]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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> | |
aa5a6662 | 29 | #include <linux/pspace.h> |
1da177e4 LT |
30 | |
31 | #define pid_hashfn(nr) hash_long((unsigned long)nr, pidhash_shift) | |
92476d7f | 32 | static struct hlist_head *pid_hash; |
1da177e4 | 33 | static int pidhash_shift; |
92476d7f | 34 | static kmem_cache_t *pid_cachep; |
1da177e4 LT |
35 | |
36 | int pid_max = PID_MAX_DEFAULT; | |
37 | int last_pid; | |
38 | ||
39 | #define RESERVED_PIDS 300 | |
40 | ||
41 | int pid_max_min = RESERVED_PIDS + 1; | |
42 | int pid_max_max = PID_MAX_LIMIT; | |
43 | ||
1da177e4 LT |
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 | */ | |
6a1f3b84 | 56 | static struct pidmap pidmap_array[PIDMAP_ENTRIES] = |
1da177e4 LT |
57 | { [ 0 ... PIDMAP_ENTRIES-1 ] = { ATOMIC_INIT(BITS_PER_PAGE), NULL } }; |
58 | ||
92476d7f EB |
59 | /* |
60 | * Note: disable interrupts while the pidmap_lock is held as an | |
61 | * interrupt might come in and do read_lock(&tasklist_lock). | |
62 | * | |
63 | * If we don't disable interrupts there is a nasty deadlock between | |
64 | * detach_pid()->free_pid() and another cpu that does | |
65 | * spin_lock(&pidmap_lock) followed by an interrupt routine that does | |
66 | * read_lock(&tasklist_lock); | |
67 | * | |
68 | * After we clean up the tasklist_lock and know there are no | |
69 | * irq handlers that take it we can leave the interrupts enabled. | |
70 | * For now it is easier to be safe than to prove it can't happen. | |
71 | */ | |
1da177e4 LT |
72 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock); |
73 | ||
92476d7f | 74 | static fastcall void free_pidmap(int pid) |
1da177e4 | 75 | { |
6a1f3b84 | 76 | struct pidmap *map = pidmap_array + pid / BITS_PER_PAGE; |
1da177e4 LT |
77 | int offset = pid & BITS_PER_PAGE_MASK; |
78 | ||
79 | clear_bit(offset, map->page); | |
80 | atomic_inc(&map->nr_free); | |
81 | } | |
82 | ||
92476d7f | 83 | static int alloc_pidmap(void) |
1da177e4 LT |
84 | { |
85 | int i, offset, max_scan, pid, last = last_pid; | |
6a1f3b84 | 86 | struct pidmap *map; |
1da177e4 LT |
87 | |
88 | pid = last + 1; | |
89 | if (pid >= pid_max) | |
90 | pid = RESERVED_PIDS; | |
91 | offset = pid & BITS_PER_PAGE_MASK; | |
92 | map = &pidmap_array[pid/BITS_PER_PAGE]; | |
93 | max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset; | |
94 | for (i = 0; i <= max_scan; ++i) { | |
95 | if (unlikely(!map->page)) { | |
96 | unsigned long page = get_zeroed_page(GFP_KERNEL); | |
97 | /* | |
98 | * Free the page if someone raced with us | |
99 | * installing it: | |
100 | */ | |
92476d7f | 101 | spin_lock_irq(&pidmap_lock); |
1da177e4 LT |
102 | if (map->page) |
103 | free_page(page); | |
104 | else | |
105 | map->page = (void *)page; | |
92476d7f | 106 | spin_unlock_irq(&pidmap_lock); |
1da177e4 LT |
107 | if (unlikely(!map->page)) |
108 | break; | |
109 | } | |
