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; | |
1da177e4 LT |
37 | |
38 | #define RESERVED_PIDS 300 | |
39 | ||
40 | int pid_max_min = RESERVED_PIDS + 1; | |
41 | int pid_max_max = PID_MAX_LIMIT; | |
42 | ||
1da177e4 LT |
43 | #define BITS_PER_PAGE (PAGE_SIZE*8) |
44 | #define BITS_PER_PAGE_MASK (BITS_PER_PAGE-1) | |
3fbc9648 SB |
45 | |
46 | static inline int mk_pid(struct pspace *pspace, struct pidmap *map, int off) | |
47 | { | |
48 | return (map - pspace->pidmap)*BITS_PER_PAGE + off; | |
49 | } | |
50 | ||
1da177e4 LT |
51 | #define find_next_offset(map, off) \ |
52 | find_next_zero_bit((map)->page, BITS_PER_PAGE, off) | |
53 | ||
54 | /* | |
55 | * PID-map pages start out as NULL, they get allocated upon | |
56 | * first use and are never deallocated. This way a low pid_max | |
57 | * value does not cause lots of bitmaps to be allocated, but | |
58 | * the scheme scales to up to 4 million PIDs, runtime. | |
59 | */ | |
3fbc9648 SB |
60 | struct pspace init_pspace = { |
61 | .pidmap = { | |
62 | [ 0 ... PIDMAP_ENTRIES-1] = { ATOMIC_INIT(BITS_PER_PAGE), NULL } | |
63 | }, | |
64 | .last_pid = 0 | |
65 | }; | |
1da177e4 | 66 | |
92476d7f EB |
67 | /* |
68 | * Note: disable interrupts while the pidmap_lock is held as an | |
69 | * interrupt might come in and do read_lock(&tasklist_lock). | |
70 | * | |
71 | * If we don't disable interrupts there is a nasty deadlock between | |
72 | * detach_pid()->free_pid() and another cpu that does | |
73 | * spin_lock(&pidmap_lock) followed by an interrupt routine that does | |
74 | * read_lock(&tasklist_lock); | |
75 | * | |
76 | * After we clean up the tasklist_lock and know there are no | |
77 | * irq handlers that take it we can leave the interrupts enabled. | |
78 | * For now it is easier to be safe than to prove it can't happen. | |
79 | */ | |
3fbc9648 | 80 | |
1da177e4 LT |
81 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock); |
82 | ||
3fbc9648 | 83 | static fastcall void free_pidmap(struct pspace *pspace, int pid) |
1da177e4 | 84 | { |
3fbc9648 | 85 | struct pidmap *map = pspace->pidmap + pid / BITS_PER_PAGE; |
1da177e4 LT |
86 | int offset = pid & BITS_PER_PAGE_MASK; |
87 | ||
88 | clear_bit(offset, map->page); | |
89 | atomic_inc(&map->nr_free); | |
90 | } | |
91 | ||
3fbc9648 | 92 | static int alloc_pidmap(struct pspace *pspace) |
1da177e4 | 93 | { |
3fbc9648 | 94 | int i, offset, max_scan, pid, last = pspace->last_pid; |
6a1f3b84 | 95 | struct pidmap *map; |
1da177e4 LT |
96 | |
97 | pid = last + 1; | |
98 | if (pid >= pid_max) | |
99 | pid = RESERVED_PIDS; | |
100 | offset = pid & BITS_PER_PAGE_MASK; | |
3fbc9648 | 101 | map = &pspace->pidmap[pid/BITS_PER_PAGE]; |
1da177e4 LT |
102 | max_scan = (pid_max + BITS_PER_PAGE - 1)/BITS_PER_PAGE - !offset; |
103 | for (i = 0; i <= max_scan; ++i) { | |
104 | if (unlikely(!map->page)) { | |
3fbc9648 | 105 | void *page = kzalloc(PAGE_SIZE, GFP_KERNEL); |
1da177e4 LT |
106 | /* |
107 | * Free the page if someone raced with us | |
108 | * installing it: | |
109 | */ | |
92476d7f | 110 | spin_lock_irq(&pidmap_lock); |
1da177e4 | 111 | if (map->page) |
3fbc9648 | 112 | kfree(page); |
1da177e4 | 113 | else |
3fbc9648 | 114 | map->page = page; |
