fd = fd;
  1627. }
  1628. task_unlock(current);
  1629. if (new_cred) {
  1630. /* Install the new user namespace */
  1631. commit_creds(new_cred);
  1632. new_cred = NULL;
  1633. }
  1634. }
  1635. bad_unshare_cleanup_cred:
  1636. if (new_cred)
  1637. put_cred(new_cred);
  1638. bad_unshare_cleanup_fd:
  1639. if (new_fd)
  1640. put_files_struct(new_fd);
  1641. bad_unshare_cleanup_fs:
  1642. if (new_fs)
  1643. free_fs_struct(new_fs);
  1644. bad_unshare_out:
  1645. return err;
  1646. }
  1647. /*
  1648. * Helper to unshare the files of the current task.
  1649. * We don't want to expose copy_files internals to
  1650. * the exec layer of the kernel.
  1651. */
  1652. int unshare_files(struct files_struct **displaced)
  1653. {
  1654. struct task_struct *task = current;
  1655. struct files_struct *copy = NULL;
  1656. int error;
  1657. error = unshare_fd(CLONE_FILES, &copy);
  1658. if (error || !copy) {
  1659. *displaced = NULL;
  1660. return error;
  1661. }
  1662. *displaced = task->files;
  1663. task_lock(task);
  1664. task->files = copy;
  1665. task_unlock(task);
  1666. return 0;
  1667. }