110 | if (likely(atomic_read(&map->nr_free))) { | |
111 | do { | |
112 | if (!test_and_set_bit(offset, map->page)) { | |
113 | atomic_dec(&map->nr_free); | |
114 | last_pid = pid; | |
115 | return pid; | |
116 | } | |
117 | offset = find_next_offset(map, offset); | |
118 | pid = mk_pid(map, offset); | |
119 | /* | |
120 | * find_next_offset() found a bit, the pid from it | |
121 | * is in-bounds, and if we fell back to the last | |
122 | * bitmap block and the final block was the same | |
123 | * as the starting point, pid is before last_pid. | |
124 | */ | |
125 | } while (offset < BITS_PER_PAGE && pid < pid_max && | |
126 | (i != max_scan || pid < last || | |
127 | !((last+1) & BITS_PER_PAGE_MASK))); | |
128 | } | |
129 | if (map < &pidmap_array[(pid_max-1)/BITS_PER_PAGE]) { | |
130 | ++map; | |
131 | offset = 0; | |
132 | } else { | |
133 | map = &pidmap_array[0]; | |
134 | offset = RESERVED_PIDS; | |
135 | if (unlikely(last == offset)) | |
136 | break; | |
137 | } | |
138 | pid = mk_pid(map, offset); | |
139 | } | |
140 | return -1; | |
141 | } | |
142 | ||
0804ef4b EB |
143 | static int next_pidmap(int last) |
144 | { | |
145 | int offset; | |
c88be3eb | 146 | struct pidmap *map; |
0804ef4b EB |
147 | |
148 | offset = (last + 1) & BITS_PER_PAGE_MASK; | |
149 | map = &pidmap_array[(last + 1)/BITS_PER_PAGE]; | |
150 | for (; map < &pidmap_array[PIDMAP_ENTRIES]; map++, offset = 0) { | |
151 | if (unlikely(!map->page)) | |
152 | continue; | |
153 | offset = find_next_bit((map)->page, BITS_PER_PAGE, offset); | |
154 | if (offset < BITS_PER_PAGE) | |
155 | return mk_pid(map, offset); | |
156 | } | |
157 | return -1; | |
158 | } | |
159 | ||
92476d7f EB |
160 | fastcall void put_pid(struct pid *pid) |
161 | { | |
162 | if (!pid) | |
163 | return; | |
164 | if ((atomic_read(&pid->count) == 1) || | |
165 | atomic_dec_and_test(&pid->count)) | |
166 | kmem_cache_free(pid_cachep, pid); | |
167 | } | |
bbf73147 | 168 | EXPORT_SYMBOL_GPL(put_pid); |
92476d7f EB |
169 | |
170 | static void delayed_put_pid(struct rcu_head *rhp) | |
171 | { | |
172 | struct pid *pid = container_of(rhp, struct pid, rcu); | |
173 | put_pid(pid); | |
174 | } | |
175 | ||
176 | fastcall void free_pid(struct pid *pid) | |
177 | { | |
178 | /* We can be called with write_lock_irq(&tasklist_lock) held */ | |
179 | unsigned long flags; | |
180 | ||
181 | spin_lock_irqsave(&pidmap_lock, flags); | |
182 | hlist_del_rcu(&pid->pid_chain); | |
183 | spin_unlock_irqrestore(&pidmap_lock, flags); | |
184 | ||
185 | free_pidmap(pid->nr); | |
186 | call_rcu(&pid->rcu, delayed_put_pid); | |
187 | } | |
188 | ||
189 | struct pid *alloc_pid(void) | |
190 | { | |
191 | struct pid *pid; | |
192 | enum pid_type type; | |
193 | int nr = -1; | |
194 | ||
195 | pid = kmem_cache_alloc(pid_cachep, GFP_KERNEL); | |
196 | if (!pid) | |
197 | goto out; | |
198 | ||
199 | nr = alloc_pidmap(); | |
200 | if (nr < 0) | |
201 | goto out_free; | |
202 | ||
203 | atomic_set(&pid->count, 1); | |