92476d7f | 115 | spin_unlock_irq(&pidmap_lock); |
1da177e4 LT |
116 | if (unlikely(!map->page)) |
117 | break; | |
118 | } | |
119 | if (likely(atomic_read(&map->nr_free))) { | |
120 | do { | |
121 | if (!test_and_set_bit(offset, map->page)) { | |
122 | atomic_dec(&map->nr_free); | |
3fbc9648 | 123 | pspace->last_pid = pid; |
1da177e4 LT |
124 | return pid; |
125 | } | |
126 | offset = find_next_offset(map, offset); | |
3fbc9648 | 127 | pid = mk_pid(pspace, map, offset); |
1da177e4 LT |
128 | /* |
129 | * find_next_offset() found a bit, the pid from it | |
130 | * is in-bounds, and if we fell back to the last | |
131 | * bitmap block and the final block was the same | |
132 | * as the starting point, pid is before last_pid. | |
133 | */ | |
134 | } while (offset < BITS_PER_PAGE && pid < pid_max && | |
135 | (i != max_scan || pid < last || | |
136 | !((last+1) & BITS_PER_PAGE_MASK))); | |
137 | } | |
3fbc9648 | 138 | if (map < &pspace->pidmap[(pid_max-1)/BITS_PER_PAGE]) { |
1da177e4 LT |
139 | ++map; |
140 | offset = 0; | |
141 | } else { | |
3fbc9648 | 142 | map = &pspace->pidmap[0]; |
1da177e4 LT |
143 | offset = RESERVED_PIDS; |
144 | if (unlikely(last == offset)) | |
145 | break; | |
146 | } | |
3fbc9648 | 147 | pid = mk_pid(pspace, map, offset); |
1da177e4 LT |
148 | } |
149 | return -1; | |
150 | } | |
151 | ||
f40f50d3 | 152 | static int next_pidmap(struct pspace *pspace, int last) |
0804ef4b EB |
153 | { |
154 | int offset; | |
f40f50d3 | 155 | struct pidmap *map, *end; |
0804ef4b EB |
156 | |
157 | offset = (last + 1) & BITS_PER_PAGE_MASK; | |
f40f50d3 EB |
158 | map = &pspace->pidmap[(last + 1)/BITS_PER_PAGE]; |
159 | end = &pspace->pidmap[PIDMAP_ENTRIES]; | |
160 | for (; map < end; map++, offset = 0) { | |
0804ef4b EB |
161 | if (unlikely(!map->page)) |
162 | continue; | |
163 | offset = find_next_bit((map)->page, BITS_PER_PAGE, offset); | |
164 | if (offset < BITS_PER_PAGE) | |
f40f50d3 | 165 | return mk_pid(pspace, map, offset); |
0804ef4b EB |
166 | } |
167 | return -1; | |
168 | } | |
169 | ||
92476d7f EB |
170 | fastcall void put_pid(struct pid *pid) |
171 | { | |
172 | if (!pid) | |
173 | return; | |
174 | if ((atomic_read(&pid->count) == 1) || | |
175 | atomic_dec_and_test(&pid->count)) | |
176 | kmem_cache_free(pid_cachep, pid); | |
177 | } | |
bbf73147 | 178 | EXPORT_SYMBOL_GPL(put_pid); |
92476d7f EB |
179 | |
180 | static void delayed_put_pid(struct rcu_head *rhp) | |
181 | { | |
182 | struct pid *pid = container_of(rhp, struct pid, rcu); | |
183 | put_pid(pid); | |
184 | } | |
185 | ||
186 | fastcall void free_pid(struct pid *pid) | |
187 | { | |
188 | /* We can be called with write_lock_irq(&tasklist_lock) held */ | |
189 | unsigned long flags; | |
190 | ||
191 | spin_lock_irqsave(&pidmap_lock, flags); | |
192 | hlist_del_rcu(&pid->pid_chain); | |
193 | spin_unlock_irqrestore(&pidmap_lock, flags); | |
194 | ||
3fbc9648 | 195 | free_pidmap(&init_pspace, pid->nr); |
92476d7f EB |
196 | call_rcu(&pid->rcu, delayed_put_pid); |
197 | } | |
198 | ||
199 | struct pid *alloc_pid(void) | |
200 | { | |
201 | struct pid *pid; | |
202 | enum pid_type type; | |
203 | int nr = -1; | |
204 | ||
205 | pid = kmem_cache_alloc(pid_cachep, GFP_KERNEL); | |
