]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Generic pidhash and scalable, time-bounded PID allocator | |
3 | * | |
6d49e352 NYC |
4 | * (C) 2002-2003 Nadia Yvette Chambers, IBM |
5 | * (C) 2004 Nadia Yvette Chambers, Oracle | |
1da177e4 LT |
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). | |
30e49c26 PE |
21 | * |
22 | * Pid namespaces: | |
23 | * (C) 2007 Pavel Emelyanov <xemul@openvz.org>, OpenVZ, SWsoft Inc. | |
24 | * (C) 2007 Sukadev Bhattiprolu <sukadev@us.ibm.com>, IBM | |
25 | * Many thanks to Oleg Nesterov for comments and help | |
26 | * | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
9984de1a | 30 | #include <linux/export.h> |
1da177e4 LT |
31 | #include <linux/slab.h> |
32 | #include <linux/init.h> | |
82524746 | 33 | #include <linux/rculist.h> |
1da177e4 LT |
34 | #include <linux/bootmem.h> |
35 | #include <linux/hash.h> | |
61a58c6c | 36 | #include <linux/pid_namespace.h> |
820e45db | 37 | #include <linux/init_task.h> |
3eb07c8c | 38 | #include <linux/syscalls.h> |
0bb80f24 | 39 | #include <linux/proc_ns.h> |
0a01f2cc | 40 | #include <linux/proc_fs.h> |
29930025 | 41 | #include <linux/sched/task.h> |
95846ecf | 42 | #include <linux/idr.h> |
1da177e4 | 43 | |
820e45db | 44 | struct pid init_struct_pid = INIT_STRUCT_PID; |
1da177e4 LT |
45 | |
46 | int pid_max = PID_MAX_DEFAULT; | |
1da177e4 LT |
47 | |
48 | #define RESERVED_PIDS 300 | |
49 | ||
50 | int pid_max_min = RESERVED_PIDS + 1; | |
51 | int pid_max_max = PID_MAX_LIMIT; | |
52 | ||
1da177e4 LT |
53 | /* |
54 | * PID-map pages start out as NULL, they get allocated upon | |
55 | * first use and are never deallocated. This way a low pid_max | |
56 | * value does not cause lots of bitmaps to be allocated, but | |
57 | * the scheme scales to up to 4 million PIDs, runtime. | |
58 | */ | |
61a58c6c | 59 | struct pid_namespace init_pid_ns = { |
1e24edca | 60 | .kref = KREF_INIT(2), |
95846ecf | 61 | .idr = IDR_INIT, |
e8cfbc24 | 62 | .pid_allocated = PIDNS_ADDING, |
faacbfd3 PE |
63 | .level = 0, |
64 | .child_reaper = &init_task, | |
49f4d8b9 | 65 | .user_ns = &init_user_ns, |
435d5f4b | 66 | .ns.inum = PROC_PID_INIT_INO, |
33c42940 AV |
67 | #ifdef CONFIG_PID_NS |
68 | .ns.ops = &pidns_operations, | |
69 | #endif | |
3fbc9648 | 70 | }; |
198fe21b | 71 | EXPORT_SYMBOL_GPL(init_pid_ns); |
1da177e4 | 72 | |
92476d7f EB |
73 | /* |
74 | * Note: disable interrupts while the pidmap_lock is held as an | |
75 | * interrupt might come in and do read_lock(&tasklist_lock). | |
76 | * | |
77 | * If we don't disable interrupts there is a nasty deadlock between | |
78 | * detach_pid()->free_pid() and another cpu that does | |
79 | * spin_lock(&pidmap_lock) followed by an interrupt routine that does | |
80 | * read_lock(&tasklist_lock); | |
81 | * | |
82 | * After we clean up the tasklist_lock and know there are no | |
83 | * irq handlers that take it we can leave the interrupts enabled. | |
84 | * For now it is easier to be safe than to prove it can't happen. | |
85 | */ | |
3fbc9648 | 86 | |
1da177e4 LT |
87 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(pidmap_lock); |
88 | ||
7ad5b3a5 | 89 | void put_pid(struct pid *pid) |
92476d7f | 90 | { |
baf8f0f8 PE |
91 | struct pid_namespace *ns; |
92 | ||
