xref: /linux/kernel/pid.c (revision 14b42963f64b98ab61fa9723c03d71aa5ef4f862)
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 
30 #define pid_hashfn(nr) hash_long((unsigned long)nr, pidhash_shift)
31 static struct hlist_head *pid_hash;
32 static int pidhash_shift;
33 static kmem_cache_t *pid_cachep;
34 
35 int pid_max = PID_MAX_DEFAULT;
36 int last_pid;
37 
38 #define RESERVED_PIDS		300
39 
40 int pid_max_min = RESERVED_PIDS + 1;
41 int pid_max_max = PID_MAX_LIMIT;
42 
43 #define PIDMAP_ENTRIES		((PID_MAX_LIMIT + 8*PAGE_SIZE - 1)/PAGE_SIZE/8)
44 #define BITS_PER_PAGE		(PAGE_SIZE*8)
45 #define BITS_PER_PAGE_MASK	(BITS_PER_PAGE-1)
46 #define mk_pid(map, off)	(((map) - pidmap_array)*BITS_PER_PAGE + (off))
47 #define find_next_offset(map, off)					\
48 		find_next_zero_bit((map)->page, BITS_PER_PAGE, off)
49 
50 /*
51  * PID-map pages start out as NULL, they get allocated upon
52  * first use and are never deallocated. This way a low pid_max
53  * value does not cause lots of bitmaps to be allocated, but
54  * the scheme scales to up to 4 million PIDs, runtime.
55  */
56 typedef struct pidmap {
57 	atomic_t nr_free;
58 	void *page;
59 } pidmap_t;
60 
61 static pidmap_t pidmap_array[PIDMAP_ENTRIES] =
62 	 { [ 0 ... PIDMAP_ENTRIES-1 ] = { ATOMIC_INIT(BITS_PER_PAGE), NULL } };
63 
64 /*
65  * Note: disable interrupts while the pidmap_lock is held as an
66  * interrupt might come in and do read_lock(&tasklist_lock).
67  *
68  * If we don't disable interrupts there is a nasty deadlock between
69  * detach_pid()->free_pid() and another cpu that does
70  * spin_lock(&pidmap_lock) followed by an interrupt routine that does
71  * read_lock(&tasklist_lock);
72  *
73  * After we clean up the tasklist_lock and know there are no
74  * irq handlers that take it we can leave the interrupts enabled.
75  * For now it is easier to be safe than to prove it can't happen.
76  */
77 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock);
78 
79 static fastcall void free_pidmap(int pid)
80 {
81 	pidmap_t *map = pidmap_array + pid / BITS_PER_PAGE;
82 	int offset = pid & BITS_PER_PAGE_MASK;
83 
84 	clear_bit(offset, map->page);
85 	atomic_inc(&map->nr_free);
86 }
87 
88 static int alloc_pidmap(void)
89 {
90 	int i, offset, max_scan, pid, last = last_pid;
91 	pidmap_t *map;
92 
93 	pid = last + 1;
94 	if (pid >= pid_max)
95 		pid = RESERVED_PIDS;
96 	offset = pid & BITS_PER_PAGE_MASK;
97 	map = &pidmap_array[pid/BITS_PER_PAGE];
98 	max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset;
99 	for (i = 0; i <= max_scan; ++i) {
100 		if (unlikely(!map->page)) {
101 			unsigned long page = get_zeroed_page(GFP_KERNEL);
102 			/*
103 			 * Free the page if someone raced with us
104 			 * installing it:
105 			 */
106 			spin_lock_irq(&pidmap_lock);
107 			if (map->page)
108 				free_page(page);
109 			else
110 				map->page = (void *)page;
111 			spin_unlock_irq(&pidmap_lock);
112 			if (unlikely(!map->page))
113 				break;
114 		}
115 		if (likely(atomic_read(&map->nr_free))) {
116 			do {
117 				if (!test_and_set_bit(offset, map->page)) {
118 					atomic_dec(&map->nr_free);
119 					last_pid = pid;
120 					return pid;
121 				}
122 				offset = find_next_offset(map, offset);
123 				pid = mk_pid(map, offset);
124 			/*
125 			 * find_next_offset() found a bit, the pid from it
126 			 * is in-bounds, and if we fell back to the last
127 			 * bitmap block and the final block was the same
128 			 * as the starting point, pid is before last_pid.
