Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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/config.h> |
| 15 | #include <linux/slab.h> |
| 16 | #include <linux/init.h> |
| 17 | #include <linux/unistd.h> |
| 18 | #include <linux/smp_lock.h> |
| 19 | #include <linux/module.h> |
| 20 | #include <linux/vmalloc.h> |
| 21 | #include <linux/completion.h> |
| 22 | #include <linux/namespace.h> |
| 23 | #include <linux/personality.h> |
| 24 | #include <linux/mempolicy.h> |
| 25 | #include <linux/sem.h> |
| 26 | #include <linux/file.h> |
| 27 | #include <linux/key.h> |
| 28 | #include <linux/binfmts.h> |
| 29 | #include <linux/mman.h> |
| 30 | #include <linux/fs.h> |
| 31 | #include <linux/cpu.h> |
| 32 | #include <linux/cpuset.h> |
| 33 | #include <linux/security.h> |
| 34 | #include <linux/swap.h> |
| 35 | #include <linux/syscalls.h> |
| 36 | #include <linux/jiffies.h> |
| 37 | #include <linux/futex.h> |
Dipankar Sarma | ab2af1f | 2005-09-09 13:04:13 -0700 | [diff] [blame] | 38 | #include <linux/rcupdate.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | #include <linux/ptrace.h> |
| 40 | #include <linux/mount.h> |
| 41 | #include <linux/audit.h> |
| 42 | #include <linux/profile.h> |
| 43 | #include <linux/rmap.h> |
| 44 | #include <linux/acct.h> |
| 45 | |
| 46 | #include <asm/pgtable.h> |
| 47 | #include <asm/pgalloc.h> |
| 48 | #include <asm/uaccess.h> |
| 49 | #include <asm/mmu_context.h> |
| 50 | #include <asm/cacheflush.h> |
| 51 | #include <asm/tlbflush.h> |
| 52 | |
| 53 | /* |
| 54 | * Protected counters by write_lock_irq(&tasklist_lock) |
| 55 | */ |
| 56 | unsigned long total_forks; /* Handle normal Linux uptimes. */ |
| 57 | int nr_threads; /* The idle threads do not count.. */ |
| 58 | |
| 59 | int max_threads; /* tunable limit on nr_threads */ |
| 60 | |
| 61 | DEFINE_PER_CPU(unsigned long, process_counts) = 0; |
| 62 | |
| 63 | __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */ |
| 64 | |
| 65 | EXPORT_SYMBOL(tasklist_lock); |
| 66 | |
| 67 | int nr_processes(void) |
| 68 | { |
| 69 | int cpu; |
| 70 | int total = 0; |
| 71 | |
| 72 | for_each_online_cpu(cpu) |
| 73 | total += per_cpu(process_counts, cpu); |
| 74 | |
| 75 | return total; |
| 76 | } |
| 77 | |
| 78 | #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR |
| 79 | # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL) |
| 80 | # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk)) |
| 81 | static kmem_cache_t *task_struct_cachep; |
| 82 | #endif |
| 83 | |
| 84 | /* SLAB cache for signal_struct structures (tsk->signal) */ |
| 85 | kmem_cache_t *signal_cachep; |
| 86 | |
| 87 | /* SLAB cache for sighand_struct structures (tsk->sighand) */ |
| 88 | kmem_cache_t *sighand_cachep; |
| 89 | |
| 90 | /* SLAB cache for files_struct structures (tsk->files) */ |
| 91 | kmem_cache_t *files_cachep; |
| 92 | |
| 93 | /* SLAB cache for fs_struct structures (tsk->fs) */ |
| 94 | kmem_cache_t *fs_cachep; |
| 95 | |
| 96 | /* SLAB cache for vm_area_struct structures */ |
| 97 | kmem_cache_t *vm_area_cachep; |
| 98 | |
| 99 | /* SLAB cache for mm_struct structures (tsk->mm) */ |
| 100 | static kmem_cache_t *mm_cachep; |
| 101 | |
| 102 | void free_task(struct task_struct *tsk) |
| 103 | { |
| 104 | free_thread_info(tsk->thread_info); |
| 105 | free_task_struct(tsk); |
| 106 | } |
| 107 | EXPORT_SYMBOL(free_task); |
| 108 | |
| 109 | void __put_task_struct(struct task_struct *tsk) |
| 110 | { |
| 111 | WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE))); |
| 112 | WARN_ON(atomic_read(&tsk->usage)); |
| 113 | WARN_ON(tsk == current); |
| 114 | |
| 115 | if (unlikely(tsk->audit_context)) |
| 116 | audit_free(tsk); |
| 117 | security_task_free(tsk); |
| 118 | free_uid(tsk->user); |
| 119 | put_group_info(tsk->group_info); |
| 120 | |
| 121 | if (!profile_handoff_task(tsk)) |
| 122 | free_task(tsk); |
| 123 | } |
| 124 | |
| 125 | void __init fork_init(unsigned long mempages) |
| 126 | { |
| 127 | #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR |
| 128 | #ifndef ARCH_MIN_TASKALIGN |
| 129 | #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES |
| 130 | #endif |
| 131 | /* create a slab on which task_structs can be allocated */ |
| 132 | task_struct_cachep = |
| 133 | kmem_cache_create("task_struct", sizeof(struct task_struct), |
| 134 | ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL); |
| 135 | #endif |
| 136 | |
| 137 | /* |
| 138 | * The default maximum number of threads is set to a safe |
| 139 | * value: the thread structures can take up at most half |
| 140 | * of memory. |
| 141 | */ |
| 142 | max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE); |
| 143 | |
| 144 | /* |
| 145 | * we need to allow at least 20 threads to boot a system |
| 146 | */ |
| 147 | if(max_threads < 20) |
| 148 | max_threads = 20; |
| 149 | |
| 150 | init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2; |
| 151 | init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2; |
| 152 | init_task.signal->rlim[RLIMIT_SIGPENDING] = |
| 153 | init_task.signal->rlim[RLIMIT_NPROC]; |
| 154 | } |
| 155 | |
| 156 | static struct task_struct *dup_task_struct(struct task_struct *orig) |
| 157 | { |
| 158 | struct task_struct *tsk; |
| 159 | struct thread_info *ti; |
| 160 | |
| 161 | prepare_to_copy(orig); |
| 162 | |
| 163 | tsk = alloc_task_struct(); |
| 164 | if (!tsk) |
| 165 | return NULL; |
| 166 | |
| 167 | ti = alloc_thread_info(tsk); |
| 168 | if (!ti) { |
| 169 | free_task_struct(tsk); |
| 170 | return NULL; |
| 171 | } |
| 172 | |
| 173 | *ti = *orig->thread_info; |
| 174 | *tsk = *orig; |
| 175 | tsk->thread_info = ti; |
| 176 | ti->task = tsk; |
| 177 | |
| 178 | /* One for us, one for whoever does the "release_task()" (usually parent) */ |
| 179 | atomic_set(&tsk->usage,2); |
Giancarlo Formicuccia | 4b5d37a | 2005-09-09 13:01:22 -0700 | [diff] [blame] | 180 | atomic_set(&tsk->fs_excl, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | return tsk; |
| 182 | } |
| 183 | |
| 184 | #ifdef CONFIG_MMU |
Hugh Dickins | fd3e42f | 2005-10-29 18:16:06 -0700 | [diff] [blame] | 185 | static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | { |
