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 static struct kmem_cache *pid_cachep;
36 struct pid init_struct_pid = INIT_STRUCT_PID;
38 int pid_max = PID_MAX_DEFAULT;
40 #define RESERVED_PIDS 300
42 int pid_max_min = RESERVED_PIDS + 1;
43 int pid_max_max = PID_MAX_LIMIT;
45 #define BITS_PER_PAGE (PAGE_SIZE*8)
46 #define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1)
48 static inline int mk_pid(struct pid_namespace *pid_ns,
49 struct pidmap *map, int off)
51 return (map - pid_ns->pidmap)*BITS_PER_PAGE + off;
54 #define find_next_offset(map, off) \
55 find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
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.
63 struct pid_namespace init_pid_ns = {
65 .refcount = ATOMIC_INIT(2),
68 [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL }
71 .child_reaper = &init_task
75 * Note: disable interrupts while the pidmap_lock is held as an
76 * interrupt might come in and do read_lock(&tasklist_lock).
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);
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.
88 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
90 static fastcall void free_pidmap(struct pid_namespace *pid_ns, int pid)
92 struct pidmap *map = pid_ns->pidmap + pid / BITS_PER_PAGE;
93 int offset = pid & BITS_PER_PAGE_MASK;
95 clear_bit(offset, map->page);
96 atomic_inc(&map->nr_free);
99 static int alloc_pidmap(struct pid_namespace *pid_ns)
101 int i, offset, max_scan, pid, last = pid_ns->last_pid;
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);
114 * Free the page if someone raced with us
117 spin_lock_irq(&pidmap_lock);
122 spin_unlock_irq(&pidmap_lock);
123 if (unlikely(!map->page))
126 if (likely(atomic_read(&map->nr_free))) {
128 if (!test_and_set_bit(offset, map->page)) {
129 atomic_dec(&map->nr_free);
130 pid_ns->last_pid = pid;
133 offset = find_next_offset(map, offset);
134 pid = mk_pid(pid_ns, map, offset);
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.
141 } while (offset < BITS_PER_PAGE && pid < pid_max &&
142 (i != max_scan || pid < last ||
143 !((last+1) & BITS_PER_PAGE_MASK)));
145 if (map < &pid_ns->pidmap[(pid_max-1)/BITS_PER_PAGE]) {
149 map = &pid_ns->pidmap[0];
150 offset = RESERVED_PIDS;
151 if (unlikely(last == offset))
154 pid = mk_pid(pid_ns, map, offset);
159 static int next_pidmap(struct pid_namespace *pid_ns, int last)
162 struct pidmap *map, *end;
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))
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);
177 fastcall void put_pid(struct pid *pid)
181 if ((atomic_read(&pid->count) == 1) ||
182 atomic_dec_and_test(&pid->count))
183 kmem_cache_free(pid_cachep, pid);
185 EXPORT_SYMBOL_GPL(put_pid);
187 static void delayed_put_pid(struct rcu_head *rhp)
189 struct pid *pid = container_of(rhp, struct pid, rcu);
193 fastcall void free_pid(struct pid *pid)
195 /* We can be called with write_lock_irq(&tasklist_lock) held */
198 spin_lock_irqsave(&pidmap_lock, flags);
199 hlist_del_rcu(&pid->pid_chain);
200 spin_unlock_irqrestore(&pidmap_lock, flags);
202 free_pidmap(&init_pid_ns, pid->nr);
203 call_rcu(&pid->rcu, delayed_put_pid);
206 struct pid *alloc_pid(void)
212 pid = kmem_cache_alloc(pid_cachep, GFP_KERNEL);
216 nr = alloc_pidmap(current->nsproxy->pid_ns);
220 atomic_set(&pid->count, 1);
222 for (type = 0; type < PIDTYPE_MAX; ++type)
223 INIT_HLIST_HEAD(&pid->tasks[type]);
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);
233 kmem_cache_free(pid_cachep, pid);
238 struct pid * fastcall find_pid(int nr)
240 struct hlist_node *elem;
243 hlist_for_each_entry_rcu(pid, elem,
244 &pid_hash[pid_hashfn(nr)], pid_chain) {
250 EXPORT_SYMBOL_GPL(find_pid);
253 * attach_pid() must be called with the tasklist_lock write-held.
255 int fastcall attach_pid(struct task_struct *task, enum pid_type type,
258 struct pid_link *link;
260 link = &task->pids[type];
262 hlist_add_head_rcu(&link->node, &pid->tasks[type]);
267 void fastcall detach_pid(struct task_struct *task, enum pid_type type)
269 struct pid_link *link;
273 link = &task->pids[type];
276 hlist_del_rcu(&link->node);
279 for (tmp = PIDTYPE_MAX; --tmp >= 0; )
280 if (!hlist_empty(&pid->tasks[tmp]))
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,
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;
295 struct task_struct * fastcall pid_task(struct pid *pid, enum pid_type type)
297 struct task_struct *result = NULL;
299 struct hlist_node *first;
300 first = rcu_dereference(pid->tasks[type].first);
302 result = hlist_entry(first, struct task_struct, pids[(type)].node);
308 * Must be called under rcu_read_lock() or with tasklist_lock read-held.
310 struct task_struct *find_task_by_pid_type(int type, int nr)
312 return pid_task(find_pid(nr), type);
315 EXPORT_SYMBOL(find_task_by_pid_type);
317 struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
321 pid = get_pid(task->pids[type].pid);
326 struct task_struct *fastcall get_pid_task(struct pid *pid, enum pid_type type)
328 struct task_struct *result;
330 result = pid_task(pid, type);
332 get_task_struct(result);
337 struct pid *find_get_pid(pid_t nr)
342 pid = get_pid(find_pid(nr));
349 * Used by proc to find the first pid that is greater then or equal to nr.
351 * If there is a pid at nr this function is exactly the same as find_pid.
353 struct pid *find_ge_pid(int nr)
361 nr = next_pidmap(current->nsproxy->pid_ns, nr);
366 EXPORT_SYMBOL_GPL(find_get_pid);
368 struct pid_namespace *copy_pid_ns(int flags, struct pid_namespace *old_ns)
375 void free_pid_ns(struct kref *kref)
377 struct pid_namespace *ns;
379 ns = container_of(kref, struct pid_namespace, kref);
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
388 void __init pidhash_init(void)
391 unsigned long megabytes = nr_kernel_pages >> (20 - PAGE_SHIFT);
393 pidhash_shift = max(4, fls(megabytes * 4));
394 pidhash_shift = min(12, pidhash_shift);
395 pidhash_size = 1 << pidhash_shift;
397 printk("PID hash table entries: %d (order: %d, %Zd bytes)\n",
398 pidhash_size, pidhash_shift,
399 pidhash_size * sizeof(struct hlist_head));
401 pid_hash = alloc_bootmem(pidhash_size * sizeof(*(pid_hash)));
403 panic("Could not alloc pidhash!\n");
404 for (i = 0; i < pidhash_size; i++)
405 INIT_HLIST_HEAD(&pid_hash[i]);
408 void __init pidmap_init(void)
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);
415 pid_cachep = KMEM_CACHE(pid, SLAB_PANIC);