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>
31 #define pid_hashfn(nr) hash_long((unsigned long)nr, pidhash_shift)
32 static struct hlist_head
*pid_hash
;
33 static int pidhash_shift
;
34 static struct kmem_cache
*pid_cachep
;
36 int pid_max
= PID_MAX_DEFAULT
;
38 #define RESERVED_PIDS 300
40 int pid_max_min
= RESERVED_PIDS
+ 1;
41 int pid_max_max
= PID_MAX_LIMIT
;
43 #define BITS_PER_PAGE (PAGE_SIZE*8)
44 #define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1)
46 static inline int mk_pid(struct pid_namespace
*pid_ns
,
47 struct pidmap
*map
, int off
)
49 return (map
- pid_ns
->pidmap
)*BITS_PER_PAGE
+ off
;
52 #define find_next_offset(map, off) \
53 find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
56 * PID-map pages start out as NULL, they get allocated upon
57 * first use and are never deallocated. This way a low pid_max
58 * value does not cause lots of bitmaps to be allocated, but
59 * the scheme scales to up to 4 million PIDs, runtime.
61 struct pid_namespace init_pid_ns
= {
63 [ 0 ... PIDMAP_ENTRIES
-1] = { ATOMIC_INIT(BITS_PER_PAGE
), NULL
}
69 * Note: disable interrupts while the pidmap_lock is held as an
70 * interrupt might come in and do read_lock(&tasklist_lock).
72 * If we don't disable interrupts there is a nasty deadlock between
73 * detach_pid()->free_pid() and another cpu that does
74 * spin_lock(&pidmap_lock) followed by an interrupt routine that does
75 * read_lock(&tasklist_lock);
77 * After we clean up the tasklist_lock and know there are no
78 * irq handlers that take it we can leave the interrupts enabled.
79 * For now it is easier to be safe than to prove it can't happen.
82 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(pidmap_lock
);
84 static fastcall
void free_pidmap(struct pid_namespace
*pid_ns
, int pid
)
86 struct pidmap
*map
= pid_ns
->pidmap
+ pid
/ BITS_PER_PAGE
;
87 int offset
= pid
& BITS_PER_PAGE_MASK
;
89 clear_bit(offset
, map
->page
);
90 atomic_inc(&map
->nr_free
);
93 static int alloc_pidmap(struct pid_namespace
*pid_ns
)
95 int i
, offset
, max_scan
, pid
, last
= pid_ns
->last_pid
;
101 offset
= pid
& BITS_PER_PAGE_MASK
;
102 map
= &pid_ns
->pidmap
[pid
/BITS_PER_PAGE
];
103 max_scan
= (pid_max
+ BITS_PER_PAGE
- 1)/BITS_PER_PAGE
- !offset
;
104 for (i
= 0; i
<= max_scan
; ++i
) {
105 if (unlikely(!map
->page
)) {
106 void *page
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
108 * Free the page if someone raced with us
111 spin_lock_irq(&pidmap_lock
);
116 spin_unlock_irq(&pidmap_lock
);
117 if (unlikely(!map
->page
))
120 if (likely(atomic_read(&map
->nr_free
))) {
122 if (!test_and_set_bit(offset
, map
->page
)) {
123 atomic_dec(&map
->nr_free
);
124 pid_ns
->last_pid
= pid
;
127 offset
= find_next_offset(map
, offset
);
128 pid
= mk_pid(pid_ns
, map
, offset
);
130 * find_next_offset() found a bit, the pid from it
131 * is in-bounds, and if we fell back to the last
132 * bitmap block and the final block was the same
133 * as the starting point, pid is before last_pid.
135 } while (offset
< BITS_PER_PAGE
&& pid
< pid_max
&&
136 (i
!= max_scan
|| pid
< last
||
137 !((last
+1) & BITS_PER_PAGE_MASK
)));
139 if (map
< &pid_ns
->pidmap
[(pid_max
-1)/BITS_PER_PAGE
]) {
143 map
= &pid_ns
->pidmap
[0];
144 offset
= RESERVED_PIDS
;
145 if (unlikely(last
== offset
))
148 pid
= mk_pid(pid_ns
, map
, offset
);
153 static int next_pidmap(struct pid_namespace
*pid_ns
, int last
)
156 struct pidmap
*map
, *end
;
158 offset
= (last
+ 1) & BITS_PER_PAGE_MASK
;
159 map
= &pid_ns
->pidmap
[(last
+ 1)/BITS_PER_PAGE
];
160 end
= &pid_ns
->pidmap
[PIDMAP_ENTRIES
];
161 for (; map
< end
; map
++, offset
= 0) {
162 if (unlikely(!map
->page
))
164 offset
= find_next_bit((map
)->page
, BITS_PER_PAGE
, offset
);
165 if (offset
< BITS_PER_PAGE
)
166 return mk_pid(pid_ns
, map
, offset
);
171 fastcall
void put_pid(struct pid
*pid
)
175 if ((atomic_read(&pid
->count
) == 1) ||
176 atomic_dec_and_test(&pid
->count
))
177 kmem_cache_free(pid_cachep
, pid
);
179 EXPORT_SYMBOL_GPL(put_pid
);
181 static void delayed_put_pid(struct rcu_head
*rhp
)
183 struct pid
*pid
