2 * Generic pidhash and scalable, time-bounded PID allocator
4 * (C) 2002-2003 William Irwin, IBM
5 * (C) 2004 William Irwin, Oracle
6 * (C) 2002-2004 Ingo Molnar, Red Hat
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.
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.
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).
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>
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 struct pid init_struct_pid = INIT_STRUCT_PID;
37 int pid_max = PID_MAX_DEFAULT;
39 #define RESERVED_PIDS 300
41 int pid_max_min = RESERVED_PIDS + 1;
42 int pid_max_max = PID_MAX_LIMIT;
44 #define BITS_PER_PAGE (PAGE_SIZE*8)
45 #define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1)
47 static inline int mk_pid(struct pid_namespace *pid_ns,
48 struct pidmap *map, int off)
50 return (map - pid_ns->pidmap)*BITS_PER_PAGE + off;
53 #define find_next_offset(map, off) \
54 find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
57 * PID-map pages start out as NULL, they get allocated upon
58 * first use and are never deallocated. This way a low pid_max
59 * value does not cause lots of bitmaps to be allocated, but
60 * the scheme scales to up to 4 million PIDs, runtime.
62 struct pid_namespace init_pid_ns = {
64 .refcount = ATOMIC_INIT(2),
67 [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL }
70 .child_reaper = &init_task
74 * Note: disable interrupts while the pidmap_lock is held as an
75 * interrupt might come in and do read_lock(&tasklist_lock).
77 * If we don't disable interrupts there is a nasty deadlock between
78 * detach_pid()->free_pid() and another cpu that does
79 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
80 * read_lock(&tasklist_lock);
82 * After we clean up the tasklist_lock and know there are no
83 * irq handlers that take it we can leave the interrupts enabled.
84 * For now it is easier to be safe than to prove it can't happen.
87 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
89 static fastcall void free_pidmap(struct pid_namespace *pid_ns, int pid)
91 struct pidmap *map = pid_ns->pidmap + pid / BITS_PER_PAGE;
92 int offset = pid & BITS_PER_PAGE_MASK;
94 clear_bit(offset, map->page);
95 atomic_inc(&map->nr_free);
98 static int alloc_pidmap(struct pid_namespace *pid_ns)
100 int i, offset, max_scan, pid, last = pid_ns->last_pid;
106 offset = pid & BITS_PER_PAGE_MASK;
107 map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
108 max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset;
109 for (i = 0; i <= max_scan; ++i) {
110 if (unlikely(!map->page)) {
111 void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
113 * Free the page if someone raced with us
116 spin_lock_irq(&pidmap_lock);
121 spin_unlock_irq(&pidmap_lock);
122 if (unlikely(!map->page))
125 if (likely(atomic_read(&map->nr_free))) {
127 if (!test_and_set_bit(offset, map->page)) {
128 atomic_dec(&map->nr_free);
129 pid_ns->last_pid = pid;
132 offset = find_next_offset(map, offset);
133 pid = mk_pid(pid_ns, map, offset);
135 * find_next_offset() found a bit, the pid from it
136 * is in-bounds, and if we fell back to the last
137 * bitmap block and the final block was the same
138 * as the starting point, pid is before last_pid.
140 } while (offset < BITS_PER_PAGE && pid < pid_max &&
141 (i != max_scan || pid < last ||
142 !((last+1) & BITS_PER_PAGE_MASK)));
144 if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) {
148 map = &pid_ns->pidmap[0];
149 offset = RESERVED_PIDS;
150 if (unlikely(last == offset))
153 pid = mk_pid(pid_ns, map, offset);
158 static int next_pidmap(struct pid_namespace *pid_ns, int last)
161 struct pidmap *map, *end;
163 offset = (last + 1) & BITS_PER_PAGE_MASK;
164 map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
165 end = &pid_ns->pidmap[PIDMAP_ENTRIES];
166 for (; map < end; map++, offset = 0) {
167 if (unlikely(!map->page))
169 offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
170 if (offset < BITS_PER_PAGE)
171 return mk_pid(pid_ns, map, offset);
176 fastcall void put_pid(struct pid *pid)
178 struct pid_namespace *ns;
183 /* FIXME - this must be the namespace this pid lives in */
185 if ((atomic_read(&pid->count) == 1) ||
186 atomic_dec_and_test(&pid->count))
187 kmem_cache_free(ns->pid_cachep, pid);
189 EXPORT_SYMBOL_GPL(put_pid);
191 static void delayed_put_pid(struct rcu_head *rhp)
193 struct pid *pid = container_of(rhp, struct pid, rcu);
197 fastcall void free_pid(struct pid *pid)
199 /* We can be called with write_lock_irq(&tasklist_lock) held */
202 spin_lock_irqsave(&pidmap_lock, flags);
203 hlist_del_rcu(&pid->pid_chain);
204 spin_unlock_irqrestore(&pidmap_lock, flags);
206 free_pidmap(&init_pid_ns, pid->nr);
207 call_rcu(&pid->rcu, delayed_put_pid);
210 struct pid *alloc_pid(void)
215 struct pid_namespace *ns;
217 ns = current->nsproxy->pid_ns;
218 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
222 nr = alloc_pidmap(ns);
226 atomic_set(&pid->count, 1);
228 for (type = 0; type < PIDTYPE_MAX; ++type)
229 INIT_HLIST_HEAD(&pid->tasks[type]);
231 spin_lock_irq(&pidmap_lock);
232 hlist_add_head_rcu(&pid->pid_chain, &pid_hash[pid_hashfn(pid->nr)]);
233 spin_unlock_irq(&pidmap_lock);
239 kmem_cache_free(ns->pid_cachep, pid);
244 struct pid * fastcall find_pid(int nr)
246 struct hlist_node *elem;
249 hlist_for_each_entry_rcu(pid, elem,
250 &pid_hash[pid_hashfn(nr)], pid_chain) {
256 EXPORT_SYMBOL_GPL(find_pid);
259 * attach_pid() must be called with the tasklist_lock write-held.