Hugh Dickins | fd3e42f | 2005-10-29 18:16:06 -0700 | [diff] [blame] | 187 | struct vm_area_struct *mpnt, *tmp, **pprev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | struct rb_node **rb_link, *rb_parent; |
| 189 | int retval; |
| 190 | unsigned long charge; |
| 191 | struct mempolicy *pol; |
| 192 | |
| 193 | down_write(&oldmm->mmap_sem); |
Hugh Dickins | fd3e42f | 2005-10-29 18:16:06 -0700 | [diff] [blame] | 194 | flush_cache_mm(oldmm); |
Hugh Dickins | 7ee7823 | 2005-10-29 18:16:08 -0700 | [diff] [blame] | 195 | down_write(&mm->mmap_sem); |
| 196 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 197 | mm->locked_vm = 0; |
| 198 | mm->mmap = NULL; |
| 199 | mm->mmap_cache = NULL; |
| 200 | mm->free_area_cache = oldmm->mmap_base; |
Wolfgang Wander | 1363c3c | 2005-06-21 17:14:49 -0700 | [diff] [blame] | 201 | mm->cached_hole_size = ~0UL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | mm->map_count = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 203 | cpus_clear(mm->cpu_vm_mask); |
| 204 | mm->mm_rb = RB_ROOT; |
| 205 | rb_link = &mm->mm_rb.rb_node; |
| 206 | rb_parent = NULL; |
| 207 | pprev = &mm->mmap; |
| 208 | |
Hugh Dickins | fd3e42f | 2005-10-29 18:16:06 -0700 | [diff] [blame] | 209 | for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | struct file *file; |
| 211 | |
| 212 | if (mpnt->vm_flags & VM_DONTCOPY) { |
Hugh Dickins | 3b6bfcd | 2005-07-12 13:58:09 -0700 | [diff] [blame] | 213 | long pages = vma_pages(mpnt); |
| 214 | mm->total_vm -= pages; |
Hugh Dickins | ab50b8e | 2005-10-29 18:15:56 -0700 | [diff] [blame] | 215 | vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file, |
Hugh Dickins | 3b6bfcd | 2005-07-12 13:58:09 -0700 | [diff] [blame] | 216 | -pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 217 | continue; |
| 218 | } |
| 219 | charge = 0; |
| 220 | if (mpnt->vm_flags & VM_ACCOUNT) { |
| 221 | unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT; |
| 222 | if (security_vm_enough_memory(len)) |
| 223 | goto fail_nomem; |
| 224 | charge = len; |
| 225 | } |
| 226 | tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL); |
| 227 | if (!tmp) |
| 228 | goto fail_nomem; |
| 229 | *tmp = *mpnt; |
| 230 | pol = mpol_copy(vma_policy(mpnt)); |
| 231 | retval = PTR_ERR(pol); |
| 232 | if (IS_ERR(pol)) |
| 233 | goto fail_nomem_policy; |
| 234 | vma_set_policy(tmp, pol); |
| 235 | tmp->vm_flags &= ~VM_LOCKED; |
| 236 | tmp->vm_mm = mm; |
| 237 | tmp->vm_next = NULL; |
| 238 | anon_vma_link(tmp); |
| 239 | file = tmp->vm_file; |
| 240 | if (file) { |
| 241 | struct inode *inode = file->f_dentry->d_inode; |
| 242 | get_file(file); |
| 243 | if (tmp->vm_flags & VM_DENYWRITE) |
| 244 | atomic_dec(&inode->i_writecount); |
| 245 | |
| 246 | /* insert tmp into the share list, just after mpnt */ |
| 247 | spin_lock(&file->f_mapping->i_mmap_lock); |
| 248 | tmp->vm_truncate_count = mpnt->vm_truncate_count; |
| 249 | flush_dcache_mmap_lock(file->f_mapping); |
| 250 | vma_prio_tree_add(tmp, mpnt); |
| 251 | flush_dcache_mmap_unlock(file->f_mapping); |
| 252 | spin_unlock(&file->f_mapping->i_mmap_lock); |
| 253 | } |
| 254 | |
| 255 | /* |
Hugh Dickins | 7ee7823 | 2005-10-29 18:16:08 -0700 | [diff] [blame] | 256 | * Link in the new vma and copy the page table entries. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 257 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 258 | *pprev = tmp; |
| 259 | pprev = &tmp->vm_next; |
| 260 | |
| 261 | __vma_link_rb(mm, tmp, rb_link, rb_parent); |
| 262 | rb_link = &tmp->vm_rb.rb_right; |
| 263 | rb_parent = &tmp->vm_rb; |
| 264 | |
| 265 | mm->map_count++; |
Hugh Dickins | fd3e42f | 2005-10-29 18:16:06 -0700 | [diff] [blame] | 266 | retval = copy_page_range(mm, oldmm, tmp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 267 | |
| 268 | if (tmp->vm_ops && tmp->vm_ops->open) |
| 269 | tmp->vm_ops->open(tmp); |
| 270 | |
| 271 | if (retval) |
| 272 | goto out; |
| 273 | } |
| 274 | retval = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | out: |
Hugh Dickins | 7ee7823 | 2005-10-29 18:16:08 -0700 | [diff] [blame] | 276 | up_write(&mm->mmap_sem); |
Hugh Dickins | fd3e42f | 2005-10-29 18:16:06 -0700 | [diff] [blame] | 277 | flush_tlb_mm(oldmm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 278 | up_write(&oldmm->mmap_sem); |
| 279 | return retval; |
| 280 | fail_nomem_policy: |
| 281 | kmem_cache_free(vm_area_cachep, tmp); |
| 282 | fail_nomem: |
| 283 | retval = -ENOMEM; |
| 284 | vm_unacct_memory(charge); |
| 285 | goto out; |
| 286 | } |
| 287 | |
| 288 | static inline int mm_alloc_pgd(struct mm_struct * mm) |
| 289 | { |
| 290 | mm->pgd = pgd_alloc(mm); |
| 291 | if (unlikely(!mm->pgd)) |
| 292 | return -ENOMEM; |
| 293 | return 0; |
| 294 | } |
| 295 | |
| 296 | static inline void mm_free_pgd(struct mm_struct * mm) |
| 297 | { |
| 298 | pgd_free(mm->pgd); |
| 299 | } |
| 300 | #else |
| 301 | #define dup_mmap(mm, oldmm) (0) |
| 302 | #define mm_alloc_pgd(mm) (0) |
| 303 | #define mm_free_pgd(mm) |
| 304 | #endif /* CONFIG_MMU */ |
| 305 | |
| 306 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock); |
| 307 | |
| 308 | #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL)) |
| 309 | #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm))) |
| 310 | |
| 311 | #include <linux/init_task.h> |
| 312 | |
| 313 | static struct mm_struct * mm_init(struct mm_struct * mm) |
| 314 | { |
| 315 | atomic_set(&mm->mm_users, 1); |
| 316 | atomic_set(&mm->mm_count, 1); |
| 317 | init_rwsem(&mm->mmap_sem); |
| 318 | INIT_LIST_HEAD(&mm->mmlist); |
| 319 | mm->core_waiters = 0; |
| 320 | mm->nr_ptes = 0; |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 321 | set_mm_counter(mm, file_rss, 0); |
Hugh Dickins | 404351e | 2005-10-29 18:16:04 -0700 | [diff] [blame] | 322 | set_mm_counter(mm, anon_rss, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 323 | spin_lock_init(&mm->page_table_lock); |
| 324 | rwlock_init(&mm->ioctx_list_lock); |
| 325 | mm->ioctx_list = NULL; |
| 326 | mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm); |
| 327 | mm->free_area_cache = TASK_UNMAPPED_BASE; |