206 | if (!pid) | |
207 | goto out; | |
208 | ||
3fbc9648 | 209 | nr = alloc_pidmap(&init_pspace); |
92476d7f EB |
210 | if (nr < 0) |
211 | goto out_free; | |
212 | ||
213 | atomic_set(&pid->count, 1); | |
214 | pid->nr = nr; | |
215 | for (type = 0; type < PIDTYPE_MAX; ++type) | |
216 | INIT_HLIST_HEAD(&pid->tasks[type]); | |
217 | ||
218 | spin_lock_irq(&pidmap_lock); | |
219 | hlist_add_head_rcu(&pid->pid_chain, &pid_hash[pid_hashfn(pid->nr)]); | |
220 | spin_unlock_irq(&pidmap_lock); | |
221 | ||
222 | out: | |
223 | return pid; | |
224 | ||
225 | out_free: | |
226 | kmem_cache_free(pid_cachep, pid); | |
227 | pid = NULL; | |
228 | goto out; | |
229 | } | |
230 | ||
231 | struct pid * fastcall find_pid(int nr) | |
1da177e4 LT |
232 | { |
233 | struct hlist_node *elem; | |
234 | struct pid *pid; | |
235 | ||
e56d0903 | 236 | hlist_for_each_entry_rcu(pid, elem, |
92476d7f | 237 | &pid_hash[pid_hashfn(nr)], pid_chain) { |
1da177e4 LT |
238 | if (pid->nr == nr) |
239 | return pid; | |
240 | } | |
241 | return NULL; | |
242 | } | |
bbf73147 | 243 | EXPORT_SYMBOL_GPL(find_pid); |
1da177e4 | 244 | |
36c8b586 | 245 | int fastcall attach_pid(struct task_struct *task, enum pid_type type, int nr) |
1da177e4 | 246 | { |
92476d7f EB |
247 | struct pid_link *link; |
248 | struct pid *pid; | |
249 | ||
92476d7f EB |
250 | link = &task->pids[type]; |
251 | link->pid = pid = find_pid(nr); | |
252 | hlist_add_head_rcu(&link->node, &pid->tasks[type]); | |
1da177e4 LT |
253 | |
254 | return 0; | |
255 | } | |
256 | ||
36c8b586 | 257 | void fastcall detach_pid(struct task_struct *task, enum pid_type type) |
1da177e4 | 258 | { |
92476d7f EB |
259 | struct pid_link *link; |
260 | struct pid *pid; | |
261 | int tmp; | |
1da177e4 | 262 | |
92476d7f EB |
263 | link = &task->pids[type]; |
264 | pid = link->pid; | |
1da177e4 | 265 | |
92476d7f EB |
266 | hlist_del_rcu(&link->node); |
267 | link->pid = NULL; | |
1da177e4 | 268 | |
92476d7f EB |
269 | for (tmp = PIDTYPE_MAX; --tmp >= 0; ) |
270 | if (!hlist_empty(&pid->tasks[tmp])) | |
271 | return; | |
1da177e4 | 272 | |
92476d7f | 273 | free_pid(pid); |
1da177e4 LT |
274 | } |
275 | ||
c18258c6 EB |
276 | /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */ |
277 | void fastcall transfer_pid(struct task_struct *old, struct task_struct *new, | |
278 | enum pid_type type) | |
279 | { | |
280 | new->pids[type].pid = old->pids[type].pid; | |
281 | hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node); | |
282 | old->pids[type].pid = NULL; | |
283 | } | |
284 | ||
92476d7f | 285 | struct task_struct * fastcall pid_task(struct pid *pid, enum pid_type type) |
1da177e4 | 286 | { |
92476d7f EB |
287 | struct task_struct *result = NULL; |
288 | if (pid) { | |
289 | struct hlist_node *first; | |
290 | first = rcu_dereference(pid->tasks[type].first); | |
291 | if (first) | |
292 | result = hlist_entry(first, struct task_struct, pids[(type)].node); | |
293 | } | |
294 | return result; | |
295 | } | |
1da177e4 | 296 | |
92476d7f EB |
297 | /* |
298 | * Must be called under rcu_read_lock() or with tasklist_lock read-held. | |
299 | */ | |
36c8b586 | 300 | struct task_struct *find_task_by_pid_type(int type, int nr) |
92476d7f EB |
