kernel/pid.c

   1 /*
   2  * Generic pidhash and scalable, time-bounded PID allocator
   3  *
   4  * (C) 2002-2003 William Irwin, IBM
   5  * (C) 2004 William Irwin, Oracle
   6  * (C) 2002-2004 Ingo Molnar, Red Hat
   7  *
   8  * pid-structures are backing objects for tasks sharing a given ID to chain
   9  * against. There is very little to them aside from hashing them and
  10  * parking tasks using given ID's on a list.
  11  *
  12  * The hash is always changed with the tasklist_lock write-acquired,
  13  * and the hash is only accessed with the tasklist_lock at least
  14  * read-acquired, so there's no additional SMP locking needed here.
  15  *
  16  * We have a list of bitmap pages, which bitmaps represent the PID space.
  17  * Allocating and freeing PIDs is completely lockless. The worst-case
  18  * allocation scenario when all but one out of 1 million PIDs possible are
  19  * allocated already: the scanning of 32 list entries and at most PAGE_SIZE
  20  * bytes. The typical fastpath is a single successful setbit. Freeing is O(1).
  21  *
  22  * Pid namespaces:
  23  *    (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc.
  24  *    (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM
  25  *     Many thanks to Oleg Nesterov for comments and help
  26  *
  27  */
  28
  29 #include <linux/mm.h>
  30 #include <linux/module.h>
  31 #include <linux/slab.h>
  32 #include <linux/init.h>
  33 #include <linux/rculist.h>
  34 #include <linux/bootmem.h>
  35 #include <linux/hash.h>
  36 #include <linux/pid_namespace.h>
  37 #include <linux/init_task.h>
  38 #include <linux/syscalls.h>
  39
  40 #define pid_hashfn(nr, ns)      \
  41         hash_long((unsigned long)nr + (unsigned long)ns, pidhash_shift)
  42 static struct hlist_head *pid_hash;
  43 static unsigned int pidhash_shift = 4;
  44 struct pid init_struct_pid = INIT_STRUCT_PID;
  45
  46 int pid_max = PID_MAX_DEFAULT;
  47
  48 #define RESERVED_PIDS           300
  49
  50 int pid_max_min = RESERVED_PIDS + 1;
  51 int pid_max_max = PID_MAX_LIMIT;
  52
  53 #define BITS_PER_PAGE           (PAGE_SIZE*8)
  54 #define BITS_PER_PAGE_MASK      (BITS_PER_PAGE-1)
  55
  56 static inline int mk_pid(struct pid_namespace *pid_ns,
  57                 struct pidmap *map, int off)
  58 {
  59         return (map - pid_ns->pidmap)*BITS_PER_PAGE + off;
  60 }
  61
  62 #define find_next_offset(map, off)                                      \
  63                 find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
  64
  65 /*
  66  * PID-map pages start out as NULL, they get allocated upon
  67  * first use and are never deallocated. This way a low pid_max
  68  * value does not cause lots of bitmaps to be allocated, but
  69  * the scheme scales to up to 4 million PIDs, runtime.
  70  */
  71 struct pid_namespace init_pid_ns = {
  72         .kref = {
  73                 .refcount       = ATOMIC_INIT(2),
  74         },
  75         .pidmap = {
  76                 [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL }
  77         },
  78         .last_pid = 0,
  79         .level = 0,
  80         .child_reaper = &init_task,
  81 };
  82 EXPORT_SYMBOL_GPL(init_pid_ns);
  83
  84 int is_container_init(struct task_struct *tsk)
  85 {
  86         int ret = 0;
  87         struct pid *pid;
  88
  89         rcu_read_lock();
  90         pid = task_pid(tsk);
  91         if (pid != NULL && pid->numbers[pid->level].nr == 1)
  92                 ret = 1;
  93         rcu_read_unlock();
  94
  95         return ret;
  96 }
  97 EXPORT_SYMBOL(is_container_init);
  98
  99 /*
 100  * Note: disable interrupts while the pidmap_lock is held as an
 101  * interrupt might come in and do read_lock(&tasklist_lock).
 102  *
 103  * If we don't disable interrupts there is a nasty deadlock between
 104  * detach_pid()->free_pid() and another cpu that does
 105  * spin_lock(&pidmap_lock) followed by an interrupt routine that does
 106  * read_lock(&tasklist_lock);
 107  *
 108  * After we clean up the tasklist_lock and know there are no
 109  * irq handlers that take it we can leave the interrupts enabled.
 110  * For now it is easier to be safe than to prove it can't happen.
