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 }
71 .child_reaper = &init_task,
74 int is_global_init(struct task_struct *tsk)
76 return tsk == init_pid_ns.child_reaper;
80 * Note: disable interrupts while the pidmap_lock is held as an
81 * interrupt might come in and do read_lock(&tasklist_lock).
83 * If we don't disable interrupts there is a nasty deadlock between
84 * detach_pid()->free_pid() and another cpu that does
85 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
86 * read_lock(&tasklist_lock);
88 * After we clean up the tasklist_lock and know there are no
89 * irq handlers that take it we can leave the interrupts enabled.
90 * For now it is easier to be safe than to prove it can't happen.
93 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
95 static fastcall void free_pidmap(struct pid_namespace *pid_ns, int pid)
97 struct pidmap *map = pid_ns->pidmap + pid / BITS_PER_PAGE;
98 int offset = pid & BITS_PER_PAGE_MASK;
100 clear_bit(offset, map->page);
101 atomic_inc(&map->nr_free);
104 static int alloc_pidmap(struct pid_namespace *pid_ns)
106 int i, offset, max_scan, pid, last = pid_ns->last_pid;
112 offset = pid & BITS_PER_PAGE_MASK;
113 map = &pid_ns->pidmap[pid/BITS_PER_PAGE];
114 max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset;
115 for (i = 0; i <= max_scan; ++i) {
116 if (unlikely(!map->page)) {
117 void *page = kzalloc(PAGE_SIZE, GFP_KERNEL);
119 * Free the page if someone raced with us
122 spin_lock_irq(&pidmap_lock);
127 spin_unlock_irq(&pidmap_lock);
128 if (unlikely(!map->page))
131 if (likely(atomic_read(&map->nr_free))) {
133 if (!test_and_set_bit(offset, map->page)) {
134 atomic_dec(&map->nr_free);
135 pid_ns->last_pid = pid;
138 offset = find_next_offset(map, offset);
139 pid = mk_pid(pid_ns, map, offset);
141 * find_next_offset() found a bit, the pid from it
142 * is in-bounds, and if we fell back to the last
143 * bitmap block and the final block was the same
144 * as the starting point, pid is before last_pid.
146 } while (offset < BITS_PER_PAGE && pid < pid_max &&
147 (i != max_scan || pid < last ||
148 !((last+1) & BITS_PER_PAGE_MASK)));
150 if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) {
154 map = &pid_ns->pidmap[0];
155 offset = RESERVED_PIDS;
156 if (unlikely(last == offset))
159 pid = mk_pid(pid_ns, map, offset);
164 static int next_pidmap(struct pid_namespace *pid_ns, int last)
167 struct pidmap *map, *end;
169 offset = (last + 1) & BITS_PER_PAGE_MASK;
170 map = &pid_ns->pidmap[(last + 1)/BITS_PER_PAGE];
171 end = &pid_ns->pidmap[PIDMAP_ENTRIES];
172 for (; map < end; map++, offset = 0) {
173 if (unlikely(!map->page))
175 offset = find_next_bit((map)->page, BITS_PER_PAGE, offset);
176 if (offset < BITS_PER_PAGE)
177 return mk_pid(pid_ns, map, offset);
182 fastcall void put_pid(struct pid *pid)
184 struct pid_namespace *ns;
189 /* FIXME - this must be the namespace this pid lives in */
191 if ((atomic_read(&pid->count) == 1) ||
192 atomic_dec_and_test(&pid->count))
193 kmem_cache_free(ns->pid_cachep, pid);
195 EXPORT_SYMBOL_GPL(put_pid);
197 static void delayed_put_pid(struct rcu_head *rhp)
199 struct pid *pid = container_of(rhp, struct pid, rcu);
203 fastcall void free_pid(struct pid *pid)
205 /* We can be called with write_lock_irq(&tasklist_lock) held */
208 spin_lock_irqsave(&pidmap_lock, flags);
209 hlist_del_rcu(&pid->pid_chain);
210 spin_unlock_irqrestore(&pidmap_lock, flags);
212 free_pidmap(&init_pid_ns, pid->nr);
213 call_rcu(&pid->rcu, delayed_put_pid);
216 struct pid *alloc_pid(void)
221 struct pid_namespace *ns;
223 ns = task_active_pid_ns(current);
224 pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL);
228 nr = alloc_pidmap(ns);
232 atomic_set(&pid->count, 1);
234 for (type = 0; type < PIDTYPE_MAX; ++type)
235 INIT_HLIST_HEAD(&pid->tasks[type]);
237 spin_lock_irq(&pidmap_lock);
238 hlist_add_head_rcu(&pid->pid_chain, &pid_hash[pid_hashfn(pid->nr)]);
239 spin_unlock_irq(&pidmap_lock);
245 kmem_cache_free(ns->pid_cachep, pid);
250 struct pid * fastcall find_pid(int nr)
252 struct hlist_node *elem;
255 hlist_for_each_entry_rcu(pid, elem,
256 &pid_hash[pid_hashfn(nr)], pid_chain) {
262 EXPORT_SYMBOL_GPL(find_pid);
265 * attach_pid() must be called with the tasklist_lock write-held.