Wolfgang Wander | 1363c3c | 2005-06-21 17:14:49 -0700 | [diff] [blame] | 328 | mm->cached_hole_size = ~0UL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 329 | |
| 330 | if (likely(!mm_alloc_pgd(mm))) { |
| 331 | mm->def_flags = 0; |
| 332 | return mm; |
| 333 | } |
| 334 | free_mm(mm); |
| 335 | return NULL; |
| 336 | } |
| 337 | |
| 338 | /* |
| 339 | * Allocate and initialize an mm_struct. |
| 340 | */ |
| 341 | struct mm_struct * mm_alloc(void) |
| 342 | { |
| 343 | struct mm_struct * mm; |
| 344 | |
| 345 | mm = allocate_mm(); |
| 346 | if (mm) { |
| 347 | memset(mm, 0, sizeof(*mm)); |
| 348 | mm = mm_init(mm); |
| 349 | } |
| 350 | return mm; |
| 351 | } |
| 352 | |
| 353 | /* |
| 354 | * Called when the last reference to the mm |
| 355 | * is dropped: either by a lazy thread or by |
| 356 | * mmput. Free the page directory and the mm. |
| 357 | */ |
| 358 | void fastcall __mmdrop(struct mm_struct *mm) |
| 359 | { |
| 360 | BUG_ON(mm == &init_mm); |
| 361 | mm_free_pgd(mm); |
| 362 | destroy_context(mm); |
| 363 | free_mm(mm); |
| 364 | } |
| 365 | |
| 366 | /* |
| 367 | * Decrement the use count and release all resources for an mm. |
| 368 | */ |
| 369 | void mmput(struct mm_struct *mm) |
| 370 | { |
| 371 | if (atomic_dec_and_test(&mm->mm_users)) { |
| 372 | exit_aio(mm); |
| 373 | exit_mmap(mm); |
| 374 | if (!list_empty(&mm->mmlist)) { |
| 375 | spin_lock(&mmlist_lock); |
| 376 | list_del(&mm->mmlist); |
| 377 | spin_unlock(&mmlist_lock); |
| 378 | } |
| 379 | put_swap_token(mm); |
| 380 | mmdrop(mm); |
| 381 | } |
| 382 | } |
| 383 | EXPORT_SYMBOL_GPL(mmput); |
| 384 | |
| 385 | /** |
| 386 | * get_task_mm - acquire a reference to the task's mm |
| 387 | * |
| 388 | * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning |
| 389 | * this kernel workthread has transiently adopted a user mm with use_mm, |
| 390 | * to do its AIO) is not set and if so returns a reference to it, after |
| 391 | * bumping up the use count. User must release the mm via mmput() |
| 392 | * after use. Typically used by /proc and ptrace. |
| 393 | */ |
| 394 | struct mm_struct *get_task_mm(struct task_struct *task) |
| 395 | { |
| 396 | struct mm_struct *mm; |
| 397 | |
| 398 | task_lock(task); |
| 399 | mm = task->mm; |
| 400 | if (mm) { |
| 401 | if (task->flags & PF_BORROWED_MM) |
| 402 | mm = NULL; |
| 403 | else |
| 404 | atomic_inc(&mm->mm_users); |
| 405 | } |
| 406 | task_unlock(task); |
| 407 | return mm; |
| 408 | } |
| 409 | EXPORT_SYMBOL_GPL(get_task_mm); |
| 410 | |
| 411 | /* Please note the differences between mmput and mm_release. |
| 412 | * mmput is called whenever we stop holding onto a mm_struct, |
| 413 | * error success whatever. |
| 414 | * |
| 415 | * mm_release is called after a mm_struct has been removed |
| 416 | * from the current process. |
| 417 | * |
| 418 | * This difference is important for error handling, when we |
| 419 | * only half set up a mm_struct for a new process and need to restore |
| 420 | * the old one. Because we mmput the new mm_struct before |
| 421 | * restoring the old one. . . |
| 422 | * Eric Biederman 10 January 1998 |
| 423 | */ |
| 424 | void mm_release(struct task_struct *tsk, struct mm_struct *mm) |
| 425 | { |
| 426 | struct completion *vfork_done = tsk->vfork_done; |
| 427 | |
| 428 | /* Get rid of any cached register state */ |
| 429 | deactivate_mm(tsk, mm); |
| 430 | |
| 431 | /* notify parent sleeping on vfork() */ |
| 432 | if (vfork_done) { |
| 433 | tsk->vfork_done = NULL; |
| 434 | complete(vfork_done); |
| 435 | } |
| 436 | if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) { |
| 437 | u32 __user * tidptr = tsk->clear_child_tid; |
| 438 | tsk->clear_child_tid = NULL; |
| 439 | |
| 440 | /* |
| 441 | * We don't check the error code - if userspace has |
| 442 | * not set up a proper pointer then tough luck. |
| 443 | */ |
| 444 | put_user(0, tidptr); |
| 445 | sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0); |
| 446 | } |
| 447 | } |
| 448 | |
| 449 | static int copy_mm(unsigned long clone_flags, struct task_struct * tsk) |
| 450 | { |
| 451 | struct mm_struct * mm, *oldmm; |
| 452 | int retval; |
| 453 | |
| 454 | tsk->min_flt = tsk->maj_flt = 0; |
| 455 | tsk->nvcsw = tsk->nivcsw = 0; |
| 456 | |
| 457 | tsk->mm = NULL; |
| 458 | tsk->active_mm = NULL; |
| 459 | |
| 460 | /* |
| 461 | * Are we cloning a kernel thread? |
| 462 | * |
| 463 | * We need to steal a active VM for that.. |
| 464 | */ |
| 465 | oldmm = current->mm; |
| 466 | if (!oldmm) |
| 467 | return 0; |
| 468 | |
| 469 | if (clone_flags & CLONE_VM) { |
| 470 | atomic_inc(&oldmm->mm_users); |
| 471 | mm = oldmm; |
| 472 | /* |
| 473 | * There are cases where the PTL is held to ensure no |
| 474 | * new threads start up in user mode using an mm, which |
| 475 | * allows optimizing out ipis; the tlb_gather_mmu code |
| 476 | * is an example. |
| 477 | */ |
| 478 | spin_unlock_wait(&oldmm->page_table_lock); |
| 479 | goto good_mm; |
| 480 | } |
| 481 | |
| 482 | retval = -ENOMEM; |
| 483 | mm = allocate_mm(); |
| 484 | if (!mm) |
| 485 | goto fail_nomem; |
| 486 | |
| 487 | /* Copy the current MM stuff.. */ |
| 488 | memcpy(mm, oldmm, sizeof(*mm)); |
| 489 | if (!mm_init(mm)) |
| 490 | goto fail_nomem; |
| 491 | |
| 492 | if (init_new_context(tsk,mm)) |
| 493 | goto fail_nocontext; |
| 494 | |
| 495 | retval = dup_mmap(mm, oldmm); |
| 496 | if (retval) |
| 497 | goto free_pt; |
| 498 | |
Hugh Dickins | 4294621 | 2005-10-29 18:16:05 -0700 | [diff] [blame] | 499 | mm->hiwater_rss = get_mm_rss(mm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 500 | mm->hiwater_vm = mm->total_vm; |
| 501 | |
| 502 | good_mm: |
| 503 | tsk->mm = mm; |
| 504 | tsk->active_mm = mm; |
| 505 | return 0; |
| 506 | |
| 507 | free_pt: |
| 508 | mmput(mm); |
| 509 | fail_nomem: |
| 510 | return retval; |
| 511 | |
| 512 | fail_nocontext: |
| 513 | /* |
| 514 | * If init_new_context() failed, we cannot use mmput() to free the mm |
| 515 | * because it calls destroy_context() |
| 516 | */ |
| 517 | mm_free_pgd(mm); |
| 518 | free_mm(mm); |
| 519 | return retval; |
| 520 | } |
| 521 | |
| 522 | static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old) |
| 523 | { |