 111  */
 112
 113 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
 114
 115 static void free_pidmap(struct upid *upid)
 116 {
 117         int nr = upid->nr;
 118         struct pidmap *map = upid->ns->pidmap + nr / BITS_PER_PAGE;
 119         int offset = nr & BITS_PER_PAGE_MASK;
 120
 121         clear_bit(offset, map->page);
 122         atomic_inc(&map->nr_free);
 123 }
 124
 125 static int alloc_pidmap(struct pid_namespace *pid_ns)
 126 {
 127         int i, offset, max_scan, pid, last = pid_ns->last_pid;
 128         struct pidmap *map;
 129
 130         pid = last + 1;
 131         if (pid >= pid_max)
 132                 pid = RESERVED_PIDS;
 133         offset = pid & BITS_PER_PAGE_MASK;
 134         map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
 135         max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset;
 136         for (i = 0; i <= max_scan; ++i) {
 137                 if (unlikely(!map->page)) {
 138                         void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 139                         /*
 140                          * Free the page if someone raced with us
 141                          * installing it:
 142                          */
 143                         spin_lock_irq(&pidmap_lock);
 144                         if (!map->page) {
 145                                 map->page = page;
 146                                 page = NULL;
 147                         }
 148                         spin_unlock_irq(&pidmap_lock);
 149                         kfree(page);
 150                         if (unlikely(!map->page))
 151                                 break;
 152                 }
 153                 if (likely(atomic_read(&map->nr_free))) {
 154                         do {
 155                                 if (!test_and_set_bit(offset, map->page)) {
 156                                         atomic_dec(&map->nr_free);
 157                                         pid_ns->last_pid = pid;
 158                                         return pid;
 159                                 }
 160                                 offset = find_next_offset(map, offset);
 161                                 pid = mk_pid(pid_ns, map, offset);
 162                         /*
 163                          * find_next_offset() found a bit, the pid from it
 164                          * is in-bounds, and if we fell back to the last
 165                          * bitmap block and the final block was the same
 166                          * as the starting point, pid is before last_pid.
 167                          */
 168                         } while (offset < BITS_PER_PAGE && pid < pid_max &&
 169                                         (i != max_scan || pid < last ||
 170                                             !((last+1) & BITS_PER_PAGE_MASK)));
 171                 }
 172                 if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) {
 173                         ++map;
 174                         offset = 0;
 175                 } else {
 176                         map = &pid_ns->pidmap[0];
 177                         offset = RESERVED_PIDS;
 178                         if (unlikely(last == offset))
 179                                 break;
 180                 }
 181                 pid = mk_pid(pid_ns, map, offset);
 182         }
 183         return -1;
 184 }
 185
 186 int next_pidmap(struct pid_namespace *pid_ns, int last)
 187 {
 188         int offset;
 189         struct pidmap *map, *end;
 190
 191         offset = (last + 1) & BITS_PER_PAGE_MASK;
 192         map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
 193         end = &pid_ns->pidmap[PIDMAP_ENTRIES];
 194         for (; map < end; map++, offset = 0) {
 195                 if (unlikely(!map->page))
 196                         continue;
 197                 offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
 198                 if (offset < BITS_PER_PAGE)
 199                         return mk_pid(pid_ns, map, offset);
 200         }
 201         return -1;
 202 }
 203
 204 void put_pid(struct pid *pid)
 205 {
 206         struct pid_namespace *ns;
 207
 208         if (!pid)
 209                 return;
 210
 211         ns = pid->numbers[pid->level].ns;
 212         if ((atomic_read(&pid->count) == 1) ||
 213              atomic_dec_and_test(&pid->count)) {
 214                 kmem_cache_free(ns->pid_cachep, pid);
 215                 put_pid_ns(ns);
 216         }
 217 }
 218 EXPORT_SYMBOL_GPL(put_pid);
 219
 220 static void delayed_put_pid(struct rcu_head *rhp)
 221 {
 222         struct pid *pid = container_of(rhp, struct pid, rcu);
 223         put_pid(pid);
 224 }
 225
 226 void free_pid(struct pid *pid)
 227 {
 228         /* We can be called with write_lock_irq(&tasklist_lock) held */
