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
  23 #include <linux/mm.h>
  24 #include <linux/module.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/bootmem.h>
  28 #include <linux/hash.h>
  29 #include <linux/pid_namespace.h>
  30 #include <linux/init_task.h>
  31
  32 #define pid_hashfn(nr) hash_long((unsigned long)nr, pidhash_shift)
  33 static struct hlist_head *pid_hash;
  34 static int pidhash_shift;
  35 static struct kmem_cache *pid_cachep;
  36 struct pid init_struct_pid = INIT_STRUCT_PID;
  37
  38 int pid_max = PID_MAX_DEFAULT;
  39
  40 #define RESERVED_PIDS           300
  41
  42 int pid_max_min = RESERVED_PIDS + 1;
  43 int pid_max_max = PID_MAX_LIMIT;
  44
  45 #define BITS_PER_PAGE           (PAGE_SIZE*8)
  46 #define BITS_PER_PAGE_MASK      (BITS_PER_PAGE-1)
  47
  48 static inline int mk_pid(struct pid_namespace *pid_ns,
  49                 struct pidmap *map, int off)
  50 {
  51         return (map - pid_ns->pidmap)*BITS_PER_PAGE + off;
  52 }
  53
  54 #define find_next_offset(map, off)                                      \
  55                 find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
  56
  57 /*
  58  * PID-map pages start out as NULL, they get allocated upon
  59  * first use and are never deallocated. This way a low pid_max
  60  * value does not cause lots of bitmaps to be allocated, but
  61  * the scheme scales to up to 4 million PIDs, runtime.
  62  */
  63 struct pid_namespace init_pid_ns = {
  64         .kref = {
  65                 .refcount       = ATOMIC_INIT(2),
  66         },
  67         .pidmap = {
  68                 [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL }
  69         },
  70         .last_pid = 0,
  71         .child_reaper = &init_task
  72 };
  73
  74 /*
  75  * Note: disable interrupts while the pidmap_lock is held as an
  76  * interrupt might come in and do read_lock(&tasklist_lock).
  77  *
  78  * If we don't disable interrupts there is a nasty deadlock between
  79  * detach_pid()->free_pid() and another cpu that does
  80  * spin_lock(&pidmap_lock) followed by an interrupt routine that does
  81  * read_lock(&tasklist_lock);
  82  *
  83  * After we clean up the tasklist_lock and know there are no
  84  * irq handlers that take it we can leave the interrupts enabled.
  85  * For now it is easier to be safe than to prove it can't happen.
  86  */
  87
  88 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
  89
  90 static fastcall void free_pidmap(struct pid_namespace *pid_ns, int pid)
  91 {
  92         struct pidmap *map = pid_ns->pidmap + pid / BITS_PER_PAGE;
  93         int offset = pid & BITS_PER_PAGE_MASK;
  94
  95         clear_bit(offset, map->page);
  96         atomic_inc(&map->nr_free);
  97 }
  98
  99 static int alloc_pidmap(struct pid_namespace *pid_ns)
 100 {
 101         int i, offset, max_scan, pid, last = pid_ns->last_pid;
 102         struct pidmap *map;
 103
 104         pid = last + 1;
 105         if (pid >= pid_max)
 106                 pid = RESERVED_PIDS;
 107         offset = pid & BITS_PER_PAGE_MASK;
 108         map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
 109         max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset;
 110         for (i = 0; i <= max_scan; ++i) {
 111                 if (unlikely(!map->page)) {
 112                         void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 113                         /*
 114                          * Free the page if someone raced with us
 115                          * installing it:
 116                          */
 117                         spin_lock_irq(&pidmap_lock);
 118                         if (map->page)
 119                                 kfree(page);
 120                         else
 121                                 map->page = page;
 122                         spin_unlock_irq(&pidmap_lock);
 123                         if (unlikely(!map->page))
 124                                 break;
 125                 }
 126                 if (likely(atomic_read(&map->nr_free))) {
 127                         do {
 128                                 if (!test_and_set_bit(offset, map->page)) {
 129                                         atomic_dec(&map->nr_free);
 130                                         pid_ns->last_pid = pid;
 131                                         return pid;
 132                                 }
 133                                 offset = find_next_offset(map, offset);
 134                                 pid = mk_pid(pid_ns, map, offset);
 135                         /*
 136                          * find_next_offset() found a bit, the pid from it
 137                          * is in-bounds, and if we fell back to the last
 138                          * bitmap block and the final block was the same
 139                          * as the starting point, pid is before last_pid.