| 524 | struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL); |
| 525 | /* We don't need to lock fs - think why ;-) */ |
| 526 | if (fs) { |
| 527 | atomic_set(&fs->count, 1); |
| 528 | rwlock_init(&fs->lock); |
| 529 | fs->umask = old->umask; |
| 530 | read_lock(&old->lock); |
| 531 | fs->rootmnt = mntget(old->rootmnt); |
| 532 | fs->root = dget(old->root); |
| 533 | fs->pwdmnt = mntget(old->pwdmnt); |
| 534 | fs->pwd = dget(old->pwd); |
| 535 | if (old->altroot) { |
| 536 | fs->altrootmnt = mntget(old->altrootmnt); |
| 537 | fs->altroot = dget(old->altroot); |
| 538 | } else { |
| 539 | fs->altrootmnt = NULL; |
| 540 | fs->altroot = NULL; |
| 541 | } |
| 542 | read_unlock(&old->lock); |
| 543 | } |
| 544 | return fs; |
| 545 | } |
| 546 | |
| 547 | struct fs_struct *copy_fs_struct(struct fs_struct *old) |
| 548 | { |
| 549 | return __copy_fs_struct(old); |
| 550 | } |
| 551 | |
| 552 | EXPORT_SYMBOL_GPL(copy_fs_struct); |
| 553 | |
| 554 | static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk) |
| 555 | { |
| 556 | if (clone_flags & CLONE_FS) { |
| 557 | atomic_inc(¤t->fs->count); |
| 558 | return 0; |
| 559 | } |
| 560 | tsk->fs = __copy_fs_struct(current->fs); |
| 561 | if (!tsk->fs) |
| 562 | return -ENOMEM; |
| 563 | return 0; |
| 564 | } |
| 565 | |
Dipankar Sarma | ab2af1f | 2005-09-09 13:04:13 -0700 | [diff] [blame] | 566 | static int count_open_files(struct fdtable *fdt) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 567 | { |
Dipankar Sarma | ab2af1f | 2005-09-09 13:04:13 -0700 | [diff] [blame] | 568 | int size = fdt->max_fdset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 569 | int i; |
| 570 | |
| 571 | /* Find the last open fd */ |
| 572 | for (i = size/(8*sizeof(long)); i > 0; ) { |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 573 | if (fdt->open_fds->fds_bits[--i]) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | break; |
| 575 | } |
| 576 | i = (i+1) * 8 * sizeof(long); |
| 577 | return i; |
| 578 | } |
| 579 | |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 580 | static struct files_struct *alloc_files(void) |
| 581 | { |
| 582 | struct files_struct *newf; |
| 583 | struct fdtable *fdt; |
| 584 | |
| 585 | newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL); |
| 586 | if (!newf) |
| 587 | goto out; |
| 588 | |
| 589 | atomic_set(&newf->count, 1); |
| 590 | |
| 591 | spin_lock_init(&newf->file_lock); |
Dipankar Sarma | ab2af1f | 2005-09-09 13:04:13 -0700 | [diff] [blame] | 592 | fdt = &newf->fdtab; |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 593 | fdt->next_fd = 0; |
| 594 | fdt->max_fds = NR_OPEN_DEFAULT; |
| 595 | fdt->max_fdset = __FD_SETSIZE; |
| 596 | fdt->close_on_exec = &newf->close_on_exec_init; |
| 597 | fdt->open_fds = &newf->open_fds_init; |
| 598 | fdt->fd = &newf->fd_array[0]; |
Dipankar Sarma | ab2af1f | 2005-09-09 13:04:13 -0700 | [diff] [blame] | 599 | INIT_RCU_HEAD(&fdt->rcu); |
| 600 | fdt->free_files = NULL; |
| 601 | fdt->next = NULL; |
| 602 | rcu_assign_pointer(newf->fdt, fdt); |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 603 | out: |
| 604 | return newf; |
| 605 | } |
| 606 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | static int copy_files(unsigned long clone_flags, struct task_struct * tsk) |
| 608 | { |
| 609 | struct files_struct *oldf, *newf; |
| 610 | struct file **old_fds, **new_fds; |
| 611 | int open_files, size, i, error = 0, expand; |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 612 | struct fdtable *old_fdt, *new_fdt; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | |
| 614 | /* |
| 615 | * A background process may not have any files ... |
| 616 | */ |
| 617 | oldf = current->files; |
| 618 | if (!oldf) |
| 619 | goto out; |
| 620 | |
| 621 | if (clone_flags & CLONE_FILES) { |
| 622 | atomic_inc(&oldf->count); |
| 623 | goto out; |
| 624 | } |
| 625 | |
| 626 | /* |
| 627 | * Note: we may be using current for both targets (See exec.c) |
| 628 | * This works because we cache current->files (old) as oldf. Don't |
| 629 | * break this. |
| 630 | */ |
| 631 | tsk->files = NULL; |
| 632 | error = -ENOMEM; |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 633 | newf = alloc_files(); |
| 634 | if (!newf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | goto out; |
| 636 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 637 | spin_lock(&oldf->file_lock); |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 638 | old_fdt = files_fdtable(oldf); |
| 639 | new_fdt = files_fdtable(newf); |
| 640 | size = old_fdt->max_fdset; |
Dipankar Sarma | ab2af1f | 2005-09-09 13:04:13 -0700 | [diff] [blame] | 641 | open_files = count_open_files(old_fdt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 642 | expand = 0; |
| 643 | |
| 644 | /* |
| 645 | * Check whether we need to allocate a larger fd array or fd set. |
| 646 | * Note: we're not a clone task, so the open count won't change. |
| 647 | */ |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 648 | if (open_files > new_fdt->max_fdset) { |
| 649 | new_fdt->max_fdset = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 650 | expand = 1; |
| 651 | } |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 652 | if (open_files > new_fdt->max_fds) { |
| 653 | new_fdt->max_fds = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 654 | expand = 1; |
| 655 | } |
| 656 | |
| 657 | /* if the old fdset gets grown now, we'll only copy up to "size" fds */ |
| 658 | if (expand) { |
| 659 | spin_unlock(&oldf->file_lock); |
| 660 | spin_lock(&newf->file_lock); |
| 661 | error = expand_files(newf, open_files-1); |
| 662 | spin_unlock(&newf->file_lock); |
| 663 | if (error < 0) |
| 664 | goto out_release; |
Dipankar Sarma | ab2af1f | 2005-09-09 13:04:13 -0700 | [diff] [blame] | 665 | new_fdt = files_fdtable(newf); |
| 666 | /* |
| 667 | * Reacquire the oldf lock and a pointer to its fd table |
| 668 | * who knows it may have a new bigger fd table. We need |
| 669 | * the latest pointer. |
| 670 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 671 | spin_lock(&oldf->file_lock); |
Dipankar Sarma | ab2af1f | 2005-09-09 13:04:13 -0700 | [diff] [blame] | 672 | old_fdt = files_fdtable(oldf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | } |
| 674 | |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 675 | old_fds = old_fdt->fd; |
| 676 | new_fds = new_fdt->fd; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 677 | |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 678 | memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8); |
| 679 | memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 680 | |
| 681 | for (i = open_files; i != 0; i--) { |
| 682 | struct file *f = *old_fds++; |
| 683 | if (f) { |
| 684 | get_file(f); |
| 685 | } else { |
| 686 | /* |
| 687 | * The fd may be claimed in the fd bitmap but not yet |
| 688 | * instantiated in the files array if a sibling thread |
| 689 | * is partway through open(). So make sure that this |
| 690 | * fd is available to the new process. |
| 691 | */ |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 692 | FD_CLR(open_files - i, new_fdt->open_fds); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | } |
Dipankar Sarma | ab2af1f | 2005-09-09 13:04:13 -0700 | [diff] [blame] | 694 | rcu_assign_pointer(*new_fds++, f); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 695 | } |
| 696 | spin_unlock(&oldf->file_lock); |
| 697 | |
| 698 | /* compute the remainder to be cleared */ |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 699 | size = (new_fdt->max_fds - open_files) * sizeof(struct file *); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 700 | |
| 701 | /* This is long word aligned thus could use a optimized version */ |
| 702 | memset(new_fds, 0, size); |
| 703 | |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 704 | if (new_fdt->max_fdset > open_files) { |
| 705 | int left = (new_fdt->max_fdset-open_files)/8; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 706 | int start = open_files / (8 * sizeof(unsigned long)); |
| 707 | |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 708 | memset(&new_fdt->open_fds->fds_bits[start], 0, left); |
| 709 | memset(&new_fdt->close_on_exec->fds_bits[start], 0, left); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 710 | } |
| 711 | |
| 712 | tsk->files = newf; |
| 713 | error = 0; |
| 714 | out: |
| 715 | return error; |
| 716 | |
| 717 | out_release: |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame] | 718 | free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset); |
| 719 | free_fdset (new_fdt->open_fds, new_fdt->max_fdset); |
| 720 | free_fd_array(new_fdt->fd, new_fdt->max_fds); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 721 | kmem_cache_free(files_cachep, newf); |
| 722 | goto out; |
| 723 | } |
| 724 | |
| 725 | /* |
| 726 | * Helper to unshare the files of the current task. |
| 727 | * We don't want to expose copy_files internals to |
| 728 | * the exec layer of the kernel. |
| 729 | */ |
| 730 | |
| 731 | int unshare_files(void) |
| 732 | { |
| 733 | struct files_struct *files = current->files; |
| 734 | int rc; |
| 735 | |
| 736 | if(!files) |
| 737 | BUG(); |
| 738 | |
| 739 | /* This can race but the race causes us to copy when we don't |
| 740 | need to and drop the copy */ |
| 741 | if(atomic_read(&files->count) == 1) |
| 742 | { |
| 743 | atomic_inc(&files->count); |
| 744 | return 0; |
| 745 | } |
| 746 | rc = copy_files(0, current); |
| 747 | if(rc) |
| 748 | current->files = files; |
| 749 | return rc; |
| 750 | } |
| 751 | |
| 752 | EXPORT_SYMBOL(unshare_files); |
| 753 | |
| 754 | static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk) |
| 755 | { |
| 756 | struct sighand_struct *sig; |
| 757 | |
| 758 | if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) { |
| 759 | atomic_inc(¤t->sighand->count); |
| 760 | return 0; |
| 761 | } |
| 762 | sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); |
| 763 | tsk->sighand = sig; |
| 764 | if (!sig) |
| 765 | return -ENOMEM; |
| 766 | spin_lock_init(&sig->siglock); |
| 767 | atomic_set(&sig->count, 1); |
| 768 | memcpy(sig->action, current->sighand->action, sizeof(sig->action)); |
| 769 | return 0; |
| 770 | } |
| 771 | |
| 772 | static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk) |
| 773 | { |
| 774 | struct signal_struct *sig; |
| 775 | int ret; |
| 776 | |
| 777 | if (clone_flags & CLONE_THREAD) { |
| 778 | atomic_inc(¤t->signal->count); |
| 779 | atomic_inc(¤t->signal->live); |
| 780 | return 0; |
| 781 | } |
| 782 | sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL); |
| 783 | tsk->signal = sig; |
| 784 | if (!sig) |
| 785 | return -ENOMEM; |
| 786 | |
| 787 | ret = copy_thread_group_keys(tsk); |
| 788 | if (ret < 0) { |
| 789 | kmem_cache_free(signal_cachep, sig); |
| 790 | return ret; |
| 791 | } |
| 792 | |
| 793 | atomic_set(&sig->count, 1); |
| 794 | atomic_set(&sig->live, 1); |
| 795 | init_waitqueue_head(&sig->wait_chldexit); |
| 796 | sig->flags = 0; |
| 797 | sig->group_exit_code = 0; |
| 798 | sig->group_exit_task = NULL; |
| 799 | sig->group_stop_count = 0; |
| 800 | sig->curr_target = NULL; |
| 801 | init_sigpending(&sig->shared_pending); |
| 802 | INIT_LIST_HEAD(&sig->posix_timers); |
| 803 | |
| 804 | sig->it_real_value = sig->it_real_incr = 0; |
| 805 | sig->real_timer.function = it_real_fn; |
| 806 | sig->real_timer.data = (unsigned long) tsk; |
| 807 | init_timer(&sig->real_timer); |
| 808 | |
| 809 | sig->it_virt_expires = cputime_zero; |
| 810 | sig->it_virt_incr = cputime_zero; |
| 811 | sig->it_prof_expires = cputime_zero; |
| 812 | sig->it_prof_incr = cputime_zero; |
| 813 | |
| 814 | sig->tty = current->signal->tty; |
| 815 | sig->pgrp = process_group(current); |
| 816 | sig->session = current->signal->session; |
| 817 | sig->leader = 0; /* session leadership doesn't inherit */ |
| 818 | sig->tty_old_pgrp = 0; |
| 819 | |
| 820 | sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero; |
| 821 | sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0; |
| 822 | sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0; |
| 823 | sig->sched_time = 0; |
| 824 | INIT_LIST_HEAD(&sig->cpu_timers[0]); |
| 825 | INIT_LIST_HEAD(&sig->cpu_timers[1]); |
| 826 | INIT_LIST_HEAD(&sig->cpu_timers[2]); |
| 827 | |
| 828 | task_lock(current->group_leader); |