92476d7f EB |
93 | if (!pid) |
94 | return; | |
baf8f0f8 | 95 | |
8ef047aa | 96 | ns = pid->numbers[pid->level].ns; |
92476d7f | 97 | if ((atomic_read(&pid->count) == 1) || |
8ef047aa | 98 | atomic_dec_and_test(&pid->count)) { |
baf8f0f8 | 99 | kmem_cache_free(ns->pid_cachep, pid); |
b461cc03 | 100 | put_pid_ns(ns); |
8ef047aa | 101 | } |
92476d7f | 102 | } |
bbf73147 | 103 | EXPORT_SYMBOL_GPL(put_pid); |
92476d7f EB |
104 | |
105 | static void delayed_put_pid(struct rcu_head *rhp) | |
106 | { | |
107 | struct pid *pid = container_of(rhp, struct pid, rcu); | |
108 | put_pid(pid); | |
109 | } | |
110 | ||
7ad5b3a5 | 111 | void free_pid(struct pid *pid) |
92476d7f EB |
112 | { |
113 | /* We can be called with write_lock_irq(&tasklist_lock) held */ | |
8ef047aa | 114 | int i; |
92476d7f EB |
115 | unsigned long flags; |
116 | ||
117 | spin_lock_irqsave(&pidmap_lock, flags); | |
0a01f2cc EB |
118 | for (i = 0; i <= pid->level; i++) { |
119 | struct upid *upid = pid->numbers + i; | |
af4b8a83 | 120 | struct pid_namespace *ns = upid->ns; |
e8cfbc24 | 121 | switch (--ns->pid_allocated) { |
a6064885 | 122 | case 2: |
af4b8a83 EB |
123 | case 1: |
124 | /* When all that is left in the pid namespace | |
125 | * is the reaper wake up the reaper. The reaper | |
126 | * may be sleeping in zap_pid_ns_processes(). | |
127 | */ | |
128 | wake_up_process(ns->child_reaper); | |
129 | break; | |
e8cfbc24 | 130 | case PIDNS_ADDING: |
314a8ad0 ON |
131 | /* Handle a fork failure of the first process */ |
132 | WARN_ON(ns->child_reaper); | |
e8cfbc24 | 133 | ns->pid_allocated = 0; |
314a8ad0 | 134 | /* fall through */ |
af4b8a83 | 135 | case 0: |
af4b8a83 EB |
136 | schedule_work(&ns->proc_work); |
137 | break; | |
5e1182de | 138 | } |
95846ecf GS |
139 | |
140 | idr_remove(&ns->idr, upid->nr); | |
0a01f2cc | 141 | } |
92476d7f EB |
142 | spin_unlock_irqrestore(&pidmap_lock, flags); |
143 | ||
92476d7f EB |
144 | call_rcu(&pid->rcu, delayed_put_pid); |
145 | } | |
146 | ||
8ef047aa | 147 | struct pid *alloc_pid(struct pid_namespace *ns) |
92476d7f EB |
148 | { |
149 | struct pid *pid; | |
150 | enum pid_type type; | |
8ef047aa PE |
151 | int i, nr; |
152 | struct pid_namespace *tmp; | |
198fe21b | 153 | struct upid *upid; |
35f71bc0 | 154 | int retval = -ENOMEM; |
92476d7f | 155 | |
baf8f0f8 | 156 | pid = kmem_cache_alloc(ns->pid_cachep, GFP_KERNEL); |
92476d7f | 157 | if (!pid) |
35f71bc0 | 158 | return ERR_PTR(retval); |
92476d7f | 159 | |
8ef047aa | 160 | tmp = ns; |
0a01f2cc | 161 | pid->level = ns->level; |
95846ecf | 162 | |
8ef047aa | 163 | for (i = ns->level; i >= 0; i--) { |
95846ecf GS |
164 | int pid_min = 1; |
165 | ||
166 | idr_preload(GFP_KERNEL); | |
167 | spin_lock_irq(&pidmap_lock); | |
168 | ||
169 | /* | |
170 | * init really needs pid 1, but after reaching the maximum | |
171 | * wrap back to RESERVED_PIDS | |
172 | */ | |
173 | if (idr_get_cursor(&tmp->idr) > RESERVED_PIDS) | |
174 | pid_min = RESERVED_PIDS; | |
175 | ||
176 | /* | |
177 | * Store a null pointer so find_pid_ns does not find | |
178 | * a partially initialized PID (see below). | |
179 | */ | |
180 | nr = idr_alloc_cyclic(&tmp->idr, NULL, pid_min, | |
181 | pid_max, GFP_ATOMIC); | |
182 | spin_unlock_irq(&pidmap_lock); | |