129 			 */
130 			} while (offset < BITS_PER_PAGE && pid < pid_max &&
131 					(i != max_scan || pid < last ||
132 					    !((last+1) & BITS_PER_PAGE_MASK)));
133 		}
134 		if (map < &pidmap_array[(pid_max-1)/BITS_PER_PAGE]) {
135 			++map;
136 			offset = 0;
137 		} else {
138 			map = &pidmap_array[0];
139 			offset = RESERVED_PIDS;
140 			if (unlikely(last == offset))
141 				break;
142 		}
143 		pid = mk_pid(map, offset);
144 	}
145 	return -1;
146 }
147 
148 fastcall void put_pid(struct pid *pid)
149 {
150 	if (!pid)
151 		return;
152 	if ((atomic_read(&pid->count) == 1) ||
153 	     atomic_dec_and_test(&pid->count))
154 		kmem_cache_free(pid_cachep, pid);
155 }
156 
157 static void delayed_put_pid(struct rcu_head *rhp)
158 {
159 	struct pid *pid = container_of(rhp, struct pid, rcu);
160 	put_pid(pid);
161 }
162 
163 fastcall void free_pid(struct pid *pid)
164 {
165 	/* We can be called with write_lock_irq(&tasklist_lock) held */
166 	unsigned long flags;
167 
168 	spin_lock_irqsave(&pidmap_lock, flags);
169 	hlist_del_rcu(&pid->pid_chain);
170 	spin_unlock_irqrestore(&pidmap_lock, flags);
171 
172 	free_pidmap(pid->nr);
173 	call_rcu(&pid->rcu, delayed_put_pid);
174 }
175 
176 struct pid *alloc_pid(void)
177 {
178 	struct pid *pid;
179 	enum pid_type type;
180 	int nr = -1;
181 
182 	pid = kmem_cache_alloc(pid_cachep, GFP_KERNEL);
183 	if (!pid)
184 		goto out;
185 
186 	nr = alloc_pidmap();
187 	if (nr < 0)
188 		goto out_free;
189 
190 	atomic_set(&pid->count, 1);
191 	pid->nr = nr;
192 	for (type = 0; type < PIDTYPE_MAX; ++type)
193 		INIT_HLIST_HEAD(&pid->tasks[type]);
194 
195 	spin_lock_irq(&pidmap_lock);
196 	hlist_add_head_rcu(&pid->pid_chain, &pid_hash[pid_hashfn(pid->nr)]);
197 	spin_unlock_irq(&pidmap_lock);
198 
199 out:
200 	return pid;
201 
202 out_free:
203 	kmem_cache_free(pid_cachep, pid);
204 	pid = NULL;
205 	goto out;
206 }
207 
208 struct pid * fastcall find_pid(int nr)
209 {
210 	struct hlist_node *elem;
211 	struct pid *pid;
212 
213 	hlist_for_each_entry_rcu(pid, elem,
214 			&pid_hash[pid_hashfn(nr)], pid_chain) {
215 		if (pid->nr == nr)
216 			return pid;
217 	}
218 	return NULL;
219 }
220 
221 int fastcall attach_pid(struct task_struct *task, enum pid_type type, int nr)
222 {
223 	struct pid_link *link;
224 	struct pid *pid;
225 
226 	WARN_ON(!task->pid); /* to be removed soon */
227 	WARN_ON(!nr); /* to be removed soon */
228 
229 	link = &task->pids[type];
230 	link->pid = pid = find_pid(nr);
231 	hlist_add_head_rcu(&link->node, &pid->tasks[type]);
232 
233 	return 0;
234 }
235 
236 void fastcall detach_pid(struct task_struct *task, enum pid_type type)
237 {
238 	struct pid_link *link;
239 	struct pid *pid;
240 	int tmp;
241 
242 	link = &task->pids[type];
243 	pid = link->pid;
244 
245 	hlist_del_rcu(&link->node);
246 	link->pid = NULL;
247 
248 	for (tmp = PIDTYPE_MAX; --tmp >= 0; )
249 		if (!hlist_empty(&pid->tasks[tmp]))
250 			return;
251 
252 	free_pid(pid);
253 }
254 
255 struct task_struct * fastcall pid_task(struct pid *pid, enum pid_type type)
256 {
257 	struct task_struct *result = NULL;
258 	if (pid) {
259 		struct hlist_node *first;
260 		first = rcu_dereference(pid->tasks[type].first);
261 		if (first)
262 			result = hlist_entry(first, struct task_struct, pids[(type)].node);
263 	}
264 	return result;
265 }
266 
267 /*
268  * Must be called under rcu_read_lock() or with tasklist_lock read-held.
269  */
270 struct task_struct *find_task_by_pid_type(int type, int nr)
271 {
272 	return pid_task(find_pid(nr), type);
273 }
274 
275 EXPORT_SYMBOL(find_task_by_pid_type);
276 
277 struct task_struct *fastcall get_pid_task(struct pid *pid, enum pid_type type)
278 {
279 	struct task_struct *result;
280 	rcu_read_lock();
281 	result = pid_task(pid, type);
282 	if (result)
283 		get_task_struct(result);
284 	rcu_read_unlock();
285 	return result;
286 }
287 
288 struct pid *find_get_pid(pid_t nr)
289 {
290 	struct pid *pid;
291 
292 	rcu_read_lock();
293 	pid = get_pid(find_pid(nr));
294 	rcu_read_unlock();
295 
296 	return pid;
297 }
298 
299 /*
300  * The pid hash table is scaled according to the amount of memory in the
301  * machine.  From a minimum of 16 slots up to 4096 slots at one gigabyte or
302  * more.
303  */
304 void __init pidhash_init(void)
305 {
306 	int i, pidhash_size;
307 	unsigned long megabytes = nr_kernel_pages >> (20 - PAGE_SHIFT);
308 
309 	pidhash_shift = max(4, fls(megabytes * 4));
310 	pidhash_shift = min(12, pidhash_shift);
311 	pidhash_size = 1 << pidhash_shift;
312 
313 	printk("PID hash table entries: %d (order: %d, %Zd bytes)\n",
314 		pidhash_size, pidhash_shift,
315 		pidhash_size * sizeof(struct hlist_head));
316 
317 	pid_hash = alloc_bootmem(pidhash_size *	sizeof(*(pid_hash)));
318 	if (!pid_hash)
319 		panic("Could not alloc pidhash!\n");
320 	for (i = 0; i < pidhash_size; i++)
321 		INIT_HLIST_HEAD(&pid_hash[i]);
322 }
323 
324 void __init pidmap_init(void)
325 {
326 	pidmap_array->page = (void *)get_zeroed_page(GFP_KERNEL);
327 	/* Reserve PID 0. We never call free_pidmap(0) */
328 	set_bit(0, pidmap_array->page);
329 	atomic_dec(&pidmap_array->nr_free);
330 
331 	pid_cachep = kmem_cache_create("pid", sizeof(struct pid),
332 					__alignof__(struct pid),
333 					SLAB_PANIC, NULL, NULL);
334 }
335