| 829 | memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); |
| 830 | task_unlock(current->group_leader); |
| 831 | |
| 832 | if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) { |
| 833 | /* |
| 834 | * New sole thread in the process gets an expiry time |
| 835 | * of the whole CPU time limit. |
| 836 | */ |
| 837 | tsk->it_prof_expires = |
| 838 | secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur); |
| 839 | } |
| 840 | |
| 841 | return 0; |
| 842 | } |
| 843 | |
| 844 | static inline void copy_flags(unsigned long clone_flags, struct task_struct *p) |
| 845 | { |
| 846 | unsigned long new_flags = p->flags; |
| 847 | |
Alan Stern | d1209d0 | 2005-10-19 21:23:51 -0700 | [diff] [blame] | 848 | new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 849 | new_flags |= PF_FORKNOEXEC; |
| 850 | if (!(clone_flags & CLONE_PTRACE)) |
| 851 | p->ptrace = 0; |
| 852 | p->flags = new_flags; |
| 853 | } |
| 854 | |
| 855 | asmlinkage long sys_set_tid_address(int __user *tidptr) |
| 856 | { |
| 857 | current->clear_child_tid = tidptr; |
| 858 | |
| 859 | return current->pid; |
| 860 | } |
| 861 | |
| 862 | /* |
| 863 | * This creates a new process as a copy of the old one, |
| 864 | * but does not actually start it yet. |
| 865 | * |
| 866 | * It copies the registers, and all the appropriate |
| 867 | * parts of the process environment (as per the clone |
| 868 | * flags). The actual kick-off is left to the caller. |
| 869 | */ |
| 870 | static task_t *copy_process(unsigned long clone_flags, |
| 871 | unsigned long stack_start, |
| 872 | struct pt_regs *regs, |
| 873 | unsigned long stack_size, |
| 874 | int __user *parent_tidptr, |
| 875 | int __user *child_tidptr, |
| 876 | int pid) |
| 877 | { |
| 878 | int retval; |
| 879 | struct task_struct *p = NULL; |
| 880 | |
| 881 | if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) |
| 882 | return ERR_PTR(-EINVAL); |
| 883 | |
| 884 | /* |
| 885 | * Thread groups must share signals as well, and detached threads |
| 886 | * can only be started up within the thread group. |
| 887 | */ |
| 888 | if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) |
| 889 | return ERR_PTR(-EINVAL); |
| 890 | |
| 891 | /* |
| 892 | * Shared signal handlers imply shared VM. By way of the above, |
| 893 | * thread groups also imply shared VM. Blocking this case allows |
| 894 | * for various simplifications in other code. |
| 895 | */ |
| 896 | if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) |
| 897 | return ERR_PTR(-EINVAL); |
| 898 | |
| 899 | retval = security_task_create(clone_flags); |
| 900 | if (retval) |
| 901 | goto fork_out; |
| 902 | |
| 903 | retval = -ENOMEM; |
| 904 | p = dup_task_struct(current); |
| 905 | if (!p) |
| 906 | goto fork_out; |
| 907 | |
| 908 | retval = -EAGAIN; |
| 909 | if (atomic_read(&p->user->processes) >= |
| 910 | p->signal->rlim[RLIMIT_NPROC].rlim_cur) { |
| 911 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) && |
| 912 | p->user != &root_user) |
| 913 | goto bad_fork_free; |
| 914 | } |
| 915 | |
| 916 | atomic_inc(&p->user->__count); |
| 917 | atomic_inc(&p->user->processes); |
| 918 | get_group_info(p->group_info); |
| 919 | |
| 920 | /* |
| 921 | * If multiple threads are within copy_process(), then this check |
| 922 | * triggers too late. This doesn't hurt, the check is only there |
| 923 | * to stop root fork bombs. |
| 924 | */ |
| 925 | if (nr_threads >= max_threads) |
| 926 | goto bad_fork_cleanup_count; |
| 927 | |
| 928 | if (!try_module_get(p->thread_info->exec_domain->module)) |
| 929 | goto bad_fork_cleanup_count; |
| 930 | |
| 931 | if (p->binfmt && !try_module_get(p->binfmt->module)) |
| 932 | goto bad_fork_cleanup_put_domain; |
| 933 | |
| 934 | p->did_exec = 0; |
| 935 | copy_flags(clone_flags, p); |
| 936 | p->pid = pid; |
| 937 | retval = -EFAULT; |
| 938 | if (clone_flags & CLONE_PARENT_SETTID) |
| 939 | if (put_user(p->pid, parent_tidptr)) |
| 940 | goto bad_fork_cleanup; |
| 941 | |
| 942 | p->proc_dentry = NULL; |
| 943 | |
| 944 | INIT_LIST_HEAD(&p->children); |
| 945 | INIT_LIST_HEAD(&p->sibling); |
| 946 | p->vfork_done = NULL; |
| 947 | spin_lock_init(&p->alloc_lock); |
| 948 | spin_lock_init(&p->proc_lock); |
| 949 | |
| 950 | clear_tsk_thread_flag(p, TIF_SIGPENDING); |
| 951 | init_sigpending(&p->pending); |
| 952 | |
| 953 | p->utime = cputime_zero; |
| 954 | p->stime = cputime_zero; |
| 955 | p->sched_time = 0; |
| 956 | p->rchar = 0; /* I/O counter: bytes read */ |
| 957 | p->wchar = 0; /* I/O counter: bytes written */ |
| 958 | p->syscr = 0; /* I/O counter: read syscalls */ |
| 959 | p->syscw = 0; /* I/O counter: write syscalls */ |
| 960 | acct_clear_integrals(p); |
| 961 | |
| 962 | p->it_virt_expires = cputime_zero; |
| 963 | p->it_prof_expires = cputime_zero; |
| 964 | p->it_sched_expires = 0; |
| 965 | INIT_LIST_HEAD(&p->cpu_timers[0]); |
| 966 | INIT_LIST_HEAD(&p->cpu_timers[1]); |
| 967 | INIT_LIST_HEAD(&p->cpu_timers[2]); |
| 968 | |
| 969 | p->lock_depth = -1; /* -1 = no lock */ |
| 970 | do_posix_clock_monotonic_gettime(&p->start_time); |
| 971 | p->security = NULL; |
| 972 | p->io_context = NULL; |
| 973 | p->io_wait = NULL; |
| 974 | p->audit_context = NULL; |
| 975 | #ifdef CONFIG_NUMA |
| 976 | p->mempolicy = mpol_copy(p->mempolicy); |
| 977 | if (IS_ERR(p->mempolicy)) { |
| 978 | retval = PTR_ERR(p->mempolicy); |
| 979 | p->mempolicy = NULL; |
| 980 | goto bad_fork_cleanup; |
| 981 | } |
| 982 | #endif |
| 983 | |
| 984 | p->tgid = p->pid; |
| 985 | if (clone_flags & CLONE_THREAD) |
| 986 | p->tgid = current->tgid; |
| 987 | |
| 988 | if ((retval = security_task_alloc(p))) |
| 989 | goto bad_fork_cleanup_policy; |
| 990 | if ((retval = audit_alloc(p))) |
| 991 | goto bad_fork_cleanup_security; |
| 992 | /* copy all the process information */ |
| 993 | if ((retval = copy_semundo(clone_flags, p))) |
| 994 | goto bad_fork_cleanup_audit; |
| 995 | if ((retval = copy_files(clone_flags, p))) |
| 996 | goto bad_fork_cleanup_semundo; |
| 997 | if ((retval = copy_fs(clone_flags, p))) |
| 998 | goto bad_fork_cleanup_files; |
| 999 | if ((retval = copy_sighand(clone_flags, p))) |