183 | idr_preload_end(); | |
184 | ||
287980e4 | 185 | if (nr < 0) { |
35f71bc0 | 186 | retval = nr; |
8ef047aa | 187 | goto out_free; |
35f71bc0 | 188 | } |
92476d7f | 189 | |
8ef047aa PE |
190 | pid->numbers[i].nr = nr; |
191 | pid->numbers[i].ns = tmp; | |
192 | tmp = tmp->parent; | |
193 | } | |
194 | ||
0a01f2cc | 195 | if (unlikely(is_child_reaper(pid))) { |
8896c23d KT |
196 | if (pid_ns_prepare_proc(ns)) { |
197 | disable_pid_allocation(ns); | |
0a01f2cc | 198 | goto out_free; |
8896c23d | 199 | } |
0a01f2cc EB |
200 | } |
201 | ||
b461cc03 | 202 | get_pid_ns(ns); |
92476d7f | 203 | atomic_set(&pid->count, 1); |
92476d7f EB |
204 | for (type = 0; type < PIDTYPE_MAX; ++type) |
205 | INIT_HLIST_HEAD(&pid->tasks[type]); | |
206 | ||
417e3152 | 207 | upid = pid->numbers + ns->level; |
92476d7f | 208 | spin_lock_irq(&pidmap_lock); |
e8cfbc24 | 209 | if (!(ns->pid_allocated & PIDNS_ADDING)) |
5e1182de | 210 | goto out_unlock; |
0a01f2cc | 211 | for ( ; upid >= pid->numbers; --upid) { |
95846ecf GS |
212 | /* Make the PID visible to find_pid_ns. */ |
213 | idr_replace(&upid->ns->idr, pid, upid->nr); | |
e8cfbc24 | 214 | upid->ns->pid_allocated++; |
0a01f2cc | 215 | } |
92476d7f EB |
216 | spin_unlock_irq(&pidmap_lock); |
217 | ||
92476d7f EB |
218 | return pid; |
219 | ||
5e1182de | 220 | out_unlock: |
6e666884 | 221 | spin_unlock_irq(&pidmap_lock); |
24c037eb ON |
222 | put_pid_ns(ns); |
223 | ||
92476d7f | 224 | out_free: |
95846ecf | 225 | spin_lock_irq(&pidmap_lock); |
b7127aa4 | 226 | while (++i <= ns->level) |
95846ecf GS |
227 | idr_remove(&ns->idr, (pid->numbers + i)->nr); |
228 | ||
229 | spin_unlock_irq(&pidmap_lock); | |
8ef047aa | 230 | |
baf8f0f8 | 231 | kmem_cache_free(ns->pid_cachep, pid); |
35f71bc0 | 232 | return ERR_PTR(retval); |
92476d7f EB |
233 | } |
234 | ||
c876ad76 EB |
235 | void disable_pid_allocation(struct pid_namespace *ns) |
236 | { | |
237 | spin_lock_irq(&pidmap_lock); | |
e8cfbc24 | 238 | ns->pid_allocated &= ~PIDNS_ADDING; |
c876ad76 EB |
239 | spin_unlock_irq(&pidmap_lock); |
240 | } | |
241 | ||
7ad5b3a5 | 242 | struct pid *find_pid_ns(int nr, struct pid_namespace *ns) |
1da177e4 | 243 | { |
e8cfbc24 | 244 | return idr_find(&ns->idr, nr); |
1da177e4 | 245 | } |
198fe21b | 246 | EXPORT_SYMBOL_GPL(find_pid_ns); |
1da177e4 | 247 | |
8990571e PE |
248 | struct pid *find_vpid(int nr) |
249 | { | |
17cf22c3 | 250 | return find_pid_ns(nr, task_active_pid_ns(current)); |
8990571e PE |
251 | } |
252 | EXPORT_SYMBOL_GPL(find_vpid); | |
253 | ||
e713d0da SB |
254 | /* |
255 | * attach_pid() must be called with the tasklist_lock write-held. | |
256 | */ | |
81907739 | 257 | void attach_pid(struct task_struct *task, enum pid_type type) |
1da177e4 | 258 | { |
81907739 ON |
259 | struct pid_link *link = &task->pids[type]; |
260 | hlist_add_head_rcu(&link->node, &link->pid->tasks[type]); | |
1da177e4 LT |
261 | } |
262 | ||
24336eae ON |
263 | static void __change_pid(struct task_struct *task, enum pid_type type, |
264 | struct pid *new) | |
1da177e4 | 265 | { |
92476d7f EB |
266 | struct pid_link *link; |
267 | struct pid *pid; | |
268 | int tmp; | |
1da177e4 | 269 | |
92476d7f EB |
270 | link = &task->pids[type]; |
271 | pid = link->pid; | |
1da177e4 | 272 | |