204 | pid->nr = nr; | |
205 | for (type = 0; type < PIDTYPE_MAX; ++type) | |
206 | INIT_HLIST_HEAD(&pid->tasks[type]); | |
207 | ||
208 | spin_lock_irq(&pidmap_lock); | |
209 | hlist_add_head_rcu(&pid->pid_chain, &pid_hash[pid_hashfn(pid->nr)]); | |
210 | spin_unlock_irq(&pidmap_lock); | |
211 | ||
212 | out: | |
213 | return pid; | |
214 | ||
215 | out_free: | |
216 | kmem_cache_free(pid_cachep, pid); | |
217 | pid = NULL; | |
218 | goto out; | |
219 | } | |
220 | ||
221 | struct pid * fastcall find_pid(int nr) | |
1da177e4 LT |
222 | { |
223 | struct hlist_node *elem; | |
224 | struct pid *pid; | |
225 | ||
e56d0903 | 226 | hlist_for_each_entry_rcu(pid, elem, |
92476d7f | 227 | &pid_hash[pid_hashfn(nr)], pid_chain) { |
1da177e4 LT |
228 | if (pid->nr == nr) |
229 | return pid; | |
230 | } | |
231 | return NULL; | |
232 | } | |
bbf73147 | 233 | EXPORT_SYMBOL_GPL(find_pid); |
1da177e4 | 234 | |
36c8b586 | 235 | int fastcall attach_pid(struct task_struct *task, enum pid_type type, int nr) |
1da177e4 | 236 | { |
92476d7f EB |
237 | struct pid_link *link; |
238 | struct pid *pid; | |
239 | ||
92476d7f EB |
240 | link = &task->pids[type]; |
241 | link->pid = pid = find_pid(nr); | |
242 | hlist_add_head_rcu(&link->node, &pid->tasks[type]); | |
1da177e4 LT |
243 | |
244 | return 0; | |
245 | } | |
246 | ||
36c8b586 | 247 | void fastcall detach_pid(struct task_struct *task, enum pid_type type) |
1da177e4 | 248 | { |
92476d7f EB |
249 | struct pid_link *link; |
250 | struct pid *pid; | |
251 | int tmp; | |
1da177e4 | 252 | |
92476d7f EB |
253 | link = &task->pids[type]; |
254 | pid = link->pid; | |
1da177e4 | 255 | |
92476d7f EB |
256 | hlist_del_rcu(&link->node); |
257 | link->pid = NULL; | |
1da177e4 | 258 | |
92476d7f EB |
259 | for (tmp = PIDTYPE_MAX; --tmp >= 0; ) |
260 | if (!hlist_empty(&pid->tasks[tmp])) | |
261 | return; | |
1da177e4 | 262 | |
92476d7f | 263 | free_pid(pid); |
1da177e4 LT |
264 | } |
265 | ||
c18258c6 EB |
266 | /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */ |
267 | void fastcall transfer_pid(struct task_struct *old, struct task_struct *new, | |
268 | enum pid_type type) | |
269 | { | |
270 | new->pids[type].pid = old->pids[type].pid; | |
271 | hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node); | |
272 | old->pids[type].pid = NULL; | |
273 | } | |
274 | ||
92476d7f | 275 | struct task_struct * fastcall pid_task(struct pid *pid, enum pid_type type) |
1da177e4 | 276 | { |
92476d7f EB |
277 | struct task_struct *result = NULL; |
278 | if (pid) { | |
279 | struct hlist_node *first; | |
280 | first = rcu_dereference(pid->tasks[type].first); | |
281 | if (first) | |
282 | result = hlist_entry(first, struct task_struct, pids[(type)].node); | |
283 | } | |
284 | return result; | |
285 | } | |
1da177e4 | 286 | |
92476d7f EB |
287 | /* |
288 | * Must be called under rcu_read_lock() or with tasklist_lock read-held. | |
289 | */ | |