| 1000 | goto bad_fork_cleanup_fs; |
| 1001 | if ((retval = copy_signal(clone_flags, p))) |
| 1002 | goto bad_fork_cleanup_sighand; |
| 1003 | if ((retval = copy_mm(clone_flags, p))) |
| 1004 | goto bad_fork_cleanup_signal; |
| 1005 | if ((retval = copy_keys(clone_flags, p))) |
| 1006 | goto bad_fork_cleanup_mm; |
| 1007 | if ((retval = copy_namespace(clone_flags, p))) |
| 1008 | goto bad_fork_cleanup_keys; |
| 1009 | retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs); |
| 1010 | if (retval) |
| 1011 | goto bad_fork_cleanup_namespace; |
| 1012 | |
| 1013 | p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL; |
| 1014 | /* |
| 1015 | * Clear TID on mm_release()? |
| 1016 | */ |
| 1017 | p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL; |
| 1018 | |
| 1019 | /* |
| 1020 | * Syscall tracing should be turned off in the child regardless |
| 1021 | * of CLONE_PTRACE. |
| 1022 | */ |
| 1023 | clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE); |
Laurent Vivier | ed75e8d | 2005-09-03 15:57:18 -0700 | [diff] [blame] | 1024 | #ifdef TIF_SYSCALL_EMU |
| 1025 | clear_tsk_thread_flag(p, TIF_SYSCALL_EMU); |
| 1026 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1027 | |
| 1028 | /* Our parent execution domain becomes current domain |
| 1029 | These must match for thread signalling to apply */ |
| 1030 | |
| 1031 | p->parent_exec_id = p->self_exec_id; |
| 1032 | |
| 1033 | /* ok, now we should be set up.. */ |
| 1034 | p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL); |
| 1035 | p->pdeath_signal = 0; |
| 1036 | p->exit_state = 0; |
| 1037 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1038 | /* |
| 1039 | * Ok, make it visible to the rest of the system. |
| 1040 | * We dont wake it up yet. |
| 1041 | */ |
| 1042 | p->group_leader = p; |
| 1043 | INIT_LIST_HEAD(&p->ptrace_children); |
| 1044 | INIT_LIST_HEAD(&p->ptrace_list); |
| 1045 | |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1046 | /* Perform scheduler related setup. Assign this task to a CPU. */ |
| 1047 | sched_fork(p, clone_flags); |
| 1048 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1049 | /* Need tasklist lock for parent etc handling! */ |
| 1050 | write_lock_irq(&tasklist_lock); |
| 1051 | |
| 1052 | /* |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1053 | * The task hasn't been attached yet, so its cpus_allowed mask will |
| 1054 | * not be changed, nor will its assigned CPU. |
| 1055 | * |
| 1056 | * The cpus_allowed mask of the parent may have changed after it was |
| 1057 | * copied first time - so re-copy it here, then check the child's CPU |
| 1058 | * to ensure it is on a valid CPU (and if not, just force it back to |
| 1059 | * parent's CPU). This avoids alot of nasty races. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1060 | */ |
| 1061 | p->cpus_allowed = current->cpus_allowed; |
Srivatsa Vaddagiri | 26ff6ad | 2005-09-16 19:27:40 -0700 | [diff] [blame] | 1062 | if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) || |
| 1063 | !cpu_online(task_cpu(p)))) |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1064 | set_task_cpu(p, smp_processor_id()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1065 | |
| 1066 | /* |
| 1067 | * Check for pending SIGKILL! The new thread should not be allowed |
| 1068 | * to slip out of an OOM kill. (or normal SIGKILL.) |
| 1069 | */ |
| 1070 | if (sigismember(¤t->pending.signal, SIGKILL)) { |
| 1071 | write_unlock_irq(&tasklist_lock); |
| 1072 | retval = -EINTR; |
| 1073 | goto bad_fork_cleanup_namespace; |
| 1074 | } |
| 1075 | |
| 1076 | /* CLONE_PARENT re-uses the old parent */ |
| 1077 | if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) |
| 1078 | p->real_parent = current->real_parent; |
| 1079 | else |
| 1080 | p->real_parent = current; |
| 1081 | p->parent = p->real_parent; |
| 1082 | |
| 1083 | if (clone_flags & CLONE_THREAD) { |
| 1084 | spin_lock(¤t->sighand->siglock); |
| 1085 | /* |
| 1086 | * Important: if an exit-all has been started then |
| 1087 | * do not create this new thread - the whole thread |
| 1088 | * group is supposed to exit anyway. |
| 1089 | */ |
| 1090 | if (current->signal->flags & SIGNAL_GROUP_EXIT) { |
| 1091 | spin_unlock(¤t->sighand->siglock); |
| 1092 | write_unlock_irq(&tasklist_lock); |
| 1093 | retval = -EAGAIN; |
| 1094 | goto bad_fork_cleanup_namespace; |
| 1095 | } |
| 1096 | p->group_leader = current->group_leader; |
| 1097 | |
| 1098 | if (current->signal->group_stop_count > 0) { |
| 1099 | /* |
| 1100 | * There is an all-stop in progress for the group. |
| 1101 | * We ourselves will stop as soon as we check signals. |
| 1102 | * Make the new thread part of that group stop too. |
| 1103 | */ |
| 1104 | current->signal->group_stop_count++; |
| 1105 | set_tsk_thread_flag(p, TIF_SIGPENDING); |
| 1106 | } |
| 1107 | |
| 1108 | if (!cputime_eq(current->signal->it_virt_expires, |
| 1109 | cputime_zero) || |
| 1110 | !cputime_eq(current->signal->it_prof_expires, |
| 1111 | cputime_zero) || |
| 1112 | current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY || |
| 1113 | !list_empty(¤t->signal->cpu_timers[0]) || |
| 1114 | !list_empty(¤t->signal->cpu_timers[1]) || |
| 1115 | !list_empty(¤t->signal->cpu_timers[2])) { |
| 1116 | /* |
| 1117 | * Have child wake up on its first tick to check |
| 1118 | * for process CPU timers. |
| 1119 | */ |
| 1120 | p->it_prof_expires = jiffies_to_cputime(1); |
| 1121 | } |
| 1122 | |
| 1123 | spin_unlock(¤t->sighand->siglock); |
| 1124 | } |
| 1125 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1126 | /* |
| 1127 | * inherit ioprio |
| 1128 | */ |
| 1129 | p->ioprio = current->ioprio; |
| 1130 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1131 | SET_LINKS(p); |
| 1132 | if (unlikely(p->ptrace & PT_PTRACED)) |
| 1133 | __ptrace_link(p, current->parent); |
| 1134 | |
| 1135 | cpuset_fork(p); |
| 1136 | |
| 1137 | attach_pid(p, PIDTYPE_PID, p->pid); |
| 1138 | attach_pid(p, PIDTYPE_TGID, p->tgid); |
| 1139 | if (thread_group_leader(p)) { |
| 1140 | attach_pid(p, PIDTYPE_PGID, process_group(p)); |
| 1141 | attach_pid(p, PIDTYPE_SID, p->signal->session); |
| 1142 | if (p->pid) |