92476d7f | 273 | hlist_del_rcu(&link->node); |
24336eae | 274 | link->pid = new; |
1da177e4 | 275 | |
92476d7f EB |
276 | for (tmp = PIDTYPE_MAX; --tmp >= 0; ) |
277 | if (!hlist_empty(&pid->tasks[tmp])) | |
278 | return; | |
1da177e4 | 279 | |
92476d7f | 280 | free_pid(pid); |
1da177e4 LT |
281 | } |
282 | ||
24336eae ON |
283 | void detach_pid(struct task_struct *task, enum pid_type type) |
284 | { | |
285 | __change_pid(task, type, NULL); | |
286 | } | |
287 | ||
288 | void change_pid(struct task_struct *task, enum pid_type type, | |
289 | struct pid *pid) | |
290 | { | |
291 | __change_pid(task, type, pid); | |
81907739 | 292 | attach_pid(task, type); |
24336eae ON |
293 | } |
294 | ||
c18258c6 | 295 | /* transfer_pid is an optimization of attach_pid(new), detach_pid(old) */ |
7ad5b3a5 | 296 | void transfer_pid(struct task_struct *old, struct task_struct *new, |
c18258c6 EB |
297 | enum pid_type type) |
298 | { | |
299 | new->pids[type].pid = old->pids[type].pid; | |
300 | hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node); | |
c18258c6 EB |
301 | } |
302 | ||
7ad5b3a5 | 303 | struct task_struct *pid_task(struct pid *pid, enum pid_type type) |
1da177e4 | 304 | { |
92476d7f EB |
305 | struct task_struct *result = NULL; |
306 | if (pid) { | |
307 | struct hlist_node *first; | |
67bdbffd | 308 | first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]), |
db1466b3 | 309 | lockdep_tasklist_lock_is_held()); |
92476d7f EB |
310 | if (first) |
311 | result = hlist_entry(first, struct task_struct, pids[(type)].node); | |
312 | } | |
313 | return result; | |
314 | } | |
eccba068 | 315 | EXPORT_SYMBOL(pid_task); |
1da177e4 | 316 | |
92476d7f | 317 | /* |
9728e5d6 | 318 | * Must be called under rcu_read_lock(). |
92476d7f | 319 | */ |
17f98dcf | 320 | struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns) |
92476d7f | 321 | { |
f78f5b90 PM |
322 | RCU_LOCKDEP_WARN(!rcu_read_lock_held(), |
323 | "find_task_by_pid_ns() needs rcu_read_lock() protection"); | |
17f98dcf | 324 | return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID); |
92476d7f | 325 | } |
1da177e4 | 326 | |
228ebcbe PE |
327 | struct task_struct *find_task_by_vpid(pid_t vnr) |
328 | { | |
17cf22c3 | 329 | return find_task_by_pid_ns(vnr, task_active_pid_ns(current)); |
228ebcbe | 330 | } |
228ebcbe | 331 | |
1a657f78 ON |
332 | struct pid *get_task_pid(struct task_struct *task, enum pid_type type) |
333 | { | |
334 | struct pid *pid; | |
335 | rcu_read_lock(); | |
2ae448ef ON |
336 | if (type != PIDTYPE_PID) |
337 | task = task->group_leader; | |
81b1a832 | 338 | pid = get_pid(rcu_dereference(task->pids[type].pid)); |
1a657f78 ON |
339 | rcu_read_unlock(); |
340 | return pid; | |
341 | } | |
77c100c8 | 342 | EXPORT_SYMBOL_GPL(get_task_pid); |
1a657f78 | 343 | |
7ad5b3a5 | 344 | struct task_struct *get_pid_task(struct pid *pid, enum pid_type type) |
92476d7f EB |
345 | { |
346 | struct task_struct *result; | |
347 | rcu_read_lock(); | |
348 | result = pid_task(pid, type); | |
349 | if (result) | |
350 | get_task_struct(result); | |
351 | rcu_read_unlock(); | |
352 | return result; | |
1da177e4 | 353 | } |
77c100c8 | 354 | EXPORT_SYMBOL_GPL(get_pid_task); |
1da177e4 | 355 | |