 229         int i;
 230         unsigned long flags;
 231
 232         spin_lock_irqsave(&pidmap_lock, flags);
 233         for (i = 0; i <= pid->level; i++)
 234                 hlist_del_rcu(&pid->numbers[i].pid_chain);
 235         spin_unlock_irqrestore(&pidmap_lock, flags);
 236
 237         for (i = 0; i <= pid->level; i++)
 238                 free_pidmap(pid->numbers + i);
 239
 240         call_rcu(&pid->rcu, delayed_put_pid);
 241 }
 242
 243 struct pid *alloc_pid(struct pid_namespace *ns)
 244 {
 245         struct pid *pid;
 246         enum pid_type type;
 247         int i, nr;
 248         struct pid_namespace *tmp;
 249         struct upid *upid;
 250
 251         pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
 252         if (!pid)
 253                 goto out;
 254
 255         tmp = ns;
 256         for (i = ns->level; i >= 0; i--) {
 257                 nr = alloc_pidmap(tmp);
 258                 if (nr < 0)
 259                         goto out_free;
 260
 261                 pid->numbers[i].nr = nr;
 262                 pid->numbers[i].ns = tmp;
 263                 tmp = tmp->parent;
 264         }
 265
 266         get_pid_ns(ns);
 267         pid->level = ns->level;
 268         atomic_set(&pid->count, 1);
 269         for (type = 0; type < PIDTYPE_MAX; ++type)
 270                 INIT_HLIST_HEAD(&pid->tasks[type]);
 271
 272         upid = pid->numbers + ns->level;
 273         spin_lock_irq(&pidmap_lock);
 274         for ( ; upid >= pid->numbers; --upid)
 275                 hlist_add_head_rcu(&upid->pid_chain,
 276                                 &pid_hash[pid_hashfn(upid->nr, upid->ns)]);
 277         spin_unlock_irq(&pidmap_lock);
 278
 279 out:
 280         return pid;
 281
 282 out_free:
 283         while (++i <= ns->level)
 284                 free_pidmap(pid->numbers + i);
 285
 286         kmem_cache_free(ns->pid_cachep, pid);
 287         pid = NULL;
 288         goto out;
 289 }
 290
 291 struct pid *find_pid_ns(int nr, struct pid_namespace *ns)
 292 {
 293         struct hlist_node *elem;
 294         struct upid *pnr;
 295
 296         hlist_for_each_entry_rcu(pnr, elem,
 297                         &pid_hash[pid_hashfn(nr, ns)], pid_chain)
 298                 if (pnr->nr == nr && pnr->ns == ns)
 299                         return container_of(pnr, struct pid,
 300                                         numbers[ns->level]);
 301
 302         return NULL;
 303 }
 304 EXPORT_SYMBOL_GPL(find_pid_ns);
 305
 306 struct pid *find_vpid(int nr)
 307 {
 308         return find_pid_ns(nr, current->nsproxy->pid_ns);
 309 }
 310 EXPORT_SYMBOL_GPL(find_vpid);
 311
 312 /*
 313  * attach_pid() must be called with the tasklist_lock write-held.
 314  */
 315 void attach_pid(struct task_struct *task, enum pid_type type,
 316                 struct pid *pid)
 317 {
 318         struct pid_link *link;
 319
 320         link = &task->pids[type];
 321         link->pid = pid;
 322         hlist_add_head_rcu(&link->node, &pid->tasks[type]);
 323 }
 324
 325 static void __change_pid(struct task_struct *task, enum pid_type type,
 326                         struct pid *new)
 327 {
 328         struct pid_link *link;
 329         struct pid *pid;
 330         int tmp;
 331
 332         link = &task->pids[type];
 333         pid = link->pid;
 334
 335         hlist_del_rcu(&link->node);
 336         link->pid = new;
 337
 338         for (tmp = PIDTYPE_MAX; --tmp >= 0; )
 339                 if (!hlist_empty(&pid->tasks[tmp]))
 340                         return;
 341
 342         free_pid(pid);
 343 }
 344
 345 void detach_pid(struct task_struct *task, enum pid_type type)
 346 {
 347         __change_pid(task, type, NULL);
 348 }
 349
 350 void change_pid(struct task_struct *task, enum pid_type type,
 351                 struct pid *pid)
 352 {
 353         __change_pid(task, type, pid);
 354         attach_pid(task, type, pid);
 355 }
 356
 357 /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
 358 void transfer_pid(struct task_struct *old, struct task_struct *new,
 359                            enum pid_type type)
 360 {
 361         new->pids[type].pid = old->pids[type].pid;
 362         hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
 363 }
 364
 365 struct task_struct *pid_task(struct pid *pid, enum pid_type type)
 366 {
 367         struct task_struct *result = NULL;
 368         if (pid) {
 369                 struct hlist_node *first;
 370                 first = rcu_dereference(pid->tasks[type].first);
 371                 if (first)
 372                         result = hlist_entry(first, struct task_struct, pids[(type)].node);
 373         }
 374         return result;
 375 }
 376 EXPORT_SYMBOL(pid_task);
 377
 378 /*
 379  * Must be called under rcu_read_lock() or with tasklist_lock read-held.