| 1143 | __get_cpu_var(process_counts)++; |
| 1144 | } |
| 1145 | |
Jason Baron | b0d62e6 | 2005-09-09 13:02:01 -0700 | [diff] [blame] | 1146 | if (!current->signal->tty && p->signal->tty) |
| 1147 | p->signal->tty = NULL; |
| 1148 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1149 | nr_threads++; |
| 1150 | total_forks++; |
| 1151 | write_unlock_irq(&tasklist_lock); |
| 1152 | retval = 0; |
| 1153 | |
| 1154 | fork_out: |
| 1155 | if (retval) |
| 1156 | return ERR_PTR(retval); |
| 1157 | return p; |
| 1158 | |
| 1159 | bad_fork_cleanup_namespace: |
| 1160 | exit_namespace(p); |
| 1161 | bad_fork_cleanup_keys: |
| 1162 | exit_keys(p); |
| 1163 | bad_fork_cleanup_mm: |
| 1164 | if (p->mm) |
| 1165 | mmput(p->mm); |
| 1166 | bad_fork_cleanup_signal: |
| 1167 | exit_signal(p); |
| 1168 | bad_fork_cleanup_sighand: |
| 1169 | exit_sighand(p); |
| 1170 | bad_fork_cleanup_fs: |
| 1171 | exit_fs(p); /* blocking */ |
| 1172 | bad_fork_cleanup_files: |
| 1173 | exit_files(p); /* blocking */ |
| 1174 | bad_fork_cleanup_semundo: |
| 1175 | exit_sem(p); |
| 1176 | bad_fork_cleanup_audit: |
| 1177 | audit_free(p); |
| 1178 | bad_fork_cleanup_security: |
| 1179 | security_task_free(p); |
| 1180 | bad_fork_cleanup_policy: |
| 1181 | #ifdef CONFIG_NUMA |
| 1182 | mpol_free(p->mempolicy); |
| 1183 | #endif |
| 1184 | bad_fork_cleanup: |
| 1185 | if (p->binfmt) |
| 1186 | module_put(p->binfmt->module); |
| 1187 | bad_fork_cleanup_put_domain: |
| 1188 | module_put(p->thread_info->exec_domain->module); |
| 1189 | bad_fork_cleanup_count: |
| 1190 | put_group_info(p->group_info); |
| 1191 | atomic_dec(&p->user->processes); |
| 1192 | free_uid(p->user); |
| 1193 | bad_fork_free: |
| 1194 | free_task(p); |
| 1195 | goto fork_out; |
| 1196 | } |
| 1197 | |
| 1198 | struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs) |
| 1199 | { |
| 1200 | memset(regs, 0, sizeof(struct pt_regs)); |
| 1201 | return regs; |
| 1202 | } |
| 1203 | |
| 1204 | task_t * __devinit fork_idle(int cpu) |
| 1205 | { |
| 1206 | task_t *task; |
| 1207 | struct pt_regs regs; |
| 1208 | |
| 1209 | task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0); |
| 1210 | if (!task) |
| 1211 | return ERR_PTR(-ENOMEM); |
| 1212 | init_idle(task, cpu); |
| 1213 | unhash_process(task); |
| 1214 | return task; |
| 1215 | } |
| 1216 | |
| 1217 | static inline int fork_traceflag (unsigned clone_flags) |
| 1218 | { |
| 1219 | if (clone_flags & CLONE_UNTRACED) |
| 1220 | return 0; |
| 1221 | else if (clone_flags & CLONE_VFORK) { |
| 1222 | if (current->ptrace & PT_TRACE_VFORK) |
| 1223 | return PTRACE_EVENT_VFORK; |
| 1224 | } else if ((clone_flags & CSIGNAL) != SIGCHLD) { |
| 1225 | if (current->ptrace & PT_TRACE_CLONE) |
| 1226 | return PTRACE_EVENT_CLONE; |
| 1227 | } else if (current->ptrace & PT_TRACE_FORK) |
| 1228 | return PTRACE_EVENT_FORK; |
| 1229 | |
| 1230 | return 0; |
| 1231 | } |
| 1232 | |
| 1233 | /* |
| 1234 | * Ok, this is the main fork-routine. |
| 1235 | * |
| 1236 | * It copies the process, and if successful kick-starts |
| 1237 | * it and waits for it to finish using the VM if required. |
| 1238 | */ |
| 1239 | long do_fork(unsigned long clone_flags, |
| 1240 | unsigned long stack_start, |
| 1241 | struct pt_regs *regs, |
| 1242 | unsigned long stack_size, |
| 1243 | int __user *parent_tidptr, |
| 1244 | int __user *child_tidptr) |
| 1245 | { |
| 1246 | struct task_struct *p; |
| 1247 | int trace = 0; |
| 1248 | long pid = alloc_pidmap(); |
| 1249 | |
| 1250 | if (pid < 0) |
| 1251 | return -EAGAIN; |
| 1252 | if (unlikely(current->ptrace)) { |
| 1253 | trace = fork_traceflag (clone_flags); |
| 1254 | if (trace) |
| 1255 | clone_flags |= CLONE_PTRACE; |
| 1256 | } |
| 1257 | |
| 1258 | p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid); |
| 1259 | /* |
| 1260 | * Do this prior waking up the new thread - the thread pointer |
| 1261 | * might get invalid after that point, if the thread exits quickly. |
| 1262 | */ |
| 1263 | if (!IS_ERR(p)) { |
| 1264 | struct completion vfork; |
| 1265 | |
| 1266 | if (clone_flags & CLONE_VFORK) { |
| 1267 | p->vfork_done = &vfork; |
| 1268 | init_completion(&vfork); |
| 1269 | } |
| 1270 | |
| 1271 | if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) { |
| 1272 | /* |
| 1273 | * We'll start up with an immediate SIGSTOP. |
| 1274 | */ |
| 1275 | sigaddset(&p->pending.signal, SIGSTOP); |
| 1276 | set_tsk_thread_flag(p, TIF_SIGPENDING); |
| 1277 | } |
| 1278 | |
| 1279 | if (!(clone_flags & CLONE_STOPPED)) |
| 1280 | wake_up_new_task(p, clone_flags); |
| 1281 | else |
| 1282 | p->state = TASK_STOPPED; |
| 1283 | |
| 1284 | if (unlikely (trace)) { |
| 1285 | current->ptrace_message = pid; |
| 1286 | ptrace_notify ((trace << 8) | SIGTRAP); |
| 1287 | } |
| 1288 | |
| 1289 | if (clone_flags & CLONE_VFORK) { |
| 1290 | wait_for_completion(&vfork); |
| 1291 | if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) |
| 1292 | ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP); |
| 1293 | } |
| 1294 | } else { |
| 1295 | free_pidmap(pid); |
| 1296 | pid = PTR_ERR(p); |
| 1297 | } |
| 1298 | return pid; |
| 1299 | } |
| 1300 | |
| 1301 | void __init proc_caches_init(void) |
| 1302 | { |
| 1303 | sighand_cachep = kmem_cache_create("sighand_cache", |
| 1304 | sizeof(struct sighand_struct), 0, |
| 1305 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); |
| 1306 | signal_cachep = kmem_cache_create("signal_cache", |
| 1307 | sizeof(struct signal_struct), 0, |
| 1308 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); |
| 1309 | files_cachep = kmem_cache_create("files_cache", |
| 1310 | sizeof(struct files_struct), 0, |
| 1311 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); |
| 1312 | fs_cachep = kmem_cache_create("fs_cache", |
| 1313 | sizeof(struct fs_struct), 0, |
| 1314 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); |
| 1315 | vm_area_cachep = kmem_cache_create("vm_area_struct", |
| 1316 | sizeof(struct vm_area_struct), 0, |
| 1317 | SLAB_PANIC, NULL, NULL); |
| 1318 | mm_cachep = kmem_cache_create("mm_struct", |
| 1319 | sizeof(struct mm_struct), 0, |
| 1320 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); |
| 1321 | } |