92476d7f | 356 | struct pid *find_get_pid(pid_t nr) |
1da177e4 LT |
357 | { |
358 | struct pid *pid; | |
359 | ||
92476d7f | 360 | rcu_read_lock(); |
198fe21b | 361 | pid = get_pid(find_vpid(nr)); |
92476d7f | 362 | rcu_read_unlock(); |
1da177e4 | 363 | |
92476d7f | 364 | return pid; |
1da177e4 | 365 | } |
339caf2a | 366 | EXPORT_SYMBOL_GPL(find_get_pid); |
1da177e4 | 367 | |
7af57294 PE |
368 | pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns) |
369 | { | |
370 | struct upid *upid; | |
371 | pid_t nr = 0; | |
372 | ||
373 | if (pid && ns->level <= pid->level) { | |
374 | upid = &pid->numbers[ns->level]; | |
375 | if (upid->ns == ns) | |
376 | nr = upid->nr; | |
377 | } | |
378 | return nr; | |
379 | } | |
4f82f457 | 380 | EXPORT_SYMBOL_GPL(pid_nr_ns); |
7af57294 | 381 | |
44c4e1b2 EB |
382 | pid_t pid_vnr(struct pid *pid) |
383 | { | |
17cf22c3 | 384 | return pid_nr_ns(pid, task_active_pid_ns(current)); |
44c4e1b2 EB |
385 | } |
386 | EXPORT_SYMBOL_GPL(pid_vnr); | |
387 | ||
52ee2dfd ON |
388 | pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type, |
389 | struct pid_namespace *ns) | |
2f2a3a46 | 390 | { |
52ee2dfd ON |
391 | pid_t nr = 0; |
392 | ||
393 | rcu_read_lock(); | |
394 | if (!ns) | |
17cf22c3 | 395 | ns = task_active_pid_ns(current); |
52ee2dfd | 396 | if (likely(pid_alive(task))) { |
dd1c1f2f ON |
397 | if (type != PIDTYPE_PID) { |
398 | if (type == __PIDTYPE_TGID) | |
399 | type = PIDTYPE_PID; | |
e8cfbc24 | 400 | |
52ee2dfd | 401 | task = task->group_leader; |
dd1c1f2f | 402 | } |
81b1a832 | 403 | nr = pid_nr_ns(rcu_dereference(task->pids[type].pid), ns); |
52ee2dfd ON |
404 | } |
405 | rcu_read_unlock(); | |
406 | ||
407 | return nr; | |
2f2a3a46 | 408 | } |
52ee2dfd | 409 | EXPORT_SYMBOL(__task_pid_nr_ns); |
2f2a3a46 | 410 | |
61bce0f1 EB |
411 | struct pid_namespace *task_active_pid_ns(struct task_struct *tsk) |
412 | { | |
413 | return ns_of_pid(task_pid(tsk)); | |
414 | } | |
415 | EXPORT_SYMBOL_GPL(task_active_pid_ns); | |
416 | ||
0804ef4b | 417 | /* |
025dfdaf | 418 | * Used by proc to find the first pid that is greater than or equal to nr. |
0804ef4b | 419 | * |
e49859e7 | 420 | * If there is a pid at nr this function is exactly the same as find_pid_ns. |
0804ef4b | 421 | */ |
198fe21b | 422 | struct pid *find_ge_pid(int nr, struct pid_namespace *ns) |
0804ef4b | 423 | { |
95846ecf | 424 | return idr_get_next(&ns->idr, &nr); |
0804ef4b EB |
425 | } |
426 | ||
95846ecf | 427 | void __init pid_idr_init(void) |
1da177e4 | 428 | { |
840d6fe7 | 429 | /* Verify no one has done anything silly: */ |
e8cfbc24 | 430 | BUILD_BUG_ON(PID_MAX_LIMIT >= PIDNS_ADDING); |
c876ad76 | 431 | |
72680a19 HB |
432 | /* bump default and minimum pid_max based on number of cpus */ |
433 | pid_max = min(pid_max_max, max_t(int, pid_max, | |
434 | PIDS_PER_CPU_DEFAULT * num_possible_cpus())); | |
435 | pid_max_min = max_t(int, pid_max_min, | |
436 | PIDS_PER_CPU_MIN * num_possible_cpus()); | |
437 | pr_info("pid_max: default: %u minimum: %u\n", pid_max, pid_max_min); | |
438 | ||
95846ecf | 439 | idr_init(&init_pid_ns.idr); |
92476d7f | 440 | |
74bd59bb | 441 | init_pid_ns.pid_cachep = KMEM_CACHE(pid, |
5d097056 | 442 | SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT); |
1da177e4 | 443 | } |