36c8b586 | 290 | struct task_struct *find_task_by_pid_type(int type, int nr) |
92476d7f EB |
291 | { |
292 | return pid_task(find_pid(nr), type); | |
293 | } | |
1da177e4 | 294 | |
92476d7f | 295 | EXPORT_SYMBOL(find_task_by_pid_type); |
1da177e4 | 296 | |
92476d7f EB |
297 | struct task_struct *fastcall get_pid_task(struct pid *pid, enum pid_type type) |
298 | { | |
299 | struct task_struct *result; | |
300 | rcu_read_lock(); | |
301 | result = pid_task(pid, type); | |
302 | if (result) | |
303 | get_task_struct(result); | |
304 | rcu_read_unlock(); | |
305 | return result; | |
1da177e4 LT |
306 | } |
307 | ||
92476d7f | 308 | struct pid *find_get_pid(pid_t nr) |
1da177e4 LT |
309 | { |
310 | struct pid *pid; | |
311 | ||
92476d7f EB |
312 | rcu_read_lock(); |
313 | pid = get_pid(find_pid(nr)); | |
314 | rcu_read_unlock(); | |
1da177e4 | 315 | |
92476d7f | 316 | return pid; |
1da177e4 LT |
317 | } |
318 | ||
0804ef4b EB |
319 | /* |
320 | * Used by proc to find the first pid that is greater then or equal to nr. | |
321 | * | |
322 | * If there is a pid at nr this function is exactly the same as find_pid. | |
323 | */ | |
324 | struct pid *find_ge_pid(int nr) | |
325 | { | |
326 | struct pid *pid; | |
327 | ||
328 | do { | |
329 | pid = find_pid(nr); | |
330 | if (pid) | |
331 | break; | |
332 | nr = next_pidmap(nr); | |
333 | } while (nr > 0); | |
334 | ||
335 | return pid; | |
336 | } | |
bbf73147 | 337 | EXPORT_SYMBOL_GPL(find_get_pid); |
0804ef4b | 338 | |
1da177e4 LT |
339 | /* |
340 | * The pid hash table is scaled according to the amount of memory in the | |
341 | * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or | |
342 | * more. | |
343 | */ | |
344 | void __init pidhash_init(void) | |
345 | { | |
92476d7f | 346 | int i, pidhash_size; |
1da177e4 LT |
347 | unsigned long megabytes = nr_kernel_pages >> (20 - PAGE_SHIFT); |
348 | ||
349 | pidhash_shift = max(4, fls(megabytes * 4)); | |
350 | pidhash_shift = min(12, pidhash_shift); | |
351 | pidhash_size = 1 << pidhash_shift; | |
352 | ||
353 | printk("PID hash table entries: %d (order: %d, %Zd bytes)\n", | |
354 | pidhash_size, pidhash_shift, | |
92476d7f EB |
355 | pidhash_size * sizeof(struct hlist_head)); |
356 | ||
357 | pid_hash = alloc_bootmem(pidhash_size * sizeof(*(pid_hash))); | |
358 | if (!pid_hash) | |
359 | panic("Could not alloc pidhash!\n"); | |
360 | for (i = 0; i < pidhash_size; i++) | |
361 | INIT_HLIST_HEAD(&pid_hash[i]); | |
1da177e4 LT |
362 | } |
363 | ||
364 | void __init pidmap_init(void) | |
365 | { | |
1da177e4 | 366 | pidmap_array->page = (void *)get_zeroed_page(GFP_KERNEL); |
73b9ebfe | 367 | /* Reserve PID 0. We never call free_pidmap(0) */ |
1da177e4 LT |
368 | set_bit(0, pidmap_array->page); |
369 | atomic_dec(&pidmap_array->nr_free); | |
92476d7f EB |
370 | |
371 | pid_cachep = kmem_cache_create("pid", sizeof(struct pid), | |
372 | __alignof__(struct pid), | |
373 | SLAB_PANIC, NULL, NULL); | |
1da177e4 | 374 | } |