 140                          */
 141                         } while (offset < BITS_PER_PAGE && pid < pid_max &&
 142                                         (i != max_scan || pid < last ||
 143                                             !((last+1) & BITS_PER_PAGE_MASK)));
 144                 }
 145                 if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) {
 146                         ++map;
 147                         offset = 0;
 148                 } else {
 149                         map = &pid_ns->pidmap[0];
 150                         offset = RESERVED_PIDS;
 151                         if (unlikely(last == offset))
 152                                 break;
 153                 }
 154                 pid = mk_pid(pid_ns, map, offset);
 155         }
 156         return -1;
 157 }
 158
 159 static int next_pidmap(struct pid_namespace *pid_ns, int last)
 160 {
 161         int offset;
 162         struct pidmap *map, *end;
 163
 164         offset = (last + 1) & BITS_PER_PAGE_MASK;
 165         map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
 166         end = &pid_ns->pidmap[PIDMAP_ENTRIES];
 167         for (; map < end; map++, offset = 0) {
 168                 if (unlikely(!map->page))
 169                         continue;
 170                 offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
 171                 if (offset < BITS_PER_PAGE)
 172                         return mk_pid(pid_ns, map, offset);
 173         }
 174         return -1;
 175 }
 176
 177 fastcall void put_pid(struct pid *pid)
 178 {
 179         if (!pid)
 180                 return;
 181         if ((atomic_read(&pid->count) == 1) ||
 182              atomic_dec_and_test(&pid->count))
 183                 kmem_cache_free(pid_cachep, pid);
 184 }
 185 EXPORT_SYMBOL_GPL(put_pid);
 186
 187 static void delayed_put_pid(struct rcu_head *rhp)
 188 {
 189         struct pid *pid = container_of(rhp, struct pid, rcu);
 190         put_pid(pid);
 191 }
 192
 193 fastcall void free_pid(struct pid *pid)
 194 {
 195         /* We can be called with write_lock_irq(&tasklist_lock) held */
 196         unsigned long flags;
 197
 198         spin_lock_irqsave(&pidmap_lock, flags);
 199         hlist_del_rcu(&pid->pid_chain);
 200         spin_unlock_irqrestore(&pidmap_lock, flags);
 201
 202         free_pidmap(&init_pid_ns, pid->nr);
 203         call_rcu(&pid->rcu, delayed_put_pid);
 204 }
 205
 206 struct pid *alloc_pid(void)
 207 {
 208         struct pid *pid;
 209         enum pid_type type;
 210         int nr = -1;
 211
 212         pid = kmem_cache_alloc(pid_cachep, GFP_KERNEL);
 213         if (!pid)
 214                 goto out;
 215
 216         nr = alloc_pidmap(current->nsproxy->pid_ns);
 217         if (nr < 0)
 218                 goto out_free;
 219
 220         atomic_set(&pid->count, 1);
 221         pid->nr = nr;
 222         for (type = 0; type < PIDTYPE_MAX; ++type)
 223                 INIT_HLIST_HEAD(&pid->tasks[type]);
 224
 225         spin_lock_irq(&pidmap_lock);
 226         hlist_add_head_rcu(&pid->pid_chain, &pid_hash[pid_hashfn(pid->nr)]);
 227         spin_unlock_irq(&pidmap_lock);
 228
 229 out:
 230         return pid;
 231
 232 out_free:
 233         kmem_cache_free(pid_cachep, pid);
 234         pid = NULL;
 235         goto out;
 236 }
 237
 238 struct pid * fastcall find_pid(int nr)
 239 {
 240         struct hlist_node *elem;
 241         struct pid *pid;
 242
 243         hlist_for_each_entry_rcu(pid, elem,
 244                         &pid_hash[pid_hashfn(nr)], pid_chain) {
 245                 if (pid->nr == nr)
 246                         return pid;
 247         }
 248         return NULL;
 249 }
 250 EXPORT_SYMBOL_GPL(find_pid);
 251
 252 /*
 253  * attach_pid() must be called with the tasklist_lock write-held.