 380  */
 381 struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
 382 {
 383         return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
 384 }
 385
 386 struct task_struct *find_task_by_vpid(pid_t vnr)
 387 {
 388         return find_task_by_pid_ns(vnr, current->nsproxy->pid_ns);
 389 }
 390
 391 struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
 392 {
 393         struct pid *pid;
 394         rcu_read_lock();
 395         if (type != PIDTYPE_PID)
 396                 task = task->group_leader;
 397         pid = get_pid(task->pids[type].pid);
 398         rcu_read_unlock();
 399         return pid;
 400 }
 401
 402 struct task_struct *get_pid_task(struct pid *pid, enum pid_type type)
 403 {
 404         struct task_struct *result;
 405         rcu_read_lock();
 406         result = pid_task(pid, type);
 407         if (result)
 408                 get_task_struct(result);
 409         rcu_read_unlock();
 410         return result;
 411 }
 412
 413 struct pid *find_get_pid(pid_t nr)
 414 {
 415         struct pid *pid;
 416
 417         rcu_read_lock();
 418         pid = get_pid(find_vpid(nr));
 419         rcu_read_unlock();
 420
 421         return pid;
 422 }
 423 EXPORT_SYMBOL_GPL(find_get_pid);
 424
 425 pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
 426 {
 427         struct upid *upid;
 428         pid_t nr = 0;
 429
 430         if (pid && ns->level <= pid->level) {
 431                 upid = &pid->numbers[ns->level];
 432                 if (upid->ns == ns)
 433                         nr = upid->nr;
 434         }
 435         return nr;
 436 }
 437
 438 pid_t pid_vnr(struct pid *pid)
 439 {
 440         return pid_nr_ns(pid, current->nsproxy->pid_ns);
 441 }
 442 EXPORT_SYMBOL_GPL(pid_vnr);
 443
 444 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
 445                         struct pid_namespace *ns)
 446 {
 447         pid_t nr = 0;
 448
 449         rcu_read_lock();
 450         if (!ns)
 451                 ns = current->nsproxy->pid_ns;
 452         if (likely(pid_alive(task))) {
 453                 if (type != PIDTYPE_PID)
 454                         task = task->group_leader;
 455                 nr = pid_nr_ns(task->pids[type].pid, ns);
 456         }
 457         rcu_read_unlock();
 458
 459         return nr;
 460 }
 461 EXPORT_SYMBOL(__task_pid_nr_ns);
 462
 463 pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
 464 {
 465         return pid_nr_ns(task_tgid(tsk), ns);
 466 }
 467 EXPORT_SYMBOL(task_tgid_nr_ns);
 468
 469 struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
 470 {
 471         return ns_of_pid(task_pid(tsk));
 472 }
 473 EXPORT_SYMBOL_GPL(task_active_pid_ns);
 474
 475 /*
 476  * Used by proc to find the first pid that is greater than or equal to nr.
 477  *
 478  * If there is a pid at nr this function is exactly the same as find_pid_ns.
 479  */
 480 struct pid *find_ge_pid(int nr, struct pid_namespace *ns)
 481 {
 482         struct pid *pid;
 483
 484         do {
 485                 pid = find_pid_ns(nr, ns);
 486                 if (pid)
 487                         break;
 488                 nr = next_pidmap(ns, nr);
 489         } while (nr > 0);
 490
 491         return pid;
 492 }
 493
 494 /*
 495  * The pid hash table is scaled according to the amount of memory in the
 496  * machine.  From a minimum of 16 slots up to 4096 slots at one gigabyte or
 497  * more.
 498  */
 499 void __init pidhash_init(void)
 500 {
 501         int i, pidhash_size;
 502
 503         pid_hash = alloc_large_system_hash("PID", sizeof(*pid_hash), 0, 18,
 504                                            HASH_EARLY | HASH_SMALL,
 505                                            &pidhash_shift, NULL, 4096);
 506         pidhash_size = 1 << pidhash_shift;
 507
 508         for (i = 0; i < pidhash_size; i++)
 509                 INIT_HLIST_HEAD(&pid_hash[i]);
 510 }
 511
 512 void __init pidmap_init(void)
 513 {
 514         init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 515         /* Reserve PID 0. We never call free_pidmap(0) */
 516         set_bit(0, init_pid_ns.pidmap[0].page);
 517         atomic_dec(&init_pid_ns.pidmap[0].nr_free);
 518
 519         init_pid_ns.pid_cachep = KMEM_CACHE(pid,
 520                         SLAB_HWCACHE_ALIGN | SLAB_PANIC);
 521 }