301 | { |
302 | return pid_task(find_pid(nr), type); | |
303 | } | |
1da177e4 | 304 | |
92476d7f | 305 | EXPORT_SYMBOL(find_task_by_pid_type); |
1da177e4 | 306 | |
1a657f78 ON |
307 | struct pid *get_task_pid(struct task_struct *task, enum pid_type type) |
308 | { | |
309 | struct pid *pid; | |
310 | rcu_read_lock(); | |
311 | pid = get_pid(task->pids[type].pid); | |
312 | rcu_read_unlock(); | |
313 | return pid; | |
314 | } | |
315 | ||
92476d7f EB |
316 | struct task_struct *fastcall get_pid_task(struct pid *pid, enum pid_type type) |
317 | { | |
318 | struct task_struct *result; | |
319 | rcu_read_lock(); | |
320 | result = pid_task(pid, type); | |
321 | if (result) | |
322 | get_task_struct(result); | |
323 | rcu_read_unlock(); | |
324 | return result; | |
1da177e4 LT |
325 | } |
326 | ||
92476d7f | 327 | struct pid *find_get_pid(pid_t nr) |
1da177e4 LT |
328 | { |
329 | struct pid *pid; | |
330 | ||
92476d7f EB |
331 | rcu_read_lock(); |
332 | pid = get_pid(find_pid(nr)); | |
333 | rcu_read_unlock(); | |
1da177e4 | 334 | |
92476d7f | 335 | return pid; |
1da177e4 LT |
336 | } |
337 | ||
0804ef4b EB |
338 | /* |
339 | * Used by proc to find the first pid that is greater then or equal to nr. | |
340 | * | |
341 | * If there is a pid at nr this function is exactly the same as find_pid. | |
342 | */ | |
343 | struct pid *find_ge_pid(int nr) | |
344 | { | |
345 | struct pid *pid; | |
346 | ||
347 | do { | |
348 | pid = find_pid(nr); | |
349 | if (pid) | |
350 | break; | |
f40f50d3 | 351 | nr = next_pidmap(&init_pspace, nr); |
0804ef4b EB |
352 | } while (nr > 0); |
353 | ||
354 | return pid; | |
355 | } | |
bbf73147 | 356 | EXPORT_SYMBOL_GPL(find_get_pid); |
0804ef4b | 357 | |
1da177e4 LT |
358 | /* |
359 | * The pid hash table is scaled according to the amount of memory in the | |
360 | * machine. From a minimum of 16 slots up to 4096 slots at one gigabyte or | |
361 | * more. | |
362 | */ | |
363 | void __init pidhash_init(void) | |
364 | { | |
92476d7f | 365 | int i, pidhash_size; |
1da177e4 LT |
366 | unsigned long megabytes = nr_kernel_pages >> (20 - PAGE_SHIFT); |
367 | ||
368 | pidhash_shift = max(4, fls(megabytes * 4)); | |
369 | pidhash_shift = min(12, pidhash_shift); | |
370 | pidhash_size = 1 << pidhash_shift; | |
371 | ||
372 | printk("PID hash table entries: %d (order: %d, %Zd bytes)\n", | |
373 | pidhash_size, pidhash_shift, | |
92476d7f EB |
374 | pidhash_size * sizeof(struct hlist_head)); |
375 | ||
376 | pid_hash = alloc_bootmem(pidhash_size * sizeof(*(pid_hash))); | |
377 | if (!pid_hash) | |
378 | panic("Could not alloc pidhash!\n"); | |
379 | for (i = 0; i < pidhash_size; i++) | |
380 | INIT_HLIST_HEAD(&pid_hash[i]); | |
1da177e4 LT |
381 | } |
382 | ||
383 | void __init pidmap_init(void) | |
384 | { | |
3fbc9648 | 385 | init_pspace.pidmap[0].page = kzalloc(PAGE_SIZE, GFP_KERNEL); |
73b9ebfe | 386 | /* Reserve PID 0. We never call free_pidmap(0) */ |
3fbc9648 SB |
387 | set_bit(0, init_pspace.pidmap[0].page); |
388 | atomic_dec(&init_pspace.pidmap[0].nr_free); | |
92476d7f EB |
389 | |
390 | pid_cachep = kmem_cache_create("pid", sizeof(struct pid), | |
391 | __alignof__(struct pid), | |
392 | SLAB_PANIC, NULL, NULL); | |
1da177e4 | 393 | } |