 254  */
 255 int fastcall attach_pid(struct task_struct *task, enum pid_type type,
 256                 struct pid *pid)
 257 {
 258         struct pid_link *link;
 259
 260         link = &task->pids[type];
 261         link->pid = pid;
 262         hlist_add_head_rcu(&link->node, &pid->tasks[type]);
 263
 264         return 0;
 265 }
 266
 267 void fastcall detach_pid(struct task_struct *task, enum pid_type type)
 268 {
 269         struct pid_link *link;
 270         struct pid *pid;
 271         int tmp;
 272
 273         link = &task->pids[type];
 274         pid = link->pid;
 275
 276         hlist_del_rcu(&link->node);
 277         link->pid = NULL;
 278
 279         for (tmp = PIDTYPE_MAX; --tmp >= 0; )
 280                 if (!hlist_empty(&pid->tasks[tmp]))
 281                         return;
 282
 283         free_pid(pid);
 284 }
 285
 286 /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
 287 void fastcall transfer_pid(struct task_struct *old, struct task_struct *new,
 288                            enum pid_type type)
 289 {
 290         new->pids[type].pid = old->pids[type].pid;
 291         hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
 292         old->pids[type].pid = NULL;
 293 }
 294
 295 struct task_struct * fastcall pid_task(struct pid *pid, enum pid_type type)
 296 {
 297         struct task_struct *result = NULL;
 298         if (pid) {
 299                 struct hlist_node *first;
 300                 first = rcu_dereference(pid->tasks[type].first);
 301                 if (first)
 302                         result = hlist_entry(first, struct task_struct, pids[(type)].node);
 303         }
 304         return result;
 305 }
 306
 307 /*
 308  * Must be called under rcu_read_lock() or with tasklist_lock read-held.
 309  */
 310 struct task_struct *find_task_by_pid_type(int type, int nr)
 311 {
 312         return pid_task(find_pid(nr), type);
 313 }
 314
 315 EXPORT_SYMBOL(find_task_by_pid_type);
 316
 317 struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
 318 {
 319         struct pid *pid;
 320         rcu_read_lock();
 321         pid = get_pid(task->pids[type].pid);
 322         rcu_read_unlock();
 323         return pid;
 324 }
 325
 326 struct task_struct *fastcall get_pid_task(struct pid *pid, enum pid_type type)
 327 {
 328         struct task_struct *result;
 329         rcu_read_lock();
 330         result = pid_task(pid, type);
 331         if (result)
 332                 get_task_struct(result);
 333         rcu_read_unlock();
 334         return result;
 335 }
 336
 337 struct pid *find_get_pid(pid_t nr)
 338 {
 339         struct pid *pid;
 340
 341         rcu_read_lock();
 342         pid = get_pid(find_pid(nr));
 343         rcu_read_unlock();
 344
 345         return pid;
 346 }
 347
 348 /*
 349  * Used by proc to find the first pid that is greater then or equal to nr.
 350  *
 351  * If there is a pid at nr this function is exactly the same as find_pid.
 352  */
 353 struct pid *find_ge_pid(int nr)
 354 {
 355         struct pid *pid;
 356
 357         do {
 358                 pid = find_pid(nr);
 359                 if (pid)
 360                         break;
 361                 nr = next_pidmap(current->nsproxy->pid_ns, nr);
 362         } while (nr > 0);
 363
 364         return pid;
 365 }
 366 EXPORT_SYMBOL_GPL(find_get_pid);
 367
 368 struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns)
 369 {
 370         BUG_ON(!old_ns);
 371         get_pid_ns(old_ns);
 372         return old_ns;
 373 }
 374
 375 void free_pid_ns(struct kref *kref)
 376 {
 377         struct pid_namespace *ns;
 378
 379         ns = container_of(kref, struct pid_namespace, kref);
 380         kfree(ns);
 381 }
 382
 383 /*
 384  * The pid hash table is scaled according to the amount of memory in the
 385  * machine.  From a minimum of 16 slots up to 4096 slots at one gigabyte or
 386  * more.
 387  */
 388 void __init pidhash_init(void)
 389 {
 390         int i, pidhash_size;
 391         unsigned long megabytes = nr_kernel_pages >> (20 - PAGE_SHIFT);
 392
 393         pidhash_shift = max(4, fls(megabytes * 4));
 394         pidhash_shift = min(12, pidhash_shift);
 395         pidhash_size = 1 << pidhash_shift;
 396
 397         printk("PID hash table entries: %d (order: %d, %Zd bytes)\n",
 398                 pidhash_size, pidhash_shift,
 399                 pidhash_size * sizeof(struct hlist_head));
 400
 401         pid_hash = alloc_bootmem(pidhash_size * sizeof(*(pid_hash)));
 402         if (!pid_hash)
 403                 panic("Could not alloc pidhash!\n");
 404         for (i = 0; i < pidhash_size; i++)
 405                 INIT_HLIST_HEAD(&pid_hash[i]);
 406 }
 407
 408 void __init pidmap_init(void)
 409 {
 410         init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
 411         /* Reserve PID 0. We never call free_pidmap(0) */
 412         set_bit(0, init_pid_ns.pidmap[0].page);
 413         atomic_dec(&init_pid_ns.pidmap[0].nr_free);
 414
 415         pid_cachep = KMEM_CACHE(pid, SLAB_PANIC);
 416 }