= container_of(rhp
, struct pid
, rcu
);
187 fastcall
void free_pid(struct pid
*pid
)
189 /* We can be called with write_lock_irq(&tasklist_lock) held */
192 spin_lock_irqsave(&pidmap_lock
, flags
);
193 hlist_del_rcu(&pid
->pid_chain
);
194 spin_unlock_irqrestore(&pidmap_lock
, flags
);
196 free_pidmap(&init_pid_ns
, pid
->nr
);
197 call_rcu(&pid
->rcu
, delayed_put_pid
);
200 struct pid
*alloc_pid(void)
206 pid
= kmem_cache_alloc(pid_cachep
, GFP_KERNEL
);
210 nr
= alloc_pidmap(&init_pid_ns
);
214 atomic_set(&pid
->count
, 1);
216 for (type
= 0; type
< PIDTYPE_MAX
; ++type
)
217 INIT_HLIST_HEAD(&pid
->tasks
[type
]);
219 spin_lock_irq(&pidmap_lock
);
220 hlist_add_head_rcu(&pid
->pid_chain
, &pid_hash
[pid_hashfn(pid
->nr
)]);
221 spin_unlock_irq(&pidmap_lock
);
227 kmem_cache_free(pid_cachep
, pid
);
232 struct pid
* fastcall
find_pid(int nr
)
234 struct hlist_node
*elem
;
237 hlist_for_each_entry_rcu(pid
, elem
,
238 &pid_hash
[pid_hashfn(nr
)], pid_chain
) {
244 EXPORT_SYMBOL_GPL(find_pid
);
246 int fastcall
attach_pid(struct task_struct
*task
, enum pid_type type
, int nr
)
248 struct pid_link
*link
;
251 link
= &task
->pids
[type
];
252 link
->pid
= pid
= find_pid(nr
);
253 hlist_add_head_rcu(&link
->node
, &pid
->tasks
[type
]);
258 void fastcall
detach_pid(struct task_struct
*task
, enum pid_type type
)
260 struct pid_link
*link
;
264 link
= &task
->pids
[type
];
267 hlist_del_rcu(&link
->node
);
270 for (tmp
= PIDTYPE_MAX
; --tmp
>= 0; )
271 if (!hlist_empty(&pid
->tasks
[tmp
]))
277 /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */
278 void fastcall
transfer_pid(struct task_struct
*old
, struct task_struct
*new,
281 new->pids
[type
].pid
= old
->pids
[type
].pid
;
282 hlist_replace_rcu(&old
->pids
[type
].node
, &new->pids
[type
].node
);
283 old
->pids
[type
].pid
= NULL
;
286 struct task_struct
* fastcall
pid_task(struct pid
*pid
, enum pid_type type
)
288 struct task_struct
*result
= NULL
;
290 struct hlist_node
*first
;
291 first
= rcu_dereference(pid
->tasks
[type
].first
);
293 result
= hlist_entry(first
, struct task_struct
, pids
[(type
)].node
);
299 * Must be called under rcu_read_lock() or with tasklist_lock read-held.
301 struct task_struct
*find_task_by_pid_type(int type
, int nr
)
303 return pid_task(find_pid(nr
), type
);
306 EXPORT_SYMBOL(find_task_by_pid_type
);
308 struct pid
*get_task_pid(struct task_struct
*task
, enum pid_type type
)
312 pid
= get_pid(task
->pids
[type
].pid
);
317 struct task_struct
*fastcall
get_pid_task(struct pid
*pid
, enum pid_type type
)
319 struct task_struct
*result
;
321 result
= pid_task(pid
, type
);
323 get_task_struct(result
);
328 struct pid
*find_get_pid(pid_t nr
)
333 pid
= get_pid(find_pid(nr
));
340 * Used by proc to find the first pid that is greater then or equal to nr.
342 * If there is a pid at nr this function is exactly the same as find_pid.
344 struct pid
*find_ge_pid(int nr
)
352 nr
= next_pidmap(&init_pid_ns
, nr
);
357 EXPORT_SYMBOL_GPL(find_get_pid
);
360 * The pid hash table is scaled according to the amount of memory in the
361 * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or
364 void __init
pidhash_init(void)
367 unsigned long megabytes
= nr_kernel_pages
>> (20 - PAGE_SHIFT
);
369 pidhash_shift
= max(4, fls(megabytes
* 4));
370 pidhash_shift
= min(12, pidhash_shift
);
371 pidhash_size
= 1 << pidhash_shift
;
373 printk("PID hash table entries: %d (order: %d, %Zd bytes)\n",
374 pidhash_size
, pidhash_shift
,
375 pidhash_size
* sizeof(struct hlist_head
));
377 pid_hash
= alloc_bootmem(pidhash_size
* sizeof(*(pid_hash
)));
379 panic("Could not alloc pidhash!\n");
380 for (i
= 0; i
< pidhash_size
; i
++)
381 INIT_HLIST_HEAD(&pid_hash
[i
]);
384 void __init
pidmap_init(void)
386 init_pid_ns
.pidmap
[0].page
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
387 /* Reserve PID 0. We never call free_pidmap(0) */
388 set_bit(0, init_pid_ns
.pidmap
[0].page
);
389 atomic_dec(&init_pid_ns
.pidmap
[0].nr_free
);
391 pid_cachep
= kmem_cache_create("pid", sizeof(struct pid
),
392 __alignof__(struct pid
),
393 SLAB_PANIC
, NULL
, NULL
);