267 int fastcall attach_pid(struct task_struct *task, enum pid_type type,
270 struct pid_link *link;
272 link = &task->pids[type];
274 hlist_add_head_rcu(&link->node, &pid->tasks[type]);
279 void fastcall detach_pid(struct task_struct *task, enum pid_type type)
281 struct pid_link *link;
285 link = &task->pids[type];
288 hlist_del_rcu(&link->node);
291 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
292 if (!hlist_empty(&pid->tasks[tmp]))
298 /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
299 void fastcall transfer_pid(struct task_struct *old, struct task_struct *new,
302 new->pids[type].pid = old->pids[type].pid;
303 hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
304 old->pids[type].pid = NULL;
307 struct task_struct * fastcall pid_task(struct pid *pid, enum pid_type type)
309 struct task_struct *result = NULL;
311 struct hlist_node *first;
312 first = rcu_dereference(pid->tasks[type].first);
314 result = hlist_entry(first, struct task_struct, pids[(type)].node);
320 * Must be called under rcu_read_lock() or with tasklist_lock read-held.
322 struct task_struct *find_task_by_pid_type(int type, int nr)
324 return pid_task(find_pid(nr), type);
327 EXPORT_SYMBOL(find_task_by_pid_type);
329 struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
333 pid = get_pid(task->pids[type].pid);
338 struct task_struct *fastcall get_pid_task(struct pid *pid, enum pid_type type)
340 struct task_struct *result;
342 result = pid_task(pid, type);
344 get_task_struct(result);
349 struct pid *find_get_pid(pid_t nr)
354 pid = get_pid(find_pid(nr));
361 * Used by proc to find the first pid that is greater then or equal to nr.
363 * If there is a pid at nr this function is exactly the same as find_pid.
365 struct pid *find_ge_pid(int nr)
373 nr = next_pidmap(task_active_pid_ns(current), nr);
378 EXPORT_SYMBOL_GPL(find_get_pid);
383 struct kmem_cache *cachep;
384 struct list_head list;
387 static LIST_HEAD(pid_caches_lh);
388 static DEFINE_MUTEX(pid_caches_mutex);
391 * creates the kmem cache to allocate pids from.
392 * @nr_ids: the number of numerical ids this pid will have to carry
395 static struct kmem_cache *create_pid_cachep(int nr_ids)
397 struct pid_cache *pcache;
398 struct kmem_cache *cachep;
400 mutex_lock(&pid_caches_mutex);
401 list_for_each_entry (pcache, &pid_caches_lh, list)
402 if (pcache->nr_ids == nr_ids)
405 pcache = kmalloc(sizeof(struct pid_cache), GFP_KERNEL);
409 snprintf(pcache->name, sizeof(pcache->name), "pid_%d", nr_ids);
410 cachep = kmem_cache_create(pcache->name,
411 /* FIXME add numerical ids here */
412 sizeof(struct pid), 0, SLAB_HWCACHE_ALIGN, NULL);
416 pcache->nr_ids = nr_ids;
417 pcache->cachep = cachep;
418 list_add(&pcache->list, &pid_caches_lh);
420 mutex_unlock(&pid_caches_mutex);
421 return pcache->cachep;
426 mutex_unlock(&pid_caches_mutex);
430 struct pid_namespace *copy_pid_ns(unsigned long flags, struct pid_namespace *old_ns)
437 void free_pid_ns(struct kref *kref)
439 struct pid_namespace *ns;
441 ns = container_of(kref, struct pid_namespace, kref);
446 * The pid hash table is scaled according to the amount of memory in the
447 * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or
450 void __init pidhash_init(void)
453 unsigned long megabytes = nr_kernel_pages >> (20 - PAGE_SHIFT);
455 pidhash_shift = max(4, fls(megabytes * 4));
456 pidhash_shift = min(12, pidhash_shift);
457 pidhash_size = 1 << pidhash_shift;
459 printk("PID hash table entries: %d (order: %d, %Zd bytes)\n",
460 pidhash_size, pidhash_shift,
461 pidhash_size * sizeof(struct hlist_head));
463 pid_hash = alloc_bootmem(pidhash_size * sizeof(*(pid_hash)));
465 panic("Could not alloc pidhash!\n");
466 for (i = 0; i < pidhash_size; i++)
467 INIT_HLIST_HEAD(&pid_hash[i]);
470 void __init pidmap_init(void)
472 init_pid_ns.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL);
473 /* Reserve PID 0. We never call free_pidmap(0) */
474 set_bit(0, init_pid_ns.pidmap[0].page);
475 atomic_dec(&init_pid_ns.pidmap[0].nr_free);
477 init_pid_ns.pid_cachep = create_pid_cachep(1);
478 if (init_pid_ns.pid_cachep == NULL)
479 panic("Can't create pid_1 cachep\n");