261 int fastcall attach_pid(struct task_struct *task, enum pid_type type,
264 struct pid_link *link;
266 link = &task->pids[type];
268 hlist_add_head_rcu(&link->node, &pid->tasks[type]);
273 void fastcall detach_pid(struct task_struct *task, enum pid_type type)
275 struct pid_link *link;
279 link = &task->pids[type];
282 hlist_del_rcu(&link->node);
285 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
286 if (!hlist_empty(&pid->tasks[tmp]))
292 /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
293 void fastcall transfer_pid(struct task_struct *old, struct task_struct *new,
296 new->pids[type].pid = old->pids[type].pid;
297 hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
298 old->pids[type].pid = NULL;
301 struct task_struct * fastcall pid_task(struct pid *pid, enum pid_type type)
303 struct task_struct *result = NULL;
305 struct hlist_node *first;
306 first = rcu_dereference(pid->tasks[type].first);
308 result = hlist_entry(first, struct task_struct, pids[(type)].node);
314 * Must be called under rcu_read_lock() or with tasklist_lock read-held.
316 struct task_struct *find_task_by_pid_type(int type, int nr)
318 return pid_task(find_pid(nr), type);
321 EXPORT_SYMBOL(find_task_by_pid_type);
323 struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
327 pid = get_pid(task->pids[type].pid);
332 struct task_struct *fastcall get_pid_task(struct pid *pid, enum pid_type type)
334 struct task_struct *result;
336 result = pid_task(pid, type);
338 get_task_struct(result);
343 struct pid *find_get_pid(pid_t nr)
348 pid = get_pid(find_pid(nr));
355 * Used by proc to find the first pid that is greater then or equal to nr.
357 * If there is a pid at nr this function is exactly the same as find_pid.
359 struct pid *find_ge_pid(int nr)
367 nr = next_pidmap(current->nsproxy->pid_ns, nr);
372 EXPORT_SYMBOL_GPL(find_get_pid);
377 struct kmem_cache *cachep;
378 struct list_head list;
381 static LIST_HEAD(pid_caches_lh);
382 static DEFINE_MUTEX(pid_caches_mutex);
385 * creates the kmem cache to allocate pids from.
386 * @nr_ids: the number of numerical ids this pid will have to carry
389 static struct kmem_cache *create_pid_cachep(int nr_ids)
391 struct pid_cache *pcache;
392 struct kmem_cache *cachep;
394 mutex_lock(&pid_caches_mutex);
395 list_for_each_entry (pcache, &pid_caches_lh, list)
396 if (pcache->nr_ids == nr_ids)
399 pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
403 snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
404 cachep = kmem_cache_create(pcache->name,
405 /* FIXME add numerical ids here */
406 sizeof(struct pid), 0, SLAB_HWCACHE_ALIGN, NULL);
410 pcache->nr_ids = nr_ids;
411 pcache->cachep = cachep;
412 list_add(&pcache->list, &pid_caches_lh);
414 mutex_unlock(&pid_caches_mutex);
415 return pcache->cachep;
420 mutex_unlock(&pid_caches_mutex);
424 struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns)
431 void free_pid_ns(struct kref *kref)
433 struct pid_namespace *ns;
435 ns = container_of(kref, struct pid_namespace, kref);
440 * The pid hash table is scaled according to the amount of memory in the
441 * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or
444 void __init pidhash_init(void)
447 unsigned long megabytes = nr_kernel_pages >> (20 - PAGE_SHIFT);
449 pidhash_shift = max(4, fls(megabytes * 4));
450 pidhash_shift = min(12, pidhash_shift);
451 pidhash_size = 1 << pidhash_shift;
453 printk("PID hash table entries: %d (order: %d, %Zd bytes)\n",
454 pidhash_size, pidhash_shift,
455 pidhash_size * sizeof(struct hlist_head));
457 pid_hash = alloc_bootmem(pidhash_size * sizeof(*(pid_hash)));
459 panic("Could not alloc pidhash!\n");
460 for (i = 0; i < pidhash_size; i++)
461 INIT_HLIST_HEAD(&pid_hash[i]);
464 void __init pidmap_init(void)
466 init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
467 /* Reserve PID 0. We never call free_pidmap(0) */
468 set_bit(0, init_pid_ns.pidmap[0].page);
469 atomic_dec(&init_pid_ns.pidmap[0].nr_free);
471 init_pid_ns.pid_cachep = create_pid_cachep(1);
472 if (init_pid_ns.pid_cachep == NULL)
473 panic("Can't create pid_1 cachep\n");